diff options
Diffstat (limited to 'erts')
111 files changed, 8286 insertions, 2811 deletions
diff --git a/erts/doc/src/Makefile b/erts/doc/src/Makefile index 21aa3db864..40f74b78ff 100644 --- a/erts/doc/src/Makefile +++ b/erts/doc/src/Makefile @@ -52,18 +52,17 @@ XML_REF3_EFILES = \ erlang.xml \ erl_tracer.xml \ init.xml \ + persistent_term.xml \ + atomics.xml \ + counters.xml \ zlib.xml XML_REF3_FILES = \ + $(XML_REF3_EFILES) \ driver_entry.xml \ erl_nif.xml \ - erl_tracer.xml \ erl_driver.xml \ - erl_prim_loader.xml \ - erlang.xml \ - erts_alloc.xml \ - init.xml \ - zlib.xml + erts_alloc.xml XML_PART_FILES = \ part.xml diff --git a/erts/doc/src/atomics.xml b/erts/doc/src/atomics.xml new file mode 100644 index 0000000000..3fca92fb97 --- /dev/null +++ b/erts/doc/src/atomics.xml @@ -0,0 +1,183 @@ +<?xml version="1.0" encoding="utf-8" ?> +<!DOCTYPE erlref SYSTEM "erlref.dtd"> + +<erlref> + <header> + <copyright> + <year>2018</year> + <holder>Ericsson AB. All Rights Reserved.</holder> + </copyright> + <legalnotice> + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. + </legalnotice> + + <title>atomics</title> + </header> + <module>atomics</module> + <modulesummary>Atomic Functions</modulesummary> + <description> + <p>This module provides a set of functions to do atomic operations towards + mutable atomic variables. The implementation utilizes only + atomic hardware instructions without any software level locking, which makes + it very efficient for concurrent access. The atomics are organized into + arrays with the follwing semantics:</p> + <list type="bulleted"> + <item> + <p>Atomics are 64 bit integers.</p> + </item> + <item> + <p>Atomics can be represented as either signed or unsigned.</p> + </item> + <item> + <p>Atomics wrap around at overflow and underflow operations.</p> + </item> + <item> + <p>All operations guarantee atomicity. No intermediate results can be + seen. The result of one mutation can only be the input to one + following mutation.</p> + </item> + <item> + <p>All atomic operations are mutually ordered. If atomic B is updated + <em>after</em> atomic A, then that is how it will appear to any + concurrent readers. No one can read the new value of B and then read the + old value of A.</p> + </item> + <item> + <p>Indexes into atomic arrays are one-based. An atomic array of + arity N contains N atomics with index from 1 to N.</p> + </item> + </list> + </description> + + <datatypes> + <datatype> + <name name="atomics_ref"/> + <desc><p>Identifies an atomic array returned from + <seealso marker="#new/2"><c>new/2</c></seealso>.</p> + </desc> + </datatype> + </datatypes> + + <funcs> + <func> + <name name="new" arity="2"/> + <fsummary>Create atomic array</fsummary> + <desc> + <p>Create a new atomic array of <c><anno>Arity</anno></c> atomics.</p> + <p>Argument <c><anno>Opts</anno></c> is a list of the following possible + options:</p> + <taglist> + <tag><c>{signed, boolean()}</c></tag> + <item><p>Indicate if the elements of the array will be treated + as signed or unsigned integers. Default is <c>true</c> (signed).</p> + <p>The integer interval for signed atomics are from <c>-(1 bsl 63)</c> + to <c>(1 bsl 63)-1</c> and for unsigned atomics from <c>0</c> to <c>(1 + bsl 64)-1</c>.</p> + </item> + </taglist> + </desc> + </func> + + <func> + <name name="put" arity="3"/> + <fsummary>Set atomic value</fsummary> + <desc> + <p>Set atomic to <c><anno>Value</anno></c>.</p> + </desc> + </func> + + <func> + <name name="get" arity="2"/> + <fsummary>Read atomic value</fsummary> + <desc> + <p>Read atomic value.</p> + </desc> + </func> + + <func> + <name name="add" arity="3"/> + <fsummary>Add to atomic</fsummary> + <desc> + <p>Add <c><anno>Incr</anno></c> to atomic.</p> + </desc> + </func> + + <func> + <name name="add_get" arity="3"/> + <fsummary>Atomic add and get</fsummary> + <desc> + <p>Atomic addition and return of the result.</p> + </desc> + </func> + + <func> + <name name="sub" arity="3"/> + <fsummary>Subtract from atomic</fsummary> + <desc> + <p>Subtract <c><anno>Decr</anno></c> from atomic.</p> + </desc> + </func> + + <func> + <name name="sub_get" arity="3"/> + <fsummary>Atomic sub and get</fsummary> + <desc> + <p>Atomic subtraction and return of the result.</p> + </desc> + </func> + + <func> + <name name="exchange" arity="3"/> + <fsummary>Atomic exchange.</fsummary> + <desc> + <p>Atomically replaces the value of the atomic with + <c><anno>Desired</anno></c> and returns the value it held + previously.</p> + </desc> + </func> + + <func> + <name name="compare_exchange" arity="4"/> + <fsummary>Atomic compare and exchange.</fsummary> + <desc> + <p>Atomically compares the atomic with <c><anno>Expected</anno></c>, + and if those are equal, set atomic to <c><anno>Desired</anno></c>. + Returns <c>ok</c> if <c><anno>Desired</anno></c> was written. Returns + the actual atomic value if not equal to <c><anno>Expected</anno></c>.</p> + </desc> + </func> + + <func> + <name name="info" arity="1"/> + <fsummary>Get information about atomic array.</fsummary> + <desc> + <p>Return information about an atomic array in a map. The map + has the following keys:</p> + <taglist> + <tag><c>size</c></tag> + <item><p>The number of atomics in the array.</p></item> + <tag><c>max</c></tag> + <item><p>The highest possible value an atomic in this array can + hold.</p></item> + <tag><c>min</c></tag> + <item><p>The lowest possible value an atomic in this array can + hold.</p></item> + <tag><c>memory</c></tag> + <item><p>Approximate memory consumption for the array in + bytes.</p></item> + </taglist> + </desc> + </func> + + </funcs> +</erlref> diff --git a/erts/doc/src/counters.xml b/erts/doc/src/counters.xml new file mode 100644 index 0000000000..85eedfdadc --- /dev/null +++ b/erts/doc/src/counters.xml @@ -0,0 +1,142 @@ +<?xml version="1.0" encoding="utf-8" ?> +<!DOCTYPE erlref SYSTEM "erlref.dtd"> + +<erlref> + <header> + <copyright> + <year>2018</year> + <holder>Ericsson AB. All Rights Reserved.</holder> + </copyright> + <legalnotice> + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. + </legalnotice> + + <title>counters</title> + </header> + <module>counters</module> + <modulesummary>Counter Functions</modulesummary> + <description> + <p>This module provides a set of functions to do operations towards + shared mutable counter variables. The implementation does not utilize any + software level locking, which makes it very efficient for concurrent + access. The counters are organized into arrays with the follwing + semantics:</p> + <list type="bulleted"> + <item> + <p>Counters are 64 bit signed integers.</p> + </item> + <item> + <p>Counters wrap around at overflow and underflow operations.</p> + </item> + <item><p>Counters are initialized to zero and can then only be written to + by adding or subtracting.</p> + </item> + <item> + <p>Write operations guarantee atomicity. No intermediate results can be + seen from a single write operation.</p> + </item> + <item> + <p>Two types of counter arrays can be created with options <c>atomics</c> or + <c>write_concurrency</c>. The <c>atomics</c> counters have good allround + performance with nice consistent semantics while + <c>write_concurrency</c> counters offers even better concurrent + write performance at the expense of some potential read + inconsistencies. See <seealso marker="#new/2"><c>new/2</c></seealso>.</p> + </item> + <item> + <p>Indexes into counter arrays are one-based. A counter array of + size N contains N counters with index from 1 to N.</p> + </item> + </list> + </description> + + <datatypes> + <datatype> + <name name="counters_ref"/> + <desc><p>Identifies a counter array returned from + <seealso marker="#new/2"><c>new/2</c></seealso>.</p> + </desc> + </datatype> + </datatypes> + + <funcs> + <func> + <name name="new" arity="2"/> + <fsummary>Create counter array</fsummary> + <desc> + <p>Create a new counter array of <c><anno>Size</anno></c> counters.</p> + <p>Argument <c><anno>Opts</anno></c> is a list of the following possible + options:</p> + <taglist> + <tag><c>atomics</c> (Default)</tag> + <item><p>Counters will be sequentially consistent. If write + operation A is done sequencially before write operation B, then a concurrent reader + may see none of them, only A, or both A and B. It cannot see only B.</p> + </item> + <tag><c>write_concurrency</c></tag> + <item><p>This is an optimization to achieve very efficient concurrent + write operations at the expense of potential read inconsistency and memory + consumption per counter.</p> + <p>Read operations may see sequentially inconsistent results with + regard to concurrent write operations. Even if write operation A is done + sequencially before write operation B, a concurrent reader may see any + combination of A and B, including only B. A read operation is only + guaranteed to see all writes done sequentially before the read. No writes + are ever lost, but will eventually all be seen.</p> + </item> + </taglist> + </desc> + </func> + + <func> + <name name="get" arity="2"/> + <fsummary>Read counter value</fsummary> + <desc> + <p>Read counter value.</p> + </desc> + </func> + + <func> + <name name="add" arity="3"/> + <fsummary>Add to counter</fsummary> + <desc> + <p>Add <c><anno>Incr</anno></c> to counter.</p> + </desc> + </func> + + <func> + <name name="sub" arity="3"/> + <fsummary>Subtract from counter</fsummary> + <desc> + <p>Subtract <c><anno>Decr</anno></c> from counter.</p> + </desc> + </func> + + <func> + <name name="info" arity="1"/> + <fsummary>Get information about counter array.</fsummary> + <desc> + <p>Return information about a counter array in a map. The map + has the following keys (at least):</p> + <taglist> + <tag><c>size</c></tag> + <item><p>The number of counters in the array.</p></item> + <tag><c>memory</c></tag> + <item><p>Approximate memory consumption for the array in + bytes.</p></item> + </taglist> + </desc> + </func> + + </funcs> +</erlref> diff --git a/erts/doc/src/erl_nif.xml b/erts/doc/src/erl_nif.xml index a5826307f7..3820132d71 100644 --- a/erts/doc/src/erl_nif.xml +++ b/erts/doc/src/erl_nif.xml @@ -1252,7 +1252,7 @@ typedef struct { <p>Similar to <c>fprintf</c> but this format string also accepts <c>"%T"</c>, which formats Erlang terms of type <seealso marker="#ERL_NIF_TERM"><c>ERL_NIF_TERM</c></seealso>.</p> - <p>This function is primarily intenden for debugging purpose. It is not + <p>This function is primarily intended for debugging purpose. It is not recommended to print very large terms with <c>%T</c>. The function may change <c>errno</c>, even if successful.</p> </desc> @@ -3216,7 +3216,7 @@ if (retval & ERL_NIF_SELECT_STOP_CALLED) { <p>Similar to <c>snprintf</c> but this format string also accepts <c>"%T"</c>, which formats Erlang terms of type <seealso marker="#ERL_NIF_TERM"><c>ERL_NIF_TERM</c></seealso>.</p> - <p>This function is primarily intenden for debugging purpose. It is not + <p>This function is primarily intended for debugging purpose. It is not recommended to print very large terms with <c>%T</c>. The function may change <c>errno</c>, even if successful.</p> </desc> diff --git a/erts/doc/src/match_spec.xml b/erts/doc/src/match_spec.xml index 5cd6dc1750..48e502739a 100644 --- a/erts/doc/src/match_spec.xml +++ b/erts/doc/src/match_spec.xml @@ -113,6 +113,7 @@ <c><![CDATA[length]]></c> | <c><![CDATA[map_get]]></c> | <c><![CDATA[map_size]]></c> | <c><![CDATA[node]]></c> | <c><![CDATA[round]]></c> | <c><![CDATA[size]]></c> | + <c><![CDATA[bit_size]]></c> | <c><![CDATA[tl]]></c> | <c><![CDATA[trunc]]></c> | <c><![CDATA['+']]></c> | <c><![CDATA['-']]></c> | <c><![CDATA['*']]></c> | <c><![CDATA['div']]></c> | @@ -194,6 +195,7 @@ <c><![CDATA[length]]></c> | <c><![CDATA[map_get]]></c> | <c><![CDATA[map_size]]></c> | <c><![CDATA[node]]></c> | <c><![CDATA[round]]></c> | <c><![CDATA[size]]></c> | + <c><![CDATA[bit_size]]></c> | <c><![CDATA[tl]]></c> | <c><![CDATA[trunc]]></c> | <c><![CDATA['+']]></c> | <c><![CDATA['-']]></c> | <c><![CDATA['*']]></c> | <c><![CDATA['div']]></c> | @@ -269,8 +271,9 @@ other <c>false</c> to return <c><![CDATA[true]]></c>; otherwise <c><![CDATA['xor']]></c> returns false.</p> </item> - <tag><c>abs</c>, <c>element</c>, <c>hd</c>, <c>length</c>, <c>node</c>, - <c>round</c>, <c>size</c>, <c>tl</c>, <c>trunc</c>, <c>'+'</c>, + <tag><c>abs</c>, <c>element</c>, <c>hd</c>, <c>length</c>, + <c>map_get</c>, <c>map_size</c>, <c>node</c>, <c>round</c>, + <c>size</c>, <c>bit_size</c>, <c>tl</c>, <c>trunc</c>, <c>'+'</c>, <c>'-'</c>, <c>'*'</c>, <c>'div'</c>, <c>'rem'</c>, <c>'band'</c>, <c>'bor'</c>, <c>'bxor'</c>, <c>'bnot'</c>, <c>'bsl'</c>, <c>'bsr'</c>, <c>'>'</c>, <c>'>='</c>, <c>'<'</c>, <c>'=<'</c>, diff --git a/erts/doc/src/notes.xml b/erts/doc/src/notes.xml index 88089ae469..6519e79d95 100644 --- a/erts/doc/src/notes.xml +++ b/erts/doc/src/notes.xml @@ -31,6 +31,34 @@ </header> <p>This document describes the changes made to the ERTS application.</p> +<section><title>Erts 10.1.1</title> + + <section><title>Fixed Bugs and Malfunctions</title> + <list> + <item> + <p> + A bug where the socket option 'pktoptions' caused a read + of uninitialized memory has been fixed. Would cause + malfunction on FreeBSD.</p> + <p> + Own Id: OTP-14297 Aux Id: OTP-15141 </p> + </item> + <item> + <p>Fixed a memory leak on errors when reading files.</p> + <p> + Own Id: OTP-15318</p> + </item> + <item> + <p>File access through UNC paths works again on Windows. + This regression was introduced in OTP 21.</p> + <p> + Own Id: OTP-15333 Aux Id: ERL-737 </p> + </item> + </list> + </section> + +</section> + <section><title>Erts 10.1</title> <section><title>Fixed Bugs and Malfunctions</title> @@ -1327,6 +1355,63 @@ </section> +<section><title>Erts 9.3.3.6</title> + + <section><title>Improvements and New Features</title> + <list> + <item> + <p>List subtraction (The <c>--</c> operator) will now + yield properly on large inputs.</p> + <p> + Own Id: OTP-15371</p> + </item> + </list> + </section> + +</section> + +<section><title>Erts 9.3.3.5</title> + + <section><title>Fixed Bugs and Malfunctions</title> + <list> + <item> + <p> + ERTS internal trees of monitor structures could get into + an inconsistent state. This could cause <c>'DOWN'</c> + messages not to be delivered when they should, as well as + delivery of <c>'DOWN'</c> messages that should not be + delivered.</p> + <p> + This bug was introduced in ERTS version 9.0 (OTP 20.0) + and was fixed in ERTS version 10.0 (OTP 21.0) due to a + rewrite of the monitor code. That is, this bug only exist + in the OTP 20 release.</p> + <p> + Own Id: OTP-15399 Aux Id: ERL-751, ERIERL-262, OTP-14205 </p> + </item> + </list> + </section> + +</section> + +<section><title>Erts 9.3.3.4</title> + + <section><title>Fixed Bugs and Malfunctions</title> + <list> + <item> + <p> + Fixed bug in <c>ets:select_replace</c> when called with a + fully bound key could cause a following call to + <c>ets:next</c> or <c>ets:prev</c> to crash the emulator + or return invalid result.</p> + <p> + Own Id: OTP-15346</p> + </item> + </list> + </section> + +</section> + <section><title>Erts 9.3.3.3</title> <section><title>Fixed Bugs and Malfunctions</title> @@ -4729,6 +4814,37 @@ </section> +<section><title>Erts 7.3.1.6</title> + + <section><title>Improvements and New Features</title> + <list> + <item> + <p>List subtraction (The <c>--</c> operator) will now + yield properly on large inputs.</p> + <p> + Own Id: OTP-15371</p> + </item> + </list> + </section> + +</section> + +<section><title>Erts 7.3.1.5</title> + + <section><title>Fixed Bugs and Malfunctions</title> + <list> + <item> + <p> + Fixed small memory leak that could occur when sending to + a terminating port.</p> + <p> + Own Id: OTP-14609 Aux Id: ERIERL-238 </p> + </item> + </list> + </section> + +</section> + <section><title>Erts 7.3.1.4</title> <section><title>Fixed Bugs and Malfunctions</title> <list> @@ -6221,6 +6337,103 @@ </section> +<section><title>Erts 6.4.1.7</title> + + <section><title>Fixed Bugs and Malfunctions</title> + <list> + <item> + <p> + A process communicating with a port via one of the + <c>erlang:port_*</c> BIFs could potentially end up in an + inconsistent state if the port terminated during the + communication. When this occurred the process could later + block in a <c>receive</c> even though it had messages + matching in its message queue.</p> + <p> + This bug was introduced in erts version 5.10 (OTP R16A).</p> + <p> + Own Id: OTP-13424 Aux Id: OTP-10336 </p> + </item> + <item> + <p> + Calls to <c>erl_drv_send_term()</c> or + <c>erl_drv_output_term()</c> from a non-scheduler thread + while the corresponding port was invalid caused the + emulator to enter an inconsistent state which eventually + caused an emulator crash.</p> + <p> + Own Id: OTP-13866</p> + </item> + <item> + <p>Driver and NIF operations accessing processes or ports + could cause an emulator crash when used from + non-scheduler threads. Those operations are:</p> <list> + <item><c>erl_drv_send_term()</c></item> + <item><c>driver_send_term()</c></item> + <item><c>erl_drv_output_term()</c></item> + <item><c>driver_output_term()</c></item> + <item><c>enif_send()</c></item> + <item><c>enif_port_command()</c></item> </list> + <p> + Own Id: OTP-13869</p> + </item> + <item> + <p> + Fix bug in <c>binary_to_term</c> for binaries created by + <c>term_to_binary </c> with option <c>compressed</c>. The + bug can cause <c>badarg</c> exception for a valid binary + when Erlang VM is linked against a <c>zlib</c> library of + version 1.2.9 or newer. Bug exists since OTP 17.0.</p> + <p> + Own Id: OTP-14159 Aux Id: ERL-340 </p> + </item> + <item> + <p> + Fixed bug in operator <c>bxor</c> causing erroneuos + result when one operand is a big <em>negative</em> + integer with the lowest <c>N*W</c> bits as zero and the + other operand not larger than <c>N*W</c> bits. <c>N</c> + is an integer of 1 or larger and <c>W</c> is 32 or 64 + depending on word size.</p> + <p> + Own Id: OTP-14514</p> + </item> + <item> + <p> + Fixed bug in <c>binary_to_term</c> and + <c>binary_to_atom</c> that could cause VM crash. + Typically happens when the last character of an UTF8 + string is in the range 128 to 255, but truncated to only + one byte. Bug exists in <c>binary_to_term</c> since ERTS + version 5.10.2 (OTP_R16B01) and <c>binary_to_atom</c> + since ERTS version 9.0 (OTP-20.0).</p> + <p> + Own Id: OTP-14590 Aux Id: ERL-474 </p> + </item> + </list> + </section> + +</section> + +<section><title>Erts 6.4.1.6</title> + + <section><title>Fixed Bugs and Malfunctions</title> + <list> + <item> + <p> + When calling <c>garbage_collect/[1,2]</c> or + <c>check_process_code/[2,3]</c> from a process with a + higher priority than the priority of the process operated + on, the run queues could end up in an inconsistent state. + This bug has now been fixed.</p> + <p> + Own Id: OTP-13298 Aux Id: OTP-11388 </p> + </item> + </list> + </section> + +</section> + <section><title>Erts 6.4.1.5</title> <section><title>Fixed Bugs and Malfunctions</title> @@ -19047,4 +19260,3 @@ </section> </section> </chapter> - diff --git a/erts/doc/src/persistent_term.xml b/erts/doc/src/persistent_term.xml new file mode 100644 index 0000000000..d2a138d65f --- /dev/null +++ b/erts/doc/src/persistent_term.xml @@ -0,0 +1,290 @@ +<?xml version="1.0" encoding="utf-8" ?> +<!DOCTYPE erlref SYSTEM "erlref.dtd"> + +<erlref> + <header> + <copyright> + <year>2018</year><year>2018</year> + <holder>Ericsson AB. All Rights Reserved.</holder> + </copyright> + <legalnotice> + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. + + </legalnotice> + + <title>persistent_term</title> + <prepared></prepared> + <docno></docno> + <date></date> + <rev></rev> + <file>persistent_term.xml</file> + </header> + <module>persistent_term</module> + <modulesummary>Persistent terms.</modulesummary> + <description> + <p>This module is similar to <seealso + marker="stdlib:ets"><c>ets</c></seealso> in that it provides a + storage for Erlang terms that can be accessed in constant time, + but with the difference that <c>persistent_term</c> has been + highly optimized for reading terms at the expense of writing and + updating terms. When a persistent term is updated or deleted, a + global garbage collection pass is run to scan all processes for + the deleted term, and to copy it into each process that still uses + it. Therefore, <c>persistent_term</c> is suitable for storing + Erlang terms that are frequently accessed but never or + infrequently updated.</p> + + <warning><p>Persistent terms is an advanced feature and is not a + general replacement for ETS tables. Before using persistent terms, + make sure to fully understand the consequence to system + performance when updating or deleting persistent terms.</p></warning> + + <p>Term lookup (using <seealso + marker="#get/1"><c>get/1</c></seealso>), is done in constant time + and without taking any locks, and the term is <strong>not</strong> + copied to the heap (as is the case with terms stored in ETS + tables).</p> + + <p>Storing or updating a term (using <seealso + marker="#put/2"><c>put/2</c></seealso>) is proportional to the + number of already created persistent terms because the hash table + holding the keys will be copied. In addition, the term itself will + be copied.</p> + + <p>When a (complex) term is deleted (using <seealso + marker="#erase/1"><c>erase/1</c></seealso>) or replaced by another + (using <seealso marker="#put/2"><c>put/2</c></seealso>), a global + garbage collection is initiated. It works like this:</p> + + <list> + <item><p>All processes in the system will be scheduled to run a + scan of their heaps for the term that has been deleted. While + such scan is relatively light-weight, if there are many + processes, the system can become less responsive until all + process have scanned theirs heaps.</p></item> + + <item><p>If the deleted term (or any part of it) is still used + by a process, that process will do a major (fullsweep) garbage + collection and copy the term into the process. However, at most + two processes at a time will be scheduled to do that kind of + garbage collection.</p></item> + </list> + + <p>Deletion of atoms and other terms that fit in one machine word + is specially optimized to avoid doing a global GC. It is still not + recommended to update persistent terms with such values too + frequently because the hash table holding the keys is copied every + time a persistent term is updated.</p> + + <p>Some examples are suitable uses for persistent terms are:</p> + + <list> + <item><p>Storing of configuration data that must be easily + accessible by all processes.</p></item> + + <item><p>Storing of references for NIF resources.</p></item> + + <item><p>Storing of references for efficient counters.</p></item> + + <item><p>Storing an atom to indicate a logging level or whether debugging + is turned on.</p></item> + </list> + + </description> + + <section> + <title>Storing Huge Persistent Terms</title> + <p>The current implementation of persistent terms uses the literal + <seealso marker="erts_alloc">allocator</seealso> also used for + literals (constant terms) in BEAM code. By default, 1 GB of + virtual address space is reserved for literals in BEAM code and + persistent terms. The amount of virtual address space reserved for + literals can be changed by using the <seealso + marker="erts_alloc#MIscs"><c>+MIscs option</c></seealso> when + starting the emulator.</p> + + <p>Here is an example how the reserved virtual address space for literals + can be raised to 2 GB (2048 MB):</p> + + <pre> + erl +MIscs 2048</pre> + </section> + + <section> + <title>Warning For Many Persistent Terms</title> + <p>The runtime system will send a warning report to the + error logger if more than 20000 persistent terms have been + created. It will look like this:</p> + +<pre> +More than 20000 persistent terms have been created. +It is recommended to avoid creating an excessive number of +persistent terms, as creation and deletion of persistent terms +will be slower as the number of persistent terms increases.</pre> + </section> + + <section> + <title>Best Practices for Using Persistent Terms</title> + + <p>It is recommended to use keys like <c>?MODULE</c> or + <c>{?MODULE,SubKey}</c> to avoid name collisions.</p> + + <p>Prefer creating a few large persistent terms to creating many + small persistent terms. The execution time for storing a + persistent term is proportional to the number of already existing + terms.</p> + + <p>Updating a persistent term with the same value as it already + has is specially optimized to do nothing quickly; thus, there is + no need compare the old and new values and avoid calling + <seealso marker="#put/2"><c>put/2</c></seealso> if the values + are equal.</p> + + <p>When atoms or other terms that fit in one machine word are + deleted, no global GC is needed. Therefore, persistent terms that + have atoms as their values can be updated more frequently, but + note that updating such persistent terms is still much more + expensive than reading them.</p> + + <p>Updating or deleting a persistent term will trigger a global GC + if the term does not fit in one machine word. Processes will be + scheduled as usual, but all processes will be made runnable at + once, which will make the system less responsive until all process + have run and scanned their heaps for the deleted terms. One way to + minimize the effects on responsiveness could be to minimize the + number of processes on the node before updating or deleting a + persistent term. It would also be wise to avoid updating terms + when the system is at peak load.</p> + + <p>Avoid storing a retrieved persistent term in a process if that + persistent term could be deleted or updated in the future. If a + process holds a reference to a persistent term when the term is + deleted, the process will be garbage collected and the term copied + to process.</p> + + <p>Avoid updating or deleting more than one persistent term at a + time. Each deleted term will trigger its own global GC. That + means that deleting N terms will make the system less responsive N + times longer than deleting a single persistent term. Therefore, + terms that are to be updated at the same time should be collected + into a larger term, for example, a map or a tuple.</p> + </section> + + <section> + <title>Example</title> + + <p>The following example shows how lock contention for ETS tables + can be minimized by having one ETS table for each scheduler. The + table identifiers for the ETS tables are stored as a single + persistent term:</p> + +<pre> + %% There is one ETS table for each scheduler. + Sid = erlang:system_info(scheduler_id), + Tid = element(Sid, persistent_term:get(?MODULE)), + ets:update_counter(Tid, Key, 1).</pre> + + </section> + + <datatypes> + <datatype> + <name name="key"/> + <desc> + <p>Any Erlang term.</p> + </desc> + </datatype> + <datatype> + <name name="value"/> + <desc> + <p>Any Erlang term.</p> + </desc> + </datatype> + </datatypes> + + <funcs> + <func> + <name name="erase" arity="1"/> + <fsummary>Erase the name for a persistent term.</fsummary> + <desc> + <p>Erase the name for the persistent term with key + <c><anno>Key</anno></c>. The return value will be <c>true</c> + if there was a persistent term with the key + <c><anno>Key</anno></c>, and <c>false</c> if there was no + persistent term associated with the key.</p> + <p>If there existed a previous persistent term associated with + key <c><anno>Key</anno></c>, a global GC has been initiated + when <c>erase/1</c> returns. See <seealso + marker="#description">Description</seealso>.</p> + </desc> + </func> + + <func> + <name name="get" arity="0"/> + <fsummary>Get all persistent terms.</fsummary> + <desc> + <p>Retrieve the keys and values for all persistent terms. + The keys will be copied to the heap for the process calling + <c>get/0</c>, but the values will not.</p> + </desc> + </func> + + <func> + <name name="get" arity="1"/> + <fsummary>Get the value for a persistent term.</fsummary> + <desc> + <p>Retrieve the value for the persistent term associated with + the key <c><anno>Key</anno></c>. The lookup will be made in + constant time and the value will not be copied to the heap + of the calling process.</p> + <p>This function fails with a <c>badarg</c> exception if no + term has been stored with the key + <c><anno>Key</anno></c>.</p> + <p>If the calling process holds on to the value of the + persistent term and the persistent term is deleted in the future, + the term will be copied to the process.</p> + </desc> + </func> + + <func> + <name name="info" arity="0"/> + <fsummary>Get information about persistent terms.</fsummary> + <desc> + <p>Return information about persistent terms in a map. The map + has the following keys:</p> + <taglist> + <tag><c>count</c></tag> + <item><p>The number of persistent terms.</p></item> + <tag><c>memory</c></tag> + <item><p>The total amount of memory (measured in bytes) + used by all persistent terms.</p></item> + </taglist> + </desc> + </func> + + <func> + <name name="put" arity="2"/> + <fsummary>Store a term.</fsummary> + <desc> + <p>Store the value <c><anno>Value</anno></c> as a persistent term and + associate it with the key <c><anno>Key</anno></c>.</p> + <p>If the value <c><anno>Value</anno></c> is equal to the value + previously stored for the key, <c>put/2</c> will do nothing and return + quickly.</p> + <p>If there existed a previous persistent term associated with + key <c><anno>Key</anno></c>, a global GC has been initiated + when <c>put/2</c> returns. See <seealso + marker="#description">Description</seealso>.</p> + </desc> + </func> + </funcs> +</erlref> diff --git a/erts/doc/src/ref_man.xml b/erts/doc/src/ref_man.xml index 0617463a7b..a78aaa449e 100644 --- a/erts/doc/src/ref_man.xml +++ b/erts/doc/src/ref_man.xml @@ -34,6 +34,7 @@ <xi:include href="erl_prim_loader.xml"/> <xi:include href="erlang.xml"/> <xi:include href="init.xml"/> + <xi:include href="persistent_term.xml"/> <xi:include href="zlib.xml"/> <xi:include href="epmd.xml"/> <xi:include href="erl.xml"/> @@ -49,5 +50,7 @@ <xi:include href="erts_alloc.xml"/> <xi:include href="erl_nif.xml"/> <xi:include href="erl_tracer.xml"/> + <xi:include href="atomics.xml"/> + <xi:include href="counters.xml"/> </application> diff --git a/erts/doc/src/specs.xml b/erts/doc/src/specs.xml index ed6be650e5..0b943e6295 100644 --- a/erts/doc/src/specs.xml +++ b/erts/doc/src/specs.xml @@ -4,5 +4,8 @@ <xi:include href="../specs/specs_erlang.xml"/> <xi:include href="../specs/specs_erl_tracer.xml"/> <xi:include href="../specs/specs_init.xml"/> + <xi:include href="../specs/specs_persistent_term.xml"/> <xi:include href="../specs/specs_zlib.xml"/> + <xi:include href="../specs/specs_atomics.xml"/> + <xi:include href="../specs/specs_counters.xml"/> </specs> diff --git a/erts/emulator/Makefile.in b/erts/emulator/Makefile.in index eeb9577a33..b74f8371f5 100644 --- a/erts/emulator/Makefile.in +++ b/erts/emulator/Makefile.in @@ -633,21 +633,24 @@ GENERATE += $(TTF_DIR)/driver_tab.c # This list must be consistent with PRE_LOADED_MODULES in # erts/preloaded/src/Makefile. -PRELOAD_BEAM = $(ERL_TOP)/erts/preloaded/ebin/otp_ring0.beam \ - $(ERL_TOP)/erts/preloaded/ebin/erts_code_purger.beam \ - $(ERL_TOP)/erts/preloaded/ebin/init.beam \ - $(ERL_TOP)/erts/preloaded/ebin/prim_buffer.beam \ - $(ERL_TOP)/erts/preloaded/ebin/prim_eval.beam \ - $(ERL_TOP)/erts/preloaded/ebin/prim_inet.beam \ - $(ERL_TOP)/erts/preloaded/ebin/prim_file.beam \ - $(ERL_TOP)/erts/preloaded/ebin/zlib.beam \ - $(ERL_TOP)/erts/preloaded/ebin/prim_zip.beam \ - $(ERL_TOP)/erts/preloaded/ebin/erl_prim_loader.beam \ - $(ERL_TOP)/erts/preloaded/ebin/erlang.beam \ - $(ERL_TOP)/erts/preloaded/ebin/erts_internal.beam \ - $(ERL_TOP)/erts/preloaded/ebin/erl_tracer.beam \ - $(ERL_TOP)/erts/preloaded/ebin/erts_literal_area_collector.beam \ - $(ERL_TOP)/erts/preloaded/ebin/erts_dirty_process_signal_handler.beam +PRELOAD_BEAM = $(ERL_TOP)/erts/preloaded/ebin/erts_code_purger.beam \ + $(ERL_TOP)/erts/preloaded/ebin/erl_init.beam \ + $(ERL_TOP)/erts/preloaded/ebin/init.beam \ + $(ERL_TOP)/erts/preloaded/ebin/prim_buffer.beam \ + $(ERL_TOP)/erts/preloaded/ebin/prim_eval.beam \ + $(ERL_TOP)/erts/preloaded/ebin/prim_inet.beam \ + $(ERL_TOP)/erts/preloaded/ebin/prim_file.beam \ + $(ERL_TOP)/erts/preloaded/ebin/zlib.beam \ + $(ERL_TOP)/erts/preloaded/ebin/prim_zip.beam \ + $(ERL_TOP)/erts/preloaded/ebin/erl_prim_loader.beam \ + $(ERL_TOP)/erts/preloaded/ebin/erlang.beam \ + $(ERL_TOP)/erts/preloaded/ebin/erts_internal.beam \ + $(ERL_TOP)/erts/preloaded/ebin/erl_tracer.beam \ + $(ERL_TOP)/erts/preloaded/ebin/erts_literal_area_collector.beam \ + $(ERL_TOP)/erts/preloaded/ebin/erts_dirty_process_signal_handler.beam \ + $(ERL_TOP)/erts/preloaded/ebin/atomics.beam \ + $(ERL_TOP)/erts/preloaded/ebin/counters.beam \ + $(ERL_TOP)/erts/preloaded/ebin/persistent_term.beam ifeq ($(TARGET),win32) # On windows the preloaded objects are in a resource object. @@ -839,6 +842,8 @@ RUN_OBJS += \ $(OBJDIR)/erl_bif_ddll.o $(OBJDIR)/erl_bif_guard.o \ $(OBJDIR)/erl_bif_info.o $(OBJDIR)/erl_bif_op.o \ $(OBJDIR)/erl_bif_os.o $(OBJDIR)/erl_bif_lists.o \ + $(OBJDIR)/erl_bif_persistent.o \ + $(OBJDIR)/erl_bif_atomics.o $(OBJDIR)/erl_bif_counters.o \ $(OBJDIR)/erl_bif_trace.o $(OBJDIR)/erl_bif_unique.o \ $(OBJDIR)/erl_bif_wrap.o $(OBJDIR)/erl_nfunc_sched.o \ $(OBJDIR)/erl_guard_bifs.o $(OBJDIR)/erl_dirty_bif_wrap.o \ diff --git a/erts/emulator/beam/atom.c b/erts/emulator/beam/atom.c index 5381611fab..59b51fd15e 100644 --- a/erts/emulator/beam/atom.c +++ b/erts/emulator/beam/atom.c @@ -174,7 +174,7 @@ atom_alloc(Atom* tmpl) /* * Precompute ordinal value of first 3 bytes + 7 bits. - * This is used by utils.c:erts_cmp_atoms(). + * This is used by erl_utils.h:erts_cmp_atoms(). * We cannot use the full 32 bits of the first 4 bytes, * since we use the sign of the difference between two * ordinal values to represent their relative order. diff --git a/erts/emulator/beam/atom.names b/erts/emulator/beam/atom.names index 45b7540aeb..3f7aa0983c 100644 --- a/erts/emulator/beam/atom.names +++ b/erts/emulator/beam/atom.names @@ -182,6 +182,7 @@ atom control atom copy atom copy_literals atom counters +atom count atom cpu atom cpu_timestamp atom cr @@ -208,7 +209,6 @@ atom dirty_nif_finalizer atom disable_trace atom disabled atom discard -atom display_items atom dist atom dist_cmd atom dist_ctrl_put_data @@ -237,6 +237,7 @@ atom eof atom eol atom Eq='=:=' atom Eqeq='==' +atom erl_init atom erl_tracer atom erlang atom erl_signal_server @@ -287,6 +288,7 @@ atom gc_minor_end atom gc_minor_start atom Ge='>=' atom generational +atom get_all_trap atom get_seq_token atom get_tcw atom gather_gc_info_result @@ -325,6 +327,7 @@ atom index atom infinity atom info atom info_msg +atom info_trap atom init atom initial_call atom input @@ -393,6 +396,7 @@ atom microsecond atom microstate_accounting atom milli_seconds atom millisecond +atom min atom min_heap_size atom min_bin_vheap_size atom minor diff --git a/erts/emulator/beam/beam_bif_load.c b/erts/emulator/beam/beam_bif_load.c index d221e6aea6..bb1b2e5b27 100644 --- a/erts/emulator/beam/beam_bif_load.c +++ b/erts/emulator/beam/beam_bif_load.c @@ -1752,29 +1752,7 @@ BIF_RETTYPE erts_internal_purge_module_2(BIF_ALIST_2) finalize_purge_operation(BIF_P, ret == am_true); if (literals) { - ErtsLiteralAreaRef *ref; - ErtsMessage *mp; - ref = erts_alloc(ERTS_ALC_T_LITERAL_REF, - sizeof(ErtsLiteralAreaRef)); - ref->literal_area = literals; - ref->next = NULL; - erts_mtx_lock(&release_literal_areas.mtx); - if (release_literal_areas.last) { - release_literal_areas.last->next = ref; - release_literal_areas.last = ref; - } - else { - release_literal_areas.first = ref; - release_literal_areas.last = ref; - } - erts_mtx_unlock(&release_literal_areas.mtx); - mp = erts_alloc_message(0, NULL); - ERL_MESSAGE_TOKEN(mp) = am_undefined; - erts_queue_proc_message(BIF_P, - erts_literal_area_collector, - 0, - mp, - am_copy_literals); + erts_queue_release_literals(BIF_P, literals); } return ret; @@ -1786,6 +1764,41 @@ BIF_RETTYPE erts_internal_purge_module_2(BIF_ALIST_2) } } +void +erts_queue_release_literals(Process* c_p, ErtsLiteralArea* literals) +{ + ErtsLiteralAreaRef *ref; + ErtsMessage *mp; + ref = erts_alloc(ERTS_ALC_T_LITERAL_REF, + sizeof(ErtsLiteralAreaRef)); + ref->literal_area = literals; + ref->next = NULL; + erts_mtx_lock(&release_literal_areas.mtx); + if (release_literal_areas.last) { + release_literal_areas.last->next = ref; + release_literal_areas.last = ref; + } else { + release_literal_areas.first = ref; + release_literal_areas.last = ref; + } + erts_mtx_unlock(&release_literal_areas.mtx); + mp = erts_alloc_message(0, NULL); + ERL_MESSAGE_TOKEN(mp) = am_undefined; + if (c_p == NULL) { + erts_queue_message(erts_literal_area_collector, + 0, + mp, + am_copy_literals, + am_system); + } else { + erts_queue_proc_message(c_p, + erts_literal_area_collector, + 0, + mp, + am_copy_literals); + } +} + /* * Move code from current to old and null all export entries for the module */ diff --git a/erts/emulator/beam/bif.c b/erts/emulator/beam/bif.c index 9ec863f259..4741d7451d 100644 --- a/erts/emulator/beam/bif.c +++ b/erts/emulator/beam/bif.c @@ -3624,6 +3624,10 @@ erts_internal_garbage_collect_1(BIF_ALIST_1) default: BIF_ERROR(BIF_P, BADARG); } erts_garbage_collect(BIF_P, 0, NULL, 0); + if (ERTS_PROC_IS_EXITING(BIF_P)) { + /* The max heap size limit was reached. */ + return THE_NON_VALUE; + } return am_true; } @@ -4440,13 +4444,6 @@ BIF_RETTYPE system_flag_2(BIF_ALIST_2) erts_proc_lock(BIF_P, ERTS_PROC_LOCK_MAIN); BIF_RET(old_value); - } else if (BIF_ARG_1 == am_display_items) { - int oval = display_items; - if (!is_small(BIF_ARG_2) || (n = signed_val(BIF_ARG_2)) < 0) { - goto error; - } - display_items = n < 32 ? 32 : n; - BIF_RET(make_small(oval)); } else if (BIF_ARG_1 == am_debug_flags) { BIF_RET(am_true); } else if (BIF_ARG_1 == am_backtrace_depth) { diff --git a/erts/emulator/beam/bif.tab b/erts/emulator/beam/bif.tab index a770524221..6015fc0691 100644 --- a/erts/emulator/beam/bif.tab +++ b/erts/emulator/beam/bif.tab @@ -40,6 +40,7 @@ # Note: Guards BIFs usually require special support in the compiler. # + gcbif erlang:abs/1 bif erlang:adler32/1 bif erlang:adler32/2 @@ -697,3 +698,28 @@ ubif erlang:map_get/2 ubif erlang:is_map_key/2 bif ets:internal_delete_all/2 bif ets:internal_select_delete/2 + +# +# New in 21.2 +# + +bif persistent_term:put/2 +bif persistent_term:get/1 +bif persistent_term:get/0 +bif persistent_term:erase/1 +bif persistent_term:info/0 +bif erts_internal:erase_persistent_terms/0 + +bif erts_internal:atomics_new/2 +bif atomics:get/2 +bif atomics:put/3 +bif atomics:add/3 +bif atomics:add_get/3 +bif atomics:exchange/3 +bif atomics:compare_exchange/4 +bif atomics:info/1 + +bif erts_internal:counters_new/1 +bif erts_internal:counters_get/2 +bif erts_internal:counters_add/3 +bif erts_internal:counters_info/1 diff --git a/erts/emulator/beam/copy.c b/erts/emulator/beam/copy.c index e7bfd04b73..e7bd046e18 100644 --- a/erts/emulator/beam/copy.c +++ b/erts/emulator/beam/copy.c @@ -1074,6 +1074,7 @@ Uint copy_shared_calculate(Eterm obj, erts_shcopy_t *info) Eterm* ptr; Eterm *lit_purge_ptr = info->lit_purge_ptr; Uint lit_purge_sz = info->lit_purge_sz; + int copy_literals = info->copy_literals; #ifdef DEBUG Eterm mypid = erts_get_current_pid(); #endif @@ -1119,7 +1120,7 @@ Uint copy_shared_calculate(Eterm obj, erts_shcopy_t *info) /* off heap list pointers are copied verbatim */ if (erts_is_literal(obj,ptr)) { VERBOSE(DEBUG_SHCOPY, ("[pid=%T] bypassed copying %p is %T\n", mypid, ptr, obj)); - if (in_literal_purge_area(ptr)) + if (copy_literals || in_literal_purge_area(ptr)) info->literal_size += size_object(obj); goto pop_next; } @@ -1170,7 +1171,7 @@ Uint copy_shared_calculate(Eterm obj, erts_shcopy_t *info) /* off heap pointers to boxes are copied verbatim */ if (erts_is_literal(obj,ptr)) { VERBOSE(DEBUG_SHCOPY, ("[pid=%T] bypassed copying %p is %T\n", mypid, ptr, obj)); - if (in_literal_purge_area(ptr)) + if (copy_literals || in_literal_purge_area(ptr)) info->literal_size += size_object(obj); goto pop_next; } @@ -1338,6 +1339,7 @@ Uint copy_shared_perform(Eterm obj, Uint size, erts_shcopy_t *info, unsigned remaining; Eterm *lit_purge_ptr = info->lit_purge_ptr; Uint lit_purge_sz = info->lit_purge_sz; + int copy_literals = info->copy_literals; #ifdef DEBUG Eterm mypid = erts_get_current_pid(); Eterm saved_obj = obj; @@ -1387,7 +1389,7 @@ Uint copy_shared_perform(Eterm obj, Uint size, erts_shcopy_t *info, ptr = list_val(obj); /* off heap list pointers are copied verbatim */ if (erts_is_literal(obj,ptr)) { - if (!in_literal_purge_area(ptr)) { + if (!(copy_literals || in_literal_purge_area(ptr))) { *resp = obj; } else { Uint bsz = 0; @@ -1455,7 +1457,7 @@ Uint copy_shared_perform(Eterm obj, Uint size, erts_shcopy_t *info, ptr = boxed_val(obj); /* off heap pointers to boxes are copied verbatim */ if (erts_is_literal(obj,ptr)) { - if (!in_literal_purge_area(ptr)) { + if (!(copy_literals || in_literal_purge_area(ptr))) { *resp = obj; } else { Uint bsz = 0; diff --git a/erts/emulator/beam/erl_alloc.c b/erts/emulator/beam/erl_alloc.c index 36c46fd7aa..e6169ebeaa 100644 --- a/erts/emulator/beam/erl_alloc.c +++ b/erts/emulator/beam/erl_alloc.c @@ -4041,7 +4041,7 @@ debug_realloc(ErtsAlcType_t type, void *extra, void *ptr, Uint size) erts_hdbg_chk_blks(); #endif - if (old_size > size) + if (ptr && old_size > size) sys_memset((void *) (((char *) ptr) + size), 0xf, sizeof(Uint) + old_size - size); @@ -4072,6 +4072,9 @@ debug_free(ErtsAlcType_t type, void *extra, void *ptr) ASSERT(ERTS_ALC_N_MIN <= n && n <= ERTS_ALC_N_MAX); + if (!ptr) + return; + dptr = check_memory_fence(ptr, &size, n, ERTS_ALC_O_FREE); #ifdef ERTS_ALC_A_EXEC diff --git a/erts/emulator/beam/erl_alloc.types b/erts/emulator/beam/erl_alloc.types index f1e99820af..08dcc5ea9a 100644 --- a/erts/emulator/beam/erl_alloc.types +++ b/erts/emulator/beam/erl_alloc.types @@ -274,9 +274,13 @@ type ML_YIELD_STATE SHORT_LIVED SYSTEM monitor_link_yield_state type ML_DIST STANDARD SYSTEM monitor_link_dist type PF3_ARGS SHORT_LIVED PROCESSES process_flag_3_arguments type SETUP_CONN_ARG SHORT_LIVED PROCESSES setup_connection_argument +type LIST_TRAP SHORT_LIVED PROCESSES list_bif_trap_state type ENVIRONMENT SYSTEM SYSTEM environment +type PERSISTENT_TERM LONG_LIVED CODE persisten_term +type PERSISTENT_LOCK_Q SHORT_LIVED SYSTEM persistent_lock_q + # # Types used for special emulators # @@ -334,6 +338,8 @@ type GC_INFO_REQ SHORT_LIVED SYSTEM gc_info_request type PORT_DATA_HEAP STANDARD SYSTEM port_data_heap type MSACC DRIVER SYSTEM microstate_accounting type SYS_CHECK_REQ SHORT_LIVED SYSTEM system_check_request +type ATOMICS STANDARD SYSTEM erl_bif_atomics +type COUNTERS STANDARD SYSTEM erl_bif_counters # # Types used by system specific code diff --git a/erts/emulator/beam/erl_bif_atomics.c b/erts/emulator/beam/erl_bif_atomics.c new file mode 100644 index 0000000000..092dbb3bd3 --- /dev/null +++ b/erts/emulator/beam/erl_bif_atomics.c @@ -0,0 +1,256 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2018. All Rights Reserved. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * %CopyrightEnd% + */ + +/* + * Purpose: High performance atomics. + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#include <stddef.h> /* offsetof */ + +#include "sys.h" +#include "export.h" +#include "bif.h" +#include "erl_threads.h" +#include "big.h" +#include "erl_binary.h" +#include "erl_bif_unique.h" +#include "erl_map.h" + +typedef struct +{ + int is_signed; + UWord vlen; + erts_atomic64_t v[1]; +}AtomicsRef; + +static int atomics_destructor(Binary *unused) +{ + return 1; +} + +#define OPT_SIGNED (1 << 0) + +BIF_RETTYPE erts_internal_atomics_new_2(BIF_ALIST_2) +{ + AtomicsRef* p; + Binary* mbin; + UWord i, cnt, opts; + Uint bytes; + Eterm* hp; + + if (!term_to_UWord(BIF_ARG_1, &cnt) + || cnt == 0 + || !term_to_UWord(BIF_ARG_2, &opts)) { + + BIF_ERROR(BIF_P, BADARG); + } + + if (cnt > (ERTS_UWORD_MAX / sizeof(p->v[0]))) + BIF_ERROR(BIF_P, SYSTEM_LIMIT); + + bytes = offsetof(AtomicsRef, v) + cnt*sizeof(p->v[0]); + mbin = erts_create_magic_binary_x(bytes, + atomics_destructor, + ERTS_ALC_T_ATOMICS, + 0); + p = ERTS_MAGIC_BIN_DATA(mbin); + p->is_signed = opts & OPT_SIGNED; + p->vlen = cnt; + for (i=0; i < cnt; i++) + erts_atomic64_init_nob(&p->v[i], 0); + hp = HAlloc(BIF_P, ERTS_MAGIC_REF_THING_SIZE); + return erts_mk_magic_ref(&hp, &MSO(BIF_P), mbin); +} + +static ERTS_INLINE int get_ref(Eterm ref, AtomicsRef** pp) +{ + Binary* mbin; + if (!is_internal_magic_ref(ref)) + return 0; + + mbin = erts_magic_ref2bin(ref); + if (ERTS_MAGIC_BIN_DESTRUCTOR(mbin) != atomics_destructor) + return 0; + *pp = ERTS_MAGIC_BIN_DATA(mbin); + return 1; +} + +static ERTS_INLINE int get_ref_ix(Eterm ref, Eterm ix, + AtomicsRef** pp, UWord* ixp) +{ + return (get_ref(ref, pp) + && term_to_UWord(ix, ixp) + && --(*ixp) < (*pp)->vlen); +} + +static ERTS_INLINE int get_value(AtomicsRef* p, Eterm term, erts_aint64_t *valp) +{ + return (p->is_signed ? + term_to_Sint64(term, (Sint64*)valp) : + term_to_Uint64(term, (Uint64*)valp)); +} + +static ERTS_INLINE int get_incr(AtomicsRef* p, Eterm term, erts_aint64_t *valp) +{ + return (term_to_Sint64(term, (Sint64*)valp) + || term_to_Uint64(term, (Uint64*)valp)); +} + +static ERTS_INLINE Eterm bld_atomic(Process* proc, AtomicsRef* p, + erts_aint64_t val) +{ + if (p->is_signed) { + if (IS_SSMALL(val)) + return make_small((Sint) val); + else { + Uint hsz = ERTS_SINT64_HEAP_SIZE(val); + Eterm* hp = HAlloc(proc, hsz); + return erts_sint64_to_big(val, &hp); + } + } + else { + if ((Uint64)val <= MAX_SMALL) + return make_small((Sint) val); + else { + Uint hsz = ERTS_UINT64_HEAP_SIZE(val); + Eterm* hp = HAlloc(proc, hsz); + return erts_uint64_to_big(val, &hp); + } + } +} + +BIF_RETTYPE atomics_put_3(BIF_ALIST_3) +{ + AtomicsRef* p; + UWord ix; + erts_aint64_t val; + + if (!get_ref_ix(BIF_ARG_1, BIF_ARG_2, &p, &ix) + || !get_value(p, BIF_ARG_3, &val)) { + BIF_ERROR(BIF_P, BADARG); + } + erts_atomic64_set_mb(&p->v[ix], val); + return am_ok; +} + +BIF_RETTYPE atomics_get_2(BIF_ALIST_2) +{ + AtomicsRef* p; + UWord ix; + + if (!get_ref_ix(BIF_ARG_1, BIF_ARG_2, &p, &ix)) { + BIF_ERROR(BIF_P, BADARG); + } + return bld_atomic(BIF_P, p, erts_atomic64_read_mb(&p->v[ix])); +} + +BIF_RETTYPE atomics_add_3(BIF_ALIST_3) +{ + AtomicsRef* p; + UWord ix; + erts_aint64_t incr; + + if (!get_ref_ix(BIF_ARG_1, BIF_ARG_2, &p, &ix) + || !get_incr(p, BIF_ARG_3, &incr)) { + BIF_ERROR(BIF_P, BADARG); + } + erts_atomic64_add_mb(&p->v[ix], incr); + return am_ok; +} + +BIF_RETTYPE atomics_add_get_3(BIF_ALIST_3) +{ + AtomicsRef* p; + UWord ix; + erts_aint64_t incr; + + if (!get_ref_ix(BIF_ARG_1, BIF_ARG_2, &p, &ix) + || !get_incr(p, BIF_ARG_3, &incr)) { + BIF_ERROR(BIF_P, BADARG); + } + return bld_atomic(BIF_P, p, erts_atomic64_add_read_mb(&p->v[ix], incr)); +} + +BIF_RETTYPE atomics_exchange_3(BIF_ALIST_3) +{ + AtomicsRef* p; + UWord ix; + erts_aint64_t desired, was; + + if (!get_ref_ix(BIF_ARG_1, BIF_ARG_2, &p, &ix) + || !get_value(p, BIF_ARG_3, &desired)) { + BIF_ERROR(BIF_P, BADARG); + } + was = erts_atomic64_xchg_mb(&p->v[ix], desired); + return bld_atomic(BIF_P, p, was); +} + +BIF_RETTYPE atomics_compare_exchange_4(BIF_ALIST_4) +{ + AtomicsRef* p; + UWord ix; + erts_aint64_t expected, desired, was; + + if (!get_ref_ix(BIF_ARG_1, BIF_ARG_2, &p, &ix) + || !get_value(p, BIF_ARG_3, &expected) + || !get_value(p, BIF_ARG_4, &desired)) { + BIF_ERROR(BIF_P, BADARG); + } + was = erts_atomic64_cmpxchg_mb(&p->v[ix], desired, expected); + return was == expected ? am_ok : bld_atomic(BIF_P, p, was); +} + +BIF_RETTYPE atomics_info_1(BIF_ALIST_1) +{ + AtomicsRef* p; + Uint hsz = MAP4_SZ; + Eterm *hp; + Uint64 max; + Sint64 min; + UWord memory; + Eterm max_val, min_val, sz_val, mem_val; + + if (!get_ref(BIF_ARG_1, &p)) + BIF_ERROR(BIF_P, BADARG); + + max = p->is_signed ? ERTS_SINT64_MAX : ERTS_UINT64_MAX; + min = p->is_signed ? ERTS_SINT64_MIN : 0; + memory = erts_magic_ref2bin(BIF_ARG_1)->orig_size; + + erts_bld_uint64(NULL, &hsz, max); + erts_bld_sint64(NULL, &hsz, min); + erts_bld_uword(NULL, &hsz, p->vlen); + erts_bld_uword(NULL, &hsz, memory); + + hp = HAlloc(BIF_P, hsz); + max_val = erts_bld_uint64(&hp, NULL, max); + min_val = erts_bld_sint64(&hp, NULL, min); + sz_val = erts_bld_uword(&hp, NULL, p->vlen); + mem_val = erts_bld_uword(&hp, NULL, memory); + + return MAP4(hp, am_max, max_val, + am_memory, mem_val, + am_min, min_val, + am_size, sz_val); +} diff --git a/erts/emulator/beam/erl_bif_binary.c b/erts/emulator/beam/erl_bif_binary.c index ff919082c3..d465f6c6b9 100644 --- a/erts/emulator/beam/erl_bif_binary.c +++ b/erts/emulator/beam/erl_bif_binary.c @@ -471,6 +471,9 @@ static BMData *create_bmdata(MyAllocator *my, byte *x, Uint len, Binary **the_bin /* out */) { Uint datasize; + BMData *bmd; + Binary *mb; + byte *data; if(len > 1) { datasize = BM_SIZE_MULTI(len); @@ -478,9 +481,8 @@ static BMData *create_bmdata(MyAllocator *my, byte *x, Uint len, datasize = BM_SIZE_SINGLE(); } - BMData *bmd; - Binary *mb = erts_create_magic_binary(datasize,cleanup_my_data_bm); - byte *data = ERTS_MAGIC_BIN_DATA(mb); + mb = erts_create_magic_binary(datasize,cleanup_my_data_bm); + data = ERTS_MAGIC_BIN_DATA(mb); init_my_allocator(my, datasize, data); bmd = my_alloc(my, sizeof(BMData)); bmd->x = my_alloc(my,len); @@ -1049,14 +1051,13 @@ static int do_binary_match_compile(Eterm argument, Eterm *tag, Binary **binp) Uint bitoffs, bitsize; byte *temp_alloc = NULL; MyAllocator my; - BMData *bmd; Binary *bin; ERTS_GET_BINARY_BYTES(comp_term, bytes, bitoffs, bitsize); if (bitoffs != 0) { bytes = erts_get_aligned_binary_bytes(comp_term, &temp_alloc); } - bmd = create_bmdata(&my, bytes, characters, &bin); + create_bmdata(&my, bytes, characters, &bin); erts_free_aligned_binary_bytes(temp_alloc); CHECK_ALLOCATOR(my); *tag = am_bm; diff --git a/erts/emulator/beam/erl_bif_counters.c b/erts/emulator/beam/erl_bif_counters.c new file mode 100644 index 0000000000..a46b462225 --- /dev/null +++ b/erts/emulator/beam/erl_bif_counters.c @@ -0,0 +1,219 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2018. All Rights Reserved. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * %CopyrightEnd% + */ + +/* + * Purpose: High performance atomics. + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#include <stddef.h> /* offsetof */ + +#include "sys.h" +#include "export.h" +#include "bif.h" +#include "erl_threads.h" +#include "big.h" +#include "erl_binary.h" +#include "erl_bif_unique.h" +#include "erl_map.h" + + +#define COUNTERS_PER_CACHE_LINE (ERTS_CACHE_LINE_SIZE / sizeof(erts_atomic64_t)) + +typedef struct +{ + UWord arity; +#ifdef DEBUG + UWord ulen; +#endif + union { + erts_atomic64_t v[COUNTERS_PER_CACHE_LINE]; + byte cache_line__[ERTS_CACHE_LINE_SIZE]; + } u[1]; +}CountersRef; + +static int counters_destructor(Binary *unused) +{ + return 1; +} + + +static UWord ERTS_INLINE div_ceil(UWord dividend, UWord divisor) +{ + return (dividend + divisor - 1) / divisor; +} + +BIF_RETTYPE erts_internal_counters_new_1(BIF_ALIST_1) +{ + CountersRef* p; + Binary* mbin; + UWord ui, vi, cnt; + Uint bytes, cache_lines; + Eterm* hp; + + if (!term_to_UWord(BIF_ARG_1, &cnt) + || cnt == 0) { + BIF_ERROR(BIF_P, BADARG); + } + + if (cnt > (ERTS_UWORD_MAX / (sizeof(erts_atomic64_t)*2*erts_no_schedulers))) + BIF_ERROR(BIF_P, SYSTEM_LIMIT); + + cache_lines = erts_no_schedulers * div_ceil(cnt, COUNTERS_PER_CACHE_LINE); + bytes = offsetof(CountersRef, u) + cache_lines * ERTS_CACHE_LINE_SIZE; + mbin = erts_create_magic_binary_x(bytes, + counters_destructor, + ERTS_ALC_T_ATOMICS, + 0); + p = ERTS_MAGIC_BIN_DATA(mbin); + p->arity = cnt; +#ifdef DEBUG + p->ulen = cache_lines; +#endif + ASSERT((byte*)&p->u[cache_lines] <= ((byte*)p + bytes)); + for (ui=0; ui < cache_lines; ui++) + for (vi=0; vi < COUNTERS_PER_CACHE_LINE; vi++) + erts_atomic64_init_nob(&p->u[ui].v[vi], 0); + hp = HAlloc(BIF_P, ERTS_MAGIC_REF_THING_SIZE); + return erts_mk_magic_ref(&hp, &MSO(BIF_P), mbin); +} + +static ERTS_INLINE int get_ref(Eterm ref, CountersRef** pp) +{ + Binary* mbin; + if (!is_internal_magic_ref(ref)) + return 0; + + mbin = erts_magic_ref2bin(ref); + if (ERTS_MAGIC_BIN_DESTRUCTOR(mbin) != counters_destructor) + return 0; + *pp = ERTS_MAGIC_BIN_DATA(mbin); + return 1; +} + +static ERTS_INLINE int get_ref_cnt(Eterm ref, Eterm index, + CountersRef** pp, + erts_atomic64_t** app, + UWord sched_ix) +{ + CountersRef* p; + UWord ix, ui, vi; + if (!get_ref(ref, &p) || !term_to_UWord(index, &ix) || --ix >= p->arity) + return 0; + ui = (ix / COUNTERS_PER_CACHE_LINE) * erts_no_schedulers + sched_ix; + vi = ix % COUNTERS_PER_CACHE_LINE; + ASSERT(ui < p->ulen); + *pp = p; + *app = &p->u[ui].v[vi]; + return 1; +} + +static ERTS_INLINE int get_ref_my_cnt(Eterm ref, Eterm index, + CountersRef** pp, + erts_atomic64_t** app) +{ + ErtsSchedulerData *esdp = erts_get_scheduler_data(); + ASSERT(esdp && !ERTS_SCHEDULER_IS_DIRTY(esdp)); + return get_ref_cnt(ref, index, pp, app, esdp->no - 1); +} + +static ERTS_INLINE int get_ref_first_cnt(Eterm ref, Eterm index, + CountersRef** pp, + erts_atomic64_t** app) +{ + return get_ref_cnt(ref, index, pp, app, 0); +} + +static ERTS_INLINE int get_incr(CountersRef* p, Eterm term, erts_aint64_t *valp) +{ + return (term_to_Sint64(term, (Sint64*)valp) + || term_to_Uint64(term, (Uint64*)valp)); +} + +static ERTS_INLINE Eterm bld_counter(Process* proc, CountersRef* p, + erts_aint64_t val) +{ + if (IS_SSMALL(val)) + return make_small((Sint) val); + else { + Uint hsz = ERTS_SINT64_HEAP_SIZE(val); + Eterm* hp = HAlloc(proc, hsz); + return erts_sint64_to_big(val, &hp); + } +} + +BIF_RETTYPE erts_internal_counters_get_2(BIF_ALIST_2) +{ + CountersRef* p; + erts_atomic64_t* ap; + erts_aint64_t acc = 0; + int j; + + if (!get_ref_first_cnt(BIF_ARG_1, BIF_ARG_2, &p, &ap)) { + BIF_ERROR(BIF_P, BADARG); + } + for (j = erts_no_schedulers; j ; --j) { + acc += erts_atomic64_read_nob(ap); + ap = (erts_atomic64_t*) ((byte*)ap + ERTS_CACHE_LINE_SIZE); + } + return bld_counter(BIF_P, p, acc); +} + +BIF_RETTYPE erts_internal_counters_add_3(BIF_ALIST_3) +{ + CountersRef* p; + erts_atomic64_t* ap; + erts_aint64_t incr, sum; + + if (!get_ref_my_cnt(BIF_ARG_1, BIF_ARG_2, &p, &ap) + || !get_incr(p, BIF_ARG_3, &incr)) { + BIF_ERROR(BIF_P, BADARG); + } + sum = incr + erts_atomic64_read_nob(ap); + erts_atomic64_set_nob(ap, sum); + return am_ok; +} + + +BIF_RETTYPE erts_internal_counters_info_1(BIF_ALIST_1) +{ + CountersRef* p; + Uint hsz = MAP2_SZ; + Eterm *hp; + UWord memory; + Eterm sz_val, mem_val; + + if (!get_ref(BIF_ARG_1, &p)) + BIF_ERROR(BIF_P, BADARG); + + memory = erts_magic_ref2bin(BIF_ARG_1)->orig_size; + erts_bld_uword(NULL, &hsz, p->arity); + erts_bld_uword(NULL, &hsz, memory); + + hp = HAlloc(BIF_P, hsz); + sz_val = erts_bld_uword(&hp, NULL, p->arity); + mem_val = erts_bld_uword(&hp, NULL, memory); + + return MAP2(hp, am_memory, mem_val, + am_size, sz_val); +} diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c index 2a8e7e8858..3b0f0d33fa 100644 --- a/erts/emulator/beam/erl_bif_info.c +++ b/erts/emulator/beam/erl_bif_info.c @@ -4669,6 +4669,14 @@ BIF_RETTYPE erts_debug_set_internal_state_2(BIF_ALIST_2) BIF_RET(am_notsup); #endif } + else if (ERTS_IS_ATOM_STR("ets_force_split", BIF_ARG_1)) { + if (is_tuple(BIF_ARG_2)) { + Eterm* tpl = tuple_val(BIF_ARG_2); + + if (erts_ets_force_split(tpl[1], tpl[2] == am_true)) + BIF_RET(am_ok); + } + } } BIF_ERROR(BIF_P, BADARG); diff --git a/erts/emulator/beam/erl_bif_lists.c b/erts/emulator/beam/erl_bif_lists.c index 395be67a90..a793b34852 100644 --- a/erts/emulator/beam/erl_bif_lists.c +++ b/erts/emulator/beam/erl_bif_lists.c @@ -29,12 +29,13 @@ #include "sys.h" #include "erl_vm.h" #include "global.h" -#include "erl_process.h" -#include "error.h" #include "bif.h" +#include "erl_binary.h" + static Eterm keyfind(int Bif, Process* p, Eterm Key, Eterm Pos, Eterm List); + static BIF_RETTYPE append(Process* p, Eterm A, Eterm B) { Eterm list; @@ -146,103 +147,724 @@ BIF_RETTYPE append_2(BIF_ALIST_2) return append(BIF_P, BIF_ARG_1, BIF_ARG_2); } -/* - * erlang:'--'/2 - */ +/* erlang:'--'/2 + * + * Subtracts a list from another (LHS -- RHS), removing the first occurrence of + * each element in LHS from RHS. There is no type coercion so the elements must + * match exactly. + * + * The BIF is broken into several stages that can all trap individually, and it + * chooses its algorithm based on input size. If either input is small it will + * use a linear scan tuned to which side it's on, and if both inputs are large + * enough it will convert RHS into a multiset to provide good asymptotic + * behavior. */ + +#define SUBTRACT_LHS_THRESHOLD 16 +#define SUBTRACT_RHS_THRESHOLD 16 + +typedef enum { + SUBTRACT_STAGE_START, + SUBTRACT_STAGE_LEN_LHS, + + /* Naive linear scan that's efficient when + * LEN_LHS <= SUBTRACT_LHS_THRESHOLD. */ + SUBTRACT_STAGE_NAIVE_LHS, + + SUBTRACT_STAGE_LEN_RHS, + + /* As SUBTRACT_STAGE_NAIVE_LHS but for RHS. */ + SUBTRACT_STAGE_NAIVE_RHS, + + /* Creates a multiset from RHS for faster lookups before sweeping through + * LHS. The set is implemented as a red-black tree and duplicate elements + * are handled by a counter on each node. */ + SUBTRACT_STAGE_SET_BUILD, + SUBTRACT_STAGE_SET_FINISH +} ErtsSubtractCtxStage; + +typedef struct subtract_node__ { + struct subtract_node__ *parent; + struct subtract_node__ *left; + struct subtract_node__ *right; + int is_red; + + Eterm key; + Uint count; +} subtract_tree_t; + +typedef struct { + ErtsSubtractCtxStage stage; + + Eterm lhs_original; + Eterm rhs_original; + + Uint lhs_remaining; + Uint rhs_remaining; + + Eterm iterator; + + Eterm *result_cdr; + Eterm result; + + union { + Eterm lhs_elements[SUBTRACT_LHS_THRESHOLD]; + Eterm rhs_elements[SUBTRACT_RHS_THRESHOLD]; + + struct { + subtract_tree_t *tree; + + /* A memory area for the tree's nodes, saving us the need to have + * one allocation per node. */ + subtract_tree_t *alloc_start; + subtract_tree_t *alloc; + } rhs_set; + } u; +} ErtsSubtractContext; + +#define ERTS_RBT_PREFIX subtract +#define ERTS_RBT_T subtract_tree_t +#define ERTS_RBT_KEY_T Eterm +#define ERTS_RBT_FLAGS_T int +#define ERTS_RBT_INIT_EMPTY_TNODE(T) \ + do { \ + (T)->parent = NULL; \ + (T)->left = NULL; \ + (T)->right = NULL; \ + } while(0) +#define ERTS_RBT_IS_RED(T) ((T)->is_red) +#define ERTS_RBT_SET_RED(T) ((T)->is_red = 1) +#define ERTS_RBT_IS_BLACK(T) (!ERTS_RBT_IS_RED(T)) +#define ERTS_RBT_SET_BLACK(T) ((T)->is_red = 0) +#define ERTS_RBT_GET_FLAGS(T) ((T)->is_red) +#define ERTS_RBT_SET_FLAGS(T, F) ((T)->is_red = F) +#define ERTS_RBT_GET_PARENT(T) ((T)->parent) +#define ERTS_RBT_SET_PARENT(T, P) ((T)->parent = P) +#define ERTS_RBT_GET_RIGHT(T) ((T)->right) +#define ERTS_RBT_SET_RIGHT(T, R) ((T)->right = (R)) +#define ERTS_RBT_GET_LEFT(T) ((T)->left) +#define ERTS_RBT_SET_LEFT(T, L) ((T)->left = (L)) +#define ERTS_RBT_GET_KEY(T) ((T)->key) +#define ERTS_RBT_CMP_KEYS(KX, KY) CMP_TERM(KX, KY) +#define ERTS_RBT_WANT_LOOKUP_INSERT +#define ERTS_RBT_WANT_LOOKUP +#define ERTS_RBT_WANT_DELETE +#define ERTS_RBT_UNDEF + +#include "erl_rbtree.h" + +static int subtract_continue(Process *p, ErtsSubtractContext *context); + +static void subtract_ctx_dtor(ErtsSubtractContext *context) { + switch (context->stage) { + case SUBTRACT_STAGE_SET_BUILD: + case SUBTRACT_STAGE_SET_FINISH: + erts_free(ERTS_ALC_T_LIST_TRAP, context->u.rhs_set.alloc_start); + break; + default: + break; + } +} -#define SMALL_VEC_SIZE 10 -static Eterm subtract(Process* p, Eterm A, Eterm B) -{ - Eterm list; - Eterm* hp; - Uint need; - Eterm res; - Eterm small_vec[SMALL_VEC_SIZE]; /* Preallocated memory for small lists */ - Eterm* vec_p; - Eterm* vp; - Sint i; - Sint n; - Sint m; - - if ((n = erts_list_length(A)) < 0) { - BIF_ERROR(p, BADARG); +static int subtract_ctx_bin_dtor(Binary *context_bin) { + ErtsSubtractContext *context = ERTS_MAGIC_BIN_DATA(context_bin); + subtract_ctx_dtor(context); + return 1; +} + +static void subtract_ctx_move(ErtsSubtractContext *from, + ErtsSubtractContext *to) { + int uses_result_cdr = 0; + + to->stage = from->stage; + + to->lhs_original = from->lhs_original; + to->rhs_original = from->rhs_original; + + to->lhs_remaining = from->lhs_remaining; + to->rhs_remaining = from->rhs_remaining; + + to->iterator = from->iterator; + to->result = from->result; + + switch (to->stage) { + case SUBTRACT_STAGE_NAIVE_LHS: + sys_memcpy(to->u.lhs_elements, + from->u.lhs_elements, + sizeof(Eterm) * to->lhs_remaining); + break; + case SUBTRACT_STAGE_NAIVE_RHS: + sys_memcpy(to->u.rhs_elements, + from->u.rhs_elements, + sizeof(Eterm) * to->rhs_remaining); + + uses_result_cdr = 1; + break; + case SUBTRACT_STAGE_SET_FINISH: + uses_result_cdr = 1; + /* FALL THROUGH */ + case SUBTRACT_STAGE_SET_BUILD: + to->u.rhs_set.alloc_start = from->u.rhs_set.alloc_start; + to->u.rhs_set.alloc = from->u.rhs_set.alloc; + to->u.rhs_set.tree = from->u.rhs_set.tree; + break; + default: + break; } - if ((m = erts_list_length(B)) < 0) { - BIF_ERROR(p, BADARG); + + if (uses_result_cdr) { + if (from->result_cdr == &from->result) { + to->result_cdr = &to->result; + } else { + to->result_cdr = from->result_cdr; + } } - - if (n == 0) - BIF_RET(NIL); - if (m == 0) - BIF_RET(A); - - /* allocate element vector */ - if (n <= SMALL_VEC_SIZE) - vec_p = small_vec; - else - vec_p = (Eterm*) erts_alloc(ERTS_ALC_T_TMP, n * sizeof(Eterm)); - - /* PUT ALL ELEMENTS IN VP */ - vp = vec_p; - list = A; - i = n; - while(i--) { - Eterm* listp = list_val(list); - *vp++ = CAR(listp); - list = CDR(listp); +} + +static Eterm subtract_create_trap_state(Process *p, + ErtsSubtractContext *context) { + Binary *state_bin; + Eterm *hp; + + state_bin = erts_create_magic_binary(sizeof(ErtsSubtractContext), + subtract_ctx_bin_dtor); + + subtract_ctx_move(context, ERTS_MAGIC_BIN_DATA(state_bin)); + + hp = HAlloc(p, ERTS_MAGIC_REF_THING_SIZE); + + return erts_mk_magic_ref(&hp, &MSO(p), state_bin); +} + +static int subtract_enter_len_lhs(Process *p, ErtsSubtractContext *context) { + context->stage = SUBTRACT_STAGE_LEN_LHS; + + context->iterator = context->lhs_original; + context->lhs_remaining = 0; + + return subtract_continue(p, context); +} + +static int subtract_enter_len_rhs(Process *p, ErtsSubtractContext *context) { + context->stage = SUBTRACT_STAGE_LEN_RHS; + + context->iterator = context->rhs_original; + context->rhs_remaining = 0; + + return subtract_continue(p, context); +} + +static int subtract_get_length(Process *p, Eterm *iterator_p, Uint *count_p) { + static const Sint ELEMENTS_PER_RED = 32; + + Sint budget, count; + Eterm iterator; + + budget = ELEMENTS_PER_RED * ERTS_BIF_REDS_LEFT(p); + iterator = *iterator_p; + +#ifdef DEBUG + budget = budget / 10 + 1; +#endif + + for (count = 0; count < budget && is_list(iterator); count++) { + iterator = CDR(list_val(iterator)); } - - /* UNMARK ALL DELETED CELLS */ - list = B; - m = 0; /* number of deleted elements */ - while(is_list(list)) { - Eterm* listp = list_val(list); - Eterm elem = CAR(listp); - i = n; - vp = vec_p; - while(i--) { - if (is_value(*vp) && eq(*vp, elem)) { - *vp = THE_NON_VALUE; - m++; - break; - } - vp++; - } - list = CDR(listp); + + if (!is_list(iterator) && !is_nil(iterator)) { + return -1; } - - if (m == n) /* All deleted ? */ - res = NIL; - else if (m == 0) /* None deleted ? */ - res = A; - else { /* REBUILD LIST */ - res = NIL; - need = 2*(n - m); - hp = HAlloc(p, need); - vp = vec_p + n - 1; - while(vp >= vec_p) { - if (is_value(*vp)) { - res = CONS(hp, *vp, res); - hp += 2; - } - vp--; - } + + BUMP_REDS(p, count / ELEMENTS_PER_RED); + + *iterator_p = iterator; + *count_p += count; + + if (is_nil(iterator)) { + return 1; } - if (vec_p != small_vec) - erts_free(ERTS_ALC_T_TMP, (void *) vec_p); - BIF_RET(res); + + return 0; } -BIF_RETTYPE ebif_minusminus_2(BIF_ALIST_2) -{ - return subtract(BIF_P, BIF_ARG_1, BIF_ARG_2); +static int subtract_enter_naive_lhs(Process *p, ErtsSubtractContext *context) { + Eterm iterator; + int i = 0; + + context->stage = SUBTRACT_STAGE_NAIVE_LHS; + + context->iterator = context->rhs_original; + context->result = NIL; + + iterator = context->lhs_original; + + while (is_list(iterator)) { + const Eterm *cell = list_val(iterator); + + ASSERT(i < SUBTRACT_LHS_THRESHOLD); + + context->u.lhs_elements[i++] = CAR(cell); + iterator = CDR(cell); + } + + ASSERT(i == context->lhs_remaining); + + return subtract_continue(p, context); } -BIF_RETTYPE subtract_2(BIF_ALIST_2) -{ - return subtract(BIF_P, BIF_ARG_1, BIF_ARG_2); +static int subtract_naive_lhs(Process *p, ErtsSubtractContext *context) { + const Sint CHECKS_PER_RED = 16; + Sint checks, budget; + + budget = CHECKS_PER_RED * ERTS_BIF_REDS_LEFT(p); + checks = 0; + + while (checks < budget && is_list(context->iterator)) { + const Eterm *cell; + Eterm value, next; + int found_at; + + cell = list_val(context->iterator); + + value = CAR(cell); + next = CDR(cell); + + for (found_at = 0; found_at < context->lhs_remaining; found_at++) { + if (EQ(value, context->u.lhs_elements[found_at])) { + /* We shift the array one step down as we have to preserve + * order. + * + * Note that we can't exit early as that would suppress errors + * in the right-hand side (this runs prior to determining the + * length of RHS). */ + + context->lhs_remaining--; + sys_memmove(&context->u.lhs_elements[found_at], + &context->u.lhs_elements[found_at + 1], + (context->lhs_remaining - found_at) * sizeof(Eterm)); + break; + } + } + + checks += MAX(1, context->lhs_remaining); + context->iterator = next; + } + + BUMP_REDS(p, MIN(checks, budget) / CHECKS_PER_RED); + + if (is_list(context->iterator)) { + return 0; + } else if (!is_nil(context->iterator)) { + return -1; + } + + if (context->lhs_remaining > 0) { + Eterm *hp; + int i; + + hp = HAlloc(p, context->lhs_remaining * 2); + + for (i = context->lhs_remaining - 1; i >= 0; i--) { + Eterm value = context->u.lhs_elements[i]; + + context->result = CONS(hp, value, context->result); + hp += 2; + } + } + + ASSERT(context->lhs_remaining > 0 || context->result == NIL); + + return 1; +} + +static int subtract_enter_naive_rhs(Process *p, ErtsSubtractContext *context) { + Eterm iterator; + int i = 0; + + context->stage = SUBTRACT_STAGE_NAIVE_RHS; + + context->iterator = context->lhs_original; + context->result_cdr = &context->result; + context->result = NIL; + + iterator = context->rhs_original; + + while (is_list(iterator)) { + const Eterm *cell = list_val(iterator); + + ASSERT(i < SUBTRACT_RHS_THRESHOLD); + + context->u.rhs_elements[i++] = CAR(cell); + iterator = CDR(cell); + } + + ASSERT(i == context->rhs_remaining); + + return subtract_continue(p, context); +} + +static int subtract_naive_rhs(Process *p, ErtsSubtractContext *context) { + const Sint CHECKS_PER_RED = 16; + Sint checks, budget; + + budget = CHECKS_PER_RED * ERTS_BIF_REDS_LEFT(p); + checks = 0; + +#ifdef DEBUG + budget = budget / 10 + 1; +#endif + + while (checks < budget && is_list(context->iterator)) { + const Eterm *cell; + Eterm value, next; + int found_at; + + cell = list_val(context->iterator); + value = CAR(cell); + next = CDR(cell); + + for (found_at = context->rhs_remaining - 1; found_at >= 0; found_at--) { + if (EQ(value, context->u.rhs_elements[found_at])) { + break; + } + } + + if (found_at < 0) { + /* Destructively add the value to the result. This is safe + * since the GC is disabled and the unfinished term is never + * leaked to the outside world. */ + Eterm *hp = HAllocX(p, 2, context->lhs_remaining * 2); + + *context->result_cdr = make_list(hp); + context->result_cdr = &CDR(hp); + + CAR(hp) = value; + } else if (found_at >= 0) { + Eterm swap; + + if (context->rhs_remaining-- == 1) { + /* We've run out of items to remove, so the rest of the + * result will be equal to the remainder of the input. We know + * that LHS is well-formed as any errors would've been reported + * during length determination. */ + *context->result_cdr = next; + + BUMP_REDS(p, MIN(budget, checks) / CHECKS_PER_RED); + + return 1; + } + + swap = context->u.rhs_elements[context->rhs_remaining]; + context->u.rhs_elements[found_at] = swap; + } + + checks += context->rhs_remaining; + context->iterator = next; + context->lhs_remaining--; + } + + /* The result only has to be terminated when returning it to the user, but + * we're doing it when trapping as well to prevent headaches when + * debugging. */ + *context->result_cdr = NIL; + + BUMP_REDS(p, MIN(budget, checks) / CHECKS_PER_RED); + + if (is_list(context->iterator)) { + ASSERT(context->lhs_remaining > 0 && context->rhs_remaining > 0); + return 0; + } + + return 1; } +static int subtract_enter_set_build(Process *p, ErtsSubtractContext *context) { + context->stage = SUBTRACT_STAGE_SET_BUILD; + + context->u.rhs_set.alloc_start = + erts_alloc(ERTS_ALC_T_LIST_TRAP, + context->rhs_remaining * sizeof(subtract_tree_t)); + + context->u.rhs_set.alloc = context->u.rhs_set.alloc_start; + context->u.rhs_set.tree = NULL; + + context->iterator = context->rhs_original; + + return subtract_continue(p, context); +} + +static int subtract_set_build(Process *p, ErtsSubtractContext *context) { + const static Sint INSERTIONS_PER_RED = 16; + Sint budget, insertions; + + budget = INSERTIONS_PER_RED * ERTS_BIF_REDS_LEFT(p); + insertions = 0; + +#ifdef DEBUG + budget = budget / 10 + 1; +#endif + + while (insertions < budget && is_list(context->iterator)) { + subtract_tree_t *existing_node, *new_node; + const Eterm *cell; + Eterm value, next; + + cell = list_val(context->iterator); + value = CAR(cell); + next = CDR(cell); + + new_node = context->u.rhs_set.alloc; + new_node->key = value; + new_node->count = 1; + + existing_node = subtract_rbt_lookup_insert(&context->u.rhs_set.tree, + new_node); + + if (existing_node != NULL) { + existing_node->count++; + } else { + context->u.rhs_set.alloc++; + } + + context->iterator = next; + insertions++; + } + + BUMP_REDS(p, insertions / INSERTIONS_PER_RED); + + ASSERT(is_list(context->iterator) || is_nil(context->iterator)); + ASSERT(context->u.rhs_set.tree != NULL); + + return is_nil(context->iterator); +} + +static int subtract_enter_set_finish(Process *p, ErtsSubtractContext *context) { + context->stage = SUBTRACT_STAGE_SET_FINISH; + + context->result_cdr = &context->result; + context->result = NIL; + + context->iterator = context->lhs_original; + + return subtract_continue(p, context); +} + +static int subtract_set_finish(Process *p, ErtsSubtractContext *context) { + const Sint CHECKS_PER_RED = 8; + Sint checks, budget; + + budget = CHECKS_PER_RED * ERTS_BIF_REDS_LEFT(p); + checks = 0; + +#ifdef DEBUG + budget = budget / 10 + 1; +#endif + + while (checks < budget && is_list(context->iterator)) { + subtract_tree_t *node; + const Eterm *cell; + Eterm value, next; + + cell = list_val(context->iterator); + value = CAR(cell); + next = CDR(cell); + + ASSERT(context->rhs_remaining > 0); + + node = subtract_rbt_lookup(context->u.rhs_set.tree, value); + + if (node == NULL) { + Eterm *hp = HAllocX(p, 2, context->lhs_remaining * 2); + + *context->result_cdr = make_list(hp); + context->result_cdr = &CDR(hp); + + CAR(hp) = value; + } else { + if (context->rhs_remaining-- == 1) { + *context->result_cdr = next; + + BUMP_REDS(p, checks / CHECKS_PER_RED); + + return 1; + } + + if (node->count-- == 1) { + subtract_rbt_delete(&context->u.rhs_set.tree, node); + } + } + + context->iterator = next; + context->lhs_remaining--; + checks++; + } + + *context->result_cdr = NIL; + + BUMP_REDS(p, checks / CHECKS_PER_RED); + + if (is_list(context->iterator)) { + ASSERT(context->lhs_remaining > 0 && context->rhs_remaining > 0); + return 0; + } + + return 1; +} + +static int subtract_continue(Process *p, ErtsSubtractContext *context) { + switch (context->stage) { + case SUBTRACT_STAGE_START: { + return subtract_enter_len_lhs(p, context); + } + + case SUBTRACT_STAGE_LEN_LHS: { + int res = subtract_get_length(p, + &context->iterator, + &context->lhs_remaining); + + if (res != 1) { + return res; + } + + if (context->lhs_remaining <= SUBTRACT_LHS_THRESHOLD) { + return subtract_enter_naive_lhs(p, context); + } + + return subtract_enter_len_rhs(p, context); + } + + case SUBTRACT_STAGE_NAIVE_LHS: { + return subtract_naive_lhs(p, context); + } + + case SUBTRACT_STAGE_LEN_RHS: { + int res = subtract_get_length(p, + &context->iterator, + &context->rhs_remaining); + + if (res != 1) { + return res; + } + + /* We've walked through both lists fully now so we no longer need + * to check for errors past this point. */ + + if (context->rhs_remaining <= SUBTRACT_RHS_THRESHOLD) { + return subtract_enter_naive_rhs(p, context); + } + + return subtract_enter_set_build(p, context); + } + + case SUBTRACT_STAGE_NAIVE_RHS: { + return subtract_naive_rhs(p, context); + } + + case SUBTRACT_STAGE_SET_BUILD: { + int res = subtract_set_build(p, context); + + if (res != 1) { + return res; + } + + return subtract_enter_set_finish(p, context); + } + + case SUBTRACT_STAGE_SET_FINISH: { + return subtract_set_finish(p, context); + } + + default: + ERTS_ASSERT(!"unreachable"); + } +} + +static int subtract_start(Process *p, Eterm lhs, Eterm rhs, + ErtsSubtractContext *context) { + context->stage = SUBTRACT_STAGE_START; + + context->lhs_original = lhs; + context->rhs_original = rhs; + + return subtract_continue(p, context); +} + +/* erlang:'--'/2 */ +static Eterm subtract(Export *bif_entry, BIF_ALIST_2) { + Eterm lhs = BIF_ARG_1, rhs = BIF_ARG_2; + + if ((is_list(lhs) || is_nil(lhs)) && (is_list(rhs) || is_nil(rhs))) { + /* We start with the context on the stack in the hopes that we won't + * have to trap. */ + ErtsSubtractContext context; + int res; + + res = subtract_start(BIF_P, lhs, rhs, &context); + + if (res == 0) { + Eterm state_mref; + + state_mref = subtract_create_trap_state(BIF_P, &context); + erts_set_gc_state(BIF_P, 0); + + BIF_TRAP2(bif_entry, BIF_P, state_mref, NIL); + } + + subtract_ctx_dtor(&context); + + if (res < 0) { + BIF_ERROR(BIF_P, BADARG); + } + + BIF_RET(context.result); + } else if (is_internal_magic_ref(lhs)) { + ErtsSubtractContext *context; + int (*dtor)(Binary*); + Binary *magic_bin; + + int res; + + magic_bin = erts_magic_ref2bin(lhs); + dtor = ERTS_MAGIC_BIN_DESTRUCTOR(magic_bin); + + if (dtor != subtract_ctx_bin_dtor) { + BIF_ERROR(BIF_P, BADARG); + } + + ASSERT(BIF_P->flags & F_DISABLE_GC); + ASSERT(rhs == NIL); + + context = ERTS_MAGIC_BIN_DATA(magic_bin); + res = subtract_continue(BIF_P, context); + + if (res == 0) { + BIF_TRAP2(bif_entry, BIF_P, lhs, NIL); + } + + erts_set_gc_state(BIF_P, 1); + + if (res < 0) { + ERTS_BIF_ERROR_TRAPPED2(BIF_P, BADARG, bif_entry, + context->lhs_original, + context->rhs_original); + } + + BIF_RET(context->result); + } + + ASSERT(!(BIF_P->flags & F_DISABLE_GC)); + + BIF_ERROR(BIF_P, BADARG); +} + +BIF_RETTYPE ebif_minusminus_2(BIF_ALIST_2) { + return subtract(bif_export[BIF_ebif_minusminus_2], BIF_CALL_ARGS); +} + +BIF_RETTYPE subtract_2(BIF_ALIST_2) { + return subtract(bif_export[BIF_subtract_2], BIF_CALL_ARGS); +} + + BIF_RETTYPE lists_member_2(BIF_ALIST_2) { Eterm term; @@ -283,7 +905,7 @@ static BIF_RETTYPE lists_reverse_alloc(Process *c_p, { static const Uint CELLS_PER_RED = 40; - Eterm *heap_top, *heap_end; + Eterm *alloc_top, *alloc_end; Uint cells_left, max_cells; Eterm list, tail; Eterm lookahead; @@ -305,18 +927,18 @@ static BIF_RETTYPE lists_reverse_alloc(Process *c_p, BIF_ERROR(c_p, BADARG); } - heap_top = HAlloc(c_p, 2 * (max_cells - cells_left)); - heap_end = heap_top + 2 * (max_cells - cells_left); + alloc_top = HAlloc(c_p, 2 * (max_cells - cells_left)); + alloc_end = alloc_top + 2 * (max_cells - cells_left); - while (heap_top < heap_end) { + while (alloc_top < alloc_end) { Eterm *pair = list_val(list); - tail = CONS(heap_top, CAR(pair), tail); + tail = CONS(alloc_top, CAR(pair), tail); list = CDR(pair); ASSERT(is_list(list) || is_nil(list)); - heap_top += 2; + alloc_top += 2; } if (is_nil(list)) { @@ -333,7 +955,7 @@ static BIF_RETTYPE lists_reverse_onheap(Process *c_p, { static const Uint CELLS_PER_RED = 60; - Eterm *heap_top, *heap_end; + Eterm *alloc_start, *alloc_top, *alloc_end; Uint cells_left, max_cells; Eterm list, tail; @@ -343,21 +965,27 @@ static BIF_RETTYPE lists_reverse_onheap(Process *c_p, cells_left = max_cells = CELLS_PER_RED * (1 + ERTS_BIF_REDS_LEFT(c_p)); ASSERT(HEAP_LIMIT(c_p) >= HEAP_TOP(c_p) + 2); - heap_end = HEAP_LIMIT(c_p) - 2; - heap_top = HEAP_TOP(c_p); + alloc_start = HEAP_TOP(c_p); + alloc_end = HEAP_LIMIT(c_p) - 2; + alloc_top = alloc_start; + + /* Don't process more cells than we have reductions for. */ + alloc_end = MIN(alloc_top + (cells_left * 2), alloc_end); - while (heap_top < heap_end && is_list(list)) { + while (alloc_top < alloc_end && is_list(list)) { Eterm *pair = list_val(list); - tail = CONS(heap_top, CAR(pair), tail); + tail = CONS(alloc_top, CAR(pair), tail); list = CDR(pair); - heap_top += 2; + alloc_top += 2; } - cells_left -= (heap_top - heap_end) / 2; + cells_left -= (alloc_top - alloc_start) / 2; + HEAP_TOP(c_p) = alloc_top; + + ASSERT(cells_left >= 0 && cells_left <= max_cells); BUMP_REDS(c_p, (max_cells - cells_left) / CELLS_PER_RED); - HEAP_TOP(c_p) = heap_top; if (is_nil(list)) { BIF_RET(tail); diff --git a/erts/emulator/beam/erl_bif_persistent.c b/erts/emulator/beam/erl_bif_persistent.c new file mode 100644 index 0000000000..9dca768a18 --- /dev/null +++ b/erts/emulator/beam/erl_bif_persistent.c @@ -0,0 +1,983 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2018. All Rights Reserved. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * %CopyrightEnd% + */ + +/* + * Purpose: Implement persistent term storage. + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#include "sys.h" +#include "erl_vm.h" +#include "global.h" +#include "erl_process.h" +#include "error.h" +#include "erl_driver.h" +#include "bif.h" +#include "erl_map.h" +#include "erl_binary.h" + +/* + * The limit for the number of persistent terms before + * a warning is issued. + */ + +#define WARNING_LIMIT 20000 +#define XSTR(s) STR(s) +#define STR(s) #s + +/* + * Parameters for the hash table. + */ +#define INITIAL_SIZE 8 +#define LOAD_FACTOR ((Uint)50) +#define MUST_GROW(t) (((Uint)100) * t->num_entries >= LOAD_FACTOR * t->allocated) +#define MUST_SHRINK(t) (((Uint)200) * t->num_entries <= LOAD_FACTOR * t->allocated && \ + t->allocated > INITIAL_SIZE) + +typedef struct hash_table { + Uint allocated; + Uint num_entries; + Uint mask; + Uint first_to_delete; + Uint num_to_delete; + erts_atomic_t refc; + struct hash_table* delete_next; + ErtsThrPrgrLaterOp thr_prog_op; + Eterm term[1]; +} HashTable; + +typedef struct trap_data { + HashTable* table; + Uint idx; + Uint remaining; + Uint memory; /* Used by info/0 to count used memory */ +} TrapData; + +/* + * Declarations of local functions. + */ + +static HashTable* create_initial_table(void); +static Uint lookup(HashTable* hash_table, Eterm key); +static HashTable* copy_table(HashTable* old_table, Uint new_size, int rehash); +static HashTable* tmp_table_copy(HashTable* old_table); +static int try_seize_update_permission(Process* c_p); +static void release_update_permission(int release_updater); +static void table_updater(void* table); +static void table_deleter(void* hash_table); +static void dec_table_refc(Process* c_p, HashTable* old_table); +static void delete_table(Process* c_p, HashTable* table); +static void mark_for_deletion(HashTable* hash_table, Uint entry_index); +static ErtsLiteralArea* term_to_area(Eterm tuple); +static void suspend_updater(Process* c_p); +static Eterm do_get_all(Process* c_p, TrapData* trap_data, Eterm res); +static Eterm do_info(Process* c_p, TrapData* trap_data); +static void append_to_delete_queue(HashTable* table); +static HashTable* next_to_delete(void); +static Eterm alloc_trap_data(Process* c_p); +static int cleanup_trap_data(Binary *bp); + +/* + * Traps + */ + +static Export persistent_term_get_all_export; +static BIF_RETTYPE persistent_term_get_all_trap(BIF_ALIST_2); +static Export persistent_term_info_export; +static BIF_RETTYPE persistent_term_info_trap(BIF_ALIST_1); + +/* + * Pointer to the current hash table. + */ + +static erts_atomic_t the_hash_table; + +/* + * Queue of processes waiting to update the hash table. + */ + +struct update_queue_item { + Process *p; + struct update_queue_item* next; +}; + +static erts_mtx_t update_table_permission_mtx; +static struct update_queue_item* update_queue = NULL; +static Process* updater_process = NULL; + +/* Protected by update_table_permission_mtx */ +static ErtsThrPrgrLaterOp thr_prog_op; +static int issued_warning = 0; + +/* + * Queue of hash tables to be deleted. + */ + +static erts_mtx_t delete_queue_mtx; +static HashTable* delete_queue_head = NULL; +static HashTable** delete_queue_tail = &delete_queue_head; + +/* + * The following variables are only used during crash dumping. They + * are intialized by erts_init_persistent_dumping(). + */ + +ErtsLiteralArea** erts_persistent_areas; +Uint erts_num_persistent_areas; + +void erts_init_bif_persistent_term(void) +{ + HashTable* hash_table; + + /* + * Initialize the mutex protecting updates. + */ + + erts_mtx_init(&update_table_permission_mtx, + "update_persistent_term_permission", + NIL, + ERTS_LOCK_FLAGS_PROPERTY_STATIC | + ERTS_LOCK_FLAGS_CATEGORY_GENERIC); + + /* + * Initialize delete queue. + */ + + erts_mtx_init(&delete_queue_mtx, + "persistent_term_delete_permission", + NIL, + ERTS_LOCK_FLAGS_PROPERTY_STATIC | + ERTS_LOCK_FLAGS_CATEGORY_GENERIC); + + /* + * Allocate a small initial hash table. + */ + + hash_table = create_initial_table(); + erts_atomic_init_nob(&the_hash_table, (erts_aint_t)hash_table); + + /* + * Initialize export entry for traps + */ + + erts_init_trap_export(&persistent_term_get_all_export, + am_persistent_term, am_get_all_trap, 2, + &persistent_term_get_all_trap); + erts_init_trap_export(&persistent_term_info_export, + am_persistent_term, am_info_trap, 1, + &persistent_term_info_trap); +} + +BIF_RETTYPE persistent_term_put_2(BIF_ALIST_2) +{ + Eterm key; + Eterm term; + Eterm heap[3]; + Eterm tuple; + HashTable* hash_table; + Uint term_size; + Uint lit_area_size; + ErlOffHeap code_off_heap; + ErtsLiteralArea* literal_area; + erts_shcopy_t info; + Eterm* ptr; + Uint entry_index; + + if (!try_seize_update_permission(BIF_P)) { + ERTS_BIF_YIELD2(bif_export[BIF_persistent_term_put_2], + BIF_P, BIF_ARG_1, BIF_ARG_2); + } + + hash_table = (HashTable *) erts_atomic_read_nob(&the_hash_table); + + key = BIF_ARG_1; + term = BIF_ARG_2; + + entry_index = lookup(hash_table, key); + + heap[0] = make_arityval(2); + heap[1] = key; + heap[2] = term; + tuple = make_tuple(heap); + + if (is_nil(hash_table->term[entry_index])) { + Uint size = hash_table->allocated; + if (MUST_GROW(hash_table)) { + size *= 2; + } + hash_table = copy_table(hash_table, size, 0); + entry_index = lookup(hash_table, key); + hash_table->num_entries++; + } else { + Eterm tuple = hash_table->term[entry_index]; + Eterm old_term; + + ASSERT(is_tuple_arity(tuple, 2)); + old_term = boxed_val(tuple)[2]; + if (EQ(term, old_term)) { + /* Same value. No need to update anything. */ + release_update_permission(0); + BIF_RET(am_ok); + } else { + /* Mark the old term for deletion. */ + mark_for_deletion(hash_table, entry_index); + hash_table = copy_table(hash_table, hash_table->allocated, 0); + } + } + + /* + * Preserve internal sharing in the term by using the + * sharing-preserving functions. However, literals must + * be copied in case the module holding them are unloaded. + */ + INITIALIZE_SHCOPY(info); + info.copy_literals = 1; + term_size = copy_shared_calculate(tuple, &info); + ERTS_INIT_OFF_HEAP(&code_off_heap); + lit_area_size = ERTS_LITERAL_AREA_ALLOC_SIZE(term_size); + literal_area = erts_alloc(ERTS_ALC_T_LITERAL, lit_area_size); + ptr = &literal_area->start[0]; + literal_area->end = ptr + term_size; + tuple = copy_shared_perform(tuple, term_size, &info, &ptr, &code_off_heap); + ASSERT(tuple_val(tuple) == literal_area->start); + literal_area->off_heap = code_off_heap.first; + DESTROY_SHCOPY(info); + erts_set_literal_tag(&tuple, literal_area->start, term_size); + hash_table->term[entry_index] = tuple; + + erts_schedule_thr_prgr_later_op(table_updater, hash_table, &thr_prog_op); + suspend_updater(BIF_P); + + /* + * Issue a warning once if the warning limit has been exceeded. + */ + + if (hash_table->num_entries > WARNING_LIMIT && issued_warning == 0) { + static char w[] = + "More than " XSTR(WARNING_LIMIT) " persistent terms " + "have been created.\n" + "It is recommended to avoid creating an excessive number of\n" + "persistent terms, as creation and deletion of persistent terms\n" + "will be slower as the number of persistent terms increases.\n"; + issued_warning = 1; + erts_send_warning_to_logger_str(BIF_P->group_leader, w); + } + + ERTS_BIF_YIELD_RETURN(BIF_P, am_ok); +} + +BIF_RETTYPE persistent_term_get_0(BIF_ALIST_0) +{ + HashTable* hash_table = (HashTable *) erts_atomic_read_nob(&the_hash_table); + TrapData* trap_data; + Eterm res = NIL; + Eterm magic_ref; + Binary* mbp; + + magic_ref = alloc_trap_data(BIF_P); + mbp = erts_magic_ref2bin(magic_ref); + trap_data = ERTS_MAGIC_BIN_DATA(mbp); + trap_data->table = hash_table; + trap_data->idx = 0; + trap_data->remaining = hash_table->num_entries; + res = do_get_all(BIF_P, trap_data, res); + if (trap_data->remaining == 0) { + BUMP_REDS(BIF_P, hash_table->num_entries); + trap_data->table = NULL; /* Prevent refc decrement */ + BIF_RET(res); + } else { + /* + * Increment the ref counter to prevent an update operation (by put/2 + * or erase/1) to delete this hash table. + */ + erts_atomic_inc_nob(&hash_table->refc); + BUMP_ALL_REDS(BIF_P); + BIF_TRAP2(&persistent_term_get_all_export, BIF_P, magic_ref, res); + } +} + +BIF_RETTYPE persistent_term_get_1(BIF_ALIST_1) +{ + Eterm key = BIF_ARG_1; + HashTable* hash_table = (HashTable *) erts_atomic_read_nob(&the_hash_table); + Uint entry_index; + Eterm term; + + entry_index = lookup(hash_table, key); + term = hash_table->term[entry_index]; + if (is_boxed(term)) { + ASSERT(is_tuple_arity(term, 2)); + BIF_RET(tuple_val(term)[2]); + } + BIF_ERROR(BIF_P, BADARG); +} + +BIF_RETTYPE persistent_term_erase_1(BIF_ALIST_1) +{ + Eterm key = BIF_ARG_1; + HashTable* old_table; + HashTable* new_table; + Uint entry_index; + Eterm old_term; + + if (!try_seize_update_permission(BIF_P)) { + ERTS_BIF_YIELD1(bif_export[BIF_persistent_term_erase_1], + BIF_P, BIF_ARG_1); + } + + old_table = (HashTable *) erts_atomic_read_nob(&the_hash_table); + entry_index = lookup(old_table, key); + old_term = old_table->term[entry_index]; + if (is_boxed(old_term)) { + Uint new_size; + HashTable* tmp_table; + + /* + * Since we don't use any delete markers, we must rehash + * the table when deleting terms to ensure that all terms + * can still be reached if there are hash collisions. + * We can't rehash in place and it would not be safe to modify + * the old table yet, so we will first need a new + * temporary table copy of the same size as the old one. + */ + + ASSERT(is_tuple_arity(old_term, 2)); + tmp_table = tmp_table_copy(old_table); + + /* + * Delete the term from the temporary table. Then copy the + * temporary table to a new table, rehashing the entries + * while copying. + */ + + tmp_table->term[entry_index] = NIL; + tmp_table->num_entries--; + new_size = tmp_table->allocated; + if (MUST_SHRINK(tmp_table)) { + new_size /= 2; + } + new_table = copy_table(tmp_table, new_size, 1); + erts_free(ERTS_ALC_T_TMP, tmp_table); + + mark_for_deletion(old_table, entry_index); + erts_schedule_thr_prgr_later_op(table_updater, new_table, &thr_prog_op); + suspend_updater(BIF_P); + ERTS_BIF_YIELD_RETURN(BIF_P, am_true); + } + + /* + * Key is not present. Nothing to do. + */ + + ASSERT(is_nil(old_term)); + release_update_permission(0); + BIF_RET(am_false); +} + +BIF_RETTYPE erts_internal_erase_persistent_terms_0(BIF_ALIST_0) +{ + HashTable* old_table; + HashTable* new_table; + + if (!try_seize_update_permission(BIF_P)) { + ERTS_BIF_YIELD0(bif_export[BIF_erts_internal_erase_persistent_terms_0], + BIF_P); + } + old_table = (HashTable *) erts_atomic_read_nob(&the_hash_table); + old_table->first_to_delete = 0; + old_table->num_to_delete = old_table->allocated; + new_table = create_initial_table(); + erts_schedule_thr_prgr_later_op(table_updater, new_table, &thr_prog_op); + suspend_updater(BIF_P); + ERTS_BIF_YIELD_RETURN(BIF_P, am_true); +} + +BIF_RETTYPE persistent_term_info_0(BIF_ALIST_0) +{ + HashTable* hash_table = (HashTable *) erts_atomic_read_nob(&the_hash_table); + TrapData* trap_data; + Eterm res = NIL; + Eterm magic_ref; + Binary* mbp; + + magic_ref = alloc_trap_data(BIF_P); + mbp = erts_magic_ref2bin(magic_ref); + trap_data = ERTS_MAGIC_BIN_DATA(mbp); + trap_data->table = hash_table; + trap_data->idx = 0; + trap_data->remaining = hash_table->num_entries; + trap_data->memory = 0; + res = do_info(BIF_P, trap_data); + if (trap_data->remaining == 0) { + BUMP_REDS(BIF_P, hash_table->num_entries); + trap_data->table = NULL; /* Prevent refc decrement */ + BIF_RET(res); + } else { + /* + * Increment the ref counter to prevent an update operation (by put/2 + * or erase/1) to delete this hash table. + */ + erts_atomic_inc_nob(&hash_table->refc); + BUMP_ALL_REDS(BIF_P); + BIF_TRAP2(&persistent_term_info_export, BIF_P, magic_ref, res); + } +} + +Uint +erts_persistent_term_count(void) +{ + HashTable* hash_table = (HashTable *) erts_atomic_read_nob(&the_hash_table); + return hash_table->num_entries; +} + +void +erts_init_persistent_dumping(void) +{ + HashTable* hash_table = (HashTable *) erts_atomic_read_nob(&the_hash_table); + ErtsLiteralArea** area_p; + Uint i; + + /* + * Overwrite the array of Eterms in the current hash table + * with pointers to literal areas. + */ + + erts_persistent_areas = (ErtsLiteralArea **) hash_table->term; + erts_num_persistent_areas = hash_table->num_entries; + area_p = erts_persistent_areas; + for (i = 0; i < hash_table->allocated; i++) { + Eterm term = hash_table->term[i]; + + if (is_boxed(term)) { + *area_p++ = term_to_area(term); + } + } +} + +/* + * Local functions. + */ + +static HashTable* +create_initial_table(void) +{ + HashTable* hash_table; + int i; + + hash_table = (HashTable *) erts_alloc(ERTS_ALC_T_PERSISTENT_TERM, + sizeof(HashTable)+sizeof(Eterm) * + (INITIAL_SIZE-1)); + hash_table->allocated = INITIAL_SIZE; + hash_table->num_entries = 0; + hash_table->mask = INITIAL_SIZE-1; + hash_table->first_to_delete = 0; + hash_table->num_to_delete = 0; + erts_atomic_init_nob(&hash_table->refc, (erts_aint_t)1); + for (i = 0; i < INITIAL_SIZE; i++) { + hash_table->term[i] = NIL; + } + return hash_table; +} + +static BIF_RETTYPE +persistent_term_get_all_trap(BIF_ALIST_2) +{ + TrapData* trap_data; + Eterm res = BIF_ARG_2; + Uint bump_reds; + Binary* mbp; + + ASSERT(is_list(BIF_ARG_2)); + mbp = erts_magic_ref2bin(BIF_ARG_1); + trap_data = ERTS_MAGIC_BIN_DATA(mbp); + bump_reds = trap_data->remaining; + res = do_get_all(BIF_P, trap_data, res); + ASSERT(is_list(res)); + if (trap_data->remaining > 0) { + BUMP_ALL_REDS(BIF_P); + BIF_TRAP2(&persistent_term_get_all_export, BIF_P, BIF_ARG_1, res); + } else { + /* + * Decrement ref count (and possibly delete the hash table + * and associated literal area). + */ + dec_table_refc(BIF_P, trap_data->table); + trap_data->table = NULL; /* Prevent refc decrement */ + BUMP_REDS(BIF_P, bump_reds); + BIF_RET(res); + } +} + +static Eterm +do_get_all(Process* c_p, TrapData* trap_data, Eterm res) +{ + HashTable* hash_table; + Uint remaining; + Uint idx; + Uint max_iter; + Uint i; + Eterm* hp; + Uint heap_size; + struct copy_term { + Uint key_size; + Eterm* tuple_ptr; + } *copy_data; + + hash_table = trap_data->table; + idx = trap_data->idx; +#if defined(DEBUG) || defined(VALGRIND) + max_iter = 50; +#else + max_iter = ERTS_BIF_REDS_LEFT(c_p); +#endif + remaining = trap_data->remaining < max_iter ? + trap_data->remaining : max_iter; + trap_data->remaining -= remaining; + + copy_data = (struct copy_term *) erts_alloc(ERTS_ALC_T_TMP, + remaining * + sizeof(struct copy_term)); + i = 0; + heap_size = (2 + 3) * remaining; + while (remaining != 0) { + Eterm term = hash_table->term[idx]; + if (is_tuple(term)) { + Uint key_size; + Eterm* tup_val; + + ASSERT(is_tuple_arity(term, 2)); + tup_val = tuple_val(term); + key_size = size_object(tup_val[1]); + copy_data[i].key_size = key_size; + copy_data[i].tuple_ptr = tup_val; + heap_size += key_size; + i++; + remaining--; + } + idx++; + } + trap_data->idx = idx; + + hp = HAlloc(c_p, heap_size); + remaining = i; + for (i = 0; i < remaining; i++) { + Eterm* tuple_ptr; + Uint key_size; + Eterm key; + Eterm tup; + + tuple_ptr = copy_data[i].tuple_ptr; + key_size = copy_data[i].key_size; + key = copy_struct(tuple_ptr[1], key_size, &hp, &c_p->off_heap); + tup = TUPLE2(hp, key, tuple_ptr[2]); + hp += 3; + res = CONS(hp, tup, res); + hp += 2; + } + erts_free(ERTS_ALC_T_TMP, copy_data); + return res; +} + +static BIF_RETTYPE +persistent_term_info_trap(BIF_ALIST_1) +{ + TrapData* trap_data = (TrapData *) BIF_ARG_1; + Eterm res; + Uint bump_reds; + Binary* mbp; + + mbp = erts_magic_ref2bin(BIF_ARG_1); + trap_data = ERTS_MAGIC_BIN_DATA(mbp); + bump_reds = trap_data->remaining; + res = do_info(BIF_P, trap_data); + if (trap_data->remaining > 0) { + ASSERT(res == am_ok); + BUMP_ALL_REDS(BIF_P); + BIF_TRAP1(&persistent_term_info_export, BIF_P, BIF_ARG_1); + } else { + /* + * Decrement ref count (and possibly delete the hash table + * and associated literal area). + */ + dec_table_refc(BIF_P, trap_data->table); + trap_data->table = NULL; /* Prevent refc decrement */ + BUMP_REDS(BIF_P, bump_reds); + ASSERT(is_map(res)); + BIF_RET(res); + } +} + +#define DECL_AM(S) Eterm AM_ ## S = am_atom_put(#S, sizeof(#S) - 1) + +static Eterm +do_info(Process* c_p, TrapData* trap_data) +{ + HashTable* hash_table; + Uint remaining; + Uint idx; + Uint max_iter; + + hash_table = trap_data->table; + idx = trap_data->idx; +#if defined(DEBUG) || defined(VALGRIND) + max_iter = 50; +#else + max_iter = ERTS_BIF_REDS_LEFT(c_p); +#endif + remaining = trap_data->remaining < max_iter ? trap_data->remaining : max_iter; + trap_data->remaining -= remaining; + while (remaining != 0) { + if (is_boxed(hash_table->term[idx])) { + ErtsLiteralArea* area; + area = term_to_area(hash_table->term[idx]); + trap_data->memory += sizeof(ErtsLiteralArea) + + sizeof(Eterm) * (area->end - area->start - 1); + remaining--; + } + idx++; + } + trap_data->idx = idx; + if (trap_data->remaining > 0) { + return am_ok; /* Dummy return value */ + } else { + Eterm* hp; + Eterm count_term; + Eterm memory_term; + Eterm res; + Uint memory; + Uint hsz = MAP_SZ(2); + + memory = sizeof(HashTable) + (trap_data->table->allocated-1) * + sizeof(Eterm) + trap_data->memory; + (void) erts_bld_uint(NULL, &hsz, hash_table->num_entries); + (void) erts_bld_uint(NULL, &hsz, memory); + hp = HAlloc(c_p, hsz); + count_term = erts_bld_uint(&hp, NULL, hash_table->num_entries); + memory_term = erts_bld_uint(&hp, NULL, memory); + res = MAP2(hp, am_count, count_term, am_memory, memory_term); + return res; + } +} + +#undef DECL_AM + +static Eterm +alloc_trap_data(Process* c_p) +{ + Binary* mbp = erts_create_magic_binary(sizeof(TrapData), + cleanup_trap_data); + Eterm* hp; + + hp = HAlloc(c_p, ERTS_MAGIC_REF_THING_SIZE); + return erts_mk_magic_ref(&hp, &MSO(c_p), mbp); +} + +static int +cleanup_trap_data(Binary *bp) +{ + TrapData* trap_data = ERTS_MAGIC_BIN_DATA(bp); + + if (trap_data->table) { + /* + * The process has been killed and is now exiting. + * Decrement the reference counter for the table. + */ + dec_table_refc(NULL, trap_data->table); + } + return 1; +} + +static Uint +lookup(HashTable* hash_table, Eterm key) +{ + Uint mask = hash_table->mask; + Eterm* table = hash_table->term; + Uint32 idx = make_internal_hash(key, 0); + Eterm term; + + do { + idx++; + term = table[idx & mask]; + } while (is_boxed(term) && !EQ(key, (tuple_val(term))[1])); + return idx & mask; +} + +static HashTable* +tmp_table_copy(HashTable* old_table) +{ + Uint size = old_table->allocated; + HashTable* tmp_table; + Uint i; + + tmp_table = (HashTable *) erts_alloc(ERTS_ALC_T_TMP, + sizeof(HashTable) + + sizeof(Eterm) * (size-1)); + *tmp_table = *old_table; + for (i = 0; i < size; i++) { + tmp_table->term[i] = old_table->term[i]; + } + return tmp_table; +} + +static HashTable* +copy_table(HashTable* old_table, Uint new_size, int rehash) +{ + HashTable* new_table; + Uint old_size = old_table->allocated; + Uint i; + + new_table = (HashTable *) erts_alloc(ERTS_ALC_T_PERSISTENT_TERM, + sizeof(HashTable) + + sizeof(Eterm) * (new_size-1)); + if (old_table->allocated == new_size && !rehash) { + /* + * Same size and no key deleted. Make an exact copy of the table. + */ + *new_table = *old_table; + for (i = 0; i < new_size; i++) { + new_table->term[i] = old_table->term[i]; + } + } else { + /* + * The size of the table has changed or an element has been + * deleted. Must rehash, by inserting all old terms into the + * new (empty) table. + */ + new_table->allocated = new_size; + new_table->num_entries = old_table->num_entries; + new_table->mask = new_size - 1; + for (i = 0; i < new_size; i++) { + new_table->term[i] = NIL; + } + for (i = 0; i < old_size; i++) { + if (is_tuple(old_table->term[i])) { + Eterm key = tuple_val(old_table->term[i])[1]; + Uint entry_index = lookup(new_table, key); + ASSERT(is_nil(new_table->term[entry_index])); + new_table->term[entry_index] = old_table->term[i]; + } + } + } + new_table->first_to_delete = 0; + new_table->num_to_delete = 0; + erts_atomic_init_nob(&new_table->refc, (erts_aint_t)1); + return new_table; +} + +static void +mark_for_deletion(HashTable* hash_table, Uint entry_index) +{ + hash_table->first_to_delete = entry_index; + hash_table->num_to_delete = 1; +} + +static ErtsLiteralArea* +term_to_area(Eterm tuple) +{ + ASSERT(is_tuple_arity(tuple, 2)); + return (ErtsLiteralArea *) (((char *) tuple_val(tuple)) - + offsetof(ErtsLiteralArea, start)); +} + +static void +table_updater(void* data) +{ + HashTable* old_table; + HashTable* new_table; + + old_table = (HashTable *) erts_atomic_read_nob(&the_hash_table); + new_table = (HashTable *) data; + ASSERT(new_table->num_to_delete == 0); + erts_atomic_set_nob(&the_hash_table, (erts_aint_t)new_table); + append_to_delete_queue(old_table); + erts_schedule_thr_prgr_later_op(table_deleter, + old_table, + &old_table->thr_prog_op); + release_update_permission(1); +} + +static void +table_deleter(void* data) +{ + HashTable* old_table = (HashTable *) data; + + dec_table_refc(NULL, old_table); +} + +static void +dec_table_refc(Process* c_p, HashTable* old_table) +{ + erts_aint_t refc = erts_atomic_dec_read_nob(&old_table->refc); + + if (refc == 0) { + HashTable* to_delete; + + while ((to_delete = next_to_delete()) != NULL) { + delete_table(c_p, to_delete); + } + } +} + +static void +delete_table(Process* c_p, HashTable* table) +{ + Uint idx = table->first_to_delete; + Uint n = table->num_to_delete; + + /* + * There are no longer any references to this hash table. + * + * Any literals pointed for deletion can be queued for + * deletion and the table itself can be deallocated. + */ + +#ifdef DEBUG + if (n == 1) { + ASSERT(is_tuple_arity(table->term[idx], 2)); + } +#endif + + while (n > 0) { + Eterm term = table->term[idx]; + + if (is_tuple_arity(term, 2)) { + if (is_immed(tuple_val(term)[2])) { + erts_release_literal_area(term_to_area(term)); + } else { + erts_queue_release_literals(c_p, term_to_area(term)); + } + } + idx++, n--; + } + erts_free(ERTS_ALC_T_PERSISTENT_TERM, table); +} + +/* + * Caller *must* yield if this function returns 0. + */ + +static int +try_seize_update_permission(Process* c_p) +{ + int success; + + ASSERT(!erts_thr_progress_is_blocking()); /* to avoid deadlock */ + ASSERT(c_p != NULL); + + erts_mtx_lock(&update_table_permission_mtx); + ASSERT(updater_process != c_p); + success = (updater_process == NULL); + if (success) { + updater_process = c_p; + } else { + struct update_queue_item* qitem; + qitem = erts_alloc(ERTS_ALC_T_PERSISTENT_LOCK_Q, sizeof(*qitem)); + qitem->p = c_p; + erts_proc_inc_refc(c_p); + qitem->next = update_queue; + update_queue = qitem; + erts_suspend(c_p, ERTS_PROC_LOCK_MAIN, NULL); + } + erts_mtx_unlock(&update_table_permission_mtx); + return success; +} + +static void +release_update_permission(int release_updater) +{ + erts_mtx_lock(&update_table_permission_mtx); + ASSERT(updater_process != NULL); + + if (release_updater) { + erts_proc_lock(updater_process, ERTS_PROC_LOCK_STATUS); + if (!ERTS_PROC_IS_EXITING(updater_process)) { + erts_resume(updater_process, ERTS_PROC_LOCK_STATUS); + } + erts_proc_unlock(updater_process, ERTS_PROC_LOCK_STATUS); + } + updater_process = NULL; + + while (update_queue != NULL) { /* Unleash the entire herd */ + struct update_queue_item* qitem = update_queue; + erts_proc_lock(qitem->p, ERTS_PROC_LOCK_STATUS); + if (!ERTS_PROC_IS_EXITING(qitem->p)) { + erts_resume(qitem->p, ERTS_PROC_LOCK_STATUS); + } + erts_proc_unlock(qitem->p, ERTS_PROC_LOCK_STATUS); + update_queue = qitem->next; + erts_proc_dec_refc(qitem->p); + erts_free(ERTS_ALC_T_PERSISTENT_LOCK_Q, qitem); + } + erts_mtx_unlock(&update_table_permission_mtx); +} + +static void +suspend_updater(Process* c_p) +{ +#ifdef DEBUG + ASSERT(c_p != NULL); + erts_mtx_lock(&update_table_permission_mtx); + ASSERT(updater_process == c_p); + erts_mtx_unlock(&update_table_permission_mtx); +#endif + erts_suspend(c_p, ERTS_PROC_LOCK_MAIN, NULL); +} + +static void +append_to_delete_queue(HashTable* table) +{ + erts_mtx_lock(&delete_queue_mtx); + table->delete_next = NULL; + *delete_queue_tail = table; + delete_queue_tail = &table->delete_next; + erts_mtx_unlock(&delete_queue_mtx); +} + +static HashTable* +next_to_delete(void) +{ + HashTable* table; + + erts_mtx_lock(&delete_queue_mtx); + table = delete_queue_head; + if (table) { + if (erts_atomic_read_nob(&table->refc)) { + /* + * This hash table is still referenced. Hash tables + * must be deleted in order, so we return a NULL + * pointer. + */ + table = NULL; + } else { + /* + * Remove the first hash table from the queue. + */ + delete_queue_head = table->delete_next; + if (delete_queue_head == NULL) { + delete_queue_tail = &delete_queue_head; + } + } + } + erts_mtx_unlock(&delete_queue_mtx); + return table; +} diff --git a/erts/emulator/beam/erl_bif_unique.h b/erts/emulator/beam/erl_bif_unique.h index 40b70667c0..944788c67c 100644 --- a/erts/emulator/beam/erl_bif_unique.h +++ b/erts/emulator/beam/erl_bif_unique.h @@ -242,11 +242,11 @@ erts_internal_ref_number_cmp(Uint32 num1[ERTS_REF_NUMBERS], Uint32 num2[ERTS_REF_NUMBERS]) { if (num1[2] != num2[2]) - return (int) ((Sint64) num1[2] - (Sint64) num2[2]); + return num1[2] > num2[2] ? 1 : -1; if (num1[1] != num2[1]) - return (int) ((Sint64) num1[1] - (Sint64) num2[1]); + return num1[1] > num2[1] ? 1 : -1; if (num1[0] != num2[0]) - return (int) ((Sint64) num1[0] - (Sint64) num2[0]); + return num1[0] > num2[0] ? 1 : -1; return 0; } diff --git a/erts/emulator/beam/erl_db.c b/erts/emulator/beam/erl_db.c index 1df972f4b6..df6f42edd3 100644 --- a/erts/emulator/beam/erl_db.c +++ b/erts/emulator/beam/erl_db.c @@ -409,30 +409,17 @@ static void free_dbtable(void *vtb) { DbTable *tb = (DbTable *) vtb; -#ifdef HARDDEBUG - if (erts_atomic_read_nob(&tb->common.memory_size) != sizeof(DbTable)) { - erts_fprintf(stderr, "ets: free_dbtable memory remain=%ld fix=%x\n", - erts_atomic_read_nob(&tb->common.memory_size)-sizeof(DbTable), - tb->common.fixations); - } -#endif - if (erts_atomic_read_nob(&tb->common.memory_size) > sizeof(DbTable)) { - /* The CA tree implementation use delayed freeing and the DbTable needs to - be freed after all other memory blocks that are allocated by the table. */ - erts_schedule_thr_prgr_later_cleanup_op(free_dbtable, - (void *) tb, - &tb->release.data, - sizeof(DbTable)); - return; - } - erts_rwmtx_destroy(&tb->common.rwlock); - erts_mtx_destroy(&tb->common.fixlock); - ASSERT(is_immed(tb->common.heir_data)); - if (tb->common.btid) - erts_bin_release(tb->common.btid); + ASSERT(erts_atomic_read_nob(&tb->common.memory_size) == sizeof(DbTable)); - erts_db_free(ERTS_ALC_T_DB_TABLE, tb, (void *) tb, sizeof(DbTable)); + erts_rwmtx_destroy(&tb->common.rwlock); + erts_mtx_destroy(&tb->common.fixlock); + ASSERT(is_immed(tb->common.heir_data)); + + if (tb->common.btid) + erts_bin_release(tb->common.btid); + + erts_db_free(ERTS_ALC_T_DB_TABLE, tb, (void *) tb, sizeof(DbTable)); } static void schedule_free_dbtable(DbTable* tb) @@ -4260,9 +4247,20 @@ static Eterm table_info(Process* p, DbTable* tb, Eterm What) make_small(stats.max_chain_len), make_small(stats.kept_items)); } - else { + else if (IS_CATREE_TABLE(tb->common.status)) { + DbCATreeStats stats; + Eterm* hp; + + db_calc_stats_catree(&tb->catree, &stats); + hp = HAlloc(p, 4); + ret = TUPLE3(hp, + make_small(stats.route_nodes), + make_small(stats.base_nodes), + make_small(stats.max_depth)); + + } + else ret = am_false; - } } return ret; } @@ -4467,3 +4465,16 @@ void erts_lcnt_update_db_locks(int enable) { #ifdef ETS_DBG_FORCE_TRAP erts_aint_t erts_ets_dbg_force_trap = 0; #endif + +int erts_ets_force_split(Eterm tid, int on) +{ + DbTable* tb = tid2tab(tid); + if (!tb || !IS_CATREE_TABLE(tb->common.type)) + return 0; + + db_lock(tb, LCK_WRITE); + if (!(tb->common.status & DB_DELETE)) + db_catree_force_split(&tb->catree, on); + db_unlock(tb, LCK_WRITE); + return 1; +} diff --git a/erts/emulator/beam/erl_db.h b/erts/emulator/beam/erl_db.h index 45d120ac0e..5955d42aae 100644 --- a/erts/emulator/beam/erl_db.h +++ b/erts/emulator/beam/erl_db.h @@ -130,6 +130,7 @@ extern Export ets_select_continue_exp; extern erts_atomic_t erts_ets_misc_mem_size; Eterm erts_ets_colliding_names(Process*, Eterm name, Uint cnt); +int erts_ets_force_split(Eterm tid, int on); Uint erts_db_get_max_tabs(void); Eterm erts_db_make_tid(Process *c_p, DbTableCommon *tb); @@ -286,6 +287,12 @@ ERTS_GLB_INLINE void erts_db_free(ErtsAlcType_t type, void *ptr, Uint size); +ERTS_GLB_INLINE void erts_schedule_db_free(DbTableCommon* tab, + void (*free_func)(void *), + void *ptr, + ErtsThrPrgrLaterOp *lop, + Uint size); + ERTS_GLB_INLINE void erts_db_free_nt(ErtsAlcType_t type, void *ptr, Uint size); @@ -306,6 +313,26 @@ erts_db_free(ErtsAlcType_t type, DbTable *tab, void *ptr, Uint size) } ERTS_GLB_INLINE void +erts_schedule_db_free(DbTableCommon* tab, + void (*free_func)(void *), + void *ptr, + ErtsThrPrgrLaterOp *lop, + Uint size) +{ + ASSERT(ptr != 0); + ASSERT(((void *) tab) != ptr); + ASSERT(size == ERTS_ALC_DBG_BLK_SZ(ptr)); + + /* + * We update table memory stats here as table may already be gone + * when 'free_func' is finally called. + */ + ERTS_DB_ALC_MEM_UPDATE_((DbTable*)tab, size, 0); + + erts_schedule_thr_prgr_later_cleanup_op(free_func, ptr, lop, size); +} + +ERTS_GLB_INLINE void erts_db_free_nt(ErtsAlcType_t type, void *ptr, Uint size) { ASSERT(ptr != 0); diff --git a/erts/emulator/beam/erl_db_catree.c b/erts/emulator/beam/erl_db_catree.c index 37a299df35..639c7e5e03 100644 --- a/erts/emulator/beam/erl_db_catree.c +++ b/erts/emulator/beam/erl_db_catree.c @@ -96,6 +96,12 @@ #include "erl_db_tree.h" #include "erl_db_tree_util.h" +#ifdef DEBUG +# define IF_DEBUG(X) X +#else +# define IF_DEBUG(X) +#endif + /* ** Forward declarations */ @@ -158,6 +164,14 @@ db_lookup_dbterm_catree(Process *, DbTable *, Eterm key, Eterm obj, DbUpdateHandle*); static void db_finalize_dbterm_catree(int cret, DbUpdateHandle *); +static void split_catree(DbTableCATree *tb, + DbTableCATreeNode* ERTS_RESTRICT base, + DbTableCATreeNode* ERTS_RESTRICT parent); +static void join_catree(DbTableCATree *tb, + DbTableCATreeNode *thiz, + DbTableCATreeNode *parent); + + /* ** External interface */ @@ -210,62 +224,37 @@ DbTableMethod db_catree = * Internal CA tree related helper functions and macros */ -#define GET_ROUTE_NODE_KEY(node) (node->baseOrRoute.route.key.tpl[0]) -#define GET_BASE_NODE_LOCK(node) (&(node->baseOrRoute.base.lock)) -#define GET_ROUTE_NODE_LOCK(node) (&(node->baseOrRoute.route.lock)) +#define GET_ROUTE_NODE_KEY(node) (node->u.route.key.term) +#define GET_BASE_NODE_LOCK(node) (&(node->u.base.lock)) +#define GET_ROUTE_NODE_LOCK(node) (&(node->u.route.lock)) /* Helpers for reading and writing shared atomic variables */ /* No memory barrier */ #define GET_ROOT(tb) ((DbTableCATreeNode*)erts_atomic_read_nob(&(tb->root))) -#define GET_LEFT(ca_tree_route_node) ((DbTableCATreeNode*)erts_atomic_read_nob(&(ca_tree_route_node->baseOrRoute.route.left))) -#define GET_RIGHT(ca_tree_route_node) ((DbTableCATreeNode*)erts_atomic_read_nob(&(ca_tree_route_node->baseOrRoute.route.right))) +#define GET_LEFT(ca_tree_route_node) ((DbTableCATreeNode*)erts_atomic_read_nob(&(ca_tree_route_node->u.route.left))) +#define GET_RIGHT(ca_tree_route_node) ((DbTableCATreeNode*)erts_atomic_read_nob(&(ca_tree_route_node->u.route.right))) #define SET_ROOT(tb, v) erts_atomic_set_nob(&((tb)->root), (erts_aint_t)(v)) -#define SET_LEFT(ca_tree_route_node, v) erts_atomic_set_nob(&(ca_tree_route_node->baseOrRoute.route.left), (erts_aint_t)(v)); -#define SET_RIGHT(ca_tree_route_node, v) erts_atomic_set_nob(&(ca_tree_route_node->baseOrRoute.route.right), (erts_aint_t)(v)); +#define SET_LEFT(ca_tree_route_node, v) erts_atomic_set_nob(&(ca_tree_route_node->u.route.left), (erts_aint_t)(v)); +#define SET_RIGHT(ca_tree_route_node, v) erts_atomic_set_nob(&(ca_tree_route_node->u.route.right), (erts_aint_t)(v)); /* Release or acquire barriers */ #define GET_ROOT_ACQB(tb) ((DbTableCATreeNode*)erts_atomic_read_acqb(&(tb->root))) -#define GET_LEFT_ACQB(ca_tree_route_node) ((DbTableCATreeNode*)erts_atomic_read_acqb(&(ca_tree_route_node->baseOrRoute.route.left))) -#define GET_RIGHT_ACQB(ca_tree_route_node) ((DbTableCATreeNode*)erts_atomic_read_acqb(&(ca_tree_route_node->baseOrRoute.route.right))) +#define GET_LEFT_ACQB(ca_tree_route_node) ((DbTableCATreeNode*)erts_atomic_read_acqb(&(ca_tree_route_node->u.route.left))) +#define GET_RIGHT_ACQB(ca_tree_route_node) ((DbTableCATreeNode*)erts_atomic_read_acqb(&(ca_tree_route_node->u.route.right))) #define SET_ROOT_RELB(tb, v) erts_atomic_set_relb(&((tb)->root), (erts_aint_t)(v)) -#define SET_LEFT_RELB(ca_tree_route_node, v) erts_atomic_set_relb(&(ca_tree_route_node->baseOrRoute.route.left), (erts_aint_t)(v)); -#define SET_RIGHT_RELB(ca_tree_route_node, v) erts_atomic_set_relb(&(ca_tree_route_node->baseOrRoute.route.right), (erts_aint_t)(v)); +#define SET_LEFT_RELB(ca_tree_route_node, v) erts_atomic_set_relb(&(ca_tree_route_node->u.route.left), (erts_aint_t)(v)); +#define SET_RIGHT_RELB(ca_tree_route_node, v) erts_atomic_set_relb(&(ca_tree_route_node->u.route.right), (erts_aint_t)(v)); /* Compares a key to the key in a route node */ -static ERTS_INLINE Sint cmp_key_route(DbTableCommon * tb, - Eterm key, +static ERTS_INLINE Sint cmp_key_route(Eterm key, DbTableCATreeNode *obj) { return CMP(key, GET_ROUTE_NODE_KEY(obj)); } -static ERTS_INLINE void push_node_dyn_array(DbTable *tb, - CATreeNodeStack *stack, - DbTableCATreeNode *node) -{ - int i; - if (stack->pos == stack->size) { - DbTableCATreeNode **newArray = - erts_db_alloc(ERTS_ALC_T_DB_STK, tb, - sizeof(DbTableCATreeNode*) * (stack->size*2)); - for (i = 0; i < stack->pos; i++) { - newArray[i] = stack->array[i]; - } - if (stack->size > STACK_NEED) { - /* Dynamically allocated array that needs to be deallocated */ - erts_db_free(ERTS_ALC_T_DB_STK, tb, - stack->array, - sizeof(DbTableCATreeNode *) * stack->size); - } - stack->array = newArray; - stack->size = stack->size*2; - } - PUSH_NODE(stack, node); -} - /* * Used by the split_tree function */ @@ -279,7 +268,7 @@ int less_than_two_elements(TreeDbTerm *root) * Inserts a TreeDbTerm into a tree. Returns the new root. */ static ERTS_INLINE -TreeDbTerm* insert_TreeDbTerm(DbTableCommon *common_table_data, +TreeDbTerm* insert_TreeDbTerm(DbTableCATree *tb, TreeDbTerm *insert_to_root, TreeDbTerm *value_to_insert) { /* Non recursive insertion in AVL tree, building our own stack */ @@ -295,7 +284,7 @@ TreeDbTerm* insert_TreeDbTerm(DbTableCommon *common_table_data, int dir; TreeDbTerm *p1, *p2, *p; - key = GETKEY(common_table_data, value_to_insert->dbterm.tpl); + key = GETKEY(tb, value_to_insert->dbterm.tpl); dstack[dpos++] = DIR_END; for (;;) @@ -305,7 +294,7 @@ TreeDbTerm* insert_TreeDbTerm(DbTableCommon *common_table_data, (*this)->balance = 0; (*this)->left = (*this)->right = NULL; break; - } else if ((c = cmp_key(common_table_data, key, *this)) < 0) { + } else if ((c = cmp_key(&tb->common, key, *this)) < 0) { /* go lefts */ dstack[dpos++] = DIR_LEFT; tstack[tpos++] = this; @@ -392,7 +381,7 @@ TreeDbTerm* insert_TreeDbTerm(DbTableCommon *common_table_data, * left_wb and the tree containing the rest of the keys in the write * back parameter right_wb. */ -static void split_tree(DbTableCommon *tb, +static void split_tree(DbTableCATree *tb, TreeDbTerm *root, TreeDbTerm **split_key_node_wb, TreeDbTerm **left_wb, @@ -413,9 +402,7 @@ static void split_tree(DbTableCommon *tb, split_node->left = NULL; right_root = split_node->right; split_node->right = NULL; - right_root = insert_TreeDbTerm(tb, - right_root, - split_node); + right_root = insert_TreeDbTerm(tb, right_root, split_node); *split_key_node_wb = split_node; *left_wb = left_root; *right_wb = right_root; @@ -581,14 +568,14 @@ static TreeDbTerm* join_trees(TreeDbTerm *left_root_param, p = *this; if (dir == DIR_LEFT) { switch (p->balance) { - case 1: + case 1: p->balance = 0; state = 0; break; - case 0: + case 0: p->balance = -1; break; - case -1: /* The icky case */ + case -1: /* The icky case */ p1 = p->left; if (p1->balance == -1) { /* Single LL rotation */ p->left = p1->right; @@ -611,14 +598,14 @@ static TreeDbTerm* join_trees(TreeDbTerm *left_root_param, } } else { /* dir == DIR_RIGHT */ switch (p->balance) { - case -1: + case -1: p->balance = 0; state = 0; break; - case 0: + case 0: p->balance = 1; break; - case 1: + case 1: p1 = p->right; if (p1->balance == 1) { /* Single RR rotation */ p->right = p1->left; @@ -649,6 +636,31 @@ static TreeDbTerm* join_trees(TreeDbTerm *left_root_param, } } +#ifdef DEBUG +# define FORCE_RANDOM_SPLIT_JOIN +#endif +#ifdef FORCE_RANDOM_SPLIT_JOIN +static int dbg_fastrand(void) +{ + static int g_seed = 648835; + g_seed = (214013*g_seed+2531011); + return (g_seed>>16)&0x7FFF; +} + +static void dbg_maybe_force_splitjoin(DbTableCATreeNode* base_node) +{ + switch (dbg_fastrand() % 8) { + case 1: + base_node->u.base.lock_statistics = 1+ERL_DB_CATREE_HIGH_CONTENTION_LIMIT; + break; + case 2: + base_node->u.base.lock_statistics = -1+ERL_DB_CATREE_LOW_CONTENTION_LIMIT; + break; + } +} +#else +# define dbg_maybe_force_splitjoin(N) +#endif /* FORCE_RANDOM_SPLIT_JOIN */ static ERTS_INLINE int try_wlock_base_node(DbTableCATreeBaseNode *base_node) @@ -660,9 +672,10 @@ int try_wlock_base_node(DbTableCATreeBaseNode *base_node) * Locks a base node without adjusting the lock statistics */ static ERTS_INLINE -void wlock_base_node_no_stats(DbTableCATreeBaseNode *base_node) +void wlock_base_node_no_stats(DbTableCATreeNode *base_node) { - erts_rwmtx_rwlock(&base_node->lock); + ASSERT(base_node->is_base_node); + erts_rwmtx_rwlock(&base_node->u.base.lock); } /* @@ -670,255 +683,345 @@ void wlock_base_node_no_stats(DbTableCATreeBaseNode *base_node) * the lock was contended or not */ static ERTS_INLINE -void wlock_base_node(DbTableCATreeBaseNode *base_node) +void wlock_base_node(DbTableCATreeNode *base_node) { - if (try_wlock_base_node(base_node)) { + ASSERT(base_node->is_base_node); + if (try_wlock_base_node(&base_node->u.base)) { /* The lock is contended */ wlock_base_node_no_stats(base_node); - base_node->lock_statistics += ERL_DB_CATREE_LOCK_FAILURE_CONTRIBUTION; + base_node->u.base.lock_statistics += ERL_DB_CATREE_LOCK_FAILURE_CONTRIBUTION; } else { - base_node->lock_statistics += ERL_DB_CATREE_LOCK_SUCCESS_CONTRIBUTION; + base_node->u.base.lock_statistics += ERL_DB_CATREE_LOCK_SUCCESS_CONTRIBUTION; } } static ERTS_INLINE -void wunlock_base_node(DbTableCATreeBaseNode *base_node) +void wunlock_base_node(DbTableCATreeNode *base_node) { - erts_rwmtx_rwunlock(&base_node->lock); + erts_rwmtx_rwunlock(&base_node->u.base.lock); } static ERTS_INLINE -void rlock_base_node(DbTableCATreeBaseNode *base_node) +void wunlock_adapt_base_node(DbTableCATree* tb, + DbTableCATreeNode* node, + DbTableCATreeNode* parent, + int current_level) +{ + dbg_maybe_force_splitjoin(node); + if ((!node->u.base.root && parent && !(tb->common.status + & DB_CATREE_FORCE_SPLIT)) + || node->u.base.lock_statistics < ERL_DB_CATREE_LOW_CONTENTION_LIMIT) { + join_catree(tb, node, parent); + } + else if (node->u.base.lock_statistics > ERL_DB_CATREE_HIGH_CONTENTION_LIMIT + && current_level < ERL_DB_CATREE_MAX_ROUTE_NODE_LAYER_HEIGHT) { + split_catree(tb, node, parent); + } + else { + wunlock_base_node(node); + } +} + +static ERTS_INLINE +void rlock_base_node(DbTableCATreeNode *base_node) { - erts_rwmtx_rlock(&base_node->lock); + ASSERT(base_node->is_base_node); + erts_rwmtx_rlock(&base_node->u.base.lock); } static ERTS_INLINE -void runlock_base_node(DbTableCATreeBaseNode *base_node) +void runlock_base_node(DbTableCATreeNode *base_node) { - erts_rwmtx_runlock(&base_node->lock); + ASSERT(base_node->is_base_node); + erts_rwmtx_runlock(&base_node->u.base.lock); } static ERTS_INLINE -void lock_route_node(DbTableCATreeRouteNode *route_node) +void lock_route_node(DbTableCATreeNode *route_node) { - erts_mtx_lock(&route_node->lock); + ASSERT(!route_node->is_base_node); + erts_mtx_lock(&route_node->u.route.lock); } static ERTS_INLINE -void unlock_route_node(DbTableCATreeRouteNode *route_node) +void unlock_route_node(DbTableCATreeNode *route_node) { - erts_mtx_unlock(&route_node->lock); + ASSERT(!route_node->is_base_node); + erts_mtx_unlock(&route_node->u.route.lock); } +static ERTS_INLINE +void init_root_iterator(DbTableCATree* tb, CATreeRootIterator* iter, + int read_only) +{ + iter->tb = tb; + iter->read_only = read_only; + iter->locked_bnode = NULL; + iter->next_route_key = THE_NON_VALUE; + iter->search_key = NULL; +} -/* - * The following macros are used to create the ETS functions that only - * need to access one element (e.g. db_put_catree, db_get_catree and - * db_erase_catree). - */ +static ERTS_INLINE +void lock_iter_base_node(CATreeRootIterator* iter, + DbTableCATreeNode *base_node, + DbTableCATreeNode *parent, + int current_level) +{ + ASSERT(!iter->locked_bnode); + if (iter->read_only) + rlock_base_node(base_node); + else { + wlock_base_node(base_node); + iter->bnode_parent = parent; + iter->bnode_level = current_level; + } + iter->locked_bnode = base_node; +} -#define ERL_DB_CATREE_CREATE_DO_OPERATION_FUNCTION_FIND_BASE_NODE_PART(LOCK,UNLOCK) \ - int retry; \ - DbTableCATreeNode *current_node; \ - DbTableCATreeNode *prev_node; \ - DbTableCommon* common_table_data = &tb->common; \ - DbTableCATreeBaseNode *base_node; \ - int current_level; \ - (void)prev_node; \ - do { \ - retry = 0; \ - current_node = GET_ROOT_ACQB(tb); \ - prev_node = NULL; \ - current_level = 0; \ - while ( ! current_node->is_base_node ) { \ - current_level = current_level + 1; \ - prev_node = current_node; \ - if (cmp_key_route(common_table_data,key,current_node) < 0) { \ - current_node = GET_LEFT_ACQB(current_node); \ - } else { \ - current_node = GET_RIGHT_ACQB(current_node); \ - } \ - } \ - base_node = ¤t_node->baseOrRoute.base; \ - LOCK(base_node); \ - if ( ! base_node->is_valid ) { \ - /* Retry */ \ - UNLOCK(base_node); \ - retry = 1; \ - } \ - } while(retry); - - -#define ERL_DB_CATREE_CREATE_DO_OPERATION_FUNCTION_ADAPT_AND_UNLOCK_PART \ - if (base_node->lock_statistics > ERL_DB_CATREE_HIGH_CONTENTION_LIMIT \ - && current_level < ERL_DB_CATREE_MAX_ROUTE_NODE_LAYER_HEIGHT) { \ - split_catree(&tb->common, prev_node, current_node); \ - } else if (base_node->lock_statistics < ERL_DB_CATREE_LOW_CONTENTION_LIMIT) { \ - join_catree(tb, prev_node, current_node); \ - } else { \ - wunlock_base_node(base_node); \ +static ERTS_INLINE +void unlock_iter_base_node(CATreeRootIterator* iter) +{ + ASSERT(iter->locked_bnode); + if (iter->read_only) + runlock_base_node(iter->locked_bnode); + else if (iter->locked_bnode->u.base.is_valid) { + wunlock_adapt_base_node(iter->tb, iter->locked_bnode, + iter->bnode_parent, iter->bnode_level); } + else + wunlock_base_node(iter->locked_bnode); + iter->locked_bnode = NULL; +} + +static ERTS_INLINE +void destroy_root_iterator(CATreeRootIterator* iter) +{ + if (iter->locked_bnode) + unlock_iter_base_node(iter); + if (iter->search_key) + erts_free(ERTS_ALC_T_DB_TMP, iter->search_key); +} + + +typedef struct +{ + DbTableCATreeNode *parent; + int current_level; +} FindBaseNode; -#define ERL_DB_CATREE_CREATE_DO_OPERATION_FUNCTION(FUN_POSTFIX,PARAMETERS,SEQUENTAIL_CALL) \ - static int erl_db_catree_do_operation_##FUN_POSTFIX(DbTableCATree *tb, \ - Eterm key, \ - PARAMETERS){ \ - int result; \ - ERL_DB_CATREE_CREATE_DO_OPERATION_FUNCTION_FIND_BASE_NODE_PART(wlock_base_node,wunlock_base_node) \ - result = SEQUENTAIL_CALL; \ - ERL_DB_CATREE_CREATE_DO_OPERATION_FUNCTION_ADAPT_AND_UNLOCK_PART \ - return result; \ +static ERTS_INLINE +DbTableCATreeNode* find_base_node(DbTableCATree* tb, Eterm key, + FindBaseNode* fbn) +{ + DbTableCATreeNode* ERTS_RESTRICT node = GET_ROOT_ACQB(tb); + if (fbn) { + fbn->parent = NULL; + fbn->current_level = 0; } + while (!node->is_base_node) { + if (fbn) { + fbn->current_level++; + fbn->parent = node; + } + if (cmp_key_route(key, node) < 0) { + node = GET_LEFT_ACQB(node); + } else { + node = GET_RIGHT_ACQB(node); + } + } + return node; +} +static ERTS_INLINE +DbTableCATreeNode* find_rlock_valid_base_node(DbTableCATree* tb, Eterm key) +{ + DbTableCATreeNode* base_node; -#define ERL_DB_CATREE_CREATE_DO_READ_OPERATION_FUNCTION(FUN_POSTFIX,PARAMETERS,SEQUENTAIL_CALL) \ - static int erl_db_catree_do_operation_##FUN_POSTFIX(DbTableCATree *tb, \ - Eterm key, \ - PARAMETERS){ \ - int result; \ - ERL_DB_CATREE_CREATE_DO_OPERATION_FUNCTION_FIND_BASE_NODE_PART(rlock_base_node,runlock_base_node) \ - result = SEQUENTAIL_CALL; \ - runlock_base_node(base_node); \ - return result; \ + while (1) { + base_node = find_base_node(tb, key, NULL); + rlock_base_node(base_node); + if (base_node->u.base.is_valid) + break; + runlock_base_node(base_node); } + return base_node; +} +static ERTS_INLINE +DbTableCATreeNode* find_wlock_valid_base_node(DbTableCATree* tb, Eterm key, + FindBaseNode* fbn) +{ + DbTableCATreeNode* base_node; + + while (1) { + base_node = find_base_node(tb, key, fbn); + wlock_base_node(base_node); + if (base_node->u.base.is_valid) + break; + wunlock_base_node(base_node); + } + return base_node; +} +static ERTS_INLINE +Eterm copy_route_key(DbRouteKey* dst, Eterm key, Uint key_size) +{ + dst->size = key_size; + if (key_size != 0) { + Eterm* hp = &dst->heap[0]; + ErlOffHeap tmp_offheap; + tmp_offheap.first = NULL; + dst->term = copy_struct(key, key_size, &hp, &tmp_offheap); + dst->oh = tmp_offheap.first; + } + else { + ASSERT(is_immed(key)); + dst->term = key; + dst->oh = NULL; + } + return dst->term; +} -static DbTableCATreeNode *create_catree_base_node(DbTableCATree *tb) +static ERTS_INLINE +void destroy_route_key(DbRouteKey* key) { - DbTableCATreeNode *new_base_node_container = - erts_db_alloc(ERTS_ALC_T_DB_TABLE, - (DbTable *) tb, - sizeof(DbTableCATreeNode)); - DbTableCATreeBaseNode *new_base_node = - &new_base_node_container->baseOrRoute.base; + if (key->oh) { + ErlOffHeap oh; + oh.first = key->oh; + erts_cleanup_offheap(&oh); + } +} + + +#ifdef ERTS_ENABLE_LOCK_CHECK +# define LC_ORDER(ORDER) ORDER +#else +# define LC_ORDER(ORDER) NIL +#endif + +#define sizeof_base_node() \ + offsetof(DbTableCATreeNode, u.base.end_of_struct__) + +static DbTableCATreeNode *create_base_node(DbTableCATree *tb, + TreeDbTerm* root) +{ + DbTableCATreeNode *p; erts_rwmtx_opt_t rwmtx_opt = ERTS_RWMTX_OPT_DEFAULT_INITER; - new_base_node_container->is_base_node = 1; - new_base_node->root = NULL; + p = erts_db_alloc(ERTS_ALC_T_DB_TABLE, (DbTable *) tb, + sizeof_base_node()); + + p->is_base_node = 1; + p->u.base.root = root; if (tb->common.type & DB_FREQ_READ) rwmtx_opt.type = ERTS_RWMTX_TYPE_FREQUENT_READ; if (erts_ets_rwmtx_spin_count >= 0) rwmtx_opt.main_spincount = erts_ets_rwmtx_spin_count; - erts_rwmtx_init_opt(&new_base_node->lock, &rwmtx_opt, - "erl_db_catree_base_node", tb->common.the_name, ERTS_LOCK_FLAGS_CATEGORY_DB); - new_base_node->lock_statistics = 0; - new_base_node->is_valid = 1; - new_base_node->tab = (DbTable *) tb; - return new_base_node_container; + + erts_rwmtx_init_opt(&p->u.base.lock, &rwmtx_opt, + "erl_db_catree_base_node", + NIL, + ERTS_LOCK_FLAGS_CATEGORY_DB); + p->u.base.lock_statistics = ((tb->common.status & DB_CATREE_FORCE_SPLIT) + ? INT_MAX : 0); + p->u.base.is_valid = 1; + return p; +} + +static ERTS_INLINE Uint sizeof_route_node(Uint key_size) +{ + return (offsetof(DbTableCATreeNode, u.route.key.heap) + + key_size*sizeof(Eterm)); } static DbTableCATreeNode* -create_catree_route_node(DbTableCommon * common_table_data, - DbTableCATreeNode *left, - DbTableCATreeNode *right, - DbTerm * keyTerm) +create_route_node(DbTableCATree *tb, + DbTableCATreeNode *left, + DbTableCATreeNode *right, + DbTerm * keyTerm, + DbTableCATreeNode* lc_parent) { - Eterm* top; - Eterm key = GETKEY(common_table_data,keyTerm->tpl); + Eterm key = GETKEY(tb,keyTerm->tpl); int key_size = size_object(key); - Uint offset = offsetof(DbTableCATreeNode,baseOrRoute) + - offsetof(DbTableCATreeRouteNode,key); - size_t route_node_container_size = - offset + - sizeof(DbTerm) + - sizeof(Eterm)*key_size; - ErlOffHeap tmp_offheap; - byte* new_route_node_container_bytes = - erts_db_alloc(ERTS_ALC_T_DB_TABLE, - (DbTable *) common_table_data, - route_node_container_size); - DbTerm* newp = (DbTerm*) (new_route_node_container_bytes + offset); - DbTableCATreeNode *new_route_node_container = - (DbTableCATreeNode*)new_route_node_container_bytes; - DbTableCATreeRouteNode *new_route_node = - &new_route_node_container->baseOrRoute.route; - new_route_node->tab = (DbTable *)common_table_data; - if (key_size != 0) { - newp->size = key_size; - top = &newp->tpl[1]; - tmp_offheap.first = NULL; - newp->tpl[0] = copy_struct(key, key_size, &top, &tmp_offheap); - newp->first_oh = tmp_offheap.first; - } else { - newp->size = key_size; - newp->first_oh = NULL; - newp->tpl[0] = key; - } - new_route_node_container->is_base_node = 0; - new_route_node->is_valid = 1; - erts_atomic_init_nob(&(new_route_node->left), (erts_aint_t)left); - erts_atomic_init_nob(&(new_route_node->right), (erts_aint_t)right); - erts_mtx_init(&new_route_node->lock, "erl_db_catree_route_node", - NIL, ERTS_LOCK_FLAGS_CATEGORY_DB); - return new_route_node_container; + DbTableCATreeNode* p = erts_db_alloc(ERTS_ALC_T_DB_TABLE, + (DbTable *) tb, + sizeof_route_node(key_size)); + + copy_route_key(&p->u.route.key, key, key_size); + p->is_base_node = 0; + p->u.route.is_valid = 1; + erts_atomic_init_nob(&p->u.route.left, (erts_aint_t)left); + erts_atomic_init_nob(&p->u.route.right, (erts_aint_t)right); +#ifdef ERTS_ENABLE_LOCK_CHECK + /* Route node lock order is inverse tree depth (from leafs toward root) */ + p->u.route.lc_order = (lc_parent == NULL ? MAX_SMALL : + lc_parent->u.route.lc_order - 1); + /* + * This assert may eventually fail as we don't increase 'lc_order' in join + * operations when route nodes move up in the tree. + * Tough luck if you run a lock-checking VM for such a long time on 32-bit. + */ + ERTS_LC_ASSERT(p->u.route.lc_order >= 0); +#endif + erts_mtx_init(&p->u.route.lock, "erl_db_catree_route_node", + LC_ORDER(make_small(p->u.route.lc_order)), + ERTS_LOCK_FLAGS_CATEGORY_DB); + return p; } -static void free_catree_base_node(void* base_node_container_ptr) +static void do_free_base_node(void* vptr) { - DbTableCATreeNode *base_node_container = - (DbTableCATreeNode *)base_node_container_ptr; - DbTableCATreeBaseNode *base_node = - &base_node_container->baseOrRoute.base; - erts_rwmtx_destroy(&base_node->lock); - erts_db_free(ERTS_ALC_T_DB_TABLE, - base_node->tab, base_node_container, - sizeof(DbTableCATreeNode)); -} - -static void free_catree_routing_node(void *route_node_container_ptr) -{ - size_t route_node_container_size; - byte* route_node_container_bytes = route_node_container_ptr; - DbTableCATreeNode *route_node_container = - (DbTableCATreeNode *)route_node_container_bytes; - DbTableCATreeRouteNode *route_node = - &route_node_container->baseOrRoute.route; - int key_size = route_node->key.size; - Uint offset = offsetof(DbTableCATreeNode,baseOrRoute) + - offsetof(DbTableCATreeRouteNode,key); - ErlOffHeap tmp_oh; - DbTerm* db_term = (DbTerm*) (route_node_container_bytes + offset); - erts_mtx_destroy(&route_node->lock); - route_node_container_size = - offset + - sizeof(DbTerm) + - sizeof(Eterm)*key_size; - if (key_size != 0) { - tmp_oh.first = db_term->first_oh; - erts_cleanup_offheap(&tmp_oh); - } - erts_db_free(ERTS_ALC_T_DB_TABLE, - route_node->tab, - route_node_container, - route_node_container_size); + DbTableCATreeNode *p = (DbTableCATreeNode *)vptr; + ASSERT(p->is_base_node); + erts_rwmtx_destroy(&p->u.base.lock); + erts_free(ERTS_ALC_T_DB_TABLE, p); +} + +static void free_catree_base_node(DbTableCATree* tb, DbTableCATreeNode* p) +{ + ASSERT(p->is_base_node); + ERTS_DB_ALC_MEM_UPDATE_(tb, sizeof_base_node(), 0); + do_free_base_node(p); +} + +static void do_free_route_node(void *vptr) +{ + DbTableCATreeNode *p = (DbTableCATreeNode *)vptr; + ASSERT(!p->is_base_node); + erts_mtx_destroy(&p->u.route.lock); + destroy_route_key(&p->u.route.key); + erts_free(ERTS_ALC_T_DB_TABLE, p); +} + +static void free_catree_route_node(DbTableCATree* tb, DbTableCATreeNode* p) +{ + ASSERT(!p->is_base_node); + ERTS_DB_ALC_MEM_UPDATE_(tb, sizeof_route_node(p->u.route.key.size), 0); + do_free_route_node(p); } + /* * Returns the parent routing node of the specified - * route_node_container if such a routing node exists or NULL if - * route_node_container is attached to the root + * route node 'child' if such a parent exists + * or NULL if 'child' is attached to the root. */ static ERTS_INLINE DbTableCATreeNode * parent_of(DbTableCATree *tb, - DbTableCATreeNode *route_node_container) + DbTableCATreeNode *child) { + Eterm key = GET_ROUTE_NODE_KEY(child); + DbTableCATreeNode *current = GET_ROOT_ACQB(tb); + DbTableCATreeNode *prev = NULL; - Eterm key = GET_ROUTE_NODE_KEY(route_node_container); - DbTableCATreeNode *current_node = GET_ROOT_ACQB(tb); - DbTableCATreeNode *prev_node = NULL; - if (current_node == route_node_container) { - return NULL; - } - while (current_node != route_node_container) { - prev_node = current_node; - if (cmp_key_route((DbTableCommon *)tb, key, current_node) < 0) { - current_node = GET_LEFT_ACQB(current_node); + while (current != child) { + prev = current; + if (cmp_key_route(key, current) < 0) { + current = GET_LEFT_ACQB(current); } else { - current_node = GET_RIGHT_ACQB(current_node); + current = GET_RIGHT_ACQB(current); } } - return prev_node; + return prev; } @@ -953,11 +1056,7 @@ leftmost_route_node(DbTableCATreeNode *root) prev_node = node; node = GET_LEFT_ACQB(node); } - if (prev_node == NULL) { - return NULL; - } else { - return prev_node; - } + return prev_node; } static ERTS_INLINE DbTableCATreeNode* @@ -969,162 +1068,7 @@ rightmost_route_node(DbTableCATreeNode *root) prev_node = node; node = GET_RIGHT_ACQB(node); } - if (prev_node == NULL) { - return NULL; - } else { - return prev_node; - } -} - -static ERTS_INLINE DbTableCATreeNode* -leftmost_base_node_and_path(DbTableCATreeNode *root, CATreeNodeStack * stack) -{ - DbTableCATreeNode * node = root; - while (!node->is_base_node) { - PUSH_NODE(stack, node); - node = GET_LEFT_ACQB(node); - } - return node; -} - -static ERTS_INLINE DbTableCATreeNode* -get_next_base_node_and_path(DbTableCommon *common_table_data, - DbTableCATreeNode *base_node, - CATreeNodeStack *stack) -{ - if (EMPTY_NODE(stack)) { /* The parent of b is the root */ - return NULL; - } else { - if (GET_LEFT(TOP_NODE(stack)) == base_node) { - return leftmost_base_node_and_path( - GET_RIGHT_ACQB(TOP_NODE(stack)), - stack); - } else { - Eterm pkey = - TOP_NODE(stack)->baseOrRoute.route.key.tpl[0]; /* pKey = key of parent */ - POP_NODE(stack); - while (!EMPTY_NODE(stack)) { - if (TOP_NODE(stack)->baseOrRoute.route.is_valid && - cmp_key_route(common_table_data, pkey, TOP_NODE(stack)) <= 0) { - return leftmost_base_node_and_path(GET_RIGHT_ACQB(TOP_NODE(stack)), stack); - } else { - POP_NODE(stack); - } - } - } - return NULL; - } -} - -static ERTS_INLINE void -clone_stack(CATreeNodeStack *search_stack_ptr, - CATreeNodeStack *search_stack_copy_ptr) -{ - int i; - search_stack_copy_ptr->pos = search_stack_ptr->pos; - for (i = 0; i < search_stack_ptr->pos; i++) { - search_stack_copy_ptr->array[i] = search_stack_ptr->array[i]; - } -} - - - -static ERTS_INLINE DbTableCATreeNode* -lock_first_base_node(DbTable *tbl, - CATreeNodeStack *search_stack_ptr, - CATreeNodeStack *locked_base_nodes_stack_ptr) -{ - int retry; - DbTableCATreeNode *current_node; - DbTableCATreeBaseNode *base_node; - DbTableCATree* tb = &tbl->catree; - do { - retry = 0; - current_node = GET_ROOT_ACQB(tb); - while ( ! current_node->is_base_node ) { - PUSH_NODE(search_stack_ptr, current_node); - current_node = GET_LEFT_ACQB(current_node); - } - base_node = ¤t_node->baseOrRoute.base; - rlock_base_node(base_node); - if ( ! base_node->is_valid ) { - /* Retry */ - runlock_base_node(base_node); - retry = 1; - } - } while(retry); - push_node_dyn_array(tbl, locked_base_nodes_stack_ptr, current_node); - return current_node; -} - -static ERTS_INLINE DbTableCATreeBaseNode* -find_and_lock_next_base_node_and_path(DbTable *tbl, - CATreeNodeStack **search_stack_ptr_ptr, - CATreeNodeStack **search_stack_copy_ptr_ptr, - CATreeNodeStack *locked_base_nodes_stack_ptr) -{ - DbTableCATreeNode *current_node; - DbTableCATreeBaseNode *base_node; - CATreeNodeStack * tmp_stack_ptr; - DbTableCommon* common_table_data; - retry_find_and_lock_next_base_node: - current_node = TOP_NODE(locked_base_nodes_stack_ptr); - common_table_data = &tbl->common; - clone_stack(*search_stack_ptr_ptr, *search_stack_copy_ptr_ptr); - current_node = - get_next_base_node_and_path(common_table_data, current_node, *search_stack_ptr_ptr); - if (current_node == NULL) { - return NULL; - } - base_node = ¤t_node->baseOrRoute.base; - rlock_base_node(base_node); - if ( ! base_node->is_valid ) { - /* Retry */ - runlock_base_node(base_node); - /* Revert to previous state */ - current_node = TOP_NODE(locked_base_nodes_stack_ptr); - tmp_stack_ptr = *search_stack_ptr_ptr; - *search_stack_ptr_ptr = *search_stack_copy_ptr_ptr; - *search_stack_copy_ptr_ptr = tmp_stack_ptr; - goto retry_find_and_lock_next_base_node; - } else { - push_node_dyn_array(tbl, locked_base_nodes_stack_ptr, current_node); - } - return base_node; -} - -static ERTS_INLINE -void unlock_and_release_locked_base_node_stack(DbTable *tbl, - CATreeNodeStack *locked_base_nodes_stack_ptr) -{ - DbTableCATreeNode *current_node; - DbTableCATreeBaseNode *base_node; - int i; - for (i = 0; i < locked_base_nodes_stack_ptr->pos; i++) { - current_node = locked_base_nodes_stack_ptr->array[i]; - base_node = ¤t_node->baseOrRoute.base; - if (locked_base_nodes_stack_ptr->pos > 1) { - base_node->lock_statistics = /* This is not atomic which is fine as */ - base_node->lock_statistics + /* correctness does not depend on that. */ - ERL_DB_CATREE_LOCK_MORE_THAN_ONE_CONTRIBUTION; - } - runlock_base_node(base_node); - } - if (locked_base_nodes_stack_ptr->size > STACK_NEED) { - erts_db_free(ERTS_ALC_T_DB_STK, tbl, - locked_base_nodes_stack_ptr->array, - sizeof(DbTableCATreeNode *) * locked_base_nodes_stack_ptr->size); - } -} - -static ERTS_INLINE -void init_stack(CATreeNodeStack *stack, - DbTableCATreeNode **stack_array, - Uint init_size) -{ - stack->array = stack_array; - stack->pos = 0; - stack->size = init_size; + return prev_node; } static ERTS_INLINE @@ -1137,404 +1081,195 @@ void init_tree_stack(DbTreeStack *stack, stack->slot = init_slot; } -#define DEC_ROUTE_NODE_STACK_AND_ARRAY(STACK_NAME) \ - CATreeNodeStack STACK_NAME; \ - CATreeNodeStack * STACK_NAME##_ptr = &(STACK_NAME); \ - DbTableCATreeNode *STACK_NAME##_array[STACK_NEED]; - -#define DECLARE_AND_INIT_BASE_NODE_SEARCH_STACKS \ -DEC_ROUTE_NODE_STACK_AND_ARRAY(search_stack) \ -DEC_ROUTE_NODE_STACK_AND_ARRAY(search_stack_copy) \ -DEC_ROUTE_NODE_STACK_AND_ARRAY(locked_base_nodes_stack) \ -init_stack(&search_stack, search_stack_array, 0); \ -init_stack(&search_stack_copy, search_stack_copy_array, 0); \ -init_stack(&locked_base_nodes_stack, locked_base_nodes_stack_array, STACK_NEED);/* Abuse as stack array size*/ - -/* - * Locks and returns the base node that contains the specified key if - * it is present. The function saves the search path to the found base - * node in search_stack_ptr and adds the found base node to - * locked_base_nodes_stack_ptr. - */ -static ERTS_INLINE DbTableCATreeBaseNode * -lock_base_node_with_key(DbTable *tbl, - Eterm key, - CATreeNodeStack * search_stack_ptr, - CATreeNodeStack * locked_base_nodes_stack_ptr) -{ - int retry; - DbTableCATreeNode *current_node; - DbTableCATreeBaseNode *base_node; - DbTableCATree* tb = &tbl->catree; - DbTableCommon* common_table_data = &tbl->common; - do { - retry = 0; - current_node = GET_ROOT_ACQB(tb); - while ( ! current_node->is_base_node ) { - PUSH_NODE(search_stack_ptr, current_node); - if( cmp_key_route(common_table_data,key,current_node) < 0 ) { - current_node = GET_LEFT_ACQB(current_node); - } else { - current_node = GET_RIGHT_ACQB(current_node); - } - } - base_node = ¤t_node->baseOrRoute.base; - rlock_base_node(base_node); - if ( ! base_node->is_valid ) { - /* Retry */ - runlock_base_node(base_node); - retry = 1; - } - } while(retry); - push_node_dyn_array(tbl, locked_base_nodes_stack_ptr, current_node); - return base_node; -} - -/* - * Joins a base node with it's right neighbor. Returns the base node - * resulting from the join in locked state or NULL if there is no base - * node to join with. - */ -static DbTableCATreeNode* -erl_db_catree_force_join_right(DbTableCommon *common_table_data, - DbTableCATreeNode *parent_container, - DbTableCATreeNode *base_container, - DbTableCATreeNode **result_parent_wb) -{ - DbTableCATreeRouteNode *parent; - DbTableCATreeNode *gparent_container; - DbTableCATreeRouteNode *gparent; - DbTableCATreeBaseNode *base = &base_container->baseOrRoute.base; - DbTableCATree *tb = (DbTableCATree *)common_table_data; - DbTableCATreeNode *neighbor_base_container; - DbTableCATreeBaseNode *neighbor_base; - DbTableCATreeNode *new_neighbor_base; - DbTableCATreeNode *neighbor_base_parent; - int neighbour_not_valid; - if (parent_container == NULL) { - return NULL; - } - parent = &parent_container->baseOrRoute.route; - do { - neighbor_base_container = leftmost_base_node(GET_RIGHT_ACQB(parent_container)); - neighbor_base = &neighbor_base_container->baseOrRoute.base; - wlock_base_node_no_stats(neighbor_base); - neighbour_not_valid = !neighbor_base->is_valid; - if (neighbour_not_valid) { - wunlock_base_node(neighbor_base); - } - } while (neighbour_not_valid); - lock_route_node(parent); - parent->is_valid = 0; - neighbor_base->is_valid = 0; - base->is_valid = 0; - gparent = NULL; - gparent_container = NULL; - do { - if (gparent != NULL) { - unlock_route_node(gparent); - } - gparent_container = parent_of(tb, parent_container); - if (gparent_container != NULL) { - gparent = &gparent_container->baseOrRoute.route; - lock_route_node(gparent); - } else { - gparent = NULL; - } - } while (gparent != NULL && !gparent->is_valid); - if (gparent == NULL) { - SET_ROOT_RELB(tb, GET_RIGHT(parent_container)); - } else if (GET_LEFT(gparent_container) == parent_container) { - SET_LEFT_RELB(gparent_container, GET_RIGHT(parent_container)); - } else { - SET_RIGHT_RELB(gparent_container, GET_RIGHT(parent_container)); - } - unlock_route_node(parent); - if (gparent != NULL) { - unlock_route_node(gparent); - } - new_neighbor_base = create_catree_base_node(tb); - new_neighbor_base->baseOrRoute.base.root = - join_trees(base->root, neighbor_base->root); - wlock_base_node_no_stats(&(new_neighbor_base->baseOrRoute.base)); - neighbor_base_parent = NULL; - if (GET_RIGHT(parent_container) == neighbor_base_container) { - neighbor_base_parent = gparent_container; - } else { - neighbor_base_parent = - leftmost_route_node(GET_RIGHT(parent_container)); - } - if(neighbor_base_parent == NULL) { - SET_ROOT_RELB(tb, new_neighbor_base); - } else if (GET_LEFT(neighbor_base_parent) == neighbor_base_container) { - SET_LEFT_RELB(neighbor_base_parent, new_neighbor_base); - } else { - SET_RIGHT_RELB(neighbor_base_parent, new_neighbor_base); - } - wunlock_base_node(base); - wunlock_base_node(neighbor_base); - /* Free the parent and base */ - erts_schedule_thr_prgr_later_op(free_catree_routing_node, - parent_container, - &parent->free_item); - erts_schedule_thr_prgr_later_op(free_catree_base_node, - base_container, - &base->free_item); - erts_schedule_thr_prgr_later_op(free_catree_base_node, - neighbor_base_container, - &neighbor_base->free_item); - - if (parent_container == neighbor_base_container) { - *result_parent_wb = gparent_container; - } else { - *result_parent_wb = neighbor_base_parent; - } - return new_neighbor_base; -} - -/* - * Used to merge together all base nodes for operations such as last - * and select_*. Returns the base node resulting from the merge in - * locked state. - */ -static DbTableCATreeNode * -merge_to_one_locked_base_node(DbTableCommon * common_table_data) -{ - DbTableCATreeNode *parent_container; - DbTableCATreeNode *new_parent_container; - DbTableCATree *tb = (DbTableCATree *)common_table_data; - DbTableCATreeNode *base_container; - DbTableCATreeNode *new_base_container; - int is_not_valid; - /* Find first base node */ - do { - parent_container = NULL; - base_container = GET_ROOT_ACQB(tb); - while ( ! base_container->is_base_node ) { - parent_container = base_container; - base_container = GET_LEFT_ACQB(base_container); - } - wlock_base_node_no_stats(&(base_container->baseOrRoute.base)); - is_not_valid = ! base_container->baseOrRoute.base.is_valid; - if (is_not_valid) { - wunlock_base_node(&(base_container->baseOrRoute.base)); - } - } while(is_not_valid); - do { - new_base_container = erl_db_catree_force_join_right(common_table_data, - parent_container, - base_container, - &new_parent_container); - if (new_base_container != NULL) { - base_container = new_base_container; - parent_container = new_parent_container; - } - } while(new_base_container != NULL); - return base_container; -} - - static void join_catree(DbTableCATree *tb, - DbTableCATreeNode *parent_container, - DbTableCATreeNode *base_container) -{ - DbTableCATreeRouteNode *parent; - DbTableCATreeNode *gparent_container; - DbTableCATreeRouteNode *gparent; - DbTableCATreeBaseNode *base = &base_container->baseOrRoute.base; - DbTableCATreeNode *neighbor_base_container; - DbTableCATreeBaseNode *neighbor_base; - DbTableCATreeNode *new_neighbor_base; - DbTableCATreeNode *neighbor_base_parent; - if (parent_container == NULL) { - base->lock_statistics = 0; - wunlock_base_node(base); + DbTableCATreeNode *thiz, + DbTableCATreeNode *parent) +{ + DbTableCATreeNode *gparent; + DbTableCATreeNode *neighbor; + DbTableCATreeNode *new_neighbor; + DbTableCATreeNode *neighbor_parent; + + ASSERT(thiz->is_base_node); + if (parent == NULL) { + thiz->u.base.lock_statistics = 0; + wunlock_base_node(thiz); return; } - parent = &parent_container->baseOrRoute.route; - if (GET_LEFT(parent_container) == base_container) { - neighbor_base_container = leftmost_base_node(GET_RIGHT_ACQB(parent_container)); - neighbor_base = &neighbor_base_container->baseOrRoute.base; - if (try_wlock_base_node(neighbor_base)) { + ASSERT(!parent->is_base_node); + if (GET_LEFT(parent) == thiz) { + neighbor = leftmost_base_node(GET_RIGHT_ACQB(parent)); + if (try_wlock_base_node(&neighbor->u.base)) { /* Failed to acquire lock */ - base->lock_statistics = 0; - wunlock_base_node(base); + thiz->u.base.lock_statistics = 0; + wunlock_base_node(thiz); return; - } else if (!neighbor_base->is_valid) { - base->lock_statistics = 0; - wunlock_base_node(base); - wunlock_base_node(neighbor_base); + } else if (!neighbor->u.base.is_valid) { + thiz->u.base.lock_statistics = 0; + wunlock_base_node(thiz); + wunlock_base_node(neighbor); return; } else { lock_route_node(parent); - parent->is_valid = 0; - neighbor_base->is_valid = 0; - base->is_valid = 0; + parent->u.route.is_valid = 0; + neighbor->u.base.is_valid = 0; + thiz->u.base.is_valid = 0; gparent = NULL; - gparent_container = NULL; do { if (gparent != NULL) { unlock_route_node(gparent); } - gparent_container = parent_of(tb, parent_container); - if (gparent_container != NULL) { - gparent = &gparent_container->baseOrRoute.route; + gparent = parent_of(tb, parent); + if (gparent != NULL) lock_route_node(gparent); - } else { - gparent = NULL; - } - } while (gparent != NULL && !gparent->is_valid); + } while (gparent != NULL && !gparent->u.route.is_valid); + if (gparent == NULL) { - SET_ROOT_RELB(tb, GET_RIGHT(parent_container)); - } else if (GET_LEFT(gparent_container) == parent_container) { - SET_LEFT_RELB(gparent_container, GET_RIGHT(parent_container)); + SET_ROOT_RELB(tb, GET_RIGHT(parent)); + } else if (GET_LEFT(gparent) == parent) { + SET_LEFT_RELB(gparent, GET_RIGHT(parent)); } else { - SET_RIGHT_RELB(gparent_container, GET_RIGHT(parent_container)); + SET_RIGHT_RELB(gparent, GET_RIGHT(parent)); } unlock_route_node(parent); if (gparent != NULL) { unlock_route_node(gparent); } - new_neighbor_base = create_catree_base_node(tb); - new_neighbor_base->baseOrRoute.base.root = - join_trees(base->root, neighbor_base->root); - neighbor_base_parent = NULL; - if (GET_RIGHT(parent_container) == neighbor_base_container) { - neighbor_base_parent = gparent_container; + { + TreeDbTerm* new_root = join_trees(thiz->u.base.root, + neighbor->u.base.root); + new_neighbor = create_base_node(tb, new_root); + } + if (GET_RIGHT(parent) == neighbor) { + neighbor_parent = gparent; } else { - neighbor_base_parent = - leftmost_route_node(GET_RIGHT(parent_container)); + neighbor_parent = leftmost_route_node(GET_RIGHT(parent)); } } } else { /* Symetric case */ - neighbor_base_container = rightmost_base_node(GET_LEFT_ACQB(parent_container)); - neighbor_base = &neighbor_base_container->baseOrRoute.base; - if (try_wlock_base_node(neighbor_base)) { + ASSERT(GET_RIGHT(parent) == thiz); + neighbor = rightmost_base_node(GET_LEFT_ACQB(parent)); + if (try_wlock_base_node(&neighbor->u.base)) { /* Failed to acquire lock */ - base->lock_statistics = 0; - wunlock_base_node(base); + thiz->u.base.lock_statistics = 0; + wunlock_base_node(thiz); return; - } else if (!neighbor_base->is_valid) { - base->lock_statistics = 0; - wunlock_base_node(base); - wunlock_base_node(neighbor_base); + } else if (!neighbor->u.base.is_valid) { + thiz->u.base.lock_statistics = 0; + wunlock_base_node(thiz); + wunlock_base_node(neighbor); return; } else { lock_route_node(parent); - parent->is_valid = 0; - neighbor_base->is_valid = 0; - base->is_valid = 0; + parent->u.route.is_valid = 0; + neighbor->u.base.is_valid = 0; + thiz->u.base.is_valid = 0; gparent = NULL; - gparent_container = NULL; do { if (gparent != NULL) { unlock_route_node(gparent); } - gparent_container = parent_of(tb, parent_container); - if (gparent_container != NULL) { - gparent = &gparent_container->baseOrRoute.route; + gparent = parent_of(tb, parent); + if (gparent != NULL) { lock_route_node(gparent); } else { gparent = NULL; } - } while (gparent != NULL && !gparent->is_valid); + } while (gparent != NULL && !gparent->u.route.is_valid); if (gparent == NULL) { - SET_ROOT_RELB(tb, GET_LEFT(parent_container)); - } else if (GET_RIGHT(gparent_container) == parent_container) { - SET_RIGHT_RELB(gparent_container, GET_LEFT(parent_container)); + SET_ROOT_RELB(tb, GET_LEFT(parent)); + } else if (GET_RIGHT(gparent) == parent) { + SET_RIGHT_RELB(gparent, GET_LEFT(parent)); } else { - SET_LEFT_RELB(gparent_container, GET_LEFT(parent_container)); + SET_LEFT_RELB(gparent, GET_LEFT(parent)); } unlock_route_node(parent); if (gparent != NULL) { unlock_route_node(gparent); } - new_neighbor_base = create_catree_base_node(tb); - new_neighbor_base->baseOrRoute.base.root = - join_trees(neighbor_base->root, base->root); - neighbor_base_parent = NULL; - if (GET_LEFT(parent_container) == neighbor_base_container) { - neighbor_base_parent = gparent_container; + { + TreeDbTerm* new_root = join_trees(neighbor->u.base.root, + thiz->u.base.root); + new_neighbor = create_base_node(tb, new_root); + } + if (GET_LEFT(parent) == neighbor) { + neighbor_parent = gparent; } else { - neighbor_base_parent = - rightmost_route_node(GET_LEFT(parent_container)); + neighbor_parent = + rightmost_route_node(GET_LEFT(parent)); } } } /* Link in new neighbor and free nodes that are no longer in the tree */ - if (neighbor_base_parent == NULL) { - SET_ROOT_RELB(tb, new_neighbor_base); - } else if (GET_LEFT(neighbor_base_parent) == neighbor_base_container) { - SET_LEFT_RELB(neighbor_base_parent, new_neighbor_base); + if (neighbor_parent == NULL) { + SET_ROOT_RELB(tb, new_neighbor); + } else if (GET_LEFT(neighbor_parent) == neighbor) { + SET_LEFT_RELB(neighbor_parent, new_neighbor); } else { - SET_RIGHT_RELB(neighbor_base_parent, new_neighbor_base); + SET_RIGHT_RELB(neighbor_parent, new_neighbor); } - wunlock_base_node(base); - wunlock_base_node(neighbor_base); + wunlock_base_node(thiz); + wunlock_base_node(neighbor); /* Free the parent and base */ - erts_schedule_thr_prgr_later_op(free_catree_routing_node, - parent_container, - &parent->free_item); - erts_schedule_thr_prgr_later_op(free_catree_base_node, - base_container, - &base->free_item); - erts_schedule_thr_prgr_later_op(free_catree_base_node, - neighbor_base_container, - &neighbor_base->free_item); -} - - -static void split_catree(DbTableCommon *tb, - DbTableCATreeNode *parent_container, - DbTableCATreeNode *base_container) { + erts_schedule_db_free(&tb->common, + do_free_route_node, + parent, + &parent->u.route.free_item, + sizeof_route_node(parent->u.route.key.size)); + erts_schedule_db_free(&tb->common, + do_free_base_node, + thiz, + &thiz->u.base.free_item, + sizeof_base_node()); + erts_schedule_db_free(&tb->common, + do_free_base_node, + neighbor, + &neighbor->u.base.free_item, + sizeof_base_node()); +} + +static void split_catree(DbTableCATree *tb, + DbTableCATreeNode* ERTS_RESTRICT base, + DbTableCATreeNode* ERTS_RESTRICT parent) +{ TreeDbTerm *splitOutWriteBack; - TreeDbTerm *leftWriteBack; - TreeDbTerm *rightWriteBack; - DbTableCATreeNode *left_base_node; - DbTableCATreeNode *right_base_node; - DbTableCATreeNode *routing_node_container; - DbTableCATreeBaseNode *base = &base_container->baseOrRoute.base; - DbTableCATreeRouteNode *parent; - if (parent_container == NULL) { - parent = NULL; - } else { - parent = &parent_container->baseOrRoute.route; - } + DbTableCATreeNode* ERTS_RESTRICT new_left; + DbTableCATreeNode* ERTS_RESTRICT new_right; + DbTableCATreeNode* ERTS_RESTRICT new_route; - if (less_than_two_elements(base->root)) { - base->lock_statistics = 0; + if (less_than_two_elements(base->u.base.root)) { + if (!(tb->common.status & DB_CATREE_FORCE_SPLIT)) + base->u.base.lock_statistics = 0; wunlock_base_node(base); return; } else { - /* Split the tree */ - split_tree(tb, - base->root, - &splitOutWriteBack, - &leftWriteBack, - &rightWriteBack); - /* Create new base nodes */ - left_base_node = - create_catree_base_node((DbTableCATree*)tb); - right_base_node = - create_catree_base_node((DbTableCATree*)tb); - left_base_node->baseOrRoute.base.root = leftWriteBack; - right_base_node->baseOrRoute.base.root = rightWriteBack; - routing_node_container = create_catree_route_node(tb, - left_base_node, - right_base_node, - &splitOutWriteBack->dbterm); + TreeDbTerm *left_tree; + TreeDbTerm *right_tree; + + split_tree(tb, base->u.base.root, &splitOutWriteBack, + &left_tree, &right_tree); + + new_left = create_base_node(tb, left_tree); + new_right = create_base_node(tb, right_tree); + new_route = create_route_node(tb, + new_left, + new_right, + &splitOutWriteBack->dbterm, + parent); if (parent == NULL) { - SET_ROOT_RELB((DbTableCATree*)tb, routing_node_container); - } else if(GET_LEFT(parent_container) == base_container) { - SET_LEFT_RELB(parent_container, routing_node_container); + SET_ROOT_RELB(tb, new_route); + } else if(GET_LEFT(parent) == base) { + SET_LEFT_RELB(parent, new_route); } else { - SET_RIGHT_RELB(parent_container, routing_node_container); + SET_RIGHT_RELB(parent, new_route); } - base->is_valid = 0; + base->u.base.is_valid = 0; wunlock_base_node(base); - erts_schedule_thr_prgr_later_op(free_catree_base_node, - base_container, - &base->free_item); + erts_schedule_db_free(&tb->common, + do_free_base_node, + base, + &base->u.base.free_item, + sizeof_base_node()); } } @@ -1546,7 +1281,7 @@ static void catree_add_base_node_to_free_list( DbTableCATree *tb, DbTableCATreeNode *base_node_container) { - base_node_container->baseOrRoute.base.next = + base_node_container->u.base.next = tb->base_nodes_to_free_list; tb->base_nodes_to_free_list = base_node_container; } @@ -1557,7 +1292,7 @@ static void catree_deque_base_node_from_free_list(DbTableCATree *tb) return; /* List empty */ } else { DbTableCATreeNode *first = tb->base_nodes_to_free_list; - tb->base_nodes_to_free_list = first->baseOrRoute.base.next; + tb->base_nodes_to_free_list = first->u.base.next; } } @@ -1596,7 +1331,7 @@ static SWord do_free_routing_nodes_catree_cont(DbTableCATree *tb, SWord num_left PUSH_NODE(&tb->free_stack_rnodes, root); root = p; } else { - free_catree_routing_node(root); + free_catree_route_node(tb, root); if (--num_left >= 0) { break; } else { @@ -1637,10 +1372,10 @@ static SWord do_free_base_node_cont(DbTableCATree *tb, SWord num_left) } } catree_deque_base_node_from_free_list(tb); - free_catree_base_node(base_node_container); + free_catree_base_node(tb, base_node_container); base_node_container = catree_first_base_node_from_free_list(tb); if (base_node_container != NULL) { - PUSH_NODE(&tb->free_stack_elems, base_node_container->baseOrRoute.base.root); + PUSH_NODE(&tb->free_stack_elems, base_node_container->u.base.root); } return num_left; } @@ -1662,7 +1397,9 @@ void db_initialize_catree(void) int db_create_catree(Process *p, DbTable *tbl) { DbTableCATree *tb = &tbl->catree; - DbTableCATreeNode *root = create_catree_base_node(tb); + DbTableCATreeNode *root; + + root = create_base_node(tb, NULL); tb->deletion = 0; tb->base_nodes_to_free_list = NULL; erts_atomic_init_relb(&(tb->root), (erts_aint_t)root); @@ -1671,177 +1408,447 @@ int db_create_catree(Process *p, DbTable *tbl) static int db_first_catree(Process *p, DbTable *tbl, Eterm *ret) { - DbTableCATreeBaseNode *base_node; + TreeDbTerm *root; + CATreeRootIterator iter; int result; - DECLARE_AND_INIT_BASE_NODE_SEARCH_STACKS; - /* Find first base node */ - base_node = &(lock_first_base_node(tbl, &search_stack, &locked_base_nodes_stack)->baseOrRoute.base); - /* Find next base node until non-empty base node is found */ - while (base_node != NULL && base_node->root == NULL) { - base_node = find_and_lock_next_base_node_and_path(tbl, &search_stack_ptr, &search_stack_copy_ptr, locked_base_nodes_stack_ptr); + + init_root_iterator(&tbl->catree, &iter, 1); + root = *catree_find_first_root(&iter); + if (!root) { + TreeDbTerm **pp = catree_find_next_root(&iter, NULL); + root = pp ? *pp : NULL; } - /* Get the return value */ - result = db_first_tree_common(p, tbl, (base_node == NULL ? NULL : base_node->root), ret, NULL); - /* Unlock base nodes */ - unlock_and_release_locked_base_node_stack(tbl, locked_base_nodes_stack_ptr); + + result = db_first_tree_common(p, tbl, root, ret, NULL); + + destroy_root_iterator(&iter); return result; } static int db_next_catree(Process *p, DbTable *tbl, Eterm key, Eterm *ret) { - DbTreeStack next_search_stack; - TreeDbTerm *next_search_stack_array[STACK_NEED]; - DbTableCATreeBaseNode *base_node; - int result = 0; - DECLARE_AND_INIT_BASE_NODE_SEARCH_STACKS; - init_tree_stack(&next_search_stack, next_search_stack_array, 0); - /* Lock base node with key */ - base_node = lock_base_node_with_key(tbl, key, &search_stack, &locked_base_nodes_stack); - /* Continue until key is not >= greatest key in base_node */ - while (base_node != NULL) { - result = db_next_tree_common(p, tbl, base_node->root, key, - ret, &next_search_stack); - if (result != DB_ERROR_NONE || *ret != am_EOT) { + DbTreeStack stack; + TreeDbTerm * stack_array[STACK_NEED]; + TreeDbTerm **rootp; + CATreeRootIterator iter; + int result; + + init_root_iterator(&tbl->catree, &iter, 1); + iter.next_route_key = key; + rootp = catree_find_next_root(&iter, NULL); + + do { + init_tree_stack(&stack, stack_array, 0); + result = db_next_tree_common(p, tbl, (rootp ? *rootp : NULL), key, ret, &stack); + if (result != DB_ERROR_NONE || *ret != am_EOT) break; - } - base_node = - find_and_lock_next_base_node_and_path(tbl, - &search_stack_ptr, - &search_stack_copy_ptr, - locked_base_nodes_stack_ptr); - } - unlock_and_release_locked_base_node_stack(tbl, &locked_base_nodes_stack); + + rootp = catree_find_next_root(&iter, NULL); + } while (rootp); + + destroy_root_iterator(&iter); return result; } static int db_last_catree(Process *p, DbTable *tbl, Eterm *ret) { - DbTableCATree *tb = &tbl->catree; - DbTableCATreeNode *base = merge_to_one_locked_base_node(&tb->common); - int result = db_last_tree_common(p, tbl, base->baseOrRoute.base.root, ret, NULL); - wunlock_base_node(&(base->baseOrRoute.base)); + TreeDbTerm *root; + CATreeRootIterator iter; + int result; + + init_root_iterator(&tbl->catree, &iter, 1); + root = *catree_find_last_root(&iter); + if (!root) { + TreeDbTerm **pp = catree_find_prev_root(&iter, NULL); + root = pp ? *pp : NULL; + } + + result = db_last_tree_common(p, tbl, root, ret, NULL); + + destroy_root_iterator(&iter); return result; - } static int db_prev_catree(Process *p, DbTable *tbl, Eterm key, Eterm *ret) { DbTreeStack stack; TreeDbTerm * stack_array[STACK_NEED]; - DbTableCATree *tb = &tbl->catree; + TreeDbTerm **rootp; + CATreeRootIterator iter; int result; - DbTableCATreeNode *base; - init_tree_stack(&stack, stack_array, 0); - base = merge_to_one_locked_base_node(&tb->common); - result = db_prev_tree_common(p, tbl, base->baseOrRoute.base.root, key, ret, &stack); - wunlock_base_node(&(base->baseOrRoute.base)); + + init_root_iterator(&tbl->catree, &iter, 1); + iter.next_route_key = key; + rootp = catree_find_prev_root(&iter, NULL); + + do { + init_tree_stack(&stack, stack_array, 0); + result = db_prev_tree_common(p, tbl, (rootp ? *rootp : NULL), key, ret, + &stack); + if (result != DB_ERROR_NONE || *ret != am_EOT) + break; + rootp = catree_find_prev_root(&iter, NULL); + } while (rootp); + + destroy_root_iterator(&iter); return result; } -#define ERL_DB_CATREE_DO_OPERATION_PUT_PARAMS Eterm obj, int key_clash_fail -ERL_DB_CATREE_CREATE_DO_OPERATION_FUNCTION(put, - ERL_DB_CATREE_DO_OPERATION_PUT_PARAMS, - db_put_tree_common(&tb->common, - &base_node->root, - obj, - key_clash_fail, - NULL);) - static int db_put_catree(DbTable *tbl, Eterm obj, int key_clash_fail) { DbTableCATree *tb = &tbl->catree; Eterm key = GETKEY(&tb->common, tuple_val(obj)); - return erl_db_catree_do_operation_put(tb, - key, - obj, - key_clash_fail); + FindBaseNode fbn; + DbTableCATreeNode* node = find_wlock_valid_base_node(tb, key, &fbn); + int result = db_put_tree_common(&tb->common, &node->u.base.root, obj, + key_clash_fail, NULL); + wunlock_adapt_base_node(tb, node, fbn.parent, fbn.current_level); + return result; } -#define ERL_DB_CATREE_DO_OPERATION_GET_PARAMS Process *p, Eterm *ret -ERL_DB_CATREE_CREATE_DO_READ_OPERATION_FUNCTION(get, - ERL_DB_CATREE_DO_OPERATION_GET_PARAMS, - db_get_tree_common(p, &tb->common, - base_node->root, - key, ret, NULL);) - static int db_get_catree(Process *p, DbTable *tbl, Eterm key, Eterm *ret) { DbTableCATree *tb = &tbl->catree; - return erl_db_catree_do_operation_get(tb, key, p, ret); + DbTableCATreeNode* node = find_rlock_valid_base_node(tb, key); + int result = db_get_tree_common(p, &tb->common, + node->u.base.root, + key, ret, NULL); + runlock_base_node(node); + return result; +} + +TreeDbTerm** catree_find_root(Eterm key, CATreeRootIterator* iter) +{ + FindBaseNode fbn; + DbTableCATreeNode* base_node; + + while (1) { + base_node = find_base_node(iter->tb, key, &fbn); + lock_iter_base_node(iter, base_node, fbn.parent, fbn.current_level); + if (base_node->u.base.is_valid) + break; + unlock_iter_base_node(iter); + } + return &base_node->u.base.root; } -#define ERL_DB_CATREE_DO_OPERATION_MEMBER_PARAMS Eterm *ret -ERL_DB_CATREE_CREATE_DO_READ_OPERATION_FUNCTION(member, - ERL_DB_CATREE_DO_OPERATION_MEMBER_PARAMS, - db_member_tree_common(&tb->common, - base_node->root, - key, - ret, - NULL);) +static Eterm save_iter_search_key(CATreeRootIterator* iter, Eterm key) +{ + Uint key_size; + + if (is_immed(key)) + return key; + + if (iter->search_key) { + if (key == iter->search_key->term) + return key; /* already saved */ + destroy_route_key(iter->search_key); + } + key_size = size_object(key); + if (!iter->search_key || key_size > iter->search_key->size) { + iter->search_key = erts_realloc(ERTS_ALC_T_DB_TMP, + iter->search_key, + (offsetof(DbRouteKey, heap) + + key_size*sizeof(Eterm))); + } + return copy_route_key(iter->search_key, key, key_size); +} + +TreeDbTerm** catree_find_nextprev_root(CATreeRootIterator *iter, + int forward, + Eterm *search_keyp) +{ +#ifdef DEBUG + DbTableCATreeNode *rejected_invalid = NULL; + DbTableCATreeNode *rejected_empty = NULL; +#endif + DbTableCATreeNode *node; + DbTableCATreeNode *parent; + DbTableCATreeNode* next_route_node; + Eterm route_key = iter->next_route_key; + int current_level; + + if (iter->locked_bnode) { + if (search_keyp) + *search_keyp = save_iter_search_key(iter, *search_keyp); + unlock_iter_base_node(iter); + } + + if (is_non_value(route_key)) + return NULL; + + while (1) { + node = GET_ROOT_ACQB(iter->tb); + current_level = 0; + parent = NULL; + next_route_node = NULL; + while (!node->is_base_node) { + current_level++; + parent = node; + if (forward) { + if (cmp_key_route(route_key,node) < 0) { + next_route_node = node; + node = GET_LEFT_ACQB(node); + } else { + node = GET_RIGHT_ACQB(node); + } + } + else { + if (cmp_key_route(route_key,node) > 0) { + next_route_node = node; + node = GET_RIGHT_ACQB(node); + } else { + node = GET_LEFT_ACQB(node); + } + } + } + ASSERT(node != rejected_invalid); + lock_iter_base_node(iter, node, parent, current_level); + if (node->u.base.is_valid) { + ASSERT(node != rejected_empty); + if (node->u.base.root) { + iter->next_route_key = (next_route_node ? + next_route_node->u.route.key.term : + THE_NON_VALUE); + iter->locked_bnode = node; + return &node->u.base.root; + } + if (!next_route_node) { + unlock_iter_base_node(iter); + return NULL; + } + route_key = next_route_node->u.route.key.term; + IF_DEBUG(rejected_empty = node); + } + else + IF_DEBUG(rejected_invalid = node); + + /* Retry */ + unlock_iter_base_node(iter); + } +} + +TreeDbTerm** catree_find_next_root(CATreeRootIterator *iter, Eterm* keyp) +{ + return catree_find_nextprev_root(iter, 1, keyp); +} + +TreeDbTerm** catree_find_prev_root(CATreeRootIterator *iter, Eterm* keyp) +{ + return catree_find_nextprev_root(iter, 0, keyp); +} + +/* @brief Find root of tree where object with smallest key of all larger than + * partially bound key may reside. Can be used as a starting point for + * a reverse iteration with pb_key. + * + * @param pb_key The partially bound key. Example {42, '$1'} + * @param iter An initialized root iterator. + * + * @return Pointer to found root pointer. May not be NULL. + */ +TreeDbTerm** catree_find_next_from_pb_key_root(Eterm pb_key, + CATreeRootIterator* iter) +{ +#ifdef DEBUG + DbTableCATreeNode *rejected_base = NULL; +#endif + DbTableCATreeNode *node; + DbTableCATreeNode *parent; + DbTableCATreeNode* next_route_node; + int current_level; + + ASSERT(!iter->locked_bnode); + + while (1) { + node = GET_ROOT_ACQB(iter->tb); + current_level = 0; + parent = NULL; + next_route_node = NULL; + while (!node->is_base_node) { + current_level++; + parent = node; + if (cmp_partly_bound(pb_key, GET_ROUTE_NODE_KEY(node)) >= 0) { + next_route_node = node; + node = GET_RIGHT_ACQB(node); + } else { + node = GET_LEFT_ACQB(node); + } + } + ASSERT(node != rejected_base); + lock_iter_base_node(iter, node, parent, current_level); + if (node->u.base.is_valid) { + iter->next_route_key = (next_route_node ? + next_route_node->u.route.key.term : + THE_NON_VALUE); + return &node->u.base.root; + } + /* Retry */ + unlock_iter_base_node(iter); +#ifdef DEBUG + rejected_base = node; +#endif + } +} + +/* @brief Find root of tree where object with largest key of all smaller than + * partially bound key may reside. Can be used as a starting point for + * a forward iteration with pb_key. + * + * @param pb_key The partially bound key. Example {42, '$1'} + * @param iter An initialized root iterator. + * + * @return Pointer to found root pointer. May not be NULL. + */ +TreeDbTerm** catree_find_prev_from_pb_key_root(Eterm key, + CATreeRootIterator* iter) +{ +#ifdef DEBUG + DbTableCATreeNode *rejected_base = NULL; +#endif + DbTableCATreeNode *node; + DbTableCATreeNode *parent; + DbTableCATreeNode* next_route_node; + int current_level; + + ASSERT(!iter->locked_bnode); + + while (1) { + node = GET_ROOT_ACQB(iter->tb); + current_level = 0; + parent = NULL; + next_route_node = NULL; + while (!node->is_base_node) { + current_level++; + parent = node; + if (cmp_partly_bound(key, GET_ROUTE_NODE_KEY(node)) <= 0) { + next_route_node = node; + node = GET_LEFT_ACQB(node); + } else { + node = GET_RIGHT_ACQB(node); + } + } + ASSERT(node != rejected_base); + lock_iter_base_node(iter, node, parent, current_level); + if (node->u.base.is_valid) { + iter->next_route_key = (next_route_node ? + next_route_node->u.route.key.term : + THE_NON_VALUE); + return &node->u.base.root; + } + /* Retry */ + unlock_iter_base_node(iter); +#ifdef DEBUG + rejected_base = node; +#endif + } +} + +static TreeDbTerm** catree_find_firstlast_root(CATreeRootIterator* iter, + int first) +{ +#ifdef DEBUG + DbTableCATreeNode *rejected_base = NULL; +#endif + DbTableCATreeNode *node; + DbTableCATreeNode* next_route_node; + int current_level; + + while (1) { + node = GET_ROOT_ACQB(iter->tb); + current_level = 0; + next_route_node = NULL; + while (!node->is_base_node) { + current_level++; + next_route_node = node; + node = first ? GET_LEFT_ACQB(node) : GET_RIGHT_ACQB(node); + } + ASSERT(node != rejected_base); + lock_iter_base_node(iter, node, next_route_node, current_level); + if (node->u.base.is_valid) { + iter->next_route_key = (next_route_node ? + next_route_node->u.route.key.term : + THE_NON_VALUE); + return &node->u.base.root; + } + /* Retry */ + unlock_iter_base_node(iter); +#ifdef DEBUG + rejected_base = node; +#endif + } +} + +TreeDbTerm** catree_find_first_root(CATreeRootIterator* iter) +{ + return catree_find_firstlast_root(iter, 1); +} + +TreeDbTerm** catree_find_last_root(CATreeRootIterator* iter) +{ + return catree_find_firstlast_root(iter, 0); +} static int db_member_catree(DbTable *tbl, Eterm key, Eterm *ret) { DbTableCATree *tb = &tbl->catree; - return erl_db_catree_do_operation_member(tb, key, ret); + DbTableCATreeNode* node = find_rlock_valid_base_node(tb, key); + int result = db_member_tree_common(&tb->common, + node->u.base.root, + key, ret, NULL); + runlock_base_node(node); + return result; } -#define ERL_DB_CATREE_DO_OPERATION_GET_ELEMENT_PARAMS Process *p, int ndex, Eterm *ret -ERL_DB_CATREE_CREATE_DO_READ_OPERATION_FUNCTION(get_element, - ERL_DB_CATREE_DO_OPERATION_GET_ELEMENT_PARAMS, - db_get_element_tree_common(p, &tb->common, - base_node->root, - key, ndex, - ret, NULL)) - static int db_get_element_catree(Process *p, DbTable *tbl, Eterm key, int ndex, Eterm *ret) { DbTableCATree *tb = &tbl->catree; - return erl_db_catree_do_operation_get_element(tb, key, p, ndex, ret); + DbTableCATreeNode* node = find_rlock_valid_base_node(tb, key); + int result = db_get_element_tree_common(p, &tb->common, + node->u.base.root, + key, ndex, ret, NULL); + runlock_base_node(node); + return result; } -#define ERL_DB_CATREE_DO_OPERATION_ERASE_PARAMS DbTable *tbl, Eterm *ret -ERL_DB_CATREE_CREATE_DO_OPERATION_FUNCTION(erase, - ERL_DB_CATREE_DO_OPERATION_ERASE_PARAMS, - db_erase_tree_common(tbl, - &base_node->root, - key, - ret, - NULL);) - static int db_erase_catree(DbTable *tbl, Eterm key, Eterm *ret) { DbTableCATree *tb = &tbl->catree; - return erl_db_catree_do_operation_erase(tb, key, tbl, ret); + FindBaseNode fbn; + DbTableCATreeNode* node = find_wlock_valid_base_node(tb, key, &fbn); + int result = db_erase_tree_common(tbl, &node->u.base.root, key, + ret, NULL); + wunlock_adapt_base_node(tb, node, fbn.parent, fbn.current_level); + return result; } -#define ERL_DB_CATREE_DO_OPERATION_ERASE_OBJECT_PARAMS DbTable *tbl, Eterm object, Eterm *ret -ERL_DB_CATREE_CREATE_DO_OPERATION_FUNCTION(erase_object, - ERL_DB_CATREE_DO_OPERATION_ERASE_OBJECT_PARAMS, - db_erase_object_tree_common(tbl, - &base_node->root, - object, - ret, - NULL);) - static int db_erase_object_catree(DbTable *tbl, Eterm object, Eterm *ret) { DbTableCATree *tb = &tbl->catree; Eterm key = GETKEY(&tb->common, tuple_val(object)); - return erl_db_catree_do_operation_erase_object(tb, key, tbl, object, ret); + FindBaseNode fbn; + DbTableCATreeNode* node = find_wlock_valid_base_node(tb, key, &fbn); + int result = db_erase_object_tree_common(tbl, + &node->u.base.root, + object, + ret, + NULL); + wunlock_adapt_base_node(tb, node, fbn.parent, fbn.current_level); + return result; } static int db_slot_catree(Process *p, DbTable *tbl, Eterm slot_term, Eterm *ret) { - DbTableCATree *tb = &tbl->catree; int result; - DbTableCATreeNode *base; - base = merge_to_one_locked_base_node(&tb->common); - result = db_slot_tree_common(p, tbl, base->baseOrRoute.base.root, - slot_term, ret, NULL); - wunlock_base_node(&(base->baseOrRoute.base)); + CATreeRootIterator iter; + + init_root_iterator(&tbl->catree, &iter, 1); + result = db_slot_tree_common(p, tbl, *catree_find_first_root(&iter), + slot_term, ret, NULL, &iter); + destroy_root_iterator(&iter); return result; } @@ -1850,28 +1857,26 @@ static int db_select_continue_catree(Process *p, Eterm continuation, Eterm *ret) { - DbTableCATree *tb = &tbl->catree; int result; - DbTableCATreeNode *base; - base = merge_to_one_locked_base_node(&tb->common); - result = db_select_continue_tree_common(p, &tb->common, &base->baseOrRoute.base.root, - continuation, ret, NULL); - wunlock_base_node(&(base->baseOrRoute.base)); + CATreeRootIterator iter; + + init_root_iterator(&tbl->catree, &iter, 1); + result = db_select_continue_tree_common(p, &tbl->common, + continuation, ret, NULL, &iter); + destroy_root_iterator(&iter); return result; } - static int db_select_catree(Process *p, DbTable *tbl, Eterm tid, Eterm pattern, int reverse, Eterm *ret) { - DbTableCATree *tb = &tbl->catree; int result; - DbTableCATreeNode *base; - base = merge_to_one_locked_base_node(&tb->common); - result = db_select_tree_common(p, &tb->common, &base->baseOrRoute.base.root, - tid, pattern, reverse, ret, - NULL); - wunlock_base_node(&(base->baseOrRoute.base)); + CATreeRootIterator iter; + + init_root_iterator(&tbl->catree, &iter, 1); + result = db_select_tree_common(p, tbl, tid, pattern, reverse, ret, + NULL, &iter); + destroy_root_iterator(&iter); return result; } @@ -1880,27 +1885,27 @@ static int db_select_count_continue_catree(Process *p, Eterm continuation, Eterm *ret) { - DbTableCATree *tb = &tbl->catree; int result; - DbTableCATreeNode *base; - base = merge_to_one_locked_base_node(&tb->common); - result = db_select_count_continue_tree_common(p, &tb->common, - &base->baseOrRoute.base.root, - continuation, ret, NULL); - wunlock_base_node(&(base->baseOrRoute.base)); + CATreeRootIterator iter; + + init_root_iterator(&tbl->catree, &iter, 1); + result = db_select_count_continue_tree_common(p, tbl, + continuation, ret, NULL, + &iter); + destroy_root_iterator(&iter); return result; } static int db_select_count_catree(Process *p, DbTable *tbl, Eterm tid, Eterm pattern, Eterm *ret) { - DbTableCATree *tb = &tbl->catree; int result; - DbTableCATreeNode *base; - base = merge_to_one_locked_base_node(&tb->common); - result = db_select_count_tree_common(p, &tb->common, &base->baseOrRoute.base.root, - tid, pattern, ret, NULL); - wunlock_base_node(&(base->baseOrRoute.base)); + CATreeRootIterator iter; + + init_root_iterator(&tbl->catree, &iter, 1); + result = db_select_count_tree_common(p, tbl, + tid, pattern, ret, NULL, &iter); + destroy_root_iterator(&iter); return result; } @@ -1908,13 +1913,14 @@ static int db_select_chunk_catree(Process *p, DbTable *tbl, Eterm tid, Eterm pattern, Sint chunk_size, int reversed, Eterm *ret) { - DbTableCATree *tb = &tbl->catree; int result; - DbTableCATreeNode *base; - base = merge_to_one_locked_base_node(&tb->common); - result = db_select_chunk_tree_common(p, &tb->common, &base->baseOrRoute.base.root, - tid, pattern, chunk_size, reversed, ret, NULL); - wunlock_base_node(&(base->baseOrRoute.base)); + CATreeRootIterator iter; + + init_root_iterator(&tbl->catree, &iter, 1); + result = db_select_chunk_tree_common(p, tbl, + tid, pattern, chunk_size, reversed, ret, + NULL, &iter); + destroy_root_iterator(&iter); return result; } @@ -1923,74 +1929,71 @@ static int db_select_delete_continue_catree(Process *p, Eterm continuation, Eterm *ret) { - DbTableCATree *tb = &tbl->catree; DbTreeStack stack; TreeDbTerm * stack_array[STACK_NEED]; int result; - DbTableCATreeNode *base; + CATreeRootIterator iter; + + init_root_iterator(&tbl->catree, &iter, 0); init_tree_stack(&stack, stack_array, 0); - base = merge_to_one_locked_base_node(&tb->common); - result = db_select_delete_continue_tree_common(p, tbl, &base->baseOrRoute.base.root, - continuation, ret, &stack); - wunlock_base_node(&(base->baseOrRoute.base)); + result = db_select_delete_continue_tree_common(p, tbl, continuation, ret, + &stack, &iter); + destroy_root_iterator(&iter); return result; } static int db_select_delete_catree(Process *p, DbTable *tbl, Eterm tid, Eterm pattern, Eterm *ret) { - DbTableCATree *tb = &tbl->catree; DbTreeStack stack; TreeDbTerm * stack_array[STACK_NEED]; int result; - DbTableCATreeNode *base; + CATreeRootIterator iter; + + init_root_iterator(&tbl->catree, &iter, 0); init_tree_stack(&stack, stack_array, 0); - base = merge_to_one_locked_base_node(&tb->common); - result = db_select_delete_tree_common(p, tbl, &base->baseOrRoute.base.root, - tid, pattern, ret, &stack); - wunlock_base_node(&(base->baseOrRoute.base)); + result = db_select_delete_tree_common(p, tbl, + tid, pattern, ret, &stack, + &iter); + destroy_root_iterator(&iter); return result; } static int db_select_replace_catree(Process *p, DbTable *tbl, Eterm tid, Eterm pattern, Eterm *ret) { - DbTableCATree *tb = &tbl->catree; int result; - DbTableCATreeNode *base; - base = merge_to_one_locked_base_node(&tb->common); - result = db_select_replace_tree_common(p, &tb->common, - &base->baseOrRoute.base.root, - tid, pattern, ret, NULL); - wunlock_base_node(&(base->baseOrRoute.base)); + CATreeRootIterator iter; + + init_root_iterator(&tbl->catree, &iter, 0); + result = db_select_replace_tree_common(p, tbl, + tid, pattern, ret, NULL, &iter); + destroy_root_iterator(&iter); return result; } static int db_select_replace_continue_catree(Process *p, DbTable *tbl, Eterm continuation, Eterm *ret) { - DbTableCATree *tb = &tbl->catree; int result; - DbTableCATreeNode *base; - base = merge_to_one_locked_base_node(&tb->common); - result = db_select_replace_continue_tree_common(p, &tb->common, - &base->baseOrRoute.base.root, - continuation, ret, NULL); - wunlock_base_node(&(base->baseOrRoute.base)); + CATreeRootIterator iter; + + init_root_iterator(&tbl->catree, &iter, 0); + result = db_select_replace_continue_tree_common(p, tbl, continuation, ret, + NULL, &iter); + destroy_root_iterator(&iter); return result; } -#define ERL_DB_CATREE_DO_OPERATION_TAKE_PARAMS Process *p, Eterm *ret, DbTable *tbl -ERL_DB_CATREE_CREATE_DO_OPERATION_FUNCTION(take, - ERL_DB_CATREE_DO_OPERATION_TAKE_PARAMS, - db_take_tree_common(p, tbl, - &base_node->root, - key, ret, NULL);) - static int db_take_catree(Process *p, DbTable *tbl, Eterm key, Eterm *ret) { DbTableCATree *tb = &tbl->catree; - return erl_db_catree_do_operation_take(tb, key, p, ret, tbl); + FindBaseNode fbn; + DbTableCATreeNode* node = find_wlock_valid_base_node(tb, key, &fbn); + int result = db_take_tree_common(p, tbl, &node->u.base.root, key, + ret, NULL); + wunlock_adapt_base_node(tb, node, fbn.parent, fbn.current_level); + return result; } /* @@ -2002,10 +2005,16 @@ static int db_take_catree(Process *p, DbTable *tbl, Eterm key, Eterm *ret) static void db_print_catree(fmtfn_t to, void *to_arg, int show, DbTable *tbl) { - DbTableCATree *tb = &tbl->catree; - DbTableCATreeNode *base = merge_to_one_locked_base_node(&tb->common); - db_print_tree_common(to, to_arg, show, base->baseOrRoute.base.root, tbl); - wunlock_base_node(&(base->baseOrRoute.base)); + CATreeRootIterator iter; + TreeDbTerm** root; + + init_root_iterator(&tbl->catree, &iter, 1); + root = catree_find_first_root(&iter); + do { + db_print_tree_common(to, to_arg, show, *root, tbl); + root = catree_find_next_root(&iter, NULL); + } while (root); + destroy_root_iterator(&iter); } /* Release all memory occupied by a single table */ @@ -2045,7 +2054,7 @@ static SWord db_free_table_continue_catree(DbTable *tbl, SWord reds) } else { tb->is_routing_nodes_freed = 1; /* Ready with the routing nodes */ first_base_node = catree_first_base_node_from_free_list(tb); - PUSH_NODE(&tb->free_stack_elems, first_base_node->baseOrRoute.base.root); + PUSH_NODE(&tb->free_stack_elems, first_base_node->u.base.root); } } while (catree_first_base_node_from_free_list(tb) != NULL) { @@ -2081,26 +2090,33 @@ static void db_foreach_offheap_catree(DbTable *tbl, void (*func)(ErlOffHeap *, void *), void *arg) { - DbTableCATree *tb = &tbl->catree; - DbTableCATreeNode *base = merge_to_one_locked_base_node(&tb->common); - db_foreach_offheap_tree_common(base->baseOrRoute.base.root, func, arg); - wunlock_base_node(&(base->baseOrRoute.base)); + CATreeRootIterator iter; + TreeDbTerm** root; + + init_root_iterator(&tbl->catree, &iter, 1); + root = catree_find_first_root(&iter); + do { + db_foreach_offheap_tree_common(*root, func, arg); + root = catree_find_next_root(&iter, NULL); + } while (root); + destroy_root_iterator(&iter); } static int db_lookup_dbterm_catree(Process *p, DbTable *tbl, Eterm key, Eterm obj, DbUpdateHandle *handle) { DbTableCATree *tb = &tbl->catree; - int res; - ERL_DB_CATREE_CREATE_DO_OPERATION_FUNCTION_FIND_BASE_NODE_PART(wlock_base_node,wunlock_base_node); - res = db_lookup_dbterm_tree_common(p, tbl, &base_node->root, key, obj, handle, NULL); + FindBaseNode fbn; + DbTableCATreeNode* node = find_wlock_valid_base_node(tb, key, &fbn); + int res = db_lookup_dbterm_tree_common(p, tbl, &node->u.base.root, key, + obj, handle, NULL); if (res == 0) { - ERL_DB_CATREE_CREATE_DO_OPERATION_FUNCTION_ADAPT_AND_UNLOCK_PART; + wunlock_adapt_base_node(tb, node, fbn.parent, fbn.current_level); } else { /* db_finalize_dbterm_catree will unlock */ - handle->lck = prev_node; - handle->lck2 = current_node; - handle->current_level = current_level; + handle->u.catree.base_node = node; + handle->u.catree.parent = fbn.parent; + handle->u.catree.current_level = fbn.current_level; } return res; } @@ -2108,12 +2124,10 @@ static int db_lookup_dbterm_catree(Process *p, DbTable *tbl, Eterm key, Eterm ob static void db_finalize_dbterm_catree(int cret, DbUpdateHandle *handle) { DbTableCATree *tb = &(handle->tb->catree); - DbTableCATreeNode *prev_node = handle->lck; - DbTableCATreeNode *current_node = handle->lck2; - int current_level = handle->current_level; - DbTableCATreeBaseNode *base_node = ¤t_node->baseOrRoute.base; db_finalize_dbterm_tree_common(cret, handle, NULL); - ERL_DB_CATREE_CREATE_DO_OPERATION_FUNCTION_ADAPT_AND_UNLOCK_PART; + wunlock_adapt_base_node(tb, handle->u.catree.base_node, + handle->u.catree.parent, + handle->u.catree.current_level); return; } @@ -2147,6 +2161,61 @@ void erts_lcnt_enable_db_catree_lock_count(DbTableCATree *tb, int enable) } #endif /* ERTS_ENABLE_LOCK_COUNT */ +void db_catree_force_split(DbTableCATree* tb, int on) +{ + CATreeRootIterator iter; + TreeDbTerm** root; + + init_root_iterator(tb, &iter, 1); + root = catree_find_first_root(&iter); + do { + iter.locked_bnode->u.base.lock_statistics = (on ? INT_MAX : 0); + root = catree_find_next_root(&iter, NULL); + } while (root); + destroy_root_iterator(&iter); + + if (on) + tb->common.status |= DB_CATREE_FORCE_SPLIT; + else + tb->common.status &= ~DB_CATREE_FORCE_SPLIT; +} + +void db_calc_stats_catree(DbTableCATree* tb, DbCATreeStats* stats) +{ + DbTableCATreeNode* stack[ERL_DB_CATREE_MAX_ROUTE_NODE_LAYER_HEIGHT]; + DbTableCATreeNode* node; + Uint depth = 0; + + stats->route_nodes = 0; + stats->base_nodes = 0; + stats->max_depth = 0; + + node = GET_ROOT(tb); + do { + while (!node->is_base_node) { + stats->route_nodes++; + ASSERT(depth < sizeof(stack)/sizeof(*stack)); + stack[depth++] = node; /* PUSH parent */ + if (stats->max_depth < depth) + stats->max_depth = depth; + node = GET_LEFT(node); + } + stats->base_nodes++; + + while (depth > 0) { + DbTableCATreeNode* parent = stack[depth-1]; + if (node == GET_LEFT(parent)) { + node = GET_RIGHT(parent); + break; + } + else { + ASSERT(node == GET_RIGHT(parent)); + node = parent; + depth--; /* POP parent */ + } + } + } while (depth > 0); +} #ifdef HARDDEBUG diff --git a/erts/emulator/beam/erl_db_catree.h b/erts/emulator/beam/erl_db_catree.h index 1f2c7da091..418837be8e 100644 --- a/erts/emulator/beam/erl_db_catree.h +++ b/erts/emulator/beam/erl_db_catree.h @@ -34,23 +34,33 @@ struct DbTableCATreeNode; typedef struct { + Eterm term; + struct erl_off_heap_header* oh; + Uint size; + Eterm heap[1]; +} DbRouteKey; + +typedef struct { erts_rwmtx_t lock; /* The lock for this base node */ Sint lock_statistics; int is_valid; /* If this base node is still valid */ TreeDbTerm *root; /* The root of the sequential tree */ - DbTable * tab; /* Table ptr, used when freeing using thread progress */ ErtsThrPrgrLaterOp free_item; /* Used when freeing using thread progress */ struct DbTableCATreeNode * next; /* Used when gradually deleting */ + + char end_of_struct__; } DbTableCATreeBaseNode; typedef struct { +#ifdef ERTS_ENABLE_LOCK_CHECK + Sint lc_order; +#endif ErtsThrPrgrLaterOp free_item; /* Used when freeing using thread progress */ - DbTable* tab; /* Table ptr, used when freeing using thread progress */ erts_mtx_t lock; /* Used when joining route nodes */ int is_valid; /* If this route node is still valid */ erts_atomic_t left; erts_atomic_t right; - DbTerm key; + DbRouteKey key; } DbTableCATreeRouteNode; typedef struct DbTableCATreeNode { @@ -58,7 +68,7 @@ typedef struct DbTableCATreeNode { union { DbTableCATreeRouteNode route; DbTableCATreeBaseNode base; - } baseOrRoute; + } u; } DbTableCATreeNode; typedef struct { @@ -79,13 +89,45 @@ typedef struct db_table_catree { int is_routing_nodes_freed; } DbTableCATree; +typedef struct { + DbTableCATree* tb; + Eterm next_route_key; + DbTableCATreeNode* locked_bnode; + DbTableCATreeNode* bnode_parent; + int bnode_level; + int read_only; + DbRouteKey* search_key; +} CATreeRootIterator; + + void db_initialize_catree(void); int db_create_catree(Process *p, DbTable *tbl); +TreeDbTerm** catree_find_root(Eterm key, CATreeRootIterator*); + +TreeDbTerm** catree_find_next_from_pb_key_root(Eterm key, CATreeRootIterator*); +TreeDbTerm** catree_find_prev_from_pb_key_root(Eterm key, CATreeRootIterator*); +TreeDbTerm** catree_find_nextprev_root(CATreeRootIterator*, int next, Eterm* keyp); +TreeDbTerm** catree_find_next_root(CATreeRootIterator*, Eterm* keyp); +TreeDbTerm** catree_find_prev_root(CATreeRootIterator*, Eterm* keyp); +TreeDbTerm** catree_find_first_root(CATreeRootIterator*); +TreeDbTerm** catree_find_last_root(CATreeRootIterator*); + + #ifdef ERTS_ENABLE_LOCK_COUNT void erts_lcnt_enable_db_catree_lock_count(DbTableCATree *tb, int enable); #endif +void db_catree_force_split(DbTableCATree*, int on); + +typedef struct { + Uint route_nodes; + Uint base_nodes; + Uint max_depth; +} DbCATreeStats; +void db_calc_stats_catree(DbTableCATree*, DbCATreeStats*); + + #endif /* _DB_CATREE_H */ diff --git a/erts/emulator/beam/erl_db_hash.c b/erts/emulator/beam/erl_db_hash.c index 752d3ae3a8..f05a3b51c9 100644 --- a/erts/emulator/beam/erl_db_hash.c +++ b/erts/emulator/beam/erl_db_hash.c @@ -2731,13 +2731,9 @@ static int free_seg(DbTableHash *tb, int free_records) * sure no lingering threads are still hanging in BUCKET macro * with an old segtab pointer. */ - Uint sz = SIZEOF_EXT_SEGTAB(est->nsegs); - ASSERT(sz == ERTS_ALC_DBG_BLK_SZ(est)); - ERTS_DB_ALC_MEM_UPDATE_(tb, sz, 0); - erts_schedule_thr_prgr_later_cleanup_op(dealloc_ext_segtab, - est, - &est->lop, - sz); + erts_schedule_db_free(&tb->common, dealloc_ext_segtab, + est, &est->lop, + SIZEOF_EXT_SEGTAB(est->nsegs)); } else erts_db_free(ERTS_ALC_T_DB_SEG, (DbTable*)tb, est, @@ -3107,7 +3103,7 @@ Ldone: handle->dbterm = &b->dbterm; handle->flags = flags; handle->new_size = b->dbterm.size; - handle->lck = lck; + handle->u.hash.lck = lck; return 1; } @@ -3120,7 +3116,7 @@ db_finalize_dbterm_hash(int cret, DbUpdateHandle* handle) DbTableHash *tb = &tbl->hash; HashDbTerm **bp = (HashDbTerm **) handle->bp; HashDbTerm *b = *bp; - erts_rwmtx_t* lck = (erts_rwmtx_t*) handle->lck; + erts_rwmtx_t* lck = handle->u.hash.lck; HashDbTerm* free_me = NULL; ERTS_LC_ASSERT(IS_HASH_WLOCKED(tb, lck)); /* locked by db_lookup_dbterm_hash */ diff --git a/erts/emulator/beam/erl_db_tree.c b/erts/emulator/beam/erl_db_tree.c index f643344477..02a5934a6e 100644 --- a/erts/emulator/beam/erl_db_tree.c +++ b/erts/emulator/beam/erl_db_tree.c @@ -70,8 +70,10 @@ */ ERTS_INLINE static DbTreeStack* get_static_stack(DbTableTree* tb) { - if (tb != NULL && !erts_atomic_xchg_acqb(&tb->is_stack_busy, 1)) { - return &tb->static_stack; + if (tb != NULL) { + ASSERT(IS_TREE_TABLE(tb->common.type)); + if (!erts_atomic_xchg_acqb(&tb->is_stack_busy, 1)) + return &tb->static_stack; } return NULL; } @@ -82,9 +84,10 @@ ERTS_INLINE static DbTreeStack* get_static_stack(DbTableTree* tb) static DbTreeStack* get_any_stack(DbTable* tb, DbTableTree* stack_container) { DbTreeStack* stack; - if (stack_container != NULL && - !erts_atomic_xchg_acqb(&stack_container->is_stack_busy, 1)) { - return &stack_container->static_stack; + if (stack_container != NULL) { + ASSERT(IS_TREE_TABLE(stack_container->common.type)); + if (!erts_atomic_xchg_acqb(&stack_container->is_stack_busy, 1)) + return &stack_container->static_stack; } stack = erts_db_alloc(ERTS_ALC_T_DB_STK, tb, sizeof(DbTreeStack) + sizeof(TreeDbTerm*) * STACK_NEED); @@ -96,14 +99,16 @@ static DbTreeStack* get_any_stack(DbTable* tb, DbTableTree* stack_container) static void release_stack(DbTable* tb, DbTableTree* stack_container, DbTreeStack* stack) { - if (stack_container != NULL && stack == &stack_container->static_stack) { - ASSERT(erts_atomic_read_nob(&stack_container->is_stack_busy) == 1); - erts_atomic_set_relb(&stack_container->is_stack_busy, 0); - } - else { - erts_db_free(ERTS_ALC_T_DB_STK, tb, - (void *) stack, sizeof(DbTreeStack) + sizeof(TreeDbTerm*) * STACK_NEED); + if (stack_container != NULL) { + ASSERT(IS_TREE_TABLE(stack_container->common.type)); + if (stack == &stack_container->static_stack) { + ASSERT(erts_atomic_read_nob(&stack_container->is_stack_busy) == 1); + erts_atomic_set_relb(&stack_container->is_stack_busy, 0); + return; + } } + erts_db_free(ERTS_ALC_T_DB_STK, tb, + (void *) stack, sizeof(DbTreeStack) + sizeof(TreeDbTerm*) * STACK_NEED); } static ERTS_INLINE void reset_stack(DbTreeStack* stack) @@ -117,6 +122,7 @@ static ERTS_INLINE void reset_stack(DbTreeStack* stack) static ERTS_INLINE void reset_static_stack(DbTableTree* tb) { if (tb != NULL) { + ASSERT(IS_TREE_TABLE(tb->common.type)); reset_stack(&tb->static_stack); } } @@ -185,31 +191,37 @@ static void do_dump_tree2(DbTableTree*, int to, void *to_arg, int show, ** Datatypes */ +enum ms_key_boundness { + /* Order significant, larger means more "boundness" => less iteration */ + MS_KEY_UNBOUND = 0, + MS_KEY_PARTIALLY_BOUND = 1, + MS_KEY_BOUND = 2, + MS_KEY_IMPOSSIBLE = 3 +}; + /* * This structure is filled in by analyze_pattern() for the select * functions. */ struct mp_info { - int something_can_match; /* The match_spec is not "impossible" */ - int some_limitation; /* There is some limitation on the search - * area, i. e. least and/or most is set.*/ - int got_partial; /* The limitation has a partially bound - * key */ + enum ms_key_boundness key_boundness; Eterm least; /* The lowest matching key (possibly * partially bound expression) */ Eterm most; /* The highest matching key (possibly * partially bound expression) */ - - TreeDbTerm **save_term; /* If the key is completely bound, this - * will be the Tree node we're searching - * for, otherwise it will be useless */ Binary *mp; /* The compiled match program */ }; +struct select_common { + TreeDbTerm **root; +}; + + /* * Used by doit_select(_chunk) */ struct select_context { + struct select_common common; Process *p; Eterm accum; Binary *mp; @@ -225,6 +237,7 @@ struct select_context { * Used by doit_select_count */ struct select_count_context { + struct select_common common; Process *p; Binary *mp; Eterm end_condition; @@ -238,9 +251,9 @@ struct select_count_context { * Used by doit_select_delete */ struct select_delete_context { + struct select_common common; Process *p; DbTableCommon *tb; - TreeDbTerm **root; DbTreeStack *stack; Uint accum; Binary *mp; @@ -255,9 +268,9 @@ struct select_delete_context { * Used by doit_select_replace */ struct select_replace_context { + struct select_common common; Process *p; DbTableCommon *tb; - TreeDbTerm **root; Binary *mp; Eterm end_condition; Eterm *lastobj; @@ -282,7 +295,8 @@ int tree_balance_left(TreeDbTerm **this); int tree_balance_right(TreeDbTerm **this); static int delsub(TreeDbTerm **this); static TreeDbTerm *slot_search(Process *p, TreeDbTerm *root, Sint slot, - DbTable *tb, DbTableTree *stack_container); + DbTable *tb, DbTableTree *stack_container, + CATreeRootIterator *iter); static TreeDbTerm *find_node(DbTableCommon *tb, TreeDbTerm *root, Eterm key, DbTableTree *stack_container); static TreeDbTerm **find_node2(DbTableCommon *tb, TreeDbTerm **root, Eterm key); @@ -292,64 +306,62 @@ static TreeDbTerm *find_next(DbTableCommon *tb, TreeDbTerm *root, DbTreeStack* stack, Eterm key); static TreeDbTerm *find_prev(DbTableCommon *tb, TreeDbTerm *root, DbTreeStack* stack, Eterm key); -static TreeDbTerm *find_next_from_pb_key(DbTableCommon *tb, TreeDbTerm *root, - DbTreeStack* stack, Eterm key); -static TreeDbTerm *find_prev_from_pb_key(DbTableCommon *tb, TreeDbTerm *root, - DbTreeStack* stack, Eterm key); +static TreeDbTerm *find_next_from_pb_key(DbTable*, TreeDbTerm*** rootpp, + DbTreeStack* stack, Eterm key, + CATreeRootIterator*); +static TreeDbTerm *find_prev_from_pb_key(DbTable*, TreeDbTerm*** rootpp, + DbTreeStack* stack, Eterm key, + CATreeRootIterator*); +typedef int traverse_doit_funcT(DbTableCommon*, TreeDbTerm*, + struct select_common*, int forward); + static void traverse_backwards(DbTableCommon *tb, - TreeDbTerm **root, DbTreeStack*, Eterm lastkey, - int (*doit)(DbTableCommon *, - TreeDbTerm *, - void *, - int), - void *context); + traverse_doit_funcT*, + struct select_common *context, + CATreeRootIterator*); static void traverse_forward(DbTableCommon *tb, - TreeDbTerm **root, DbTreeStack*, Eterm lastkey, - int (*doit)(DbTableCommon *tb, - TreeDbTerm *, - void *, - int), - void *context); + traverse_doit_funcT*, + struct select_common *context, + CATreeRootIterator*); static void traverse_update_backwards(DbTableCommon *tb, - TreeDbTerm **root, DbTreeStack*, Eterm lastkey, int (*doit)(DbTableCommon *tb, TreeDbTerm **, // out - void *, + struct select_common*, int), - void *context); -static int key_given(DbTableCommon *tb, TreeDbTerm **root, Eterm pattern, - TreeDbTerm ***ret, Eterm *partly_bound); -static Sint cmp_partly_bound(Eterm partly_bound_key, Eterm bound_key); + struct select_common*, + CATreeRootIterator*); +static enum ms_key_boundness key_boundness(DbTableCommon *tb, + Eterm pattern, Eterm *keyp); static Sint do_cmp_partly_bound(Eterm a, Eterm b, int *done); -static int analyze_pattern(DbTableCommon *tb, TreeDbTerm **root, Eterm pattern, +static int analyze_pattern(DbTableCommon *tb, Eterm pattern, extra_match_validator_t extra_validator, /* Optional callback */ struct mp_info *mpi); static int doit_select(DbTableCommon *tb, - TreeDbTerm *this, - void *ptr, + TreeDbTerm *this, + struct select_common* ptr, int forward); static int doit_select_count(DbTableCommon *tb, TreeDbTerm *this, - void *ptr, + struct select_common*, int forward); static int doit_select_chunk(DbTableCommon *tb, TreeDbTerm *this, - void *ptr, + struct select_common*, int forward); static int doit_select_delete(DbTableCommon *tb, TreeDbTerm *this, - void *ptr, + struct select_common*, int forward); static int doit_select_replace(DbTableCommon *tb, TreeDbTerm **this_ptr, - void *ptr, + struct select_common*, int forward); static int partly_bound_can_match_lesser(Eterm partly_bound_1, @@ -536,7 +548,7 @@ int db_next_tree_common(Process *p, DbTable *tbl, { TreeDbTerm *this; - if (is_atom(key) && key == am_EOT) + if (key == am_EOT) return DB_ERROR_BADKEY; this = find_next(&tbl->common, root, stack, key); if (this == NULL) { @@ -594,7 +606,7 @@ int db_prev_tree_common(Process *p, DbTable *tbl, TreeDbTerm *root, Eterm key, { TreeDbTerm *this; - if (is_atom(key) && key == am_EOT) + if (key == am_EOT) return DB_ERROR_BADKEY; this = find_prev(&tbl->common, root, stack, key); if (this == NULL) { @@ -862,7 +874,8 @@ static int db_erase_object_tree(DbTable *tbl, Eterm object, Eterm *ret) int db_slot_tree_common(Process *p, DbTable *tbl, TreeDbTerm *root, Eterm slot_term, Eterm *ret, - DbTableTree *stack_container) + DbTableTree *stack_container, + CATreeRootIterator *iter) { Sint slot; TreeDbTerm *st; @@ -892,7 +905,7 @@ int db_slot_tree_common(Process *p, DbTable *tbl, TreeDbTerm *root, * are counted from 1 and up. */ ++slot; - st = slot_search(p, root, slot, tbl, stack_container); + st = slot_search(p, root, slot, tbl, stack_container, iter); if (st == NULL) { *ret = am_false; return DB_ERROR_UNSPEC; @@ -910,7 +923,7 @@ static int db_slot_tree(Process *p, DbTable *tbl, Eterm slot_term, Eterm *ret) { DbTableTree *tb = &tbl->tree; - return db_slot_tree_common(p, tbl, tb->root, slot_term, ret, tb); + return db_slot_tree_common(p, tbl, tb->root, slot_term, ret, tb, NULL); } @@ -981,10 +994,10 @@ static BIF_RETTYPE bif_trap3(Export *bif, int db_select_continue_tree_common(Process *p, DbTableCommon *tb, - TreeDbTerm **root, Eterm continuation, Eterm *ret, - DbTableTree *stack_container) + DbTableTree *stack_container, + CATreeRootIterator* iter) { DbTreeStack* stack; struct select_context sc; @@ -998,7 +1011,6 @@ int db_select_continue_tree_common(Process *p, Sint chunk_size; Sint reverse; - #define RET_TO_BIF(Term, State) do { *ret = (Term); return State; } while(0); /* Decode continuation. We know it's a tuple but not the arity or @@ -1032,23 +1044,32 @@ int db_select_continue_tree_common(Process *p, reverse = unsigned_val(tptr[7]); sc.got = signed_val(tptr[8]); - stack = get_any_stack((DbTable*)tb, stack_container); - if (chunk_size) { - if (reverse) { - traverse_backwards(tb, root, stack, lastkey, &doit_select_chunk, &sc); - } else { - traverse_forward(tb, root, stack, lastkey, &doit_select_chunk, &sc); - } - } else { - if (reverse) { - traverse_forward(tb, root, stack, lastkey, &doit_select, &sc); - } else { - traverse_backwards(tb, root, stack, lastkey, &doit_select, &sc); - } + if (iter) { + iter->next_route_key = lastkey; + sc.common.root = catree_find_nextprev_root(iter, !!reverse != !!chunk_size, NULL); } - release_stack((DbTable*)tb,stack_container,stack); + else + sc.common.root = &((DbTableTree*)tb)->root; + + if (sc.common.root) { + stack = get_any_stack((DbTable*)tb, stack_container); + if (chunk_size) { + if (reverse) { + traverse_backwards(tb, stack, lastkey, &doit_select_chunk, &sc.common, iter); + } else { + traverse_forward(tb, stack, lastkey, &doit_select_chunk, &sc.common, iter); + } + } else { + if (reverse) { + traverse_forward(tb, stack, lastkey, &doit_select, &sc.common, iter); + } else { + traverse_backwards(tb, stack, lastkey, &doit_select, &sc.common, iter); + } + } + release_stack((DbTable*)tb,stack_container,stack); - BUMP_REDS(p, 1000 - sc.max); + BUMP_REDS(p, 1000 - sc.max); + } if (sc.max > 0 || (chunk_size && sc.got == chunk_size)) { if (chunk_size) { @@ -1142,13 +1163,14 @@ static int db_select_continue_tree(Process *p, Eterm *ret) { DbTableTree *tb = &tbl->tree; - return db_select_continue_tree_common(p, &tb->common, &tb->root, - continuation, ret, tb); + return db_select_continue_tree_common(p, &tb->common, + continuation, ret, tb, NULL); } -int db_select_tree_common(Process *p, DbTableCommon *tb, TreeDbTerm **root, +int db_select_tree_common(Process *p, DbTable *tb, Eterm tid, Eterm pattern, int reverse, Eterm *ret, - DbTableTree *stack_container) + DbTableTree *stack_container, + CATreeRootIterator* iter) { /* Strategy: Traverse backwards to build resulting list from tail to head */ DbTreeStack* stack; @@ -1179,44 +1201,64 @@ int db_select_tree_common(Process *p, DbTableCommon *tb, TreeDbTerm **root, sc.p = p; sc.max = 1000; sc.end_condition = NIL; - sc.keypos = tb->keypos; + sc.keypos = tb->common.keypos; sc.got = 0; sc.chunk_size = 0; - if ((errcode = analyze_pattern(tb, root, pattern, NULL, &mpi)) != DB_ERROR_NONE) { + if ((errcode = analyze_pattern(&tb->common, pattern, NULL, &mpi)) != DB_ERROR_NONE) { RET_TO_BIF(NIL,errcode); } - if (!mpi.something_can_match) { + if (mpi.key_boundness == MS_KEY_IMPOSSIBLE) { RET_TO_BIF(NIL,DB_ERROR_NONE); /* can't possibly match anything */ } sc.mp = mpi.mp; - if (!mpi.got_partial && mpi.some_limitation && - CMP_EQ(mpi.least,mpi.most)) { - doit_select(tb,*(mpi.save_term),&sc,0 /* direction doesn't matter */); + if (mpi.key_boundness == MS_KEY_BOUND) { + ASSERT(CMP_EQ(mpi.least, mpi.most)); + if (iter) + sc.common.root = catree_find_root(mpi.least, iter); + else + sc.common.root = &tb->tree.root; + this = find_node(&tb->common, *sc.common.root, mpi.least, NULL); + if (this) + doit_select(&tb->common, this, &sc.common, 0 /* direction doesn't matter */); RET_TO_BIF(sc.accum,DB_ERROR_NONE); } stack = get_any_stack((DbTable*)tb,stack_container); if (reverse) { - if (mpi.some_limitation) { - if ((this = find_prev_from_pb_key(tb, *root, stack, mpi.least)) != NULL) { + if (mpi.key_boundness == MS_KEY_PARTIALLY_BOUND) { + this = find_prev_from_pb_key(tb, &sc.common.root, stack, mpi.least, iter); + if (this) lastkey = GETKEY(tb, this->dbterm.tpl); - } sc.end_condition = mpi.most; } - traverse_forward(tb, root, stack, lastkey, &doit_select, &sc); + else { + ASSERT(mpi.key_boundness == MS_KEY_UNBOUND); + if (iter) + sc.common.root = catree_find_first_root(iter); + else + sc.common.root = &tb->tree.root; + } + traverse_forward(&tb->common, stack, lastkey, &doit_select, &sc.common, iter); } else { - if (mpi.some_limitation) { - if ((this = find_next_from_pb_key(tb, *root, stack, mpi.most)) != NULL) { - lastkey = GETKEY(tb, this->dbterm.tpl); - } + if (mpi.key_boundness == MS_KEY_PARTIALLY_BOUND) { + this = find_next_from_pb_key(tb, &sc.common.root, stack, mpi.most, iter); + if (this) + lastkey = GETKEY(tb, this->dbterm.tpl); sc.end_condition = mpi.least; } - traverse_backwards(tb, root, stack, lastkey, &doit_select, &sc); + else { + ASSERT(mpi.key_boundness == MS_KEY_UNBOUND); + if (iter) + sc.common.root = catree_find_last_root(iter); + else + sc.common.root = &tb->tree.root; + } + traverse_backwards(&tb->common, stack, lastkey, &doit_select, &sc.common, iter); } release_stack((DbTable*)tb,stack_container,stack); #ifdef HARDDEBUG @@ -1257,17 +1299,16 @@ int db_select_tree_common(Process *p, DbTableCommon *tb, TreeDbTerm **root, static int db_select_tree(Process *p, DbTable *tbl, Eterm tid, Eterm pattern, int reverse, Eterm *ret) { - DbTableTree *tb = &tbl->tree; - return db_select_tree_common(p, &tb->common, &tb->root, tid, - pattern, reverse, ret, tb); + return db_select_tree_common(p, tbl, tid, + pattern, reverse, ret, &tbl->tree, NULL); } int db_select_count_continue_tree_common(Process *p, - DbTableCommon *tb, - TreeDbTerm **root, + DbTable *tb, Eterm continuation, Eterm *ret, - DbTableTree *stack_container) + DbTableTree *stack_container, + CATreeRootIterator* iter) { DbTreeStack* stack; struct select_count_context sc; @@ -1280,7 +1321,6 @@ int db_select_count_continue_tree_common(Process *p, Eterm *tptr; Eterm egot; - #define RET_TO_BIF(Term, State) do { *ret = (Term); return State; } while(0); /* Decode continuation. We know it's a tuple and everything else as @@ -1305,18 +1345,28 @@ int db_select_count_continue_tree_common(Process *p, sc.end_condition = NIL; sc.lastobj = NULL; sc.max = 1000; - sc.keypos = tb->keypos; + sc.keypos = tb->common.keypos; if (is_big(tptr[5])) { sc.got = big_to_uint32(tptr[5]); } else { sc.got = unsigned_val(tptr[5]); } - stack = get_any_stack((DbTable*)tb, stack_container); - traverse_backwards(tb, root, stack, lastkey, &doit_select_count, &sc); - release_stack((DbTable*)tb,stack_container,stack); + if (iter) { + iter->next_route_key = lastkey; + sc.common.root = catree_find_prev_root(iter, NULL); + } + else { + sc.common.root = &tb->tree.root; + } - BUMP_REDS(p, 1000 - sc.max); + if (sc.common.root) { + stack = get_any_stack(tb, stack_container); + traverse_backwards(&tb->common, stack, lastkey, &doit_select_count, &sc.common, iter); + release_stack(tb,stack_container,stack); + + BUMP_REDS(p, 1000 - sc.max); + } if (sc.max > 0) { RET_TO_BIF(erts_make_integer(sc.got,p), DB_ERROR_NONE); @@ -1361,14 +1411,15 @@ static int db_select_count_continue_tree(Process *p, Eterm *ret) { DbTableTree *tb = &tbl->tree; - return db_select_count_continue_tree_common(p, &tb->common, &tb->root, - continuation, ret, tb); + return db_select_count_continue_tree_common(p, tbl, + continuation, ret, tb, NULL); } -int db_select_count_tree_common(Process *p, DbTableCommon *tb, TreeDbTerm **root, +int db_select_count_tree_common(Process *p, DbTable *tb, Eterm tid, Eterm pattern, Eterm *ret, - DbTableTree *stack_container) + DbTableTree *stack_container, + CATreeRootIterator* iter) { DbTreeStack* stack; struct select_count_context sc; @@ -1383,7 +1434,6 @@ int db_select_count_tree_common(Process *p, DbTableCommon *tb, TreeDbTerm **root Eterm egot; Eterm mpb; - #define RET_TO_BIF(Term,RetVal) do { \ if (mpi.mp != NULL) { \ erts_bin_free(mpi.mp); \ @@ -1398,35 +1448,48 @@ int db_select_count_tree_common(Process *p, DbTableCommon *tb, TreeDbTerm **root sc.p = p; sc.max = 1000; sc.end_condition = NIL; - sc.keypos = tb->keypos; + sc.keypos = tb->common.keypos; sc.got = 0; - if ((errcode = analyze_pattern(tb, root, pattern, NULL, &mpi)) != DB_ERROR_NONE) { + if ((errcode = analyze_pattern(&tb->common, pattern, NULL, &mpi)) != DB_ERROR_NONE) { RET_TO_BIF(NIL,errcode); } - if (!mpi.something_can_match) { + if (mpi.key_boundness == MS_KEY_IMPOSSIBLE) { RET_TO_BIF(make_small(0),DB_ERROR_NONE); /* can't possibly match anything */ } sc.mp = mpi.mp; - if (!mpi.got_partial && mpi.some_limitation && - CMP_EQ(mpi.least,mpi.most)) { - doit_select_count(tb,*(mpi.save_term),&sc,0 /* dummy */); + if (mpi.key_boundness == MS_KEY_BOUND) { + ASSERT(CMP_EQ(mpi.least, mpi.most)); + if (iter) + sc.common.root = catree_find_root(mpi.least, iter); + else + sc.common.root = &((DbTable*)tb)->tree.root; + this = find_node(&tb->common, *sc.common.root, mpi.least, NULL); + if (this) + doit_select_count(&tb->common, this, &sc.common, 0 /* dummy */); RET_TO_BIF(erts_make_integer(sc.got,p),DB_ERROR_NONE); } stack = get_any_stack((DbTable*)tb, stack_container); - if (mpi.some_limitation) { - if ((this = find_next_from_pb_key(tb, *root, stack, mpi.most)) != NULL) { - lastkey = GETKEY(tb, this->dbterm.tpl); - } + if (mpi.key_boundness == MS_KEY_PARTIALLY_BOUND) { + this = find_next_from_pb_key(tb, &sc.common.root, stack, mpi.most, iter); + if (this) + lastkey = GETKEY(tb, this->dbterm.tpl); sc.end_condition = mpi.least; } + else { + ASSERT(mpi.key_boundness == MS_KEY_UNBOUND); + if (iter) + sc.common.root = catree_find_last_root(iter); + else + sc.common.root = &tb->tree.root; + } - traverse_backwards(tb, root, stack, lastkey, &doit_select_count, &sc); + traverse_backwards(&tb->common, stack, lastkey, &doit_select_count, &sc.common, iter); release_stack((DbTable*)tb,stack_container,stack); BUMP_REDS(p, 1000 - sc.max); if (sc.max > 0) { @@ -1467,15 +1530,16 @@ static int db_select_count_tree(Process *p, DbTable *tbl, Eterm tid, Eterm pattern, Eterm *ret) { DbTableTree *tb = &tbl->tree; - return db_select_count_tree_common(p, &tb->common, &tb->root, - tid, pattern, ret, tb); + return db_select_count_tree_common(p, tbl, + tid, pattern, ret, tb, NULL); } -int db_select_chunk_tree_common(Process *p, DbTableCommon *tb, TreeDbTerm **root, +int db_select_chunk_tree_common(Process *p, DbTable *tb, Eterm tid, Eterm pattern, Sint chunk_size, int reverse, Eterm *ret, - DbTableTree *stack_container) + DbTableTree *stack_container, + CATreeRootIterator* iter) { DbTreeStack* stack; struct select_context sc; @@ -1489,7 +1553,6 @@ int db_select_chunk_tree_common(Process *p, DbTableCommon *tb, TreeDbTerm **root int errcode; Eterm mpb; - #define RET_TO_BIF(Term,RetVal) do { \ if (mpi.mp != NULL) { \ erts_bin_free(mpi.mp); \ @@ -1505,24 +1568,30 @@ int db_select_chunk_tree_common(Process *p, DbTableCommon *tb, TreeDbTerm **root sc.p = p; sc.max = 1000; sc.end_condition = NIL; - sc.keypos = tb->keypos; + sc.keypos = tb->common.keypos; sc.got = 0; sc.chunk_size = chunk_size; - if ((errcode = analyze_pattern(tb, root, pattern, NULL, &mpi)) != DB_ERROR_NONE) { + if ((errcode = analyze_pattern(&tb->common, pattern, NULL, &mpi)) != DB_ERROR_NONE) { RET_TO_BIF(NIL,errcode); } - if (!mpi.something_can_match) { + if (mpi.key_boundness == MS_KEY_IMPOSSIBLE) { RET_TO_BIF(am_EOT,DB_ERROR_NONE); /* can't possibly match anything */ } sc.mp = mpi.mp; - if (!mpi.got_partial && mpi.some_limitation && - CMP_EQ(mpi.least,mpi.most)) { - doit_select(tb,*(mpi.save_term),&sc, 0 /* direction doesn't matter */); + if (mpi.key_boundness == MS_KEY_BOUND) { + ASSERT(CMP_EQ(mpi.least, mpi.most)); + if (iter) + sc.common.root = catree_find_root(mpi.least, iter); + else + sc.common.root = &tb->tree.root; + this = find_node(&tb->common, *sc.common.root, mpi.least, NULL); + if (this) + doit_select(&tb->common, this, &sc.common, 0 /* direction doesn't matter */); if (sc.accum != NIL) { hp=HAlloc(p, 3); RET_TO_BIF(TUPLE2(hp,sc.accum,am_EOT),DB_ERROR_NONE); @@ -1533,21 +1602,35 @@ int db_select_chunk_tree_common(Process *p, DbTableCommon *tb, TreeDbTerm **root stack = get_any_stack((DbTable*)tb,stack_container); if (reverse) { - if (mpi.some_limitation) { - if ((this = find_next_from_pb_key(tb, *root, stack, mpi.most)) != NULL) { - lastkey = GETKEY(tb, this->dbterm.tpl); - } + if (mpi.key_boundness == MS_KEY_PARTIALLY_BOUND) { + this = find_next_from_pb_key(tb, &sc.common.root, stack, mpi.most, iter); + if (this) + lastkey = GETKEY(tb, this->dbterm.tpl); sc.end_condition = mpi.least; } - traverse_backwards(tb, root, stack, lastkey, &doit_select_chunk, &sc); + else { + ASSERT(mpi.key_boundness == MS_KEY_UNBOUND); + if (iter) + sc.common.root = catree_find_last_root(iter); + else + sc.common.root = &tb->tree.root; + } + traverse_backwards(&tb->common, stack, lastkey, &doit_select_chunk, &sc.common, iter); } else { - if (mpi.some_limitation) { - if ((this = find_prev_from_pb_key(tb, *root, stack, mpi.least)) != NULL) { - lastkey = GETKEY(tb, this->dbterm.tpl); - } + if (mpi.key_boundness == MS_KEY_PARTIALLY_BOUND) { + this = find_prev_from_pb_key(tb, &sc.common.root, stack, mpi.least, iter); + if (this) + lastkey = GETKEY(tb, this->dbterm.tpl); sc.end_condition = mpi.most; } - traverse_forward(tb, root, stack, lastkey, &doit_select_chunk, &sc); + else { + ASSERT(mpi.key_boundness == MS_KEY_UNBOUND); + if (iter) + sc.common.root = catree_find_first_root(iter); + else + sc.common.root = &tb->tree.root; + } + traverse_forward(&tb->common, stack, lastkey, &doit_select_chunk, &sc.common, iter); } release_stack((DbTable*)tb,stack_container,stack); @@ -1624,18 +1707,18 @@ static int db_select_chunk_tree(Process *p, DbTable *tbl, Eterm tid, Eterm *ret) { DbTableTree *tb = &tbl->tree; - return db_select_chunk_tree_common(p, &tb->common, &tb->root, + return db_select_chunk_tree_common(p, tbl, tid, pattern, chunk_size, - reverse, ret, tb); + reverse, ret, tb, NULL); } int db_select_delete_continue_tree_common(Process *p, DbTable *tbl, - TreeDbTerm **root, Eterm continuation, Eterm *ret, - DbTreeStack* stack) + DbTreeStack* stack, + CATreeRootIterator* iter) { struct select_delete_context sc; unsigned sz; @@ -1647,7 +1730,6 @@ int db_select_delete_continue_tree_common(Process *p, Eterm *tptr; Eterm eaccsum; - #define RET_TO_BIF(Term, State) do { \ if (sc.erase_lastterm) { \ free_term(tbl, sc.lastterm); \ @@ -1670,7 +1752,6 @@ int db_select_delete_continue_tree_common(Process *p, mp = erts_db_get_match_prog_binary_unchecked(tptr[4]); sc.p = p; sc.tb = &tbl->common; - sc.root = root; sc.stack = stack; if (is_big(tptr[5])) { sc.accum = big_to_uint32(tptr[5]); @@ -1682,9 +1763,19 @@ int db_select_delete_continue_tree_common(Process *p, sc.max = 1000; sc.keypos = tbl->common.keypos; - traverse_backwards(&tbl->common, root, stack, lastkey, &doit_select_delete, &sc); + if (iter) { + iter->next_route_key = lastkey; + sc.common.root = catree_find_prev_root(iter, NULL); + } + else { + sc.common.root = &tbl->tree.root; + } - BUMP_REDS(p, 1000 - sc.max); + if (sc.common.root) { + traverse_backwards(&tbl->common, stack, lastkey, &doit_select_delete, &sc.common, iter); + + BUMP_REDS(p, 1000 - sc.max); + } if (sc.max > 0) { RET_TO_BIF(erts_make_integer(sc.accum, p), DB_ERROR_NONE); @@ -1726,16 +1817,15 @@ static int db_select_delete_continue_tree(Process *p, { DbTableTree *tb = &tbl->tree; ASSERT(!erts_atomic_read_nob(&tb->is_stack_busy)); - return db_select_delete_continue_tree_common(p, tbl, &tb->root, - continuation, ret, - &tb->static_stack); + return db_select_delete_continue_tree_common(p, tbl, continuation, ret, + &tb->static_stack, NULL); } int db_select_delete_tree_common(Process *p, DbTable *tbl, - TreeDbTerm **root, Eterm tid, Eterm pattern, Eterm *ret, - DbTreeStack* stack) + DbTreeStack* stack, + CATreeRootIterator* iter) { struct select_delete_context sc; struct mp_info mpi; @@ -1770,35 +1860,48 @@ int db_select_delete_tree_common(Process *p, DbTable *tbl, sc.end_condition = NIL; sc.keypos = tbl->common.keypos; sc.tb = &tbl->common; - sc.root = root; sc.stack = stack; - if ((errcode = analyze_pattern(&tbl->common, root, pattern, NULL, &mpi)) != DB_ERROR_NONE) { + if ((errcode = analyze_pattern(&tbl->common, pattern, NULL, &mpi)) != DB_ERROR_NONE) { RET_TO_BIF(0,errcode); } - if (!mpi.something_can_match) { + if (mpi.key_boundness == MS_KEY_IMPOSSIBLE) { RET_TO_BIF(make_small(0),DB_ERROR_NONE); /* can't possibly match anything */ } sc.mp = mpi.mp; - if (!mpi.got_partial && mpi.some_limitation && - CMP_EQ(mpi.least,mpi.most)) { - doit_select_delete(&tbl->common,*(mpi.save_term),&sc, 0 /* direction doesn't + if (mpi.key_boundness == MS_KEY_BOUND) { + ASSERT(CMP_EQ(mpi.least, mpi.most)); + if (iter) + sc.common.root = catree_find_root(mpi.least, iter); + else + sc.common.root = &tbl->tree.root; + this = find_node(&tbl->common, *sc.common.root, mpi.least, NULL); + if (this) + doit_select_delete(&tbl->common, this, &sc.common, 0 /* direction doesn't matter */); RET_TO_BIF(erts_make_integer(sc.accum,p),DB_ERROR_NONE); } - if (mpi.some_limitation) { - if ((this = find_next_from_pb_key(&tbl->common, *root, stack, mpi.most)) != NULL) { - lastkey = GETKEY(&tbl->common, this->dbterm.tpl); - } + if (mpi.key_boundness == MS_KEY_PARTIALLY_BOUND) { + this = find_next_from_pb_key(tbl, &sc.common.root, stack, mpi.most, iter); + if (this) + lastkey = GETKEY(&tbl->common, this->dbterm.tpl); sc.end_condition = mpi.least; } + else { + ASSERT(mpi.key_boundness == MS_KEY_UNBOUND); + if (iter) + sc.common.root = catree_find_last_root(iter); + else + sc.common.root = &tbl->tree.root; + } - traverse_backwards(&tbl->common, root, stack, lastkey, &doit_select_delete, &sc); + traverse_backwards(&tbl->common, stack, lastkey, + &doit_select_delete, &sc.common, iter); BUMP_REDS(p, 1000 - sc.max); if (sc.max > 0) { @@ -1842,16 +1945,16 @@ static int db_select_delete_tree(Process *p, DbTable *tbl, Eterm tid, Eterm pattern, Eterm *ret) { DbTableTree *tb = &tbl->tree; - return db_select_delete_tree_common(p, tbl, &tb->root, tid, pattern, ret, - &tb->static_stack); + return db_select_delete_tree_common(p, tbl, tid, pattern, ret, + &tb->static_stack, NULL); } int db_select_replace_continue_tree_common(Process *p, - DbTableCommon *tb, - TreeDbTerm **root, + DbTable *tbl, Eterm continuation, Eterm *ret, - DbTableTree *stack_container) + DbTableTree *stack_container, + CATreeRootIterator* iter) { DbTreeStack* stack; struct select_replace_context sc; @@ -1888,7 +1991,7 @@ int db_select_replace_continue_tree_common(Process *p, sc.end_condition = NIL; sc.lastobj = NULL; sc.max = 1000; - sc.keypos = tb->keypos; + sc.keypos = tbl->common.keypos; if (is_big(tptr[5])) { sc.replaced = big_to_uint32(tptr[5]); } else { @@ -1896,9 +1999,18 @@ int db_select_replace_continue_tree_common(Process *p, } prev_replaced = sc.replaced; - stack = get_any_stack((DbTable*)tb,stack_container); - traverse_update_backwards(tb, root, stack, lastkey, &doit_select_replace, &sc); - release_stack((DbTable*)tb,stack_container,stack); + if (iter) { + iter->next_route_key = lastkey; + sc.common.root = catree_find_prev_root(iter, NULL); + } + else { + sc.common.root = &tbl->tree.root; + } + + stack = get_any_stack(tbl, stack_container); + traverse_update_backwards(&tbl->common, stack, lastkey, &doit_select_replace, + &sc.common, iter); + release_stack(tbl, stack_container,stack); // the more objects we've replaced, the more reductions we've consumed BUMP_REDS(p, MIN(2000, (1000 - sc.max) + (sc.replaced - prev_replaced))); @@ -1906,7 +2018,7 @@ int db_select_replace_continue_tree_common(Process *p, if (sc.max > 0) { RET_TO_BIF(erts_make_integer(sc.replaced,p), DB_ERROR_NONE); } - key = GETKEY(tb, sc.lastobj); + key = GETKEY(tbl, sc.lastobj); if (end_condition != NIL && (cmp_partly_bound(end_condition,key) > 0)) { /* done anyway */ @@ -1942,14 +2054,14 @@ static int db_select_replace_continue_tree(Process *p, Eterm continuation, Eterm *ret) { - DbTableTree *tb = &tbl->tree; - return db_select_replace_continue_tree_common(p, &tb->common, &tb->root, - continuation, ret, tb); + return db_select_replace_continue_tree_common(p, tbl, continuation, ret, + &tbl->tree, NULL); } -int db_select_replace_tree_common(Process *p, DbTableCommon *tb, TreeDbTerm **root, +int db_select_replace_tree_common(Process *p, DbTable *tbl, Eterm tid, Eterm pattern, Eterm *ret, - DbTableTree *stack_container) + DbTableTree *stack_container, + CATreeRootIterator* iter) { DbTreeStack* stack; struct select_replace_context sc; @@ -1977,57 +2089,64 @@ int db_select_replace_tree_common(Process *p, DbTableCommon *tb, TreeDbTerm **ro sc.lastobj = NULL; sc.p = p; - sc.tb = tb; - sc.root = root; + sc.tb = &tbl->common; sc.max = 1000; sc.end_condition = NIL; - sc.keypos = tb->keypos; + sc.keypos = tbl->common.keypos; sc.replaced = 0; - if ((errcode = analyze_pattern(tb, root, pattern, db_match_keeps_key, &mpi)) != DB_ERROR_NONE) { + if ((errcode = analyze_pattern(&tbl->common, pattern, db_match_keeps_key, &mpi)) != DB_ERROR_NONE) { RET_TO_BIF(NIL,errcode); } - if (!mpi.something_can_match) { + if (mpi.key_boundness == MS_KEY_IMPOSSIBLE) { RET_TO_BIF(make_small(0),DB_ERROR_NONE); /* can't possibly match anything */ } sc.mp = mpi.mp; - stack = get_static_stack(stack_container); - if (!mpi.got_partial && mpi.some_limitation && - CMP_EQ(mpi.least,mpi.most)) { - TreeDbTerm* term = *(mpi.save_term); - doit_select_replace(tb,mpi.save_term,&sc,0 /* dummy */); - if (stack != NULL) { - if (TOP_NODE(stack) == term) - // throw away potentially invalid reference - REPLACE_TOP_NODE(stack, *(mpi.save_term)); - release_stack((DbTable*)tb,stack_container, stack); + if (mpi.key_boundness == MS_KEY_BOUND) { + TreeDbTerm** pp; + ASSERT(CMP_EQ(mpi.least, mpi.most)); + if (iter) + sc.common.root = catree_find_root(mpi.least, iter); + else + sc.common.root = &tbl->tree.root; + pp = find_node2(&tbl->common, sc.common.root, mpi.least); + if (pp) { + doit_select_replace(&tbl->common, pp, &sc.common, 0 /* dummy */); + reset_static_stack(stack_container); /* may refer replaced term */ } RET_TO_BIF(erts_make_integer(sc.replaced,p),DB_ERROR_NONE); } - if (stack == NULL) - stack = get_any_stack((DbTable*)tb,stack_container); + stack = get_any_stack(tbl,stack_container); - if (mpi.some_limitation) { - if ((this = find_next_from_pb_key(tb, *root, stack, mpi.most)) != NULL) { - lastkey = GETKEY(tb, this->dbterm.tpl); - } + if (mpi.key_boundness == MS_KEY_PARTIALLY_BOUND) { + this = find_next_from_pb_key(tbl, &sc.common.root, stack, mpi.most, iter); + if (this) + lastkey = GETKEY(tbl, this->dbterm.tpl); sc.end_condition = mpi.least; } + else { + ASSERT(mpi.key_boundness == MS_KEY_UNBOUND); + if (iter) + sc.common.root = catree_find_last_root(iter); + else + sc.common.root = &tbl->tree.root; + } - traverse_update_backwards(tb, root, stack, lastkey, &doit_select_replace, &sc); - release_stack((DbTable*)tb,stack_container,stack); + traverse_update_backwards(&tbl->common, stack, lastkey, &doit_select_replace, + &sc.common, iter); + release_stack(tbl,stack_container,stack); // the more objects we've replaced, the more reductions we've consumed BUMP_REDS(p, MIN(2000, (1000 - sc.max) + sc.replaced)); if (sc.max > 0) { RET_TO_BIF(erts_make_integer(sc.replaced,p),DB_ERROR_NONE); } - key = GETKEY(tb, sc.lastobj); + key = GETKEY(tbl, sc.lastobj); sz = size_object(key); if (IS_USMALL(0, sc.replaced)) { hp = HAlloc(p, sz + ERTS_MAGIC_REF_THING_SIZE + 6); @@ -2060,9 +2179,8 @@ int db_select_replace_tree_common(Process *p, DbTableCommon *tb, TreeDbTerm **ro static int db_select_replace_tree(Process *p, DbTable *tbl, Eterm tid, Eterm pattern, Eterm *ret) { - DbTableTree *tb = &tbl->tree; - return db_select_replace_tree_common(p, &tb->common, &tb->root, - tid, pattern, ret, tb); + return db_select_replace_tree_common(p, tbl, tid, pattern, ret, + &tbl->tree, NULL); } int db_take_tree_common(Process *p, DbTable *tbl, TreeDbTerm **root, @@ -2336,7 +2454,7 @@ static TreeDbTerm *linkout_object_tree(DbTableCommon *tb, TreeDbTerm **root, ** For the select functions, analyzes the pattern and determines which ** part of the tree should be searched. Also compiles the match program */ -static int analyze_pattern(DbTableCommon *tb, TreeDbTerm **root, Eterm pattern, +static int analyze_pattern(DbTableCommon *tb, Eterm pattern, extra_match_validator_t extra_validator, /* Optional callback */ struct mp_info *mpi) { @@ -2347,17 +2465,12 @@ static int analyze_pattern(DbTableCommon *tb, TreeDbTerm **root, Eterm pattern, Eterm *ptpl; int i; int num_heads = 0; - Eterm key; - Eterm partly_bound; - int res; - Eterm least = 0; - Eterm most = 0; + Eterm least = THE_NON_VALUE; + Eterm most = THE_NON_VALUE; + enum ms_key_boundness boundness; - mpi->some_limitation = 1; - mpi->got_partial = 0; - mpi->something_can_match = 0; + mpi->key_boundness = MS_KEY_IMPOSSIBLE; mpi->mp = NULL; - mpi->save_term = NULL; for (lst = pattern; is_list(lst); lst = CDR(list_val(lst))) ++num_heads; @@ -2378,6 +2491,7 @@ static int analyze_pattern(DbTableCommon *tb, TreeDbTerm **root, Eterm pattern, Eterm match; Eterm guard; Eterm body; + Eterm key; ttpl = CAR(list_val(lst)); if (!is_tuple(ttpl)) { @@ -2409,30 +2523,29 @@ static int analyze_pattern(DbTableCommon *tb, TreeDbTerm **root, Eterm pattern, } ++i; - partly_bound = NIL; - res = key_given(tb, root, tpl, &(mpi->save_term), &partly_bound); - if ( res >= 0 ) { /* Can match something */ - key = 0; - mpi->something_can_match = 1; - if (res > 0) { - key = GETKEY(tb,tuple_val(tpl)); - } else if (partly_bound != NIL) { - mpi->got_partial = 1; - key = partly_bound; - } else { - mpi->some_limitation = 0; - } - if (key != 0) { - if (least == 0 || - partly_bound_can_match_lesser(key,least)) { - least = key; - } - if (most == 0 || - partly_bound_can_match_greater(key,most)) { - most = key; - } - } - } + boundness = key_boundness(tb, tpl, &key); + switch (boundness) + { + case MS_KEY_BOUND: + case MS_KEY_PARTIALLY_BOUND: + if (is_non_value(least) || partly_bound_can_match_lesser(key,least)) { + least = key; + } + if (is_non_value(most) || partly_bound_can_match_greater(key,most)) { + most = key; + } + break; + case MS_KEY_IMPOSSIBLE: + case MS_KEY_UNBOUND: + break; + } + if (mpi->key_boundness > boundness) + mpi->key_boundness = boundness; + } + + if (mpi->key_boundness == MS_KEY_BOUND && !CMP_EQ(least, most)) { + /* Several different bound keys */ + mpi->key_boundness = MS_KEY_PARTIALLY_BOUND; } mpi->least = least; mpi->most = most; @@ -2616,11 +2729,27 @@ static int delsub(TreeDbTerm **this) static TreeDbTerm *slot_search(Process *p, TreeDbTerm *root, Sint slot, DbTable *tb, - DbTableTree *stack_container) + DbTableTree *stack_container, + CATreeRootIterator *iter) { TreeDbTerm *this; TreeDbTerm *tmp; - DbTreeStack* stack = get_any_stack(tb,stack_container); + TreeDbTerm *lastobj; + Eterm lastkey; + TreeDbTerm **pp; + DbTreeStack* stack; + + if (iter) { + /* Find first non-empty tree */ + while (!root) { + TreeDbTerm** pp = catree_find_next_root(iter, NULL); + if (!pp) + return NULL; + root = *pp; + } + } + + stack = get_any_stack(tb,stack_container); ASSERT(stack != NULL); if (slot == 1) { /* Don't search from where we are if we are @@ -2632,56 +2761,83 @@ static TreeDbTerm *slot_search(Process *p, TreeDbTerm *root, are not recorded */ stack->pos = 0; } - if (EMPTY_NODE(stack)) { - this = root; - if (this == NULL) - goto done; - while (this->left != NULL){ - PUSH_NODE(stack, this); - this = this->left; - } - PUSH_NODE(stack, this); - stack->slot = 1; - } - this = TOP_NODE(stack); - while (stack->slot != slot && this != NULL) { - if (slot > stack->slot) { - if (this->right != NULL) { - this = this->right; - while (this->left != NULL) { - PUSH_NODE(stack, this); - this = this->left; - } - PUSH_NODE(stack, this); - } else { - for (;;) { - tmp = POP_NODE(stack); - this = TOP_NODE(stack); - if (this == NULL || this->left == tmp) - break; - } - } - ++(stack->slot); - } else { - if (this->left != NULL) { - this = this->left; - while (this->right != NULL) { - PUSH_NODE(stack, this); - this = this->right; - } - PUSH_NODE(stack, this); - } else { - for (;;) { - tmp = POP_NODE(stack); - this = TOP_NODE(stack); - if (this == NULL || this->right == tmp) - break; - } - } - --(stack->slot); - } + while (1) { + if (EMPTY_NODE(stack)) { + this = root; + if (this == NULL) + goto next_root; + while (this->left != NULL){ + PUSH_NODE(stack, this); + this = this->left; + } + PUSH_NODE(stack, this); + stack->slot++; + } + this = TOP_NODE(stack); + while (stack->slot != slot) { + ASSERT(this); + lastobj = this; + if (slot > stack->slot) { + if (this->right != NULL) { + this = this->right; + while (this->left != NULL) { + PUSH_NODE(stack, this); + this = this->left; + } + PUSH_NODE(stack, this); + } else { + for (;;) { + tmp = POP_NODE(stack); + this = TOP_NODE(stack); + if (!this) + goto next_root; + if (this->left == tmp) + break; + } + } + ++(stack->slot); + } else { + if (this->left != NULL) { + this = this->left; + while (this->right != NULL) { + PUSH_NODE(stack, this); + this = this->right; + } + PUSH_NODE(stack, this); + } else { + for (;;) { + tmp = POP_NODE(stack); + this = TOP_NODE(stack); + if (!this) + goto next_root; + if (this->right == tmp) + break; + } + } + --(stack->slot); + } + } + /* Found slot */ + ASSERT(this); + break; + +next_root: + if (!iter) + break; /* EOT */ + + ASSERT(slot > stack->slot); + if (lastobj) { + lastkey = GETKEY(tb, lastobj->dbterm.tpl); + lastobj = NULL; + } + pp = catree_find_next_root(iter, &lastkey); + if (!pp) + break; /* EOT */ + root = *pp; + stack->pos = 0; + find_next(&tb->common, root, stack, lastkey); } -done: + release_stack(tb,stack_container,stack); return this; } @@ -2716,7 +2872,7 @@ static TreeDbTerm *find_next(DbTableCommon *tb, TreeDbTerm *root, this = this->right; } else if (c < 0) { if (this->left == NULL) /* Done */ - return this; + goto found_next; else this = this->left; } else @@ -2731,8 +2887,6 @@ static TreeDbTerm *find_next(DbTableCommon *tb, TreeDbTerm *root, this = this->left; PUSH_NODE(stack, this); } - if (stack->slot > 0) - ++(stack->slot); } else { do { tmp = POP_NODE(stack); @@ -2741,9 +2895,12 @@ static TreeDbTerm *find_next(DbTableCommon *tb, TreeDbTerm *root, return NULL; } } while (this->right == tmp); - if (stack->slot > 0) - ++(stack->slot); } + +found_next: + if (stack->slot > 0) + ++(stack->slot); + return this; } @@ -2773,7 +2930,7 @@ static TreeDbTerm *find_prev(DbTableCommon *tb, TreeDbTerm *root, this = this->left; } else if (c > 0) { if (this->right == NULL) /* Done */ - return this; + goto found_prev; else this = this->right; } else @@ -2788,8 +2945,6 @@ static TreeDbTerm *find_prev(DbTableCommon *tb, TreeDbTerm *root, this = this->right; PUSH_NODE(stack, this); } - if (stack->slot > 0) - --(stack->slot); } else { do { tmp = POP_NODE(stack); @@ -2798,74 +2953,112 @@ static TreeDbTerm *find_prev(DbTableCommon *tb, TreeDbTerm *root, return NULL; } } while (this->left == tmp); - if (stack->slot > 0) - --(stack->slot); } + +found_prev: + if (stack->slot > 0) + --(stack->slot); + return this; } -static TreeDbTerm *find_next_from_pb_key(DbTableCommon *tb, TreeDbTerm *root, - DbTreeStack* stack, Eterm key) + +/* @brief Find object with smallest key of all larger than partially bound key. + * Can be used as a starting point for a reverse iteration with pb_key. + * + * @param pb_key The partially bound key. Example {42, '$1'} + * @param *rootpp Will return pointer to root pointer of tree with found object. + * @param iter Root iterator or NULL for plain DbTableTree. + * @param stack A stack to use. Will be cleared. + * + * @return found object or NULL if no such key exists. + */ +static TreeDbTerm *find_next_from_pb_key(DbTable *tbl, TreeDbTerm*** rootpp, + DbTreeStack* stack, Eterm pb_key, + CATreeRootIterator* iter) { + TreeDbTerm* root; TreeDbTerm *this; - TreeDbTerm *tmp; + Uint candidate = 0; Sint c; + if (iter) { + *rootpp = catree_find_next_from_pb_key_root(pb_key, iter); + ASSERT(*rootpp); + root = **rootpp; + } + else { + *rootpp = &tbl->tree.root; + root = tbl->tree.root; + } + /* spool the stack, we have to "re-search" */ stack->pos = stack->slot = 0; if (( this = root ) == NULL) return NULL; for (;;) { PUSH_NODE(stack, this); - if (( c = cmp_partly_bound(key,GETKEY(tb, this->dbterm.tpl))) >= 0) { + if (( c = cmp_partly_bound(pb_key,GETKEY(tbl, this->dbterm.tpl))) >= 0) { if (this->right == NULL) { - do { - tmp = POP_NODE(stack); - if (( this = TOP_NODE(stack)) == NULL) { - return NULL; - } - } while (this->right == tmp); - return this; - } else - this = this->right; + stack->pos = candidate; + return TOP_NODE(stack); + } + this = this->right; } else /*if (c < 0)*/ { if (this->left == NULL) /* Done */ return this; - else - this = this->left; + candidate = stack->pos; + this = this->left; } } } -static TreeDbTerm *find_prev_from_pb_key(DbTableCommon *tb, TreeDbTerm *root, - DbTreeStack* stack, Eterm key) +/* @brief Find object with largest key of all smaller than partially bound key. + * Can be used as a starting point for a forward iteration with pb_key. + * + * @param pb_key The partially bound key. Example {42, '$1'} + * @param *rootpp Will return pointer to root pointer of found object. + * @param iter Root iterator or NULL for plain DbTableTree. + * @param stack A stack to use. Will be cleared. + * + * @return found object or NULL if no such key exists. + */ +static TreeDbTerm *find_prev_from_pb_key(DbTable *tbl, TreeDbTerm*** rootpp, + DbTreeStack* stack, Eterm pb_key, + CATreeRootIterator* iter) { + TreeDbTerm* root; TreeDbTerm *this; - TreeDbTerm *tmp; + Uint candidate = 0; Sint c; + if (iter) { + *rootpp = catree_find_prev_from_pb_key_root(pb_key, iter); + ASSERT(*rootpp); + root = **rootpp; + } + else { + *rootpp = &tbl->tree.root; + root = tbl->tree.root; + } + /* spool the stack, we have to "re-search" */ stack->pos = stack->slot = 0; if (( this = root ) == NULL) return NULL; for (;;) { PUSH_NODE(stack, this); - if (( c = cmp_partly_bound(key,GETKEY(tb, this->dbterm.tpl))) <= 0) { + if (( c = cmp_partly_bound(pb_key,GETKEY(tbl, this->dbterm.tpl))) <= 0) { if (this->left == NULL) { - do { - tmp = POP_NODE(stack); - if (( this = TOP_NODE(stack)) == NULL) { - return NULL; - } - } while (this->left == tmp); - return this; - } else - this = this->left; - } else /*if (c < 0)*/ { + stack->pos = candidate; + return TOP_NODE(stack); + } + this = this->left; + } else /*if (c > 0)*/ { if (this->right == NULL) /* Done */ return this; - else - this = this->right; + candidate = stack->pos; + this = this->right; } } } @@ -3054,38 +3247,53 @@ db_finalize_dbterm_tree(int cret, DbUpdateHandle *handle) * Traverse the tree with a callback function, used by db_match_xxx */ static void traverse_backwards(DbTableCommon *tb, - TreeDbTerm **root, DbTreeStack* stack, Eterm lastkey, - int (*doit)(DbTableCommon *, - TreeDbTerm *, - void *, - int), - void *context) + traverse_doit_funcT* doit, + struct select_common *context, + CATreeRootIterator* iter) { TreeDbTerm *this, *next; + TreeDbTerm** root = context->root; if (lastkey == THE_NON_VALUE) { - stack->pos = stack->slot = 0; - if (( this = *root ) == NULL) { - return; - } - while (this != NULL) { - PUSH_NODE(stack, this); - this = this->right; - } - this = TOP_NODE(stack); - next = find_prev(tb, *root, stack, GETKEY(tb, this->dbterm.tpl)); - if (!((*doit)(tb, this, context, 0))) - return; + if (iter) { + while (*root == NULL) { + root = catree_find_prev_root(iter, NULL); + if (!root) + return; + } + context->root = root; + } + stack->pos = stack->slot = 0; + next = *root; + while (next != NULL) { + PUSH_NODE(stack, next); + next = next->right; + } + next = TOP_NODE(stack); } else { - next = find_prev(tb, *root, stack, lastkey); + next = find_prev(tb, *root, stack, lastkey); } - while ((this = next) != NULL) { - next = find_prev(tb, *root, stack, GETKEY(tb, this->dbterm.tpl)); - if (!((*doit)(tb, this, context, 0))) - return; + while (1) { + while (next) { + this = next; + lastkey = GETKEY(tb, this->dbterm.tpl); + next = find_prev(tb, *root, stack, lastkey); + if (!((*doit)(tb, this, context, 0))) + return; + } + + if (!iter) + return; + ASSERT(is_value(lastkey)); + root = catree_find_prev_root(iter, &lastkey); + if (!root) + return; + context->root = root; + stack->pos = stack->slot = 0; + next = find_prev(tb, *root, stack, lastkey); } } @@ -3093,38 +3301,53 @@ static void traverse_backwards(DbTableCommon *tb, * Traverse the tree with a callback function, used by db_match_xxx */ static void traverse_forward(DbTableCommon *tb, - TreeDbTerm **root, DbTreeStack* stack, Eterm lastkey, - int (*doit)(DbTableCommon *, - TreeDbTerm *, - void *, - int), - void *context) + traverse_doit_funcT* doit, + struct select_common *context, + CATreeRootIterator* iter) { TreeDbTerm *this, *next; + TreeDbTerm **root = context->root; if (lastkey == THE_NON_VALUE) { - stack->pos = stack->slot = 0; - if (( this = *root ) == NULL) { - return; - } - while (this != NULL) { - PUSH_NODE(stack, this); - this = this->left; - } - this = TOP_NODE(stack); - next = find_next(tb, *root, stack, GETKEY(tb, this->dbterm.tpl)); - if (!((*doit)(tb, this, context, 1))) - return; + if (iter) { + while (*root == NULL) { + root = catree_find_next_root(iter, NULL); + if (!root) + return; + } + context->root = root; + } + stack->pos = stack->slot = 0; + next = *root; + while (next != NULL) { + PUSH_NODE(stack, next); + next = next->left; + } + next = TOP_NODE(stack); } else { - next = find_next(tb, *root, stack, lastkey); + next = find_next(tb, *root, stack, lastkey); } - while ((this = next) != NULL) { - next = find_next(tb, *root, stack, GETKEY(tb, this->dbterm.tpl)); - if (!((*doit)(tb, this, context, 1))) - return; + while (1) { + while (next) { + this = next; + lastkey = GETKEY(tb, this->dbterm.tpl); + next = find_next(tb, *root, stack, lastkey); + if (!((*doit)(tb, this, context, 1))) + return; + } + + if (!iter) + return; + ASSERT(is_value(lastkey)); + root = catree_find_next_root(iter, &lastkey); + if (!root) + return; + context->root = root; + stack->pos = stack->slot = 0; + next = find_next(tb, *root, stack, lastkey); } } @@ -3132,77 +3355,87 @@ static void traverse_forward(DbTableCommon *tb, * Traverse the tree with an update callback function, used by db_select_replace */ static void traverse_update_backwards(DbTableCommon *tb, - TreeDbTerm **root, DbTreeStack* stack, Eterm lastkey, int (*doit)(DbTableCommon*, TreeDbTerm**, - void*, + struct select_common*, int), - void* context) + struct select_common* context, + CATreeRootIterator* iter) { int res; TreeDbTerm *this, *next, **this_ptr; + TreeDbTerm** root = context->root; if (lastkey == THE_NON_VALUE) { - stack->pos = stack->slot = 0; - if (( this = *root ) == NULL) { - return; + if (iter) { + while (*root == NULL) { + root = catree_find_prev_root(iter, NULL); + if (!root) + return; + context->root = root; + } } - while (this != NULL) { - PUSH_NODE(stack, this); - this = this->right; + stack->pos = stack->slot = 0; + next = *root; + while (next) { + PUSH_NODE(stack, next); + next = next->right; } - this = TOP_NODE(stack); - this_ptr = find_ptr(tb, root, stack, this); - ASSERT(this_ptr != NULL); - res = (*doit)(tb, this_ptr, context, 0); - REPLACE_TOP_NODE(stack, *this_ptr); - next = find_prev(tb, *root, stack, GETKEY(tb, (*this_ptr)->dbterm.tpl)); - if (!res) - return; - } else { - next = find_prev(tb, *root, stack, lastkey); + next = TOP_NODE(stack); } + else + next = find_prev(tb, *root, stack, lastkey); + - while ((this = next) != NULL) { - this_ptr = find_ptr(tb, root, stack, this); - ASSERT(this_ptr != NULL); - res = (*doit)(tb, this_ptr, context, 0); - REPLACE_TOP_NODE(stack, *this_ptr); - next = find_prev(tb, *root, stack, GETKEY(tb, (*this_ptr)->dbterm.tpl)); - if (!res) + while (1) { + while (next) { + this = next; + this_ptr = find_ptr(tb, root, stack, this); + ASSERT(this_ptr != NULL); + res = (*doit)(tb, this_ptr, context, 0); + this = *this_ptr; + REPLACE_TOP_NODE(stack, this); + if (!res) + return; + lastkey = GETKEY(tb, this->dbterm.tpl); + next = find_prev(tb, *root, stack, lastkey); + } + + if (!iter) + return; + ASSERT(is_value(lastkey)); + root = catree_find_prev_root(iter, &lastkey); + if (!root) return; + context->root = root; + stack->pos = stack->slot = 0; + next = find_prev(tb, *root, stack, lastkey); } } -/* - * Returns 0 if not given 1 if given and -1 on no possible match - * if key is given; *ret is set to point to the object concerned. - */ -static int key_given(DbTableCommon *tb, TreeDbTerm **root, Eterm pattern, - TreeDbTerm ***ret, Eterm *partly_bound) +static enum ms_key_boundness key_boundness(DbTableCommon *tb, + Eterm pattern, Eterm *keyp) { - TreeDbTerm **this; Eterm key; - ASSERT(ret != NULL); if (pattern == am_Underscore || db_is_variable(pattern) != -1) - return 0; + return MS_KEY_UNBOUND; key = db_getkey(tb->keypos, pattern); if (is_non_value(key)) - return -1; /* can't possibly match anything */ + return MS_KEY_IMPOSSIBLE; /* can't possibly match anything */ if (!db_has_variable(key)) { /* Bound key */ - if (( this = find_node2(tb, root, key) ) == NULL) { - return -1; - } - *ret = this; - return 1; - } else if (partly_bound != NULL && key != am_Underscore && - db_is_variable(key) < 0 && !db_has_map(key)) - *partly_bound = key; + *keyp = key; + return MS_KEY_BOUND; + } else if (key != am_Underscore && + db_is_variable(key) < 0 && !db_has_map(key)) { + + *keyp = key; + return MS_KEY_PARTIALLY_BOUND; + } - return 0; + return MS_KEY_UNBOUND; } @@ -3270,7 +3503,8 @@ static Sint do_cmp_partly_bound(Eterm a, Eterm b, int *done) } } -static Sint cmp_partly_bound(Eterm partly_bound_key, Eterm bound_key) { +Sint cmp_partly_bound(Eterm partly_bound_key, Eterm bound_key) +{ int done = 0; Sint ret = do_cmp_partly_bound(partly_bound_key, bound_key, &done); #ifdef HARDDEBUG @@ -3486,7 +3720,8 @@ static int do_partly_bound_can_match_greater(Eterm a, Eterm b, * Callback functions for the different match functions */ -static int doit_select(DbTableCommon *tb, TreeDbTerm *this, void *ptr, +static int doit_select(DbTableCommon *tb, TreeDbTerm *this, + struct select_common* ptr, int forward) { struct select_context *sc = (struct select_context *) ptr; @@ -3521,7 +3756,8 @@ static int doit_select(DbTableCommon *tb, TreeDbTerm *this, void *ptr, return 1; } -static int doit_select_count(DbTableCommon *tb, TreeDbTerm *this, void *ptr, +static int doit_select_count(DbTableCommon *tb, TreeDbTerm *this, + struct select_common* ptr, int forward) { struct select_count_context *sc = (struct select_count_context *) ptr; @@ -3545,7 +3781,8 @@ static int doit_select_count(DbTableCommon *tb, TreeDbTerm *this, void *ptr, return 1; } -static int doit_select_chunk(DbTableCommon *tb, TreeDbTerm *this, void *ptr, +static int doit_select_chunk(DbTableCommon *tb, TreeDbTerm *this, + struct select_common* ptr, int forward) { struct select_context *sc = (struct select_context *) ptr; @@ -3583,7 +3820,8 @@ static int doit_select_chunk(DbTableCommon *tb, TreeDbTerm *this, void *ptr, } -static int doit_select_delete(DbTableCommon *tb, TreeDbTerm *this, void *ptr, +static int doit_select_delete(DbTableCommon *tb, TreeDbTerm *this, + struct select_common *ptr, int forward) { struct select_delete_context *sc = (struct select_delete_context *) ptr; @@ -3602,7 +3840,7 @@ static int doit_select_delete(DbTableCommon *tb, TreeDbTerm *this, void *ptr, ret = db_match_dbterm(tb, sc->p, sc->mp, &this->dbterm, NULL, 0); if (ret == am_true) { key = GETKEY(sc->tb, this->dbterm.tpl); - linkout_tree(sc->tb, sc->root, key, sc->stack); + linkout_tree(sc->tb, sc->common.root, key, sc->stack); sc->erase_lastterm = 1; ++sc->accum; } @@ -3612,7 +3850,8 @@ static int doit_select_delete(DbTableCommon *tb, TreeDbTerm *this, void *ptr, return 1; } -static int doit_select_replace(DbTableCommon *tb, TreeDbTerm **this, void *ptr, +static int doit_select_replace(DbTableCommon *tb, TreeDbTerm **this, + struct select_common* ptr, int forward) { struct select_replace_context *sc = (struct select_replace_context *) ptr; diff --git a/erts/emulator/beam/erl_db_tree_util.h b/erts/emulator/beam/erl_db_tree_util.h index fbd9a9124a..02df74678d 100644 --- a/erts/emulator/beam/erl_db_tree_util.h +++ b/erts/emulator/beam/erl_db_tree_util.h @@ -92,49 +92,54 @@ int db_erase_object_tree_common(DbTable *tbl, TreeDbTerm **root, Eterm object, Eterm *ret, DbTableTree *stack_container); int db_slot_tree_common(Process *p, DbTable *tbl, TreeDbTerm *root, Eterm slot_term, Eterm *ret, - DbTableTree *stack_container); -int db_select_chunk_tree_common(Process *p, DbTableCommon *tb, TreeDbTerm **root, + DbTableTree *stack_container, + CATreeRootIterator*); +int db_select_chunk_tree_common(Process *p, DbTable *tb, Eterm tid, Eterm pattern, Sint chunk_size, int reverse, Eterm *ret, - DbTableTree *stack_container); -int db_select_tree_common(Process *p, DbTableCommon *tb, TreeDbTerm **root, + DbTableTree *stack_container, + CATreeRootIterator*); +int db_select_tree_common(Process *p, DbTable *tb, Eterm tid, Eterm pattern, int reverse, Eterm *ret, - DbTableTree *stack_container); + DbTableTree *stack_container, + CATreeRootIterator*); int db_select_delete_tree_common(Process *p, DbTable *tbl, - TreeDbTerm **root, Eterm tid, Eterm pattern, Eterm *ret, - DbTreeStack* stack); + DbTreeStack* stack, + CATreeRootIterator* iter); int db_select_continue_tree_common(Process *p, DbTableCommon *tb, - TreeDbTerm **root, Eterm continuation, Eterm *ret, - DbTableTree *stack_container); + DbTableTree *stack_container, + CATreeRootIterator* iter); int db_select_delete_continue_tree_common(Process *p, DbTable *tbl, - TreeDbTerm **root, Eterm continuation, Eterm *ret, - DbTreeStack* stack); -int db_select_count_tree_common(Process *p, DbTableCommon *tb, TreeDbTerm **root, + DbTreeStack* stack, + CATreeRootIterator* iter); +int db_select_count_tree_common(Process *p, DbTable *tb, Eterm tid, Eterm pattern, Eterm *ret, - DbTableTree *stack_container); + DbTableTree *stack_container, + CATreeRootIterator* iter); int db_select_count_continue_tree_common(Process *p, - DbTableCommon *tb, - TreeDbTerm **root, + DbTable *tb, Eterm continuation, Eterm *ret, - DbTableTree *stack_container); -int db_select_replace_tree_common(Process *p, DbTableCommon *tb, TreeDbTerm **root, + DbTableTree *stack_container, + CATreeRootIterator* iter); +int db_select_replace_tree_common(Process *p, DbTable*, Eterm tid, Eterm pattern, Eterm *ret, - DbTableTree *stack_container); + DbTableTree *stack_container, + CATreeRootIterator* iter); int db_select_replace_continue_tree_common(Process *p, - DbTableCommon *tb, - TreeDbTerm **root, + DbTable*, Eterm continuation, Eterm *ret, - DbTableTree *stack_container); + DbTableTree *stack_container, + CATreeRootIterator* iter); int db_take_tree_common(Process *p, DbTable *tbl, TreeDbTerm **root, Eterm key, Eterm *ret, DbTreeStack *stack /* NULL if no static stack */); @@ -148,4 +153,6 @@ int db_lookup_dbterm_tree_common(Process *p, DbTable *tbl, TreeDbTerm **root, DbTableTree *stack_container); void db_finalize_dbterm_tree_common(int cret, DbUpdateHandle *handle, DbTableTree *stack_container); +Sint cmp_partly_bound(Eterm partly_bound_key, Eterm bound_key); + #endif /* _DB_TREE_UTIL_H */ diff --git a/erts/emulator/beam/erl_db_util.h b/erts/emulator/beam/erl_db_util.h index be74bc795c..e1af9210ea 100644 --- a/erts/emulator/beam/erl_db_util.h +++ b/erts/emulator/beam/erl_db_util.h @@ -89,9 +89,16 @@ typedef struct { void** bp; /* {Hash|Tree}DbTerm** */ Uint new_size; int flags; - void* lck; - void* lck2; - int current_level; + union { + struct { + erts_rwmtx_t* lck; + } hash; + struct { + struct DbTableCATreeNode* base_node; + struct DbTableCATreeNode* parent; + int current_level; + } catree; + } u; } DbUpdateHandle; @@ -290,6 +297,8 @@ typedef struct db_table_common { #define DB_NAMED_TABLE (1 << 11) #define DB_BUSY (1 << 12) +#define DB_CATREE_FORCE_SPLIT (1 << 31) /* erts_debug */ + #define IS_HASH_TABLE(Status) (!!((Status) & \ (DB_BAG | DB_SET | DB_DUPLICATE_BAG))) #define IS_TREE_TABLE(Status) (!!((Status) & \ diff --git a/erts/emulator/beam/erl_dirty_bif.tab b/erts/emulator/beam/erl_dirty_bif.tab index 086275fbe5..20299ff604 100644 --- a/erts/emulator/beam/erl_dirty_bif.tab +++ b/erts/emulator/beam/erl_dirty_bif.tab @@ -59,8 +59,6 @@ dirty-cpu erts_debug:lcnt_clear/0 dirty-cpu-test erlang:'++'/2 dirty-cpu-test erlang:append/2 -dirty-cpu-test erlang:'--'/2 -dirty-cpu-test erlang:subtract/2 dirty-cpu-test erlang:iolist_size/1 dirty-cpu-test erlang:make_tuple/2 dirty-cpu-test erlang:make_tuple/3 diff --git a/erts/emulator/beam/erl_gc.c b/erts/emulator/beam/erl_gc.c index a65dbbf42b..47dd115c82 100644 --- a/erts/emulator/beam/erl_gc.c +++ b/erts/emulator/beam/erl_gc.c @@ -1133,9 +1133,28 @@ erts_garbage_collect_literals(Process* p, Eterm* literals, reds = (Sint64) garbage_collect(p, ERTS_INVALID_HFRAG_PTR, 0, p->arg_reg, p->arity, fcalls, ygen_usage); + if (ERTS_PROC_IS_EXITING(p)) { + return 0; + } ASSERT(!(p->flags & (F_DIRTY_MAJOR_GC|F_DIRTY_MINOR_GC))); + if (MAX_HEAP_SIZE_GET(p)) { + Uint new_heap_size; + Uint old_heap_size; + Uint total_heap_size; + + new_heap_size = HEAP_END(p) - HEAP_START(p); + old_heap_size = erts_next_heap_size(lit_size, 0); + total_heap_size = new_heap_size + old_heap_size; + if (MAX_HEAP_SIZE_GET(p) < total_heap_size && + reached_max_heap_size(p, total_heap_size, + new_heap_size, old_heap_size)) { + erts_set_self_exiting(p, am_killed); + return 0; + } + } + /* * Set GC state. */ diff --git a/erts/emulator/beam/erl_init.c b/erts/emulator/beam/erl_init.c index 5da7b43b9e..dcf9f90c99 100644 --- a/erts/emulator/beam/erl_init.c +++ b/erts/emulator/beam/erl_init.c @@ -163,7 +163,6 @@ int erts_initialized = 0; * Configurable parameters. */ -Uint display_items; /* no of items to display in traces etc */ int H_MIN_SIZE; /* The minimum heap grain */ int BIN_VH_MIN_SIZE; /* The minimum binary virtual*/ int H_MAX_SIZE; /* The maximum heap size */ @@ -354,6 +353,7 @@ erl_init(int ncpu, erts_init_bif(); erts_init_bif_chksum(); erts_init_bif_binary(); + erts_init_bif_persistent_term(); erts_init_bif_re(); erts_init_unicode(); /* after RE to get access to PCRE unicode */ erts_init_external(); @@ -375,21 +375,19 @@ erl_init(int ncpu, } static Eterm -erl_first_process_otp(char* modname, void* code, unsigned size, int argc, char** argv) +erl_first_process_otp(char* mod_name, int argc, char** argv) { int i; - Eterm start_mod; Eterm args; Eterm res; Eterm* hp; Process parent; ErlSpawnOpts so; - Eterm env; - - start_mod = erts_atom_put((byte *) modname, sys_strlen(modname), ERTS_ATOM_ENC_LATIN1, 1); - if (erts_find_function(start_mod, am_start, 2, + Eterm boot_mod; + + if (erts_find_function(am_erl_init, am_start, 2, erts_active_code_ix()) == NULL) { - erts_exit(ERTS_ERROR_EXIT, "No function %s:start/2\n", modname); + erts_exit(ERTS_ERROR_EXIT, "No function erl_init:start/2\n"); } /* @@ -405,13 +403,13 @@ erl_first_process_otp(char* modname, void* code, unsigned size, int argc, char** args = CONS(hp, new_binary(&parent, (byte*)argv[i], len), args); hp += 2; } - env = new_binary(&parent, code, size); + boot_mod = erts_atom_put((byte *) mod_name, sys_strlen(mod_name), ERTS_ATOM_ENC_LATIN1, 1); args = CONS(hp, args, NIL); hp += 2; - args = CONS(hp, env, args); + args = CONS(hp, boot_mod, args); so.flags = erts_default_spo_flags|SPO_SYSTEM_PROC; - res = erl_create_process(&parent, start_mod, am_start, args, &so); + res = erl_create_process(&parent, am_erl_init, am_start, args, &so); erts_proc_unlock(&parent, ERTS_PROC_LOCK_MAIN); erts_cleanup_empty_process(&parent); return res; @@ -481,7 +479,6 @@ erts_preloaded(Process* p) /* static variables that must not change (use same values at restart) */ static char* program; static char* init = "init"; -static char* boot = "boot"; static int boot_argc; static char** boot_argv; @@ -535,9 +532,6 @@ void erts_usage(void) int this_rel = this_rel_num(); erts_fprintf(stderr, "Usage: %s [flags] [ -- [init_args] ]\n", progname(program)); erts_fprintf(stderr, "The flags are:\n\n"); - - /* erts_fprintf(stderr, "-# number set the number of items to be used in traces etc\n"); */ - erts_fprintf(stderr, "-a size suggested stack size in kilo words for threads\n"); erts_fprintf(stderr, " in the async-thread pool, valid range is [%d-%d]\n", ERTS_ASYNC_THREAD_MIN_STACK_SIZE, @@ -545,13 +539,9 @@ void erts_usage(void) erts_fprintf(stderr, "-A number set number of threads in async thread pool,\n"); erts_fprintf(stderr, " valid range is [0-%d]\n", ERTS_MAX_NO_OF_ASYNC_THREADS); - erts_fprintf(stderr, "-B[c|d|i] c to have Ctrl-c interrupt the Erlang shell,\n"); erts_fprintf(stderr, " d (or no extra option) to disable the break\n"); erts_fprintf(stderr, " handler, i to ignore break signals\n"); - - /* erts_fprintf(stderr, "-b func set the boot function (default boot)\n"); */ - erts_fprintf(stderr, "-c bool enable or disable time correction\n"); erts_fprintf(stderr, "-C mode set time warp mode; valid modes are:\n"); erts_fprintf(stderr, " no_time_warp|single_time_warp|multi_time_warp\n"); @@ -570,7 +560,6 @@ void erts_usage(void) erts_pd_initial_size); erts_fprintf(stderr, "-hmqd val set default message queue data flag for processes,\n"); erts_fprintf(stderr, " valid values are: off_heap | on_heap\n"); - erts_fprintf(stderr, "-IOp number set number of pollsets to be used to poll for I/O,\n"); erts_fprintf(stderr, " This value has to be equal or smaller than the\n"); erts_fprintf(stderr, " number of poll threads. If the current platform\n"); @@ -581,9 +570,7 @@ void erts_usage(void) erts_fprintf(stderr, " number of poll threads."); erts_fprintf(stderr, "-IOPt number set number of threads to be used to poll for I/O\n"); erts_fprintf(stderr, " as a percentage of the number of schedulers."); - - /* erts_fprintf(stderr, "-i module set the boot module (default init)\n"); */ - + erts_fprintf(stderr, "-i module set the boot module (default init)\n"); erts_fprintf(stderr, "-n[s|a|d] Control behavior of signals to ports\n"); erts_fprintf(stderr, " Note that this flag is deprecated!\n"); erts_fprintf(stderr, "-M<X> <Y> memory allocator switches,\n"); @@ -598,7 +585,6 @@ void erts_usage(void) erts_fprintf(stderr, "-R number set compatibility release number,\n"); erts_fprintf(stderr, " valid range [%d-%d]\n", this_rel-2, this_rel); - erts_fprintf(stderr, "-r force ets memory block to be moved on realloc\n"); erts_fprintf(stderr, "-rg amount set reader groups limit\n"); erts_fprintf(stderr, "-sbt type set scheduler bind type, valid types are:\n"); @@ -669,9 +655,7 @@ void erts_usage(void) erts_fprintf(stderr, "-T number set modified timing level, valid range is [0-%d]\n", ERTS_MODIFIED_TIMING_LEVELS-1); erts_fprintf(stderr, "-V print Erlang version\n"); - erts_fprintf(stderr, "-v turn on chatty mode (GCs will be reported etc)\n"); - erts_fprintf(stderr, "-W<i|w|e> set error logger warnings mapping,\n"); erts_fprintf(stderr, " see error_logger documentation for details\n"); erts_fprintf(stderr, "-zdbbl size set the distribution buffer busy limit in kilobytes\n"); @@ -762,7 +746,6 @@ early_init(int *argc, char **argv) /* erts_sched_compact_load = 1; erts_printf_eterm_func = erts_printf_term; - display_items = 200; erts_backtrace_depth = DEFAULT_BACKTRACE_SIZE; erts_async_max_threads = ERTS_DEFAULT_NO_ASYNC_THREADS; erts_async_thread_suggested_stack_size = ERTS_ASYNC_THREAD_MIN_STACK_SIZE; @@ -1269,25 +1252,9 @@ erl_start(int argc, char **argv) /* * NOTE: -M flags are handled (and removed from argv) by - * erts_alloc_init(). - * - * The -d, -m, -S, -t, and -T flags was removed in - * Erlang 5.3/OTP R9C. - * - * -S, and -T has been reused in Erlang 5.5/OTP R11B. - * - * -d has been reused in a patch R12B-4. + * erts_alloc_init(). */ - case '#' : - arg = get_arg(argv[i]+2, argv[i+1], &i); - if ((display_items = atoi(arg)) == 0) { - erts_fprintf(stderr, "bad display items%s\n", arg); - erts_usage(); - } - VERBOSE(DEBUG_SYSTEM, - ("using display items %d\n",display_items)); - break; case 'p': if (!sys_strncmp(argv[i],"-pc",3)) { int printable_chars = ERL_PRINTABLE_CHARACTERS_LATIN1; @@ -1566,11 +1533,6 @@ erl_start(int argc, char **argv) init = get_arg(argv[i]+2, argv[i+1], &i); break; - case 'b': - /* define name of initial function */ - boot = get_arg(argv[i]+2, argv[i+1], &i); - break; - case 'B': if (argv[i][2] == 'i') /* +Bi */ ignore_break = 1; @@ -2256,9 +2218,8 @@ erl_start(int argc, char **argv) erts_initialized = 1; - erts_init_process_id = erl_first_process_otp("otp_ring0", NULL, 0, - boot_argc, boot_argv); - ASSERT(erts_init_process_id != ERTS_INVALID_PID); + erts_init_process_id = erl_first_process_otp(init, boot_argc, boot_argv); + ASSERT(erts_init_process_id != ERTS_INVALID_PID); { /* diff --git a/erts/emulator/beam/erl_lock_check.c b/erts/emulator/beam/erl_lock_check.c index 3b10ef8c25..3aab4828cc 100644 --- a/erts/emulator/beam/erl_lock_check.c +++ b/erts/emulator/beam/erl_lock_check.c @@ -42,6 +42,7 @@ #include "erl_term.h" #include "erl_threads.h" #include "erl_atom_table.h" +#include "erl_utils.h" typedef struct { char *name; @@ -91,14 +92,16 @@ static erts_lc_lock_order_t erts_lock_order[] = { { "db_tab", "address" }, { "db_tab_fix", "address" }, { "db_hash_slot", "address" }, - { "erl_db_catree_base_node", "dynamic" }, - { "erl_db_catree_route_node", "dynamic" }, + { "erl_db_catree_base_node", NULL }, + { "erl_db_catree_route_node", "index" }, { "resource_monitors", "address" }, { "driver_list", NULL }, { "proc_msgq", "pid" }, { "proc_btm", "pid" }, { "dist_entry", "address" }, { "dist_entry_links", "address" }, + { "update_persistent_term_permission", NULL }, + { "persistent_term_delete_permission", NULL }, { "code_write_permission", NULL }, { "purge_state", NULL }, { "proc_status", "pid" }, @@ -193,6 +196,12 @@ struct lc_locked_lock_t_ { unsigned int line; erts_lock_flags_t flags; erts_lock_options_t taken_options; + /* + * Pointer back to the lock instance if it exists or NULL for proc locks. + * If set, we use it to allow trylock of other lock instance + * but with identical lock order as an already locked lock. + */ + erts_lc_lock_t *lck; }; typedef struct { @@ -406,6 +415,10 @@ new_locked_lock(lc_thread_t* thr, ll->line = line; ll->flags = lck->flags; ll->taken_options = options; + if ((lck->flags & ERTS_LOCK_FLAGS_MASK_TYPE) == ERTS_LOCK_FLAGS_TYPE_PROCLOCK) + ll->lck = NULL; + else + ll->lck = lck; return ll; } @@ -709,26 +722,14 @@ erts_lc_get_lock_order_id(char *name) return (Sint16) -1; } -int -erts_lc_is_check_order(char *name) +static int +lc_is_term_order(Sint16 id) { - int i; - if (!name || name[0] == '\0') - erts_fprintf(stderr, "Missing lock name\n"); - - for (i = 0; i < ERTS_LOCK_ORDER_SIZE; i++) { - if (sys_strcmp(erts_lock_order[i].name, name) == 0) { - if (erts_lock_order[i].internal_order != NULL && - sys_strcmp(erts_lock_order[i].internal_order, "dynamic") == 0) { - return 0; - }else{ - return 1; - } - } - } - return 1; + return erts_lock_order[id].internal_order != NULL + && sys_strcmp(erts_lock_order[id].internal_order, "term") == 0; } + static int compare_locked_by_id(lc_locked_lock_t *locked_lock, erts_lc_lock_t *comparand) { if(locked_lock->id < comparand->id) { @@ -740,18 +741,23 @@ static int compare_locked_by_id(lc_locked_lock_t *locked_lock, erts_lc_lock_t *c return 0; } -static int compare_locked_by_id_extra(lc_locked_lock_t *locked_lock, erts_lc_lock_t *comparand) +static int compare_locked_by_id_extra(lc_locked_lock_t *ll, erts_lc_lock_t *comparand) { - int order = compare_locked_by_id(locked_lock, comparand); + int order = compare_locked_by_id(ll, comparand); if(order) { return order; - } else if(locked_lock->extra < comparand->extra) { + } + if (ll->flags & ERTS_LOCK_FLAGS_PROPERTY_TERM_ORDER) { + ASSERT(!is_header(ll->extra) && !is_header(comparand->extra)); + return CMP(ll->extra, comparand->extra); + } + + if(ll->extra < comparand->extra) { return -1; - } else if(locked_lock->extra > comparand->extra) { + } else if(ll->extra > comparand->extra) { return 1; } - return 0; } @@ -990,7 +996,8 @@ erts_lc_trylock_force_busy_flg(erts_lc_lock_t *lck, erts_lock_options_t options) return 0; } else { - lc_locked_lock_t *tl_lck; + lc_locked_lock_t *ll; + int order; ASSERT(thr->locked.last); @@ -999,27 +1006,25 @@ erts_lc_trylock_force_busy_flg(erts_lc_lock_t *lck, erts_lock_options_t options) type_order_violation("trylocking ", thr, lck); #endif - if (thr->locked.last->id < lck->id - || (thr->locked.last->id == lck->id - && thr->locked.last->extra < lck->extra)) - return 0; + ll = thr->locked.last; + order = compare_locked_by_id_extra(ll, lck); + + if (order < 0) + return 0; /* - * Lock order violation + * TryLock order violation */ - - if (lck->check_order) { - /* Check that we are not trying to lock this lock twice */ - for (tl_lck = thr->locked.last; tl_lck; tl_lck = tl_lck->prev) { - if (tl_lck->id < lck->id - || (tl_lck->id == lck->id && tl_lck->extra <= lck->extra)) { - if (tl_lck->id == lck->id && tl_lck->extra == lck->extra) - lock_twice("Trylocking", thr, lck, options); - break; - } - } - } + /* Check that we are not trying to lock this lock twice */ + do { + if (order == 0 && (ll->lck == lck || !ll->lck)) + lock_twice("Trylocking", thr, lck, options); + ll = ll->prev; + if (!ll) + break; + order = compare_locked_by_id_extra(ll, lck); + } while (order >= 0); #ifndef ERTS_LC_ALLWAYS_FORCE_BUSY_TRYLOCK_ON_LOCK_ORDER_VIOLATION /* We only force busy if a lock order violation would occur @@ -1066,10 +1071,10 @@ void erts_lc_trylock_flg_x(int locked, erts_lc_lock_t *lck, erts_lock_options_t #endif for (tl_lck = thr->locked.last; tl_lck; tl_lck = tl_lck->prev) { - if (tl_lck->id < lck->id - || (tl_lck->id == lck->id && tl_lck->extra <= lck->extra)) { - if (tl_lck->id == lck->id && tl_lck->extra == lck->extra && lck->check_order) - lock_twice("Trylocking", thr, lck, options); + int order = compare_locked_by_id_extra(tl_lck, lck); + if (order <= 0) { + if (order == 0 && (tl_lck->lck == lck || !tl_lck->lck)) + lock_twice("Trylocking", thr, lck, options); if (locked) { ll->next = tl_lck->next; ll->prev = tl_lck; @@ -1111,10 +1116,10 @@ void erts_lc_require_lock_flg(erts_lc_lock_t *lck, erts_lock_options_t options, for (l_lck2 = thr->required.last; l_lck2; l_lck2 = l_lck2->prev) { - if (l_lck2->id < lck->id - || (l_lck2->id == lck->id && l_lck2->extra < lck->extra)) + int order = compare_locked_by_id_extra(l_lck2, lck); + if (order < 0) break; - else if (l_lck2->id == lck->id && l_lck2->extra == lck->extra) + if (order == 0) require_twice(thr, lck); } if (!l_lck2) { @@ -1172,6 +1177,7 @@ void erts_lc_lock_flg_x(erts_lc_lock_t *lck, erts_lock_options_t options, { lc_thread_t *thr; lc_locked_lock_t *new_ll; + int order; if (lck->inited != ERTS_LC_INITITALIZED) uninitialized_lock(); @@ -1187,11 +1193,10 @@ void erts_lc_lock_flg_x(erts_lc_lock_t *lck, erts_lock_options_t options, thr->locked.last = thr->locked.first = new_ll; ASSERT(0 < lck->id && lck->id < ERTS_LOCK_ORDER_SIZE); thr->matrix.m[lck->id][0] = 1; + return; } - else if (( ! lck->check_order && thr->locked.last->id == lck->id) || - (thr->locked.last->id < lck->id - || (thr->locked.last->id == lck->id - && thr->locked.last->extra < lck->extra))) { + order = compare_locked_by_id_extra(thr->locked.last, lck); + if (order < 0) { lc_locked_lock_t* ll; if (LOCK_IS_TYPE_ORDER_VIOLATION(lck->flags, thr->locked.last->flags)) { type_order_violation("locking ", thr, lck); @@ -1209,7 +1214,7 @@ void erts_lc_lock_flg_x(erts_lc_lock_t *lck, erts_lock_options_t options, thr->locked.last->next = new_ll; thr->locked.last = new_ll; } - else if (thr->locked.last->id == lck->id && thr->locked.last->extra == lck->extra) + else if (order == 0) lock_twice("Locking", thr, lck, options); else lock_order_violation(thr, lck); @@ -1321,7 +1326,6 @@ void erts_lc_init_lock(erts_lc_lock_t *lck, char *name, erts_lock_flags_t flags) { lck->id = erts_lc_get_lock_order_id(name); - lck->check_order = erts_lc_is_check_order(name); lck->extra = (UWord) &lck->extra; ASSERT(is_not_immed(lck->extra)); lck->flags = flags; @@ -1333,10 +1337,14 @@ void erts_lc_init_lock_x(erts_lc_lock_t *lck, char *name, erts_lock_flags_t flags, Eterm extra) { lck->id = erts_lc_get_lock_order_id(name); - lck->check_order = erts_lc_is_check_order(name); lck->extra = extra; - ASSERT(is_immed(lck->extra)); lck->flags = flags; + if (lc_is_term_order(lck->id)) { + lck->flags |= ERTS_LOCK_FLAGS_PROPERTY_TERM_ORDER; + ASSERT(!is_header(lck->extra)); + } + else + ASSERT(is_immed(lck->extra)); lck->taken_options = 0; lck->inited = ERTS_LC_INITITALIZED; } diff --git a/erts/emulator/beam/erl_lock_check.h b/erts/emulator/beam/erl_lock_check.h index 472e88ad65..b32f27d9f9 100644 --- a/erts/emulator/beam/erl_lock_check.h +++ b/erts/emulator/beam/erl_lock_check.h @@ -46,7 +46,6 @@ typedef struct { int inited; Sint16 id; - int check_order; erts_lock_flags_t flags; erts_lock_options_t taken_options; UWord extra; @@ -54,12 +53,11 @@ typedef struct { #define ERTS_LC_INITITALIZED 0x7f7f7f7f -#define ERTS_LC_LOCK_INIT(ID, X, F) {ERTS_LC_INITITALIZED, (ID), 1, (F), 0, (X)} +#define ERTS_LC_LOCK_INIT(ID, X, F) {ERTS_LC_INITITALIZED, (ID), (F), 0, (X)} void erts_lc_init(void); void erts_lc_late_init(void); Sint16 erts_lc_get_lock_order_id(char *name); -int erts_lc_is_check_order(char *name); void erts_lc_check(erts_lc_lock_t *have, int have_len, erts_lc_lock_t *have_not, int have_not_len); void erts_lc_check_exact(erts_lc_lock_t *have, int have_len); @@ -106,7 +104,7 @@ Eterm erts_lc_dump_graph(void); #define erts_lc_lock(lck) erts_lc_lock_x(lck,__FILE__,__LINE__) #define erts_lc_trylock(res,lck) erts_lc_trylock_x(res,lck,__FILE__,__LINE__) -#define erts_lc_lock_flg(lck) erts_lc_lock_flg_x(lck,__FILE__,__LINE__) -#define erts_lc_trylock_flg(res,lck) erts_lc_trylock_flg_x(res,lck,__FILE__,__LINE__) +#define erts_lc_lock_flg(lck,flg) erts_lc_lock_flg_x(lck,flg,__FILE__,__LINE__) +#define erts_lc_trylock_flg(res,lck,flg) erts_lc_trylock_flg_x(res,lck,flg,__FILE__,__LINE__) #endif /* #ifndef ERTS_LOCK_CHECK_H__ */ diff --git a/erts/emulator/beam/erl_lock_flags.h b/erts/emulator/beam/erl_lock_flags.h index d711f69456..2db133b598 100644 --- a/erts/emulator/beam/erl_lock_flags.h +++ b/erts/emulator/beam/erl_lock_flags.h @@ -28,15 +28,17 @@ /* Property/category are bitfields to simplify their use in masks. */ #define ERTS_LOCK_FLAGS_MASK_CATEGORY (0xFFC0) -#define ERTS_LOCK_FLAGS_MASK_PROPERTY (0x0030) +#define ERTS_LOCK_FLAGS_MASK_PROPERTY (0x0038) /* Type is a plain number. */ -#define ERTS_LOCK_FLAGS_MASK_TYPE (0x000F) +#define ERTS_LOCK_FLAGS_MASK_TYPE (0x0007) #define ERTS_LOCK_FLAGS_TYPE_SPINLOCK (1) #define ERTS_LOCK_FLAGS_TYPE_MUTEX (2) #define ERTS_LOCK_FLAGS_TYPE_PROCLOCK (3) +/* Lock checker use real term order instead of raw word compare */ +#define ERTS_LOCK_FLAGS_PROPERTY_TERM_ORDER (1 << 3) /* "Static" guarantees that the lock will never be destroyed once created. */ #define ERTS_LOCK_FLAGS_PROPERTY_STATIC (1 << 4) #define ERTS_LOCK_FLAGS_PROPERTY_READ_WRITE (1 << 5) diff --git a/erts/emulator/beam/erl_monitor_link.h b/erts/emulator/beam/erl_monitor_link.h index 9ff8aa509a..ed7bf7d54a 100644 --- a/erts/emulator/beam/erl_monitor_link.h +++ b/erts/emulator/beam/erl_monitor_link.h @@ -1387,11 +1387,14 @@ ERTS_GLB_INLINE void erts_monitor_release(ErtsMonitor *mon) { ErtsMonitorData *mdp = erts_monitor_to_data(mon); - ERTS_ML_ASSERT(!(mon->flags & ERTS_ML_FLG_IN_TABLE)); ERTS_ML_ASSERT(erts_atomic32_read_nob(&mdp->refc) > 0); - if (erts_atomic32_dec_read_nob(&mdp->refc) == 0) + if (erts_atomic32_dec_read_nob(&mdp->refc) == 0) { + ERTS_ML_ASSERT(!(mdp->origin.flags & ERTS_ML_FLG_IN_TABLE)); + ERTS_ML_ASSERT(!(mdp->target.flags & ERTS_ML_FLG_IN_TABLE)); + erts_monitor_destroy__(mdp); + } } ERTS_GLB_INLINE void @@ -1399,12 +1402,14 @@ erts_monitor_release_both(ErtsMonitorData *mdp) { ERTS_ML_ASSERT((mdp->origin.flags & ERTS_ML_FLGS_SAME) == (mdp->target.flags & ERTS_ML_FLGS_SAME)); - ERTS_ML_ASSERT(!(mdp->origin.flags & ERTS_ML_FLG_IN_TABLE)); - ERTS_ML_ASSERT(!(mdp->target.flags & ERTS_ML_FLG_IN_TABLE)); ERTS_ML_ASSERT(erts_atomic32_read_nob(&mdp->refc) >= 2); - if (erts_atomic32_add_read_nob(&mdp->refc, (erts_aint32_t) -2) == 0) + if (erts_atomic32_add_read_nob(&mdp->refc, (erts_aint32_t) -2) == 0) { + ERTS_ML_ASSERT(!(mdp->origin.flags & ERTS_ML_FLG_IN_TABLE)); + ERTS_ML_ASSERT(!(mdp->target.flags & ERTS_ML_FLG_IN_TABLE)); + erts_monitor_destroy__(mdp); + } } ERTS_GLB_INLINE int diff --git a/erts/emulator/beam/erl_nif.h b/erts/emulator/beam/erl_nif.h index c5227a0c23..58a217c20b 100644 --- a/erts/emulator/beam/erl_nif.h +++ b/erts/emulator/beam/erl_nif.h @@ -54,10 +54,16 @@ ** 2.13: 20.1 add enif_ioq ** 2.14: 21.0 add enif_ioq_peek_head, enif_(mutex|cond|rwlock|thread)_name ** enif_vfprintf, enif_vsnprintf, enif_make_map_from_arrays +** 2.15: 22.0 ERL_NIF_SELECT_CANCEL */ #define ERL_NIF_MAJOR_VERSION 2 -#define ERL_NIF_MINOR_VERSION 14 -#define ERL_NIF_MIN_ERTS_VERSION "erts-10.0 (OTP-21)" +#define ERL_NIF_MINOR_VERSION 15 +/* + * WHEN CHANGING INTERFACE VERSION, also replace erts version below + * with ticket syntax like "erts-@OTP-12345@", or a temporary placeholder + * between two @ like "erts-@MyName@", if you don't know what a ticket is. + */ +#define ERL_NIF_MIN_ERTS_VERSION "erts-@OTP-15095@ (OTP-22)" /* * The emulator will refuse to load a nif-lib with a major version diff --git a/erts/emulator/beam/erl_process_dump.c b/erts/emulator/beam/erl_process_dump.c index 706530023b..0286f6a0d2 100644 --- a/erts/emulator/beam/erl_process_dump.c +++ b/erts/emulator/beam/erl_process_dump.c @@ -58,6 +58,7 @@ static void dump_externally(fmtfn_t to, void *to_arg, Eterm term); static void mark_literal(Eterm* ptr); static void init_literal_areas(void); static void dump_literals(fmtfn_t to, void *to_arg); +static void dump_persistent_terms(fmtfn_t to, void *to_arg); static void dump_module_literals(fmtfn_t to, void *to_arg, ErtsLiteralArea* lit_area); @@ -74,6 +75,7 @@ erts_deep_process_dump(fmtfn_t to, void *to_arg) all_binaries = NULL; init_literal_areas(); + erts_init_persistent_dumping(); for (i = 0; i < max; i++) { Process *p = erts_pix2proc(i); @@ -93,6 +95,7 @@ erts_deep_process_dump(fmtfn_t to, void *to_arg) } } + dump_persistent_terms(to, to_arg); dump_literals(to, to_arg); dump_binaries(to, to_arg, all_binaries); } @@ -775,6 +778,9 @@ init_literal_areas(void) qsort(lit_areas, num_lit_areas, sizeof(ErtsLiteralArea *), compare_areas); + qsort(erts_persistent_areas, erts_num_persistent_areas, + sizeof(ErtsLiteralArea *), compare_areas); + erts_runlock_old_code(code_ix); } @@ -796,6 +802,13 @@ static void mark_literal(Eterm* ptr) ap = bsearch(ptr, lit_areas, num_lit_areas, sizeof(ErtsLiteralArea*), search_areas); + if (ap == 0) { + ap = bsearch(ptr, erts_persistent_areas, + erts_num_persistent_areas, + sizeof(ErtsLiteralArea*), + search_areas); + } + /* * If the literal was created by native code, this search will not @@ -807,12 +820,12 @@ static void mark_literal(Eterm* ptr) } } - static void dump_literals(fmtfn_t to, void *to_arg) { ErtsCodeIndex code_ix; int i; + Uint idx; code_ix = erts_active_code_ix(); erts_rlock_old_code(code_ix); @@ -825,6 +838,28 @@ dump_literals(fmtfn_t to, void *to_arg) } erts_runlock_old_code(code_ix); + + for (idx = 0; idx < erts_num_persistent_areas; idx++) { + dump_module_literals(to, to_arg, erts_persistent_areas[idx]); + } +} + +static void +dump_persistent_terms(fmtfn_t to, void *to_arg) +{ + Uint idx; + + erts_print(to, to_arg, "=persistent_terms\n"); + + for (idx = 0; idx < erts_num_persistent_areas; idx++) { + ErtsLiteralArea* ap = erts_persistent_areas[idx]; + Eterm tuple = make_tuple(ap->start); + Eterm* tup_val = tuple_val(tuple); + + dump_element(to, to_arg, tup_val[1]); + erts_putc(to, to_arg, '|'); + dump_element_nl(to, to_arg, tup_val[2]); + } } static void diff --git a/erts/emulator/beam/erl_utils.h b/erts/emulator/beam/erl_utils.h index b3bfa69052..880febba8b 100644 --- a/erts/emulator/beam/erl_utils.h +++ b/erts/emulator/beam/erl_utils.h @@ -22,6 +22,7 @@ #define ERL_UTILS_H__ #include "sys.h" +#include "atom.h" #include "erl_printf.h" struct process; @@ -112,10 +113,12 @@ int eq(Eterm, Eterm); #define EQ(x,y) (((x) == (y)) || (is_not_both_immed((x),(y)) && eq((x),(y)))) -int erts_cmp_atoms(Eterm a, Eterm b); -Sint erts_cmp(Eterm, Eterm, int, int); -Sint erts_cmp_compound(Eterm, Eterm, int, int); +ERTS_GLB_INLINE Sint erts_cmp(Eterm, Eterm, int, int); +ERTS_GLB_INLINE int erts_cmp_atoms(Eterm a, Eterm b); + Sint cmp(Eterm a, Eterm b); +Sint erts_cmp_compound(Eterm, Eterm, int, int); + #define CMP(A,B) erts_cmp(A,B,0,0) #define CMP_TERM(A,B) erts_cmp(A,B,1,0) #define CMP_EQ_ONLY(A,B) erts_cmp(A,B,0,1) @@ -150,4 +153,56 @@ Sint cmp(Eterm a, Eterm b); if (erts_cmp_compound(X,Y,0,EqOnly) Op 0) { Action; }; \ } +#define erts_float_comp(x,y) (((x)<(y)) ? -1 : (((x)==(y)) ? 0 : 1)) + +#if ERTS_GLB_INLINE_INCL_FUNC_DEF + +ERTS_GLB_INLINE int erts_cmp_atoms(Eterm a, Eterm b) { + Atom *aa = atom_tab(atom_val(a)); + Atom *bb = atom_tab(atom_val(b)); + + byte *name_a, *name_b; + int len_a, len_b, diff; + + diff = aa->ord0 - bb->ord0; + + if (diff != 0) { + return diff; + } + + name_a = &aa->name[3]; + name_b = &bb->name[3]; + len_a = aa->len-3; + len_b = bb->len-3; + + if (len_a > 0 && len_b > 0) { + diff = sys_memcmp(name_a, name_b, MIN(len_a, len_b)); + + if (diff != 0) { + return diff; + } + } + + return len_a - len_b; +} + +ERTS_GLB_INLINE Sint erts_cmp(Eterm a, Eterm b, int exact, int eq_only) { + if (is_atom(a) && is_atom(b)) { + return erts_cmp_atoms(a, b); + } else if (is_both_small(a, b)) { + return (signed_val(a) - signed_val(b)); + } else if (is_float(a) && is_float(b)) { + FloatDef af, bf; + + GET_DOUBLE(a, af); + GET_DOUBLE(b, bf); + + return erts_float_comp(af.fd, bf.fd); + } + + return erts_cmp_compound(a,b,exact,eq_only); +} + +#endif /* ERTS_GLB_INLINE_INCL_FUNC_DEF */ + #endif diff --git a/erts/emulator/beam/external.c b/erts/emulator/beam/external.c index 621ba108ba..9a66e491f3 100644 --- a/erts/emulator/beam/external.c +++ b/erts/emulator/beam/external.c @@ -1953,7 +1953,8 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla #define RETURN_STATE() \ do { \ - hp = HAlloc(p, ERTS_MAGIC_REF_THING_SIZE+3); \ + static const int TUPLE2_SIZE = 2 + 1; \ + hp = HAlloc(p, ERTS_MAGIC_REF_THING_SIZE + TUPLE2_SIZE); \ c_term = erts_mk_magic_ref(&hp, &MSO(p), context_b); \ res = TUPLE2(hp, Term, c_term); \ BUMP_ALL_REDS(p); \ diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h index 21ae205237..322981ca1d 100644 --- a/erts/emulator/beam/global.h +++ b/erts/emulator/beam/global.h @@ -292,7 +292,6 @@ union erl_off_heap_ptr { /* controls warning mapping in error_logger */ extern Eterm node_cookie; -extern Uint display_items; /* no of items to display in traces etc */ extern int erts_backtrace_depth; extern erts_atomic32_t erts_max_gen_gcs; @@ -906,6 +905,8 @@ typedef struct ErtsLiteralArea_ { Eterm start[1]; /* beginning of area */ } ErtsLiteralArea; +void erts_queue_release_literals(Process *c_p, ErtsLiteralArea* literals); + #define ERTS_LITERAL_AREA_ALLOC_SIZE(N) \ (sizeof(ErtsLiteralArea) + sizeof(Eterm)*((N) - 1)) @@ -1001,6 +1002,7 @@ typedef struct { Uint literal_size; Eterm *lit_purge_ptr; Uint lit_purge_sz; + int copy_literals; } erts_shcopy_t; #define INITIALIZE_SHCOPY(info) \ @@ -1010,6 +1012,7 @@ typedef struct { info.bitstore_start = info.bitstore_default; \ info.shtable_start = info.shtable_default; \ info.literal_size = 0; \ + info.copy_literals = 0; \ if (larea__) { \ info.lit_purge_ptr = &larea__->start[0]; \ info.lit_purge_sz = larea__->end - info.lit_purge_ptr; \ @@ -1238,6 +1241,13 @@ Sint erts_re_set_loop_limit(Sint limit); void erts_init_bif_binary(void); Sint erts_binary_set_loop_limit(Sint limit); +/* erl_bif_persistent.c */ +void erts_init_bif_persistent_term(void); +Uint erts_persistent_term_count(void); +void erts_init_persistent_dumping(void); +extern ErtsLiteralArea** erts_persistent_areas; +extern Uint erts_num_persistent_areas; + /* external.c */ void erts_init_external(void); diff --git a/erts/emulator/beam/sys.h b/erts/emulator/beam/sys.h index bb22548587..869a575cb4 100644 --- a/erts/emulator/beam/sys.h +++ b/erts/emulator/beam/sys.h @@ -325,6 +325,7 @@ typedef long Sint erts_align_attribute(sizeof(long)); #define UWORD_CONSTANT(Const) Const##UL #define ERTS_UWORD_MAX ULONG_MAX #define ERTS_SWORD_MAX LONG_MAX +#define ERTS_SWORD_MIN LONG_MIN #define ERTS_SIZEOF_ETERM SIZEOF_LONG #define ErtsStrToSint strtol #elif SIZEOF_VOID_P == SIZEOF_INT @@ -335,6 +336,7 @@ typedef int Sint erts_align_attribute(sizeof(int)); #define UWORD_CONSTANT(Const) Const##U #define ERTS_UWORD_MAX UINT_MAX #define ERTS_SWORD_MAX INT_MAX +#define ERTS_SWORD_MIN INT_MIN #define ERTS_SIZEOF_ETERM SIZEOF_INT #define ErtsStrToSint strtol #elif SIZEOF_VOID_P == SIZEOF_LONG_LONG @@ -345,6 +347,7 @@ typedef long long Sint erts_align_attribute(sizeof(long long)); #define UWORD_CONSTANT(Const) Const##ULL #define ERTS_UWORD_MAX ULLONG_MAX #define ERTS_SWORD_MAX LLONG_MAX +#define ERTS_SWORD_MIN LLONG_MIN #define ERTS_SIZEOF_ETERM SIZEOF_LONG_LONG #if defined(__WIN32__) #define ErtsStrToSint _strtoi64 diff --git a/erts/emulator/beam/utils.c b/erts/emulator/beam/utils.c index 996757ef43..f70f5e73d4 100644 --- a/erts/emulator/beam/utils.c +++ b/erts/emulator/beam/utils.c @@ -2615,27 +2615,6 @@ not_equal: } -/* - * Lexically compare two strings of bytes (string s1 length l1 and s2 l2). - * - * s1 < s2 return -1 - * s1 = s2 return 0 - * s1 > s2 return +1 - */ -static int cmpbytes(byte *s1, int l1, byte *s2, int l2) -{ - int i; - i = 0; - while((i < l1) && (i < l2)) { - if (s1[i] < s2[i]) return(-1); - if (s1[i] > s2[i]) return(1); - i++; - } - if (l1 < l2) return(-1); - if (l1 > l2) return(1); - return(0); -} - /* * Compare objects. @@ -2649,20 +2628,6 @@ static int cmpbytes(byte *s1, int l1, byte *s2, int l2) * */ - -#define float_comp(x,y) (((x)<(y)) ? -1 : (((x)==(y)) ? 0 : 1)) - -int erts_cmp_atoms(Eterm a, Eterm b) -{ - Atom *aa = atom_tab(atom_val(a)); - Atom *bb = atom_tab(atom_val(b)); - int diff = aa->ord0 - bb->ord0; - if (diff) - return diff; - return cmpbytes(aa->name+3, aa->len-3, - bb->name+3, bb->len-3); -} - /* cmp(Eterm a, Eterm b) * For compatibility with HiPE - arith-based compare. */ @@ -2673,22 +2638,6 @@ Sint cmp(Eterm a, Eterm b) Sint erts_cmp_compound(Eterm a, Eterm b, int exact, int eq_only); -Sint erts_cmp(Eterm a, Eterm b, int exact, int eq_only) -{ - if (is_atom(a) && is_atom(b)) { - return erts_cmp_atoms(a, b); - } else if (is_both_small(a, b)) { - return (signed_val(a) - signed_val(b)); - } else if (is_float(a) && is_float(b)) { - FloatDef af, bf; - GET_DOUBLE(a, af); - GET_DOUBLE(b, bf); - return float_comp(af.fd, bf.fd); - } - return erts_cmp_compound(a,b,exact,eq_only); -} - - /* erts_cmp(Eterm a, Eterm b, int exact) * exact = 1 -> term-based compare * exact = 0 -> arith-based compare @@ -2985,7 +2934,7 @@ tailrecur_ne: GET_DOUBLE(a, af); GET_DOUBLE(b, bf); - ON_CMP_GOTO(float_comp(af.fd, bf.fd)); + ON_CMP_GOTO(erts_float_comp(af.fd, bf.fd)); } case (_TAG_HEADER_POS_BIG >> _TAG_PRIMARY_SIZE): case (_TAG_HEADER_NEG_BIG >> _TAG_PRIMARY_SIZE): @@ -3022,10 +2971,7 @@ tailrecur_ne: ErlFunThing* f2 = (ErlFunThing *) fun_val(b); Sint diff; - diff = cmpbytes(atom_tab(atom_val(f1->fe->module))->name, - atom_tab(atom_val(f1->fe->module))->len, - atom_tab(atom_val(f2->fe->module))->name, - atom_tab(atom_val(f2->fe->module))->len); + diff = erts_cmp_atoms((f1->fe)->module, (f2->fe)->module); if (diff != 0) { RETURN_NEQ(diff); } @@ -3219,7 +3165,7 @@ tailrecur_ne: if (f2.fd < MAX_LOSSLESS_FLOAT && f2.fd > MIN_LOSSLESS_FLOAT) { /* Float is within the no loss limit */ f1.fd = signed_val(aw); - j = float_comp(f1.fd, f2.fd); + j = erts_float_comp(f1.fd, f2.fd); } #if ERTS_SIZEOF_ETERM == 8 else if (f2.fd > (double) (MAX_SMALL + 1)) { @@ -3266,7 +3212,7 @@ tailrecur_ne: if (big_to_double(aw, &f1.fd) < 0) { j = big_sign(aw) ? -1 : 1; } else { - j = float_comp(f1.fd, f2.fd); + j = erts_float_comp(f1.fd, f2.fd); } } else { big = double_to_big(f2.fd, big_buf, sizeof(big_buf)/sizeof(Eterm)); @@ -3282,7 +3228,7 @@ tailrecur_ne: if (f1.fd < MAX_LOSSLESS_FLOAT && f1.fd > MIN_LOSSLESS_FLOAT) { /* Float is within the no loss limit */ f2.fd = signed_val(bw); - j = float_comp(f1.fd, f2.fd); + j = erts_float_comp(f1.fd, f2.fd); } #if ERTS_SIZEOF_ETERM == 8 else if (f1.fd > (double) (MAX_SMALL + 1)) { diff --git a/erts/emulator/drivers/common/inet_drv.c b/erts/emulator/drivers/common/inet_drv.c index c9f7006384..dbe0201caf 100644 --- a/erts/emulator/drivers/common/inet_drv.c +++ b/erts/emulator/drivers/common/inet_drv.c @@ -812,6 +812,7 @@ static size_t my_strnlen(const char *s, size_t maxlen) #define INET_OPT_PKTOPTIONS 45 /* IP(V6)_PKTOPTIONS get ancillary data */ #define INET_OPT_TTL 46 /* IP_TTL */ #define INET_OPT_RECVTTL 47 /* IP_RECVTTL ancillary data */ +#define TCP_OPT_NOPUSH 48 /* super-Nagle, aka TCP_CORK */ /* SCTP options: a separate range, from 100: */ #define SCTP_OPT_RTOINFO 100 #define SCTP_OPT_ASSOCINFO 101 @@ -955,6 +956,12 @@ static size_t my_strnlen(const char *s, size_t maxlen) #endif +#if defined(TCP_CORK) +#define INET_TCP_NOPUSH TCP_CORK +#elif defined(TCP_NOPUSH) && !defined(__DARWIN__) +#define INET_TCP_NOPUSH TCP_NOPUSH +#endif + #define BIN_REALLOC_MARGIN(x) ((x)/4) /* 25% */ /* The general purpose sockaddr */ @@ -5178,6 +5185,71 @@ static int hwaddr_libdlpi_lookup(const char *ifnm, } #endif +#ifdef HAVE_GETIFADDRS +/* Returns 0 for success and errno() for failure */ +static int call_getifaddrs(inet_descriptor* desc_p, struct ifaddrs **ifa_pp) +{ + int result, save_errno; +#ifdef HAVE_SETNS + int current_ns; + + current_ns = 0; + if (desc_p->netns != NULL) { + int new_ns; + /* Temporarily change network namespace for this thread + * over the getifaddrs() call + */ + current_ns = open("/proc/self/ns/net", O_RDONLY); + if (current_ns == INVALID_SOCKET) + return sock_errno(); + new_ns = open(desc_p->netns, O_RDONLY); + if (new_ns == INVALID_SOCKET) { + save_errno = sock_errno(); + while (close(current_ns) == INVALID_SOCKET && + sock_errno() == EINTR); + return save_errno; + } + if (setns(new_ns, CLONE_NEWNET) != 0) { + save_errno = sock_errno(); + while (close(new_ns) == INVALID_SOCKET && + sock_errno() == EINTR); + while (close(current_ns) == INVALID_SOCKET && + sock_errno() == EINTR); + return save_errno; + } + else { + while (close(new_ns) == INVALID_SOCKET && + sock_errno() == EINTR); + } + } +#endif + save_errno = 0; + result = getifaddrs(ifa_pp); + if (result < 0) + save_errno = sock_errno(); +#ifdef HAVE_SETNS + if (desc_p->netns != NULL) { + /* Restore network namespace */ + if (setns(current_ns, CLONE_NEWNET) != 0) { + /* XXX Failed to restore network namespace. + * What to do? Tidy up and return an error... + * Note that the thread now might still be in the set namespace. + * Can this even happen? Should the emulator be aborted? + */ + if (result >= 0) { + /* We got a result but have to waste it */ + save_errno = sock_errno(); + freeifaddrs(*ifa_pp); + } + } + while (close(current_ns) == INVALID_SOCKET && + sock_errno() == EINTR); + } +#endif + return save_errno; +} +#endif /* #ifdef HAVE_GETIFADDRS */ + /* FIXME: temporary hack */ #ifndef IFHWADDRLEN #define IFHWADDRLEN 6 @@ -5255,8 +5327,8 @@ static ErlDrvSSizeT inet_ctl_ifget(inet_descriptor* desc, struct sockaddr_dl *sdlp; int found = 0; - if (getifaddrs(&ifa) == -1) - goto error; + if (call_getifaddrs(desc, &ifa) != 0) + goto error; for (ifp = ifa; ifp; ifp = ifp->ifa_next) { if ((ifp->ifa_addr->sa_family == AF_LINK) && @@ -5974,6 +6046,7 @@ static ErlDrvSSizeT inet_ctl_getifaddrs(inet_descriptor* desc_p, ErlDrvSizeT buf_size; char *buf_p; char *buf_alloc_p; + int save_errno; buf_size = GETIFADDRS_BUFSZ; buf_alloc_p = ALLOC(GETIFADDRS_BUFSZ); @@ -6008,9 +6081,9 @@ static ErlDrvSSizeT inet_ctl_getifaddrs(inet_descriptor* desc_p, } \ } while (0) - if (getifaddrs(&ifa_p) < 0) { - return ctl_error(sock_errno(), rbuf_pp, rsize); - } + if ((save_errno = call_getifaddrs(desc_p, &ifa_p)) != 0) + return ctl_error(save_errno, rbuf_pp, rsize); + ifa_free_p = ifa_p; *buf_p++ = INET_REP_OK; for (; ifa_p; ifa_p = ifa_p->ifa_next) { @@ -6532,6 +6605,19 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len) (long)desc->port, desc->s, ival)); break; + case TCP_OPT_NOPUSH: +#if defined(INET_TCP_NOPUSH) + proto = IPPROTO_TCP; + type = INET_TCP_NOPUSH; + DEBUGF(("inet_set_opts(%ld): s=%d, t=%d TCP_NOPUSH=%d\r\n", + (long)desc->port, desc->s, type, ival)); + break; +#else + /* inet_fill_opts always returns a value for this option, + * so we need to ignore it if not implemented, just in case */ + continue; +#endif + #if defined(HAVE_MULTICAST_SUPPORT) && defined(IPPROTO_IP) case UDP_OPT_MULTICAST_TTL: @@ -7693,6 +7779,16 @@ static ErlDrvSSizeT inet_fill_opts(inet_descriptor* desc, proto = IPPROTO_TCP; type = TCP_NODELAY; break; + case TCP_OPT_NOPUSH: +#if defined(INET_TCP_NOPUSH) + proto = IPPROTO_TCP; + type = INET_TCP_NOPUSH; + break; +#else + *ptr++ = opt; + put_int32(0, ptr); + continue; +#endif #if defined(HAVE_MULTICAST_SUPPORT) && defined(IPPROTO_IP) case UDP_OPT_MULTICAST_TTL: @@ -7839,8 +7935,8 @@ static ErlDrvSSizeT inet_fill_opts(inet_descriptor* desc, * cmsg options and values */ PLACE_FOR(1+4, ptr); - *ptr = opt; - arg_ptr = ptr+1; /* Where to put total length */ + *ptr++ = opt; + arg_ptr = ptr; /* Where to put total length */ arg_sz = 0; /* Total length */ for (cmsg_top = (struct cmsghdr*)(cmsgbuf.buf + cmsg_sz), cmsg = (struct cmsghdr*)cmsgbuf.buf; @@ -7852,7 +7948,6 @@ static ErlDrvSSizeT inet_fill_opts(inet_descriptor* desc, PLACE_FOR(1+4, ptr); \ *ptr++ = OPT; \ put_cmsg_int32(cmsg, ptr); \ - ptr += 4; \ arg_sz += 1+4; \ continue; \ } @@ -7866,7 +7961,6 @@ static ErlDrvSSizeT inet_fill_opts(inet_descriptor* desc, PUT_CMSG_INT32(IPPROTO_IP, IP_TTL, INET_OPT_TTL); #endif /* BSD uses the RECV* names in CMSG fields */ - } #if defined(IPPROTO_IP) && defined(IP_RECVTOS) PUT_CMSG_INT32(IPPROTO_IP, IP_RECVTOS, INET_OPT_TOS); #endif @@ -7877,6 +7971,7 @@ static ErlDrvSSizeT inet_fill_opts(inet_descriptor* desc, PUT_CMSG_INT32(IPPROTO_IP, IP_RECVTTL, INET_OPT_TTL); #endif #undef PUT_CMSG_INT32 + } put_int32(arg_sz, arg_ptr); /* Put total length */ continue; } diff --git a/erts/emulator/nifs/common/prim_file_nif.c b/erts/emulator/nifs/common/prim_file_nif.c index 9b98ac3f9a..0ada345442 100644 --- a/erts/emulator/nifs/common/prim_file_nif.c +++ b/erts/emulator/nifs/common/prim_file_nif.c @@ -38,6 +38,9 @@ static void unload(ErlNifEnv *env, void* priv_data); static ErlNifResourceType *efile_resource_type; +static ERL_NIF_TERM am_erts_prim_file; +static ERL_NIF_TERM am_close; + static ERL_NIF_TERM am_ok; static ERL_NIF_TERM am_error; static ERL_NIF_TERM am_continue; @@ -96,11 +99,14 @@ static ERL_NIF_TERM set_cwd_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM arg static ERL_NIF_TERM read_file_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]); +static ERL_NIF_TERM open_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]); +static ERL_NIF_TERM close_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]); + +/* Internal ops */ +static ERL_NIF_TERM delayed_close_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]); static ERL_NIF_TERM get_handle_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]); static ERL_NIF_TERM altname_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]); -static ERL_NIF_TERM open_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]); - /* All file handle operations are passed through a wrapper that handles state * transitions, marking it as busy during the course of the operation, and * closing on completion if the owner died in the middle of an operation. @@ -128,7 +134,11 @@ static ERL_NIF_TERM open_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[] * * CLOSE_PENDING -> * CLOSED (file_handle_wrapper) - */ + * + * Should the owner of a file die, we can't close it immediately as that could + * potentially block a normal scheduler. When entering the CLOSED state from + * owner_death_callback, we will instead send a message to the erts_prim_file + * process that will then close the file through delayed_close_nif. */ typedef ERL_NIF_TERM (*file_op_impl_t)(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]); @@ -142,7 +152,6 @@ static ERL_NIF_TERM file_handle_wrapper(file_op_impl_t operation, ErlNifEnv *env return file_handle_wrapper( name ## _impl , env, argc, argv); \ } -WRAP_FILE_HANDLE_EXPORT(close_nif) WRAP_FILE_HANDLE_EXPORT(read_nif) WRAP_FILE_HANDLE_EXPORT(write_nif) WRAP_FILE_HANDLE_EXPORT(pread_nif) @@ -193,18 +202,27 @@ static ErlNifFunc nif_funcs[] = { /* Internal ops. */ {"get_handle_nif", 1, get_handle_nif}, + {"delayed_close_nif", 1, delayed_close_nif, ERL_NIF_DIRTY_JOB_IO_BOUND}, {"altname_nif", 1, altname_nif, ERL_NIF_DIRTY_JOB_IO_BOUND}, }; ERL_NIF_INIT(prim_file, nif_funcs, load, NULL, upgrade, unload) +static ErlNifPid erts_prim_file_pid; + static void owner_death_callback(ErlNifEnv* env, void* obj, ErlNifPid* pid, ErlNifMonitor* mon); -static void gc_callback(ErlNifEnv *env, void* data); -static int load(ErlNifEnv *env, void** priv_data, ERL_NIF_TERM load_info) +static int load(ErlNifEnv *env, void** priv_data, ERL_NIF_TERM prim_file_pid) { ErlNifResourceTypeInit callbacks; + if(!enif_get_local_pid(env, prim_file_pid, &erts_prim_file_pid)) { + ASSERT(!"bad pid passed to prim_file_nif"); + } + + am_erts_prim_file = enif_make_atom(env, "erts_prim_file"); + am_close = enif_make_atom(env, "close"); + am_ok = enif_make_atom(env, "ok"); am_error = enif_make_atom(env, "error"); am_continue = enif_make_atom(env, "continue"); @@ -239,7 +257,7 @@ static int load(ErlNifEnv *env, void** priv_data, ERL_NIF_TERM load_info) am_eof = enif_make_atom(env, "eof"); callbacks.down = owner_death_callback; - callbacks.dtor = gc_callback; + callbacks.dtor = NULL; callbacks.stop = NULL; efile_resource_type = enif_open_resource_type_x(env, "efile", &callbacks, @@ -305,8 +323,10 @@ static ERL_NIF_TERM file_handle_wrapper(file_op_impl_t operation, ErlNifEnv *env /* This is the only point where a change from CLOSE_PENDING is * possible, and we're running synchronously, so we can't race with * anything else here. */ + posix_errno_t ignored; + erts_atomic32_set_acqb(&d->state, EFILE_STATE_CLOSED); - efile_close(d); + efile_close(d, &ignored); } } else { /* CLOSE_PENDING should be impossible at this point since it requires @@ -319,6 +339,24 @@ static ERL_NIF_TERM file_handle_wrapper(file_op_impl_t operation, ErlNifEnv *env return result; } +/* This is a special close operation used by the erts_prim_file process for + * cleaning up orphaned files. It differs from the ordinary close_nif in that + * it only works for files that have already entered the CLOSED state. */ +static ERL_NIF_TERM delayed_close_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) { + posix_errno_t ignored; + efile_data_t *d; + + ASSERT(argc == 1); + if(!get_file_data(env, argv[0], &d)) { + return enif_make_badarg(env); + } + + ASSERT(erts_atomic32_read_acqb(&d->state) == EFILE_STATE_CLOSED); + efile_close(d, &ignored); + + return am_ok; +} + static void owner_death_callback(ErlNifEnv* env, void* obj, ErlNifPid* pid, ErlNifMonitor* mon) { efile_data_t *d = (efile_data_t*)obj; @@ -334,8 +372,24 @@ static void owner_death_callback(ErlNifEnv* env, void* obj, ErlNifPid* pid, ErlN switch(previous_state) { case EFILE_STATE_IDLE: - efile_close(d); - return; + { + /* We cannot close the file here as that could block a normal + * scheduler, so we tell erts_prim_file to do it for us. + * + * This can in turn become a bottleneck (especially in cases + * like NFS failure), but it's less problematic than blocking + * thread progress. */ + ERL_NIF_TERM message, file_ref; + + file_ref = enif_make_resource(env, d); + message = enif_make_tuple2(env, am_close, file_ref); + + if(!enif_send(env, &erts_prim_file_pid, NULL, message)) { + ERTS_INTERNAL_ERROR("Failed to defer prim_file close."); + } + + return; + } case EFILE_STATE_CLOSE_PENDING: case EFILE_STATE_CLOSED: /* We're either already closed or managed to mark ourselves for @@ -352,24 +406,6 @@ static void owner_death_callback(ErlNifEnv* env, void* obj, ErlNifPid* pid, ErlN } } -static void gc_callback(ErlNifEnv *env, void* data) { - efile_data_t *d = (efile_data_t*)data; - - enum efile_state_t previous_state; - - (void)env; - - previous_state = erts_atomic32_cmpxchg_acqb(&d->state, - EFILE_STATE_CLOSED, EFILE_STATE_IDLE); - - ASSERT(previous_state != EFILE_STATE_CLOSE_PENDING && - previous_state != EFILE_STATE_BUSY); - - if(previous_state == EFILE_STATE_IDLE) { - efile_close(d); - } -} - static ERL_NIF_TERM efile_filetype_to_atom(enum efile_filetype_t type) { switch(type) { case EFILE_FILETYPE_DEVICE: return am_device; @@ -455,38 +491,62 @@ static ERL_NIF_TERM open_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[] return posix_error_to_tuple(env, posix_errno); } - result = enif_make_resource(env, d); - enif_release_resource(d); - enif_self(env, &controlling_process); if(enif_monitor_process(env, d, &controlling_process, &d->monitor)) { + /* We need to close the file manually as we haven't registered a + * destructor. */ + posix_errno_t ignored; + + erts_atomic32_set_acqb(&d->state, EFILE_STATE_CLOSED); + efile_close(d, &ignored); + return posix_error_to_tuple(env, EINVAL); } + /* Note that we do not call enif_release_resource at this point. While it's + * normally safe to leave resource management to the GC, efile_close is a + * blocking operation which must not be done in the GC callback, and we + * can't defer it as the resource is gone as soon as it returns. + * + * We instead keep the resource alive until efile_close is called, after + * which it's safe to leave things to the GC. If the controlling process + * were to die before the user had a chance to close their file, the above + * monitor will tell the erts_prim_file process to close it for them. */ + result = enif_make_resource(env, d); + return enif_make_tuple2(env, am_ok, result); } -static ERL_NIF_TERM close_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) { +static ERL_NIF_TERM close_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) { enum efile_state_t previous_state; + efile_data_t *d; - ASSERT(argc == 0); + ASSERT(argc == 1); + if(!get_file_data(env, argv[0], &d)) { + return enif_make_badarg(env); + } previous_state = erts_atomic32_cmpxchg_acqb(&d->state, - EFILE_STATE_CLOSED, EFILE_STATE_BUSY); + EFILE_STATE_CLOSED, EFILE_STATE_IDLE); - ASSERT(previous_state == EFILE_STATE_CLOSE_PENDING || - previous_state == EFILE_STATE_BUSY); + if(previous_state == EFILE_STATE_IDLE) { + posix_errno_t error; - if(previous_state == EFILE_STATE_BUSY) { enif_demonitor_process(env, d, &d->monitor); - if(!efile_close(d)) { - return posix_error_to_tuple(env, d->posix_errno); + if(!efile_close(d, &error)) { + return posix_error_to_tuple(env, error); } - } - return am_ok; + return am_ok; + } else { + /* CLOSE_PENDING should be impossible at this point since it requires + * a transition from BUSY; the only valid state here is CLOSED. */ + ASSERT(previous_state == EFILE_STATE_CLOSED); + + return posix_error_to_tuple(env, EINVAL); + } } static ERL_NIF_TERM read_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) { @@ -514,6 +574,7 @@ static ERL_NIF_TERM read_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, con ASSERT(bytes_read <= block_size); if(bytes_read < 0) { + enif_release_binary(&result); return posix_error_to_tuple(env, d->posix_errno); } else if(bytes_read == 0) { enif_release_binary(&result); @@ -577,6 +638,7 @@ static ERL_NIF_TERM pread_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, co bytes_read = efile_preadv(d, offset, read_vec, 1); if(bytes_read < 0) { + enif_release_binary(&result); return posix_error_to_tuple(env, d->posix_errno); } else if(bytes_read == 0) { enif_release_binary(&result); @@ -804,6 +866,7 @@ static ERL_NIF_TERM ipread_s32bu_p32bu_nif_impl(efile_data_t *d, ErlNifEnv *env, bytes_read = efile_preadv(d, payload_offset, read_vec, 1); if(bytes_read < 0) { + enif_release_binary(&payload); return posix_error_to_tuple(env, d->posix_errno); } else if(bytes_read == 0) { enif_release_binary(&payload); @@ -1171,7 +1234,7 @@ static posix_errno_t read_file(efile_data_t *d, size_t size, ErlNifBinary *resul } static ERL_NIF_TERM read_file_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) { - posix_errno_t posix_errno; + posix_errno_t posix_errno, ignored; efile_fileinfo_t info = {0}; efile_path_t path; @@ -1190,7 +1253,9 @@ static ERL_NIF_TERM read_file_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM a } posix_errno = read_file(d, info.size, &result); - enif_release_resource(d); + + erts_atomic32_set_acqb(&d->state, EFILE_STATE_CLOSED); + efile_close(d, &ignored); if(posix_errno) { return posix_error_to_tuple(env, posix_errno); diff --git a/erts/emulator/nifs/common/prim_file_nif.h b/erts/emulator/nifs/common/prim_file_nif.h index 099c06c48c..b2e30c59dd 100644 --- a/erts/emulator/nifs/common/prim_file_nif.h +++ b/erts/emulator/nifs/common/prim_file_nif.h @@ -159,8 +159,11 @@ posix_errno_t efile_open(const efile_path_t *path, enum efile_modes_t modes, ErlNifResourceType *nif_type, efile_data_t **d); /** @brief Closes a file. The file must have entered the CLOSED state prior to - * calling this to prevent double close. */ -int efile_close(efile_data_t *d); + * calling this to prevent double close. + * + * Note that the file is completely invalid after this point, so the error code + * is provided in \c error rather than d->posix_errno. */ +int efile_close(efile_data_t *d, posix_errno_t *error); /* **** **** **** **** **** **** **** **** **** **** **** **** **** **** **** */ diff --git a/erts/emulator/nifs/unix/unix_prim_file.c b/erts/emulator/nifs/unix/unix_prim_file.c index dea73db18a..169b193993 100644 --- a/erts/emulator/nifs/unix/unix_prim_file.c +++ b/erts/emulator/nifs/unix/unix_prim_file.c @@ -202,21 +202,24 @@ posix_errno_t efile_open(const efile_path_t *path, enum efile_modes_t modes, return errno; } -int efile_close(efile_data_t *d) { +int efile_close(efile_data_t *d, posix_errno_t *error) { efile_unix_t *u = (efile_unix_t*)d; int fd; + ASSERT(enif_thread_type() == ERL_NIF_THR_DIRTY_IO_SCHEDULER); ASSERT(erts_atomic32_read_nob(&d->state) == EFILE_STATE_CLOSED); ASSERT(u->fd != -1); fd = u->fd; u->fd = -1; + enif_release_resource(d); + /* close(2) either always closes (*BSD, Linux) or leaves the fd in an * undefined state (POSIX 2008, Solaris), so we must not retry on EINTR. */ if(close(fd) < 0) { - u->common.posix_errno = errno; + *error = errno; return 0; } diff --git a/erts/emulator/nifs/win32/win_prim_file.c b/erts/emulator/nifs/win32/win_prim_file.c index f7fae3c637..d0aa70542f 100644 --- a/erts/emulator/nifs/win32/win_prim_file.c +++ b/erts/emulator/nifs/win32/win_prim_file.c @@ -33,16 +33,32 @@ #define FILE_SHARE_FLAGS (FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE) -#define LP_PREFIX L"\\\\?\\" -#define LP_PREFIX_SIZE (sizeof(LP_PREFIX) - sizeof(WCHAR)) +/* Long paths can either be in the file (?) or the device (.) namespace. UNC + * paths are always in the file namespace. */ +#define LP_FILE_PREFIX L"\\\\?\\" +#define LP_DEV_PREFIX L"\\\\.\\" +#define LP_UNC_PREFIX (LP_FILE_PREFIX L"UNC\\") + +#define LP_PREFIX_SIZE (sizeof(LP_FILE_PREFIX) - sizeof(WCHAR)) #define LP_PREFIX_LENGTH (LP_PREFIX_SIZE / sizeof(WCHAR)) +#define LP_UNC_PREFIX_SIZE (sizeof(LP_UNC_PREFIX) - sizeof(WCHAR)) +#define LP_UNC_PREFIX_LENGTH (LP_UNC_PREFIX_SIZE / sizeof(WCHAR)) + +#define IS_LONG_PATH(length, data) \ + ((length) >= LP_PREFIX_LENGTH && \ + (!sys_memcmp((data), LP_FILE_PREFIX, LP_PREFIX_SIZE) || \ + !sys_memcmp((data), LP_DEV_PREFIX, LP_PREFIX_SIZE))) + +#define IS_LONG_UNC_PATH(length, data) \ + ((length) >= LP_UNC_PREFIX_LENGTH && \ + !sys_memcmp((data), LP_UNC_PREFIX, LP_UNC_PREFIX_SIZE)) + #define PATH_LENGTH(path) (path->size / sizeof(WCHAR) - 1) #define ASSERT_PATH_FORMAT(path) \ do { \ - ASSERT(PATH_LENGTH(path) >= 4 && \ - !memcmp(path->data, LP_PREFIX, LP_PREFIX_SIZE)); \ + ASSERT(IS_LONG_PATH(PATH_LENGTH(path), (path)->data)); \ ASSERT(PATH_LENGTH(path) == wcslen((WCHAR*)path->data)); \ } while(0) @@ -106,7 +122,7 @@ static posix_errno_t get_full_path(ErlNifEnv *env, WCHAR *input, efile_path_t *r return ENOENT; } - maximum_length += LP_PREFIX_LENGTH; + maximum_length += MAX(LP_PREFIX_LENGTH, LP_UNC_PREFIX_LENGTH); if(!enif_alloc_binary(maximum_length * sizeof(WCHAR), result)) { return ENOMEM; @@ -115,18 +131,28 @@ static posix_errno_t get_full_path(ErlNifEnv *env, WCHAR *input, efile_path_t *r actual_length = GetFullPathNameW(input, maximum_length, (WCHAR*)result->data, NULL); if(actual_length < maximum_length) { - int has_long_path_prefix; + int is_long_path, maybe_unc_path; WCHAR *path_start; - /* Make sure we have a long-path prefix; GetFullPathNameW only adds one - * if the path is relative. */ - has_long_path_prefix = actual_length >= LP_PREFIX_LENGTH && - !sys_memcmp(result->data, LP_PREFIX, LP_PREFIX_SIZE); - - if(!has_long_path_prefix) { + /* The APIs we use have varying path length limits and sometimes + * behave differently when given a long-path prefix, so it's simplest + * to always use long paths. */ + + is_long_path = IS_LONG_PATH(actual_length, result->data); + maybe_unc_path = !sys_memcmp(result->data, L"\\\\", sizeof(WCHAR) * 2); + + if(maybe_unc_path && !is_long_path) { + /* \\localhost\c$\gurka -> \\?\UNC\localhost\c$\gurka */ + sys_memmove(result->data + LP_UNC_PREFIX_SIZE, + &((WCHAR*)result->data)[2], + (actual_length - 1) * sizeof(WCHAR)); + sys_memcpy(result->data, LP_UNC_PREFIX, LP_UNC_PREFIX_SIZE); + actual_length += LP_UNC_PREFIX_LENGTH; + } else if(!is_long_path) { + /* C:\gurka -> \\?\C:\gurka */ sys_memmove(result->data + LP_PREFIX_SIZE, result->data, (actual_length + 1) * sizeof(WCHAR)); - sys_memcpy(result->data, LP_PREFIX, LP_PREFIX_SIZE); + sys_memcpy(result->data, LP_FILE_PREFIX, LP_PREFIX_SIZE); actual_length += LP_PREFIX_LENGTH; } @@ -200,13 +226,19 @@ static int normalize_path_result(ErlNifBinary *path) { ASSERT(length < path->size / sizeof(WCHAR)); /* Get rid of the long-path prefix, if present. */ - if(length >= LP_PREFIX_LENGTH) { - if(!sys_memcmp(path_start, LP_PREFIX, LP_PREFIX_SIZE)) { - length -= LP_PREFIX_LENGTH; - sys_memmove(path_start, &path_start[LP_PREFIX_LENGTH], - length * sizeof(WCHAR)); - } + if(IS_LONG_UNC_PATH(length, path_start)) { + /* The first two characters (\\) are the same for both long and short + * UNC paths. */ + sys_memmove(&path_start[2], &path_start[LP_UNC_PREFIX_LENGTH], + (length - LP_UNC_PREFIX_LENGTH) * sizeof(WCHAR)); + + length -= LP_UNC_PREFIX_LENGTH - 2; + } else if(IS_LONG_PATH(length, path_start)) { + length -= LP_PREFIX_LENGTH; + + sys_memmove(path_start, &path_start[LP_PREFIX_LENGTH], + length * sizeof(WCHAR)); } path_end = &path_start[length]; @@ -318,49 +350,55 @@ static int has_same_mount_point(const efile_path_t *path_a, const efile_path_t * /* Mirrors the PathIsRootW function of the shell API, but doesn't choke on * paths longer than MAX_PATH. */ static int is_path_root(const efile_path_t *path) { - const WCHAR *path_start, *path_end; + const WCHAR *path_start, *path_end, *path_iterator; int length; ASSERT_PATH_FORMAT(path); - path_start = (WCHAR*)path->data + LP_PREFIX_LENGTH; - length = PATH_LENGTH(path) - LP_PREFIX_LENGTH; - - path_end = &path_start[length]; + if(!IS_LONG_UNC_PATH(PATH_LENGTH(path), path->data)) { + path_start = (WCHAR*)path->data + LP_PREFIX_LENGTH; + length = PATH_LENGTH(path) - LP_PREFIX_LENGTH; - if(length == 1) { /* A single \ refers to the root of the current working directory. */ - return IS_SLASH(path_start[0]); - } else if(length == 3 && iswalpha(path_start[0]) && path_start[1] == L':') { - /* Drive letter. */ - return IS_SLASH(path_start[2]); - } else if(length >= 4) { - /* Check whether we're a UNC root, eg. \\server, \\server\share */ - const WCHAR *path_iterator; + if(length == 1) { + return IS_SLASH(path_start[0]); + } - if(!IS_SLASH(path_start[0]) || !IS_SLASH(path_start[1])) { - return 0; + /* Drive letter. */ + if(length == 3 && iswalpha(path_start[0]) && path_start[1] == L':') { + return IS_SLASH(path_start[2]); } - path_iterator = path_start + 2; + return 0; + } - /* Slide to the slash between the server and share names, if present. */ - while(path_iterator < path_end && !IS_SLASH(*path_iterator)) { - path_iterator++; - } + /* Check whether we're a UNC root, eg. \\server, \\server\share */ - /* Slide past the end of the string, stopping at the first slash we - * encounter. */ - do { - path_iterator++; - } while(path_iterator < path_end && !IS_SLASH(*path_iterator)); + path_start = (WCHAR*)path->data + LP_UNC_PREFIX_LENGTH; + length = PATH_LENGTH(path) - LP_UNC_PREFIX_LENGTH; - /* If we're past the end of the string and it didnt't end with a slash, - * then we're a root path. */ - return path_iterator >= path_end && !IS_SLASH(path_start[length - 1]); + path_end = &path_start[length]; + path_iterator = path_start; + + /* Server name must be at least one character. */ + if(length <= 1) { + return 0; } - return 0; + /* Slide to the slash between the server and share names, if present. */ + while(path_iterator < path_end && !IS_SLASH(*path_iterator)) { + path_iterator++; + } + + /* Slide past the end of the string, stopping at the first slash we + * encounter. */ + do { + path_iterator++; + } while(path_iterator < path_end && !IS_SLASH(*path_iterator)); + + /* If we're past the end of the string and it didnt't end with a slash, + * then we're a root path. */ + return path_iterator >= path_end && !IS_SLASH(path_start[length - 1]); } posix_errno_t efile_open(const efile_path_t *path, enum efile_modes_t modes, @@ -428,18 +466,21 @@ posix_errno_t efile_open(const efile_path_t *path, enum efile_modes_t modes, } } -int efile_close(efile_data_t *d) { +int efile_close(efile_data_t *d, posix_errno_t *error) { efile_win_t *w = (efile_win_t*)d; HANDLE handle; + ASSERT(enif_thread_type() == ERL_NIF_THR_DIRTY_IO_SCHEDULER); ASSERT(erts_atomic32_read_nob(&d->state) == EFILE_STATE_CLOSED); ASSERT(w->handle != INVALID_HANDLE_VALUE); handle = w->handle; w->handle = INVALID_HANDLE_VALUE; + enif_release_resource(d); + if(!CloseHandle(handle)) { - w->common.posix_errno = windows_to_posix_errno(GetLastError()); + *error = windows_to_posix_errno(GetLastError()); return 0; } @@ -687,7 +728,7 @@ static int is_name_surrogate(const efile_path_t *path) { if(handle != INVALID_HANDLE_VALUE) { REPARSE_GUID_DATA_BUFFER reparse_buffer; - LPDWORD unused_length; + DWORD unused_length; BOOL success; success = DeviceIoControl(handle, @@ -1248,11 +1289,22 @@ posix_errno_t efile_set_cwd(const efile_path_t *path) { /* We have to use _wchdir since that's the only function that updates the * per-drive working directory, but it naively assumes that all paths - * starting with \\ are UNC paths, so we have to skip the \\?\-prefix. */ - path_start = (WCHAR*)path->data + LP_PREFIX_LENGTH; + * starting with \\ are UNC paths, so we have to skip the long-path prefix. + * + * _wchdir doesn't handle long-prefixed UNC paths either so we hand those + * to SetCurrentDirectoryW instead. The per-drive working directory is + * irrelevant for such paths anyway. */ - if(_wchdir(path_start)) { - return windows_to_posix_errno(GetLastError()); + if(!IS_LONG_UNC_PATH(PATH_LENGTH(path), path->data)) { + path_start = (WCHAR*)path->data + LP_PREFIX_LENGTH; + + if(_wchdir(path_start)) { + return windows_to_posix_errno(GetLastError()); + } + } else { + if(!SetCurrentDirectoryW((WCHAR*)path->data)) { + return windows_to_posix_errno(GetLastError()); + } } return 0; @@ -1333,7 +1385,7 @@ posix_errno_t efile_altname(ErlNifEnv *env, const efile_path_t *path, ERL_NIF_TE int name_length; /* Reject path wildcards. */ - if(wcspbrk(&((const WCHAR*)path->data)[4], L"?*")) { + if(wcspbrk(&((const WCHAR*)path->data)[LP_PREFIX_LENGTH], L"?*")) { return ENOENT; } diff --git a/erts/emulator/sys/unix/sys_uds.c b/erts/emulator/sys/unix/sys_uds.c index 39a4866065..c9f73622ba 100644 --- a/erts/emulator/sys/unix/sys_uds.c +++ b/erts/emulator/sys/unix/sys_uds.c @@ -88,8 +88,9 @@ sys_uds_readv(int fd, struct iovec *iov, size_t iov_len, if((msg.msg_flags & MSG_CTRUNC) == MSG_CTRUNC) { /* We assume that we have given enough space for any header - that are sent to us. So the only remaining reason to get - this flag set is if the caller has run out of file descriptors. + that are sent to us. So the only remaining reasons to get + this flag set is if the caller has run out of file descriptors + or an SELinux policy prunes the response (eg. O_APPEND on STDERR). */ errno = EMFILE; return -1; diff --git a/erts/emulator/test/Makefile b/erts/emulator/test/Makefile index bf00de2204..6a064ec8d4 100644 --- a/erts/emulator/test/Makefile +++ b/erts/emulator/test/Makefile @@ -33,6 +33,7 @@ MODULES= \ after_SUITE \ alloc_SUITE \ async_ports_SUITE \ + atomics_SUITE \ beam_SUITE \ beam_literals_SUITE \ bif_SUITE \ @@ -50,6 +51,7 @@ MODULES= \ call_trace_SUITE \ code_SUITE \ code_parallel_load_SUITE \ + counters_SUITE \ crypto_SUITE \ ddll_SUITE \ decode_packet_SUITE \ @@ -92,6 +94,7 @@ MODULES= \ port_SUITE \ port_bif_SUITE \ prim_eval_SUITE \ + persistent_term_SUITE \ process_SUITE \ pseudoknot_SUITE \ receive_SUITE \ diff --git a/erts/emulator/test/atomics_SUITE.erl b/erts/emulator/test/atomics_SUITE.erl new file mode 100644 index 0000000000..8c42354770 --- /dev/null +++ b/erts/emulator/test/atomics_SUITE.erl @@ -0,0 +1,147 @@ +%% +%% %CopyrightBegin% +%% +%% Copyright Ericsson AB 2018. All Rights Reserved. +%% +%% Licensed under the Apache License, Version 2.0 (the "License"); +%% you may not use this file except in compliance with the License. +%% You may obtain a copy of the License at +%% +%% http://www.apache.org/licenses/LICENSE-2.0 +%% +%% Unless required by applicable law or agreed to in writing, software +%% distributed under the License is distributed on an "AS IS" BASIS, +%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +%% See the License for the specific language governing permissions and +%% limitations under the License. +%% +%% %CopyrightEnd% +%% +-module(atomics_SUITE). + +-include_lib("common_test/include/ct.hrl"). + +-compile(export_all). + +suite() -> [{ct_hooks,[ts_install_cth]}]. + +all() -> + [signed, unsigned, bad, signed_limits, unsigned_limits]. + +signed(Config) when is_list(Config) -> + Size = 10, + Ref = atomics:new(Size,[]), + #{size:=Size, memory:=Memory} = atomics:info(Ref), + {_,true} = {Memory, Memory > Size*8}, + {_,true} = {Memory, Memory < Size*max_atomic_sz() + 100}, + [signed_do(Ref, Ix) || Ix <- lists:seq(1, Size)], + ok. + +signed_do(Ref, Ix) -> + 0 = atomics:get(Ref, Ix), + ok = atomics:put(Ref, Ix, 3), + ok = atomics:add(Ref, Ix, 14), + 17 = atomics:get(Ref, Ix), + 20 = atomics:add_get(Ref, Ix, 3), + -3 = atomics:add_get(Ref, Ix, -23), + 17 = atomics:add_get(Ref, Ix, 20), + ok = atomics:sub(Ref, Ix, 4), + 13 = atomics:get(Ref, Ix), + -7 = atomics:sub_get(Ref, Ix, 20), + 3 = atomics:sub_get(Ref, Ix, -10), + 3 = atomics:exchange(Ref, Ix, 666), + ok = atomics:compare_exchange(Ref, Ix, 666, 777), + 777 = atomics:compare_exchange(Ref, Ix, 666, -666), + ok. + +unsigned(Config) when is_list(Config) -> + Size = 10, + Ref = atomics:new(Size,[{signed, false}]), + #{size:=Size, memory:=Memory} = atomics:info(Ref), + true = Memory > Size*8, + true = Memory < Size*max_atomic_sz() + 100, + [unsigned_do(Ref, Ix) || Ix <- lists:seq(1, Size)], + ok. + +unsigned_do(Ref, Ix) -> + 0 = atomics:get(Ref, Ix), + ok = atomics:put(Ref, Ix, 3), + ok = atomics:add(Ref, Ix, 14), + 17 = atomics:get(Ref, Ix), + 20 = atomics:add_get(Ref, Ix, 3), + ok = atomics:sub(Ref, Ix, 7), + 13 = atomics:get(Ref, Ix), + 3 = atomics:sub_get(Ref, Ix, 10), + 3 = atomics:exchange(Ref, Ix, 666), + ok = atomics:compare_exchange(Ref, Ix, 666, 777), + 777 = atomics:compare_exchange(Ref, Ix, 666, 888), + ok. + +bad(Config) when is_list(Config) -> + {'EXIT',{badarg,_}} = (catch atomics:new(0,[])), + {'EXIT',{badarg,_}} = (catch atomics:new(10,[bad])), + {'EXIT',{badarg,_}} = (catch atomics:new(10,[{signed,bad}])), + {'EXIT',{badarg,_}} = (catch atomics:new(10,[{signed,true}, bad])), + {'EXIT',{badarg,_}} = (catch atomics:new(10,[{signed,false} | bad])), + Ref = atomics:new(10,[]), + {'EXIT',{badarg,_}} = (catch atomics:get(1742, 7)), + {'EXIT',{badarg,_}} = (catch atomics:get(make_ref(), 7)), + {'EXIT',{badarg,_}} = (catch atomics:get(Ref, -1)), + {'EXIT',{badarg,_}} = (catch atomics:get(Ref, 0)), + {'EXIT',{badarg,_}} = (catch atomics:get(Ref, 11)), + {'EXIT',{badarg,_}} = (catch atomics:get(Ref, 7.0)), + ok. + + +signed_limits(Config) when is_list(Config) -> + Bits = 64, + Max = (1 bsl (Bits-1)) - 1, + Min = -(1 bsl (Bits-1)), + + Ref = atomics:new(1,[{signed, true}]), + #{max:=Max, min:=Min} = atomics:info(Ref), + 0 = atomics:get(Ref, 1), + ok = atomics:add(Ref, 1, Max), + Min = atomics:add_get(Ref, 1, 1), + Max = atomics:sub_get(Ref, 1, 1), + + IncrMax = (Max bsl 1) bor 1, + ok = atomics:put(Ref, 1, 0), + ok = atomics:add(Ref, 1, IncrMax), + -1 = atomics:get(Ref, 1), + {'EXIT',{badarg,_}} = (catch atomics:add(Ref, 1, IncrMax+1)), + {'EXIT',{badarg,_}} = (catch atomics:add(Ref, 1, Min-1)), + + ok. + +unsigned_limits(Config) when is_list(Config) -> + Bits = 64, + Max = (1 bsl Bits) - 1, + Min = 0, + + Ref = atomics:new(1,[{signed,false}]), + #{max:=Max, min:=Min} = atomics:info(Ref), + 0 = atomics:get(Ref, 1), + ok = atomics:add(Ref, 1, Max), + Min = atomics:add_get(Ref, 1, 1), + Max = atomics:sub_get(Ref, 1, 1), + + {'EXIT',{badarg,_}} = (catch atomics:add(Ref, 1, Max+1)), + IncrMin = -(1 bsl (Bits-1)), + ok = atomics:put(Ref, 1, -IncrMin), + ok = atomics:add(Ref, 1, IncrMin), + 0 = atomics:get(Ref, 1), + {'EXIT',{badarg,_}} = (catch atomics:add(Ref, 1, IncrMin-1)), + + ok. + +max_atomic_sz() -> + case erlang:system_info({wordsize, external}) of + 4 -> 16; + 8 -> + EI = erlang:system_info(ethread_info), + case lists:keyfind("64-bit native atomics", 1, EI) of + {_, "no", _} -> 16; + _ -> 8 + end + end. diff --git a/erts/emulator/test/code_SUITE.erl b/erts/emulator/test/code_SUITE.erl index 7e690fd870..493c6ebe99 100644 --- a/erts/emulator/test/code_SUITE.erl +++ b/erts/emulator/test/code_SUITE.erl @@ -28,7 +28,7 @@ fake_literals/1, false_dependency/1,coverage/1,fun_confusion/1, t_copy_literals/1, t_copy_literals_frags/1, - erl_544/1]). + erl_544/1, max_heap_size/1]). -define(line_trace, 1). -include_lib("common_test/include/ct.hrl"). @@ -43,7 +43,7 @@ all() -> constant_pools, constant_refc_binaries, fake_literals, false_dependency, coverage, fun_confusion, t_copy_literals, t_copy_literals_frags, - erl_544]. + erl_544, max_heap_size]. init_per_suite(Config) -> erts_debug:set_internal_state(available_internal_state, true), @@ -978,6 +978,39 @@ erl_544(Config) when is_list(Config) -> {skipped, "Only run when native file name encoding is utf8"} end. +%% Test that the copying of literals to a process during purging of +%% literals will cause the process to be killed if the max heap size +%% is exceeded. +max_heap_size(_Config) -> + Mod = ?FUNCTION_NAME, + Value = [I || I <- lists:seq(1, 5000)], + Code = gen_lit(Mod, [{term,Value}]), + {module,Mod} = erlang:load_module(Mod, Code), + SpawnOpts = [monitor, + {max_heap_size, + #{size=>1024, + kill=>true, + error_logger=>true}}], + {Pid,Ref} = spawn_opt(fun() -> + max_heap_size_proc(Mod) + end, SpawnOpts), + receive + {'DOWN',Ref,process,Pid,Reason} -> + killed = Reason; + Other -> + ct:fail({unexpected_message,Other}) + after 10000 -> + ct:fail({process_did_not_die, Pid, erlang:process_info(Pid)}) + end. + +max_heap_size_proc(Mod) -> + Value = Mod:term(), + code:delete(Mod), + code:purge(Mod), + receive + _ -> Value + end. + %% Utilities. make_sub_binary(Bin) when is_binary(Bin) -> diff --git a/erts/emulator/test/counters_SUITE.erl b/erts/emulator/test/counters_SUITE.erl new file mode 100644 index 0000000000..7de164096b --- /dev/null +++ b/erts/emulator/test/counters_SUITE.erl @@ -0,0 +1,112 @@ +%% +%% %CopyrightBegin% +%% +%% Copyright Ericsson AB 2018. All Rights Reserved. +%% +%% Licensed under the Apache License, Version 2.0 (the "License"); +%% you may not use this file except in compliance with the License. +%% You may obtain a copy of the License at +%% +%% http://www.apache.org/licenses/LICENSE-2.0 +%% +%% Unless required by applicable law or agreed to in writing, software +%% distributed under the License is distributed on an "AS IS" BASIS, +%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +%% See the License for the specific language governing permissions and +%% limitations under the License. +%% +%% %CopyrightEnd% +%% +-module(counters_SUITE). + +-include_lib("common_test/include/ct.hrl"). + +-compile(export_all). + +suite() -> [{ct_hooks,[ts_install_cth]}]. + +all() -> + [basic, bad, limits]. + +basic(Config) when is_list(Config) -> + Size = 10, + [begin + Ref = counters:new(Size,[Type]), + #{size:=Size, memory:=Memory} = counters:info(Ref), + check_memory(Type, Memory, Size), + [basic_do(Ref, Ix) || Ix <- lists:seq(1, Size)] + end + || Type <- [atomics, write_concurrency]], + ok. + +basic_do(Ref, Ix) -> + 0 = counters:get(Ref, Ix), + ok = counters:add(Ref, Ix, 3), + 3 = counters:get(Ref, Ix), + ok = counters:add(Ref, Ix, 14), + 17 = counters:get(Ref, Ix), + ok = counters:add(Ref, Ix, -20), + -3 = counters:get(Ref, Ix), + ok = counters:add(Ref, Ix, 100), + 97 = counters:get(Ref, Ix), + ok = counters:sub(Ref, Ix, 20), + 77 = counters:get(Ref, Ix), + ok = counters:sub(Ref, Ix, -10), + 87 = counters:get(Ref, Ix), + ok. + +check_memory(atomics, Memory, Size) -> + {_,true} = {Memory, Memory > Size*8}, + {_,true} = {Memory, Memory < Size*max_atomic_sz() + 100}; +check_memory(write_concurrency, Memory, Size) -> + NScheds = erlang:system_info(schedulers), + {_,true} = {Memory, Memory > NScheds*Size*8}, + {_,true} = {Memory, Memory < NScheds*(Size+7)*max_atomic_sz() + 100}. + +max_atomic_sz() -> + case erlang:system_info({wordsize, external}) of + 4 -> 16; + 8 -> + EI = erlang:system_info(ethread_info), + case lists:keyfind("64-bit native atomics", 1, EI) of + {_, "no", _} -> 16; + _ -> 8 + end + end. + +bad(Config) when is_list(Config) -> + {'EXIT',{badarg,_}} = (catch counters:new(0,[])), + {'EXIT',{badarg,_}} = (catch counters:new(10,[bad])), + {'EXIT',{badarg,_}} = (catch counters:new(10,[atomic, bad])), + {'EXIT',{badarg,_}} = (catch counters:new(10,[write_concurrency | bad])), + Ref = counters:new(10,[]), + {'EXIT',{badarg,_}} = (catch counters:get(1742, 7)), + {'EXIT',{badarg,_}} = (catch counters:get(make_ref(), 7)), + {'EXIT',{badarg,_}} = (catch counters:get(Ref, -1)), + {'EXIT',{badarg,_}} = (catch counters:get(Ref, 0)), + {'EXIT',{badarg,_}} = (catch counters:get(Ref, 11)), + {'EXIT',{badarg,_}} = (catch counters:get(Ref, 7.0)), + ok. + + +limits(Config) when is_list(Config) -> + Bits = 64, + Max = (1 bsl (Bits-1)) - 1, + Min = -(1 bsl (Bits-1)), + + Ref = counters:new(1,[]), + 0 = counters:get(Ref, 1), + ok = counters:add(Ref, 1, Max), + ok = counters:add(Ref, 1, 1), + Min = counters:get(Ref, 1), + ok = counters:sub(Ref, 1, 1), + Max = counters:get(Ref, 1), + + IncrMax = (Max bsl 1) bor 1, + ok = counters:sub(Ref, 1, counters:get(Ref, 1)), + ok = counters:add(Ref, 1, IncrMax), + -1 = counters:get(Ref, 1), + {'EXIT',{badarg,_}} = (catch counters:add(Ref, 1, IncrMax+1)), + {'EXIT',{badarg,_}} = (catch counters:add(Ref, 1, Min-1)), + + ok. diff --git a/erts/emulator/test/persistent_term_SUITE.erl b/erts/emulator/test/persistent_term_SUITE.erl new file mode 100644 index 0000000000..58cd3276b0 --- /dev/null +++ b/erts/emulator/test/persistent_term_SUITE.erl @@ -0,0 +1,614 @@ +%% +%% %CopyrightBegin% +%% +%% Copyright Ericsson AB 2017. All Rights Reserved. +%% +%% Licensed under the Apache License, Version 2.0 (the "License"); +%% you may not use this file except in compliance with the License. +%% You may obtain a copy of the License at +%5 +%% http://www.apache.org/licenses/LICENSE-2.0 +%% +%% Unless required by applicable law or agreed to in writing, software +%% distributed under the License is distributed on an "AS IS" BASIS, +%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +%% See the License for the specific language governing permissions and +%% limitations under the License. +%% +%% %CopyrightEnd% +%% + +-module(persistent_term_SUITE). +-include_lib("common_test/include/ct.hrl"). + +-export([all/0,suite/0, + basic/1,purging/1,sharing/1,get_trapping/1, + info/1,info_trapping/1,killed_while_trapping/1, + off_heap_values/1,keys/1,collisions/1, + init_restart/1]). + +%% +-export([test_init_restart_cmd/1]). + +suite() -> + [{ct_hooks,[ts_install_cth]}, + {timetrap,{minutes,10}}]. + +all() -> + [basic,purging,sharing,get_trapping,info,info_trapping, + killed_while_trapping,off_heap_values,keys,collisions, + init_restart]. + +basic(_Config) -> + Chk = chk(), + N = 777, + Seq = lists:seq(1, N), + par(2, N, Seq), + seq(3, Seq), + seq(3, Seq), %Same values. + _ = [begin + Key = {?MODULE,{key,I}}, + true = persistent_term:erase(Key), + false = persistent_term:erase(Key), + {'EXIT',{badarg,_}} = (catch persistent_term:get(Key)) + end || I <- Seq], + [] = [P || {{?MODULE,_},_}=P <- persistent_term:get()], + chk(Chk). + +par(C, N, Seq) -> + _ = [spawn_link(fun() -> + ok = persistent_term:put({?MODULE,{key,I}}, + {value,C*I}) + end) || I <- Seq], + Result = wait(N), + _ = [begin + Double = C*I, + {{?MODULE,{key,I}},{value,Double}} = Res + end || {I,Res} <- lists:zip(Seq, Result)], + ok. + +seq(C, Seq) -> + _ = [ok = persistent_term:put({?MODULE,{key,I}}, {value,C*I}) || + I <- Seq], + All = persistent_term:get(), + All = [P || {{?MODULE,_},_}=P <- persistent_term:get()], + All = [{Key,persistent_term:get(Key)} || {Key,_} <- All], + Result = lists:sort(All), + _ = [begin + Double = C*I, + {{?MODULE,{key,I}},{value,Double}} = Res + end || {I,Res} <- lists:zip(Seq, Result)], + ok. + +wait(N) -> + All = [P || {{?MODULE,_},_}=P <- persistent_term:get()], + case length(All) of + N -> + All = [{Key,persistent_term:get(Key)} || {Key,_} <- All], + lists:sort(All); + _ -> + receive after 10 -> ok end, + wait(N) + end. + +%% Make sure that terms that have been erased are copied into all +%% processes that still hold a pointer to them. + +purging(_Config) -> + Chk = chk(), + do_purging(fun(K) -> persistent_term:put(K, {?MODULE,new}) end, + replaced), + do_purging(fun persistent_term:erase/1, erased), + chk(Chk). + +do_purging(Eraser, Type) -> + Parent = self(), + Key = {?MODULE,?FUNCTION_NAME}, + ok = persistent_term:put(Key, {term,[<<"abc",0:777/unit:8>>]}), + Ps0 = [spawn_monitor(fun() -> purging_tester(Parent, Key) end) || + _ <- lists:seq(1, 50)], + Ps = maps:from_list(Ps0), + purging_recv(gotten, Ps), + Eraser(Key), + _ = [P ! {Parent,Type} || P <- maps:keys(Ps)], + purging_wait(Ps). + +purging_recv(Tag, Ps) when map_size(Ps) > 0 -> + receive + {Pid,Tag} -> + true = is_map_key(Pid, Ps), + purging_recv(Tag, maps:remove(Pid, Ps)) + end; +purging_recv(_, _) -> ok. + +purging_wait(Ps) when map_size(Ps) > 0 -> + receive + {'DOWN',Ref,process,Pid,Reason} -> + normal = Reason, + Ref = map_get(Pid, Ps), + purging_wait(maps:remove(Pid, Ps)) + end; +purging_wait(_) -> ok. + +purging_tester(Parent, Key) -> + Term = persistent_term:get(Key), + purging_check_term(Term), + 0 = erts_debug:size_shared(Term), + Parent ! {self(),gotten}, + receive + {Parent,erased} -> + {'EXIT',{badarg,_}} = (catch persistent_term:get(Key)), + purging_tester_1(Term); + {Parent,replaced} -> + {?MODULE,new} = persistent_term:get(Key), + purging_tester_1(Term) + end. + +%% Wait for the term to be copied into this process. +purging_tester_1(Term) -> + purging_check_term(Term), + receive after 1 -> ok end, + case erts_debug:size_shared(Term) of + 0 -> + purging_tester_1(Term); + Size -> + %% The term has been copied into this process. + purging_check_term(Term), + Size = erts_debug:size(Term) + end. + +purging_check_term({term,[<<"abc",0:777/unit:8>>]}) -> + ok. + +%% Test that sharing is preserved when storing terms. + +sharing(_Config) -> + Chk = chk(), + Depth = 10, + Size = 2*Depth, + Shared = lists:foldl(fun(_, A) -> [A|A] end, + [], lists:seq(1, Depth)), + Size = erts_debug:size(Shared), + Key = {?MODULE,?FUNCTION_NAME}, + ok = persistent_term:put(Key, Shared), + SharedStored = persistent_term:get(Key), + Size = erts_debug:size(SharedStored), + 0 = erts_debug:size_shared(SharedStored), + + {Pid,Ref} = spawn_monitor(fun() -> + Term = persistent_term:get(Key), + Size = erts_debug:size(Term), + 0 = erts_debug:size_shared(Term), + true = Term =:= SharedStored + end), + receive + {'DOWN',Ref,process,Pid,normal} -> + true = persistent_term:erase(Key), + Size = erts_debug:size(SharedStored), + chk(Chk) + end. + +%% Test trapping of persistent_term:get/0. + +get_trapping(_Config) -> + Chk = chk(), + + %% Assume that the get/0 traps after 4000 iterations + %% in a non-debug emulator. + N = case test_server:timetrap_scale_factor() of + 1 -> 10000; + _ -> 1000 + end, + spawn_link(fun() -> get_trapping_create(N) end), + All = do_get_trapping(N, []), + N = get_trapping_check_result(lists:sort(All), 1), + erlang:garbage_collect(), + get_trapping_erase(N), + chk(Chk). + +do_get_trapping(N, Prev) -> + case persistent_term:get() of + Prev when length(Prev) >= N -> + All = [P || {{?MODULE,{get_trapping,_}},_}=P <- Prev], + case length(All) of + N -> All; + _ -> do_get_trapping(N, Prev) + end; + New -> + receive after 1 -> ok end, + do_get_trapping(N, New) + end. + +get_trapping_create(0) -> + ok; +get_trapping_create(N) -> + ok = persistent_term:put({?MODULE,{get_trapping,N}}, N), + get_trapping_create(N-1). + +get_trapping_check_result([{{?MODULE,{get_trapping,N}},N}|T], N) -> + get_trapping_check_result(T, N+1); +get_trapping_check_result([], N) -> N-1. + +get_trapping_erase(0) -> + ok; +get_trapping_erase(N) -> + true = persistent_term:erase({?MODULE,{get_trapping,N}}), + get_trapping_erase(N-1). + +%% Test retrieving information about persistent terms. + +info(_Config) -> + Chk = chk(), + + %% White box test of info/0. + N = 100, + try + Overhead = info_literal_area_overhead(), + io:format("Overhead = ~p\n", [Overhead]), + info_wb(N, Overhead, info_info()) + after + _ = [_ = persistent_term:erase({?MODULE,I}) || + I <- lists:seq(1, N)] + end, + + chk(Chk). + +%% White box test of persistent_term:info/0. We take into account +%% that there might already exist persistent terms (created by the +%% OTP standard libraries), but we assume that they are not +%% changed during the execution of this test case. + +info_wb(0, _, _) -> + ok; +info_wb(N, Overhead, {BaseCount,BaseMemory}) -> + Key = {?MODULE,N}, + Value = lists:seq(1, N), + ok = persistent_term:put(Key, Value), + + %% Calculate the extra memory needed for this term. + WordSize = erlang:system_info(wordsize), + ExtraMemory = Overhead + 2 * N * WordSize, + + %% Call persistent_term:info/0. + {Count,Memory} = info_info(), + + %% There should be one more persistent term. + Count = BaseCount + 1, + + %% Verify that the amount of memory is correct. + case BaseMemory + ExtraMemory of + Memory -> + %% Exactly right. The size of the hash table was not changed. + ok; + Expected -> + %% The size of the hash table has been doubled to avoid filling + %% the table to more than 50 percent. The previous number + %% of entries must have been exactly half the size of the + %% hash table. The expected number of extra words added by + %% the resizing will be twice that number. + ExtraWords = BaseCount * 2, + true = ExtraWords * WordSize =:= (Memory - Expected) + end, + info_wb(N-1, Overhead, {Count,Memory}). + +info_info() -> + #{count:=Count,memory:=Memory} = persistent_term:info(), + true = is_integer(Count) andalso Count >= 0, + true = is_integer(Memory) andalso Memory >= 0, + {Count,Memory}. + +%% Calculate the number of extra bytes needed for storing each term in +%% the literal, assuming that the key is a tuple of size 2 with +%% immediate elements. The calculated number is the size of the +%% ErtsLiteralArea struct excluding the storage for the literal term +%% itself. + +info_literal_area_overhead() -> + Key1 = {?MODULE,1}, + Key2 = {?MODULE,2}, + #{memory:=Mem0} = persistent_term:info(), + ok = persistent_term:put(Key1, literal), + #{memory:=Mem1} = persistent_term:info(), + ok = persistent_term:put(Key2, literal), + #{memory:=Mem2} = persistent_term:info(), + true = persistent_term:erase(Key1), + true = persistent_term:erase(Key2), + + %% The size of the hash table may have doubled when inserting + %% one of the keys. To avoiding counting the change in the hash + %% table size, take the smaller size increase. + min(Mem2-Mem1, Mem1-Mem0). + +%% Test trapping of persistent_term:info/0. + +info_trapping(_Config) -> + Chk = chk(), + + %% Assume that the info/0 traps after 4000 iterations + %% in a non-debug emulator. + N = case test_server:timetrap_scale_factor() of + 1 -> 10000; + _ -> 1000 + end, + spawn_link(fun() -> info_trapping_create(N) end), + All = do_info_trapping(N, 0), + N = info_trapping_check_result(lists:sort(All), 1), + erlang:garbage_collect(), + info_trapping_erase(N), + chk(Chk). + +do_info_trapping(N, PrevMem) -> + case info_info() of + {N,Mem} -> + true = Mem >= PrevMem, + All = [P || {{?MODULE,{info_trapping,_}},_}=P <- persistent_term:get()], + case length(All) of + N -> All; + _ -> do_info_trapping(N, PrevMem) + end; + {_,Mem} -> + true = Mem >= PrevMem, + receive after 1 -> ok end, + do_info_trapping(N, Mem) + end. + +info_trapping_create(0) -> + ok; +info_trapping_create(N) -> + ok = persistent_term:put({?MODULE,{info_trapping,N}}, N), + info_trapping_create(N-1). + +info_trapping_check_result([{{?MODULE,{info_trapping,N}},N}|T], N) -> + info_trapping_check_result(T, N+1); +info_trapping_check_result([], N) -> N-1. + +info_trapping_erase(0) -> + ok; +info_trapping_erase(N) -> + true = persistent_term:erase({?MODULE,{info_trapping,N}}), + info_trapping_erase(N-1). + +%% Test that hash tables are deallocated if a process running +%% persistent_term:get/0 is killed. + +killed_while_trapping(_Config) -> + Chk = chk(), + N = case test_server:timetrap_scale_factor() of + 1 -> 20000; + _ -> 2000 + end, + kwt_put(N), + kwt_spawn(10), + kwt_erase(N), + chk(Chk). + +kwt_put(0) -> + ok; +kwt_put(N) -> + ok = persistent_term:put({?MODULE,{kwt,N}}, N), + kwt_put(N-1). + +kwt_spawn(0) -> + ok; +kwt_spawn(N) -> + Pids = [spawn(fun kwt_getter/0) || _ <- lists:seq(1, 20)], + erlang:yield(), + _ = [exit(Pid, kill) || Pid <- Pids], + kwt_spawn(N-1). + +kwt_getter() -> + _ = persistent_term:get(), + kwt_getter(). + +kwt_erase(0) -> + ok; +kwt_erase(N) -> + true = persistent_term:erase({?MODULE,{kwt,N}}), + kwt_erase(N-1). + +%% Test storing off heap values (such as ref-counted binaries). + +off_heap_values(_Config) -> + Chk = chk(), + Key = {?MODULE,?FUNCTION_NAME}, + Val = {a,list_to_binary(lists:seq(0, 255)),make_ref(),fun() -> ok end}, + ok = persistent_term:put(Key, Val), + FetchedVal = persistent_term:get(Key), + Val = FetchedVal, + true = persistent_term:erase(Key), + off_heap_values_wait(FetchedVal, Val), + chk(Chk). + +off_heap_values_wait(FetchedVal, Val) -> + case erts_debug:size_shared(FetchedVal) of + 0 -> + Val = FetchedVal, + ok; + _ -> + erlang:yield(), + off_heap_values_wait(FetchedVal, Val) + end. + +%% Test some more data types as keys. Use the module name as a key +%% to minimize the risk of collision with any key used +%% by the OTP libraries. + +keys(_Config) -> + Chk = chk(), + do_key(?MODULE), + do_key([?MODULE]), + do_key(?MODULE_STRING), + do_key(list_to_binary(?MODULE_STRING)), + chk(Chk). + +do_key(Key) -> + Val = term_to_binary(Key), + ok = persistent_term:put(Key, Val), + StoredVal = persistent_term:get(Key), + Val = StoredVal, + true = persistent_term:erase(Key). + +%% Create persistent terms with keys that are known to collide. +%% Delete them in random order, making sure that all others +%% terms can still be found. + +collisions(_Config) -> + Chk = chk(), + + %% Create persistent terms with random keys. + Keys = lists:flatten(colliding_keys()), + Kvs = [{K,rand:uniform(1000)} || K <- Keys], + _ = [ok = persistent_term:put(K, V) || {K,V} <- Kvs], + _ = [V = persistent_term:get(K) || {K,V} <- Kvs], + + %% Now delete the persistent terms in random order. + collisions_delete(lists:keysort(2, Kvs)), + + chk(Chk). + +collisions_delete([{Key,Val}|Kvs]) -> + Val = persistent_term:get(Key), + true = persistent_term:erase(Key), + true = lists:sort(persistent_term:get()) =:= lists:sort(Kvs), + _ = [V = persistent_term:get(K) || {K,V} <- Kvs], + collisions_delete(Kvs); +collisions_delete([]) -> + ok. + +colliding_keys() -> + %% Collisions found by Jesper L. Andersen for breaking maps. + L = [[764492191,2361333849], + [49527266765044,90940896816021,20062927283041,267080852079651], + [249858369443708,206247021789428,20287304470696,25847120931175], + [10645228898670,224705626119556,267405565521452,258214397180678], + [264783762221048,166955943492306,98802957003141,102012488332476], + [69425677456944,177142907243411,137138950917722,228865047699598], + [116031213307147,29203342183358,37406949328742,255198080174323], + [200358182338308,235207156008390,120922906095920,116215987197289], + [58728890318426,68877471005069,176496507286088,221041411345780], + [91094120814795,50665258299931,256093108116737,19777509566621], + [74646746200247,98350487270564,154448261001199,39881047281135], + [23408943649483,164410325820923,248161749770122,274558342231648], + [169531547115055,213630535746863,235098262267796,200508473898303], + [235098564415817,85039146398174,51721575960328,173069189684390], + [176136386396069,155368359051606,147817099696487,265419485459634], + [137542881551462,40028925519736,70525669519846,63445773516557], + [173854695142814,114282444507812,149945832627054,99605565798831], + [177686773562184,127158716984798,132495543008547], + [227073396444896,139667311071766,158915951283562], + [26212438434289,94902985796531,198145776057315], + [266279278943923,58550737262493,74297973216378], + [32373606512065,131854353044428,184642643042326], + [34335377662439,85341895822066,273492717750246]], + + %% Verify that the keys still collide (this will fail if the + %% internal hash function has been changed). + erts_debug:set_internal_state(available_internal_state, true), + try + case erlang:system_info(wordsize) of + 8 -> + verify_colliding_keys(L); + 4 -> + %% Not guaranteed to collide on a 32-bit system. + ok + end + after + erts_debug:set_internal_state(available_internal_state, false) + end, + + L. + +verify_colliding_keys([[K|Ks]|Gs]) -> + Hash = internal_hash(K), + [Hash] = lists:usort([internal_hash(Key) || Key <- Ks]), + verify_colliding_keys(Gs); +verify_colliding_keys([]) -> + ok. + +internal_hash(Term) -> + erts_debug:get_internal_state({internal_hash,Term}). + +%% Test that all persistent terms are erased by init:restart/0. + +init_restart(_Config) -> + File = "command_file", + ok = file:write_file(File, term_to_binary(restart)), + {ok,[[Erl]]} = init:get_argument(progname), + ModPath = filename:dirname(code:which(?MODULE)), + Cmd = Erl ++ " -pa " ++ ModPath ++ " -noshell " + "-run " ++ ?MODULE_STRING ++ " test_init_restart_cmd " ++ + File, + io:format("~s\n", [Cmd]), + Expected = "12ok", + case os:cmd(Cmd) of + Expected -> + ok; + Actual -> + io:format("Expected: ~s", [Expected]), + io:format("Actual: ~s\n", [Actual]), + ct:fail(unexpected_output) + end. + +test_init_restart_cmd([File]) -> + try + do_test_init_restart_cmd(File) + catch + C:R -> + io:format("\n~p ~p\n", [C,R]), + halt() + end, + receive + _ -> ok + end. + +do_test_init_restart_cmd(File) -> + {ok,Bin} = file:read_file(File), + Seq = lists:seq(1, 50), + case binary_to_term(Bin) of + restart -> + _ = [persistent_term:put({?MODULE,I}, {value,I}) || + I <- Seq], + ok = file:write_file(File, term_to_binary(was_restarted)), + io:put_chars("1"), + init:restart(), + receive + _ -> ok + end; + was_restarted -> + io:put_chars("2"), + ok = file:delete(File), + _ = [begin + Key = {?MODULE,I}, + {'EXIT',{badarg,_}} = (catch persistent_term:get(Key)) + end || I <- Seq], + io:put_chars("ok"), + init:stop() + end. + +%% Check that there is the same number of persistents terms before +%% and after each test case. + +chk() -> + persistent_term:info(). + +chk(Chk) -> + Chk = persistent_term:info(), + Key = {?MODULE,?FUNCTION_NAME}, + ok = persistent_term:put(Key, {term,Chk}), + Term = persistent_term:get(Key), + true = persistent_term:erase(Key), + chk_not_stuck(Term), + ok. + +chk_not_stuck(Term) -> + %% Hash tables to be deleted are put onto a queue. + %% Make sure that the queue isn't stuck by a table with + %% a non-zero ref count. + + case erts_debug:size_shared(Term) of + 0 -> + erlang:yield(), + chk_not_stuck(Term); + _ -> + ok + end. diff --git a/erts/emulator/test/smoke_test_SUITE.erl b/erts/emulator/test/smoke_test_SUITE.erl index 26c610e3a8..5b46342127 100644 --- a/erts/emulator/test/smoke_test_SUITE.erl +++ b/erts/emulator/test/smoke_test_SUITE.erl @@ -56,7 +56,7 @@ end_per_testcase(_Case, Config) when is_list(Config) -> %%% boot_combo(Config) when is_list(Config) -> - ZFlags = os:getenv("ERL_ZFLAGS"), + ZFlags = os:getenv("ERL_ZFLAGS", ""), NOOP = fun () -> ok end, A42 = fun () -> case erlang:system_info(threads) of @@ -87,10 +87,7 @@ boot_combo(Config) when is_list(Config) -> %% A lot more combos could be implemented... ok after - os:putenv("ERL_ZFLAGS", case ZFlags of - false -> ""; - _ -> ZFlags - end) + os:putenv("ERL_ZFLAGS", ZFlags) end. native_atomics(Config) when is_list(Config) -> diff --git a/erts/emulator/zlib/adler32.c b/erts/emulator/zlib/adler32.c index c693a42b7c..d0be4380a3 100644 --- a/erts/emulator/zlib/adler32.c +++ b/erts/emulator/zlib/adler32.c @@ -1,20 +1,15 @@ /* adler32.c -- compute the Adler-32 checksum of a data stream - * Copyright (C) 1995-2011 Mark Adler + * Copyright (C) 1995-2011, 2016 Mark Adler * For conditions of distribution and use, see copyright notice in zlib.h */ /* @(#) $Id$ */ -#ifdef HAVE_CONFIG_H -# include "config.h" -#endif #include "zutil.h" -#define local static - local uLong adler32_combine_ OF((uLong adler1, uLong adler2, z_off64_t len2)); -#define BASE 65521 /* largest prime smaller than 65536 */ +#define BASE 65521U /* largest prime smaller than 65536 */ #define NMAX 5552 /* NMAX is the largest n such that 255n(n+1)/2 + (n+1)(BASE-1) <= 2^32-1 */ @@ -65,10 +60,10 @@ local uLong adler32_combine_ OF((uLong adler1, uLong adler2, z_off64_t len2)); #endif /* ========================================================================= */ -uLong ZEXPORT adler32(adler, buf, len) +uLong ZEXPORT adler32_z(adler, buf, len) uLong adler; const Bytef *buf; - uInt len; + z_size_t len; { unsigned long sum2; unsigned n; @@ -136,6 +131,15 @@ uLong ZEXPORT adler32(adler, buf, len) } /* ========================================================================= */ +uLong ZEXPORT adler32(adler, buf, len) + uLong adler; + const Bytef *buf; + uInt len; +{ + return adler32_z(adler, buf, len); +} + +/* ========================================================================= */ local uLong adler32_combine_(adler1, adler2, len2) uLong adler1; uLong adler2; @@ -159,7 +163,7 @@ local uLong adler32_combine_(adler1, adler2, len2) sum2 += ((adler1 >> 16) & 0xffff) + ((adler2 >> 16) & 0xffff) + BASE - rem; if (sum1 >= BASE) sum1 -= BASE; if (sum1 >= BASE) sum1 -= BASE; - if (sum2 >= (BASE << 1)) sum2 -= (BASE << 1); + if (sum2 >= ((unsigned long)BASE << 1)) sum2 -= ((unsigned long)BASE << 1); if (sum2 >= BASE) sum2 -= BASE; return sum1 | (sum2 << 16); } diff --git a/erts/emulator/zlib/compress.c b/erts/emulator/zlib/compress.c index 8ecef0f790..e2db404abf 100644 --- a/erts/emulator/zlib/compress.c +++ b/erts/emulator/zlib/compress.c @@ -1,13 +1,10 @@ /* compress.c -- compress a memory buffer - * Copyright (C) 1995-2005 Jean-loup Gailly. + * Copyright (C) 1995-2005, 2014, 2016 Jean-loup Gailly, Mark Adler * For conditions of distribution and use, see copyright notice in zlib.h */ /* @(#) $Id$ */ -#ifdef HAVE_CONFIG_H -# include "config.h" -#endif #define ZLIB_INTERNAL #include "zlib.h" @@ -31,16 +28,11 @@ int ZEXPORT compress2 (dest, destLen, source, sourceLen, level) { z_stream stream; int err; + const uInt max = (uInt)-1; + uLong left; - stream.next_in = (z_const Bytef *)source; - stream.avail_in = (uInt)sourceLen; -#ifdef MAXSEG_64K - /* Check for source > 64K on 16-bit machine: */ - if ((uLong)stream.avail_in != sourceLen) return Z_BUF_ERROR; -#endif - stream.next_out = dest; - stream.avail_out = (uInt)*destLen; - if ((uLong)stream.avail_out != *destLen) return Z_BUF_ERROR; + left = *destLen; + *destLen = 0; stream.zalloc = (alloc_func)0; stream.zfree = (free_func)0; @@ -49,15 +41,26 @@ int ZEXPORT compress2 (dest, destLen, source, sourceLen, level) err = deflateInit(&stream, level); if (err != Z_OK) return err; - err = deflate(&stream, Z_FINISH); - if (err != Z_STREAM_END) { - deflateEnd(&stream); - return err == Z_OK ? Z_BUF_ERROR : err; - } - *destLen = stream.total_out; + stream.next_out = dest; + stream.avail_out = 0; + stream.next_in = (z_const Bytef *)source; + stream.avail_in = 0; + + do { + if (stream.avail_out == 0) { + stream.avail_out = left > (uLong)max ? max : (uInt)left; + left -= stream.avail_out; + } + if (stream.avail_in == 0) { + stream.avail_in = sourceLen > (uLong)max ? max : (uInt)sourceLen; + sourceLen -= stream.avail_in; + } + err = deflate(&stream, sourceLen ? Z_NO_FLUSH : Z_FINISH); + } while (err == Z_OK); - err = deflateEnd(&stream); - return err; + *destLen = stream.total_out; + deflateEnd(&stream); + return err == Z_STREAM_END ? Z_OK : err; } /* =========================================================================== diff --git a/erts/emulator/zlib/crc32.c b/erts/emulator/zlib/crc32.c index ba506d8dd3..9580440c0e 100644 --- a/erts/emulator/zlib/crc32.c +++ b/erts/emulator/zlib/crc32.c @@ -1,5 +1,5 @@ /* crc32.c -- compute the CRC-32 of a data stream - * Copyright (C) 1995-2006, 2010, 2011, 2012 Mark Adler + * Copyright (C) 1995-2006, 2010, 2011, 2012, 2016 Mark Adler * For conditions of distribution and use, see copyright notice in zlib.h * * Thanks to Rodney Brown <[email protected]> for his contribution of faster @@ -28,23 +28,17 @@ # endif /* !DYNAMIC_CRC_TABLE */ #endif /* MAKECRCH */ -#ifdef HAVE_CONFIG_H -# include "config.h" -#endif - #include "zutil.h" /* for STDC and FAR definitions */ -#define local static - /* Definitions for doing the crc four data bytes at a time. */ #if !defined(NOBYFOUR) && defined(Z_U4) # define BYFOUR #endif #ifdef BYFOUR local unsigned long crc32_little OF((unsigned long, - const unsigned char FAR *, unsigned)); + const unsigned char FAR *, z_size_t)); local unsigned long crc32_big OF((unsigned long, - const unsigned char FAR *, unsigned)); + const unsigned char FAR *, z_size_t)); # define TBLS 8 #else # define TBLS 1 @@ -205,10 +199,10 @@ const z_crc_t FAR * ZEXPORT get_crc_table() #define DO8 DO1; DO1; DO1; DO1; DO1; DO1; DO1; DO1 /* ========================================================================= */ -unsigned long ZEXPORT crc32(crc, buf, len) +unsigned long ZEXPORT crc32_z(crc, buf, len) unsigned long crc; const unsigned char FAR *buf; - uInt len; + z_size_t len; { if (buf == Z_NULL) return 0UL; @@ -239,8 +233,29 @@ unsigned long ZEXPORT crc32(crc, buf, len) return crc ^ 0xffffffffUL; } +/* ========================================================================= */ +unsigned long ZEXPORT crc32(crc, buf, len) + unsigned long crc; + const unsigned char FAR *buf; + uInt len; +{ + return crc32_z(crc, buf, len); +} + #ifdef BYFOUR +/* + This BYFOUR code accesses the passed unsigned char * buffer with a 32-bit + integer pointer type. This violates the strict aliasing rule, where a + compiler can assume, for optimization purposes, that two pointers to + fundamentally different types won't ever point to the same memory. This can + manifest as a problem only if one of the pointers is written to. This code + only reads from those pointers. So long as this code remains isolated in + this compilation unit, there won't be a problem. For this reason, this code + should not be copied and pasted into a compilation unit in which other code + writes to the buffer that is passed to these routines. + */ + /* ========================================================================= */ #define DOLIT4 c ^= *buf4++; \ c = crc_table[3][c & 0xff] ^ crc_table[2][(c >> 8) & 0xff] ^ \ @@ -251,7 +266,7 @@ unsigned long ZEXPORT crc32(crc, buf, len) local unsigned long crc32_little(crc, buf, len) unsigned long crc; const unsigned char FAR *buf; - unsigned len; + z_size_t len; { register z_crc_t c; register const z_crc_t FAR *buf4; @@ -282,7 +297,7 @@ local unsigned long crc32_little(crc, buf, len) } /* ========================================================================= */ -#define DOBIG4 c ^= *++buf4; \ +#define DOBIG4 c ^= *buf4++; \ c = crc_table[4][c & 0xff] ^ crc_table[5][(c >> 8) & 0xff] ^ \ crc_table[6][(c >> 16) & 0xff] ^ crc_table[7][c >> 24] #define DOBIG32 DOBIG4; DOBIG4; DOBIG4; DOBIG4; DOBIG4; DOBIG4; DOBIG4; DOBIG4 @@ -291,7 +306,7 @@ local unsigned long crc32_little(crc, buf, len) local unsigned long crc32_big(crc, buf, len) unsigned long crc; const unsigned char FAR *buf; - unsigned len; + z_size_t len; { register z_crc_t c; register const z_crc_t FAR *buf4; @@ -304,7 +319,6 @@ local unsigned long crc32_big(crc, buf, len) } buf4 = (const z_crc_t FAR *)(const void FAR *)buf; - buf4--; while (len >= 32) { DOBIG32; len -= 32; @@ -313,7 +327,6 @@ local unsigned long crc32_big(crc, buf, len) DOBIG4; len -= 4; } - buf4++; buf = (const unsigned char FAR *)buf4; if (len) do { diff --git a/erts/emulator/zlib/deflate.c b/erts/emulator/zlib/deflate.c index 943c26dfb2..1ec761448d 100644 --- a/erts/emulator/zlib/deflate.c +++ b/erts/emulator/zlib/deflate.c @@ -1,5 +1,5 @@ /* deflate.c -- compress data using the deflation algorithm - * Copyright (C) 1995-2013 Jean-loup Gailly and Mark Adler + * Copyright (C) 1995-2017 Jean-loup Gailly and Mark Adler * For conditions of distribution and use, see copyright notice in zlib.h */ @@ -49,13 +49,10 @@ /* @(#) $Id$ */ -#ifdef HAVE_CONFIG_H -# include "config.h" -#endif #include "deflate.h" const char deflate_copyright[] = - " deflate 1.2.8 Copyright 1995-2013 Jean-loup Gailly and Mark Adler "; + " deflate 1.2.11 Copyright 1995-2017 Jean-loup Gailly and Mark Adler "; /* If you use the zlib library in a product, an acknowledgment is welcome in the documentation of your product. If for some reason you cannot @@ -76,6 +73,8 @@ typedef enum { typedef block_state (*compress_func) OF((deflate_state *s, int flush)); /* Compression function. Returns the block state after the call. */ +local int deflateStateCheck OF((z_streamp strm)); +local void slide_hash OF((deflate_state *s)); local void fill_window OF((deflate_state *s)); local block_state deflate_stored OF((deflate_state *s, int flush)); local block_state deflate_fast OF((deflate_state *s, int flush)); @@ -87,15 +86,16 @@ local block_state deflate_huff OF((deflate_state *s, int flush)); local void lm_init OF((deflate_state *s)); local void putShortMSB OF((deflate_state *s, uInt b)); local void flush_pending OF((z_streamp strm)); -local int read_buf OF((z_streamp strm, Bytef *buf, unsigned size)); +local unsigned read_buf OF((z_streamp strm, Bytef *buf, unsigned size)); #ifdef ASMV +# pragma message("Assembler code may have bugs -- use at your own risk") void match_init OF((void)); /* asm code initialization */ uInt longest_match OF((deflate_state *s, IPos cur_match)); #else local uInt longest_match OF((deflate_state *s, IPos cur_match)); #endif -#ifdef DEBUG +#ifdef ZLIB_DEBUG local void check_match OF((deflate_state *s, IPos start, IPos match, int length)); #endif @@ -151,21 +151,14 @@ local const config configuration_table[10] = { * meaning. */ -#define EQUAL 0 -/* result of memcmp for equal strings */ - -#ifndef NO_DUMMY_DECL -struct static_tree_desc_s {int dummy;}; /* for buggy compilers */ -#endif - /* rank Z_BLOCK between Z_NO_FLUSH and Z_PARTIAL_FLUSH */ -#define RANK(f) (((f) << 1) - ((f) > 4 ? 9 : 0)) +#define RANK(f) (((f) * 2) - ((f) > 4 ? 9 : 0)) /* =========================================================================== * Update a hash value with the given input byte - * IN assertion: all calls to to UPDATE_HASH are made with consecutive - * input characters, so that a running hash key can be computed from the - * previous key instead of complete recalculation each time. + * IN assertion: all calls to UPDATE_HASH are made with consecutive input + * characters, so that a running hash key can be computed from the previous + * key instead of complete recalculation each time. */ #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask) @@ -176,9 +169,9 @@ struct static_tree_desc_s {int dummy;}; /* for buggy compilers */ * the previous length of the hash chain. * If this file is compiled with -DFASTEST, the compression level is forced * to 1, and no hash chains are maintained. - * IN assertion: all calls to to INSERT_STRING are made with consecutive - * input characters and the first MIN_MATCH bytes of str are valid - * (except for the last MIN_MATCH-1 bytes of the input file). + * IN assertion: all calls to INSERT_STRING are made with consecutive input + * characters and the first MIN_MATCH bytes of str are valid (except for + * the last MIN_MATCH-1 bytes of the input file). */ #ifdef FASTEST #define INSERT_STRING(s, str, match_head) \ @@ -200,6 +193,37 @@ struct static_tree_desc_s {int dummy;}; /* for buggy compilers */ s->head[s->hash_size-1] = NIL; \ zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head)); +/* =========================================================================== + * Slide the hash table when sliding the window down (could be avoided with 32 + * bit values at the expense of memory usage). We slide even when level == 0 to + * keep the hash table consistent if we switch back to level > 0 later. + */ +local void slide_hash(s) + deflate_state *s; +{ + unsigned n, m; + Posf *p; + uInt wsize = s->w_size; + + n = s->hash_size; + p = &s->head[n]; + do { + m = *--p; + *p = (Pos)(m >= wsize ? m - wsize : NIL); + } while (--n); + n = wsize; +#ifndef FASTEST + p = &s->prev[n]; + do { + m = *--p; + *p = (Pos)(m >= wsize ? m - wsize : NIL); + /* If n is not on any hash chain, prev[n] is garbage but + * its value will never be used. + */ + } while (--n); +#endif +} + /* ========================================================================= */ int ZEXPORT deflateInit_(strm, level, version, stream_size) z_streamp strm; @@ -273,7 +297,7 @@ int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy, #endif if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED || windowBits < 8 || windowBits > 15 || level < 0 || level > 9 || - strategy < 0 || strategy > Z_FIXED) { + strategy < 0 || strategy > Z_FIXED || (windowBits == 8 && wrap != 1)) { return Z_STREAM_ERROR; } if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */ @@ -281,14 +305,15 @@ int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy, if (s == Z_NULL) return Z_MEM_ERROR; strm->state = (struct internal_state FAR *)s; s->strm = strm; + s->status = INIT_STATE; /* to pass state test in deflateReset() */ s->wrap = wrap; s->gzhead = Z_NULL; - s->w_bits = windowBits; + s->w_bits = (uInt)windowBits; s->w_size = 1 << s->w_bits; s->w_mask = s->w_size - 1; - s->hash_bits = memLevel + 7; + s->hash_bits = (uInt)memLevel + 7; s->hash_size = 1 << s->hash_bits; s->hash_mask = s->hash_size - 1; s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH); @@ -322,6 +347,31 @@ int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy, return deflateReset(strm); } +/* ========================================================================= + * Check for a valid deflate stream state. Return 0 if ok, 1 if not. + */ +local int deflateStateCheck (strm) + z_streamp strm; +{ + deflate_state *s; + if (strm == Z_NULL || + strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) + return 1; + s = strm->state; + if (s == Z_NULL || s->strm != strm || (s->status != INIT_STATE && +#ifdef GZIP + s->status != GZIP_STATE && +#endif + s->status != EXTRA_STATE && + s->status != NAME_STATE && + s->status != COMMENT_STATE && + s->status != HCRC_STATE && + s->status != BUSY_STATE && + s->status != FINISH_STATE)) + return 1; + return 0; +} + /* ========================================================================= */ int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength) z_streamp strm; @@ -334,7 +384,7 @@ int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength) unsigned avail; z_const unsigned char *next; - if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL) + if (deflateStateCheck(strm) || dictionary == Z_NULL) return Z_STREAM_ERROR; s = strm->state; wrap = s->wrap; @@ -392,13 +442,34 @@ int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength) } /* ========================================================================= */ +int ZEXPORT deflateGetDictionary (strm, dictionary, dictLength) + z_streamp strm; + Bytef *dictionary; + uInt *dictLength; +{ + deflate_state *s; + uInt len; + + if (deflateStateCheck(strm)) + return Z_STREAM_ERROR; + s = strm->state; + len = s->strstart + s->lookahead; + if (len > s->w_size) + len = s->w_size; + if (dictionary != Z_NULL && len) + zmemcpy(dictionary, s->window + s->strstart + s->lookahead - len, len); + if (dictLength != Z_NULL) + *dictLength = len; + return Z_OK; +} + +/* ========================================================================= */ int ZEXPORT deflateResetKeep (strm) z_streamp strm; { deflate_state *s; - if (strm == Z_NULL || strm->state == Z_NULL || - strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) { + if (deflateStateCheck(strm)) { return Z_STREAM_ERROR; } @@ -413,7 +484,11 @@ int ZEXPORT deflateResetKeep (strm) if (s->wrap < 0) { s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */ } - s->status = s->wrap ? INIT_STATE : BUSY_STATE; + s->status = +#ifdef GZIP + s->wrap == 2 ? GZIP_STATE : +#endif + s->wrap ? INIT_STATE : BUSY_STATE; strm->adler = #ifdef GZIP s->wrap == 2 ? crc32(0L, Z_NULL, 0) : @@ -443,8 +518,8 @@ int ZEXPORT deflateSetHeader (strm, head) z_streamp strm; gz_headerp head; { - if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; - if (strm->state->wrap != 2) return Z_STREAM_ERROR; + if (deflateStateCheck(strm) || strm->state->wrap != 2) + return Z_STREAM_ERROR; strm->state->gzhead = head; return Z_OK; } @@ -455,7 +530,7 @@ int ZEXPORT deflatePending (strm, pending, bits) int *bits; z_streamp strm; { - if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; + if (deflateStateCheck(strm)) return Z_STREAM_ERROR; if (pending != Z_NULL) *pending = strm->state->pending; if (bits != Z_NULL) @@ -472,7 +547,7 @@ int ZEXPORT deflatePrime (strm, bits, value) deflate_state *s; int put; - if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; + if (deflateStateCheck(strm)) return Z_STREAM_ERROR; s = strm->state; if ((Bytef *)(s->d_buf) < s->pending_out + ((Buf_size + 7) >> 3)) return Z_BUF_ERROR; @@ -497,9 +572,8 @@ int ZEXPORT deflateParams(strm, level, strategy) { deflate_state *s; compress_func func; - int err = Z_OK; - if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; + if (deflateStateCheck(strm)) return Z_STREAM_ERROR; s = strm->state; #ifdef FASTEST @@ -513,13 +587,22 @@ int ZEXPORT deflateParams(strm, level, strategy) func = configuration_table[s->level].func; if ((strategy != s->strategy || func != configuration_table[level].func) && - strm->total_in != 0) { + s->high_water) { /* Flush the last buffer: */ - err = deflate(strm, Z_BLOCK); - if (err == Z_BUF_ERROR && s->pending == 0) - err = Z_OK; + int err = deflate(strm, Z_BLOCK); + if (err == Z_STREAM_ERROR) + return err; + if (strm->avail_out == 0) + return Z_BUF_ERROR; } if (s->level != level) { + if (s->level == 0 && s->matches != 0) { + if (s->matches == 1) + slide_hash(s); + else + CLEAR_HASH(s); + s->matches = 0; + } s->level = level; s->max_lazy_match = configuration_table[level].max_lazy; s->good_match = configuration_table[level].good_length; @@ -527,7 +610,7 @@ int ZEXPORT deflateParams(strm, level, strategy) s->max_chain_length = configuration_table[level].max_chain; } s->strategy = strategy; - return err; + return Z_OK; } /* ========================================================================= */ @@ -540,12 +623,12 @@ int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain) { deflate_state *s; - if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; + if (deflateStateCheck(strm)) return Z_STREAM_ERROR; s = strm->state; - s->good_match = good_length; - s->max_lazy_match = max_lazy; + s->good_match = (uInt)good_length; + s->max_lazy_match = (uInt)max_lazy; s->nice_match = nice_length; - s->max_chain_length = max_chain; + s->max_chain_length = (uInt)max_chain; return Z_OK; } @@ -572,14 +655,13 @@ uLong ZEXPORT deflateBound(strm, sourceLen) { deflate_state *s; uLong complen, wraplen; - Bytef *str; /* conservative upper bound for compressed data */ complen = sourceLen + ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5; /* if can't get parameters, return conservative bound plus zlib wrapper */ - if (strm == Z_NULL || strm->state == Z_NULL) + if (deflateStateCheck(strm)) return complen + 6; /* compute wrapper length */ @@ -591,9 +673,11 @@ uLong ZEXPORT deflateBound(strm, sourceLen) case 1: /* zlib wrapper */ wraplen = 6 + (s->strstart ? 4 : 0); break; +#ifdef GZIP case 2: /* gzip wrapper */ wraplen = 18; if (s->gzhead != Z_NULL) { /* user-supplied gzip header */ + Bytef *str; if (s->gzhead->extra != Z_NULL) wraplen += 2 + s->gzhead->extra_len; str = s->gzhead->name; @@ -610,6 +694,7 @@ uLong ZEXPORT deflateBound(strm, sourceLen) wraplen += 2; } break; +#endif default: /* for compiler happiness */ wraplen = 6; } @@ -637,10 +722,10 @@ local void putShortMSB (s, b) } /* ========================================================================= - * Flush as much pending output as possible. All deflate() output goes - * through this function so some applications may wish to modify it - * to avoid allocating a large strm->next_out buffer and copying into it. - * (See also read_buf()). + * Flush as much pending output as possible. All deflate() output, except for + * some deflate_stored() output, goes through this function so some + * applications may wish to modify it to avoid allocating a large + * strm->next_out buffer and copying into it. (See also read_buf()). */ local void flush_pending(strm) z_streamp strm; @@ -657,13 +742,23 @@ local void flush_pending(strm) strm->next_out += len; s->pending_out += len; strm->total_out += len; - strm->avail_out -= len; - s->pending -= len; + strm->avail_out -= len; + s->pending -= len; if (s->pending == 0) { s->pending_out = s->pending_buf; } } +/* =========================================================================== + * Update the header CRC with the bytes s->pending_buf[beg..s->pending - 1]. + */ +#define HCRC_UPDATE(beg) \ + do { \ + if (s->gzhead->hcrc && s->pending > (beg)) \ + strm->adler = crc32(strm->adler, s->pending_buf + (beg), \ + s->pending - (beg)); \ + } while (0) + /* ========================================================================= */ int ZEXPORT deflate (strm, flush) z_streamp strm; @@ -672,230 +767,229 @@ int ZEXPORT deflate (strm, flush) int old_flush; /* value of flush param for previous deflate call */ deflate_state *s; - if (strm == Z_NULL || strm->state == Z_NULL || - flush > Z_BLOCK || flush < 0) { + if (deflateStateCheck(strm) || flush > Z_BLOCK || flush < 0) { return Z_STREAM_ERROR; } s = strm->state; if (strm->next_out == Z_NULL || - (strm->next_in == Z_NULL && strm->avail_in != 0) || + (strm->avail_in != 0 && strm->next_in == Z_NULL) || (s->status == FINISH_STATE && flush != Z_FINISH)) { ERR_RETURN(strm, Z_STREAM_ERROR); } if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR); - s->strm = strm; /* just in case */ old_flush = s->last_flush; s->last_flush = flush; + /* Flush as much pending output as possible */ + if (s->pending != 0) { + flush_pending(strm); + if (strm->avail_out == 0) { + /* Since avail_out is 0, deflate will be called again with + * more output space, but possibly with both pending and + * avail_in equal to zero. There won't be anything to do, + * but this is not an error situation so make sure we + * return OK instead of BUF_ERROR at next call of deflate: + */ + s->last_flush = -1; + return Z_OK; + } + + /* Make sure there is something to do and avoid duplicate consecutive + * flushes. For repeated and useless calls with Z_FINISH, we keep + * returning Z_STREAM_END instead of Z_BUF_ERROR. + */ + } else if (strm->avail_in == 0 && RANK(flush) <= RANK(old_flush) && + flush != Z_FINISH) { + ERR_RETURN(strm, Z_BUF_ERROR); + } + + /* User must not provide more input after the first FINISH: */ + if (s->status == FINISH_STATE && strm->avail_in != 0) { + ERR_RETURN(strm, Z_BUF_ERROR); + } + /* Write the header */ if (s->status == INIT_STATE) { -#ifdef GZIP - if (s->wrap == 2) { - strm->adler = crc32(0L, Z_NULL, 0); - put_byte(s, 31); - put_byte(s, 139); - put_byte(s, 8); - if (s->gzhead == Z_NULL) { - put_byte(s, 0); - put_byte(s, 0); - put_byte(s, 0); - put_byte(s, 0); - put_byte(s, 0); - put_byte(s, s->level == 9 ? 2 : - (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ? - 4 : 0)); - put_byte(s, OS_CODE); - s->status = BUSY_STATE; - } - else { - put_byte(s, (s->gzhead->text ? 1 : 0) + - (s->gzhead->hcrc ? 2 : 0) + - (s->gzhead->extra == Z_NULL ? 0 : 4) + - (s->gzhead->name == Z_NULL ? 0 : 8) + - (s->gzhead->comment == Z_NULL ? 0 : 16) - ); - put_byte(s, (Byte)(s->gzhead->time & 0xff)); - put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff)); - put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff)); - put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff)); - put_byte(s, s->level == 9 ? 2 : - (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ? - 4 : 0)); - put_byte(s, s->gzhead->os & 0xff); - if (s->gzhead->extra != Z_NULL) { - put_byte(s, s->gzhead->extra_len & 0xff); - put_byte(s, (s->gzhead->extra_len >> 8) & 0xff); - } - if (s->gzhead->hcrc) - strm->adler = crc32(strm->adler, s->pending_buf, - s->pending); - s->gzindex = 0; - s->status = EXTRA_STATE; - } - } + /* zlib header */ + uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8; + uInt level_flags; + + if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2) + level_flags = 0; + else if (s->level < 6) + level_flags = 1; + else if (s->level == 6) + level_flags = 2; else -#endif - { - uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8; - uInt level_flags; - - if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2) - level_flags = 0; - else if (s->level < 6) - level_flags = 1; - else if (s->level == 6) - level_flags = 2; - else - level_flags = 3; - header |= (level_flags << 6); - if (s->strstart != 0) header |= PRESET_DICT; - header += 31 - (header % 31); + level_flags = 3; + header |= (level_flags << 6); + if (s->strstart != 0) header |= PRESET_DICT; + header += 31 - (header % 31); + + putShortMSB(s, header); + + /* Save the adler32 of the preset dictionary: */ + if (s->strstart != 0) { + putShortMSB(s, (uInt)(strm->adler >> 16)); + putShortMSB(s, (uInt)(strm->adler & 0xffff)); + } + strm->adler = adler32(0L, Z_NULL, 0); + s->status = BUSY_STATE; + /* Compression must start with an empty pending buffer */ + flush_pending(strm); + if (s->pending != 0) { + s->last_flush = -1; + return Z_OK; + } + } +#ifdef GZIP + if (s->status == GZIP_STATE) { + /* gzip header */ + strm->adler = crc32(0L, Z_NULL, 0); + put_byte(s, 31); + put_byte(s, 139); + put_byte(s, 8); + if (s->gzhead == Z_NULL) { + put_byte(s, 0); + put_byte(s, 0); + put_byte(s, 0); + put_byte(s, 0); + put_byte(s, 0); + put_byte(s, s->level == 9 ? 2 : + (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ? + 4 : 0)); + put_byte(s, OS_CODE); s->status = BUSY_STATE; - putShortMSB(s, header); - /* Save the adler32 of the preset dictionary: */ - if (s->strstart != 0) { - putShortMSB(s, (uInt)(strm->adler >> 16)); - putShortMSB(s, (uInt)(strm->adler & 0xffff)); + /* Compression must start with an empty pending buffer */ + flush_pending(strm); + if (s->pending != 0) { + s->last_flush = -1; + return Z_OK; + } + } + else { + put_byte(s, (s->gzhead->text ? 1 : 0) + + (s->gzhead->hcrc ? 2 : 0) + + (s->gzhead->extra == Z_NULL ? 0 : 4) + + (s->gzhead->name == Z_NULL ? 0 : 8) + + (s->gzhead->comment == Z_NULL ? 0 : 16) + ); + put_byte(s, (Byte)(s->gzhead->time & 0xff)); + put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff)); + put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff)); + put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff)); + put_byte(s, s->level == 9 ? 2 : + (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ? + 4 : 0)); + put_byte(s, s->gzhead->os & 0xff); + if (s->gzhead->extra != Z_NULL) { + put_byte(s, s->gzhead->extra_len & 0xff); + put_byte(s, (s->gzhead->extra_len >> 8) & 0xff); } - strm->adler = adler32(0L, Z_NULL, 0); + if (s->gzhead->hcrc) + strm->adler = crc32(strm->adler, s->pending_buf, + s->pending); + s->gzindex = 0; + s->status = EXTRA_STATE; } } -#ifdef GZIP if (s->status == EXTRA_STATE) { if (s->gzhead->extra != Z_NULL) { - uInt beg = s->pending; /* start of bytes to update crc */ - - while (s->gzindex < (s->gzhead->extra_len & 0xffff)) { - if (s->pending == s->pending_buf_size) { - if (s->gzhead->hcrc && s->pending > beg) - strm->adler = crc32(strm->adler, s->pending_buf + beg, - s->pending - beg); - flush_pending(strm); - beg = s->pending; - if (s->pending == s->pending_buf_size) - break; + ulg beg = s->pending; /* start of bytes to update crc */ + uInt left = (s->gzhead->extra_len & 0xffff) - s->gzindex; + while (s->pending + left > s->pending_buf_size) { + uInt copy = s->pending_buf_size - s->pending; + zmemcpy(s->pending_buf + s->pending, + s->gzhead->extra + s->gzindex, copy); + s->pending = s->pending_buf_size; + HCRC_UPDATE(beg); + s->gzindex += copy; + flush_pending(strm); + if (s->pending != 0) { + s->last_flush = -1; + return Z_OK; } - put_byte(s, s->gzhead->extra[s->gzindex]); - s->gzindex++; - } - if (s->gzhead->hcrc && s->pending > beg) - strm->adler = crc32(strm->adler, s->pending_buf + beg, - s->pending - beg); - if (s->gzindex == s->gzhead->extra_len) { - s->gzindex = 0; - s->status = NAME_STATE; + beg = 0; + left -= copy; } + zmemcpy(s->pending_buf + s->pending, + s->gzhead->extra + s->gzindex, left); + s->pending += left; + HCRC_UPDATE(beg); + s->gzindex = 0; } - else - s->status = NAME_STATE; + s->status = NAME_STATE; } if (s->status == NAME_STATE) { if (s->gzhead->name != Z_NULL) { - uInt beg = s->pending; /* start of bytes to update crc */ + ulg beg = s->pending; /* start of bytes to update crc */ int val; - do { if (s->pending == s->pending_buf_size) { - if (s->gzhead->hcrc && s->pending > beg) - strm->adler = crc32(strm->adler, s->pending_buf + beg, - s->pending - beg); + HCRC_UPDATE(beg); flush_pending(strm); - beg = s->pending; - if (s->pending == s->pending_buf_size) { - val = 1; - break; + if (s->pending != 0) { + s->last_flush = -1; + return Z_OK; } + beg = 0; } val = s->gzhead->name[s->gzindex++]; put_byte(s, val); } while (val != 0); - if (s->gzhead->hcrc && s->pending > beg) - strm->adler = crc32(strm->adler, s->pending_buf + beg, - s->pending - beg); - if (val == 0) { - s->gzindex = 0; - s->status = COMMENT_STATE; - } + HCRC_UPDATE(beg); + s->gzindex = 0; } - else - s->status = COMMENT_STATE; + s->status = COMMENT_STATE; } if (s->status == COMMENT_STATE) { if (s->gzhead->comment != Z_NULL) { - uInt beg = s->pending; /* start of bytes to update crc */ + ulg beg = s->pending; /* start of bytes to update crc */ int val; - do { if (s->pending == s->pending_buf_size) { - if (s->gzhead->hcrc && s->pending > beg) - strm->adler = crc32(strm->adler, s->pending_buf + beg, - s->pending - beg); + HCRC_UPDATE(beg); flush_pending(strm); - beg = s->pending; - if (s->pending == s->pending_buf_size) { - val = 1; - break; + if (s->pending != 0) { + s->last_flush = -1; + return Z_OK; } + beg = 0; } val = s->gzhead->comment[s->gzindex++]; put_byte(s, val); } while (val != 0); - if (s->gzhead->hcrc && s->pending > beg) - strm->adler = crc32(strm->adler, s->pending_buf + beg, - s->pending - beg); - if (val == 0) - s->status = HCRC_STATE; + HCRC_UPDATE(beg); } - else - s->status = HCRC_STATE; + s->status = HCRC_STATE; } if (s->status == HCRC_STATE) { if (s->gzhead->hcrc) { - if (s->pending + 2 > s->pending_buf_size) + if (s->pending + 2 > s->pending_buf_size) { flush_pending(strm); - if (s->pending + 2 <= s->pending_buf_size) { - put_byte(s, (Byte)(strm->adler & 0xff)); - put_byte(s, (Byte)((strm->adler >> 8) & 0xff)); - strm->adler = crc32(0L, Z_NULL, 0); - s->status = BUSY_STATE; + if (s->pending != 0) { + s->last_flush = -1; + return Z_OK; + } } + put_byte(s, (Byte)(strm->adler & 0xff)); + put_byte(s, (Byte)((strm->adler >> 8) & 0xff)); + strm->adler = crc32(0L, Z_NULL, 0); } - else - s->status = BUSY_STATE; - } -#endif + s->status = BUSY_STATE; - /* Flush as much pending output as possible */ - if (s->pending != 0) { + /* Compression must start with an empty pending buffer */ flush_pending(strm); - if (strm->avail_out == 0) { - /* Since avail_out is 0, deflate will be called again with - * more output space, but possibly with both pending and - * avail_in equal to zero. There won't be anything to do, - * but this is not an error situation so make sure we - * return OK instead of BUF_ERROR at next call of deflate: - */ + if (s->pending != 0) { s->last_flush = -1; return Z_OK; } - - /* Make sure there is something to do and avoid duplicate consecutive - * flushes. For repeated and useless calls with Z_FINISH, we keep - * returning Z_STREAM_END instead of Z_BUF_ERROR. - */ - } else if (strm->avail_in == 0 && RANK(flush) <= RANK(old_flush) && - flush != Z_FINISH) { - ERR_RETURN(strm, Z_BUF_ERROR); - } - - /* User must not provide more input after the first FINISH: */ - if (s->status == FINISH_STATE && strm->avail_in != 0) { - ERR_RETURN(strm, Z_BUF_ERROR); } +#endif /* Start a new block or continue the current one. */ @@ -903,9 +997,10 @@ int ZEXPORT deflate (strm, flush) (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) { block_state bstate; - bstate = s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) : - (s->strategy == Z_RLE ? deflate_rle(s, flush) : - (*(configuration_table[s->level].func))(s, flush)); + bstate = s->level == 0 ? deflate_stored(s, flush) : + s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) : + s->strategy == Z_RLE ? deflate_rle(s, flush) : + (*(configuration_table[s->level].func))(s, flush); if (bstate == finish_started || bstate == finish_done) { s->status = FINISH_STATE; @@ -947,7 +1042,6 @@ int ZEXPORT deflate (strm, flush) } } } - Assert(strm->avail_out > 0, "bug2"); if (flush != Z_FINISH) return Z_OK; if (s->wrap <= 0) return Z_STREAM_END; @@ -984,18 +1078,9 @@ int ZEXPORT deflateEnd (strm) { int status; - if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; + if (deflateStateCheck(strm)) return Z_STREAM_ERROR; status = strm->state->status; - if (status != INIT_STATE && - status != EXTRA_STATE && - status != NAME_STATE && - status != COMMENT_STATE && - status != HCRC_STATE && - status != BUSY_STATE && - status != FINISH_STATE) { - return Z_STREAM_ERROR; - } /* Deallocate in reverse order of allocations: */ TRY_FREE(strm, strm->state->pending_buf); @@ -1026,7 +1111,7 @@ int ZEXPORT deflateCopy (dest, source) ushf *overlay; - if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) { + if (deflateStateCheck(source) || dest == Z_NULL) { return Z_STREAM_ERROR; } @@ -1076,7 +1161,7 @@ int ZEXPORT deflateCopy (dest, source) * allocating a large strm->next_in buffer and copying from it. * (See also flush_pending()). */ -local int read_buf(strm, buf, size) +local unsigned read_buf(strm, buf, size) z_streamp strm; Bytef *buf; unsigned size; @@ -1100,7 +1185,7 @@ local int read_buf(strm, buf, size) strm->next_in += len; strm->total_in += len; - return (int)len; + return len; } /* =========================================================================== @@ -1154,9 +1239,9 @@ local uInt longest_match(s, cur_match) { unsigned chain_length = s->max_chain_length;/* max hash chain length */ register Bytef *scan = s->window + s->strstart; /* current string */ - register Bytef *match; /* matched string */ + register Bytef *match; /* matched string */ register int len; /* length of current match */ - int best_len = s->prev_length; /* best match length so far */ + int best_len = (int)s->prev_length; /* best match length so far */ int nice_match = s->nice_match; /* stop if match long enough */ IPos limit = s->strstart > (IPos)MAX_DIST(s) ? s->strstart - (IPos)MAX_DIST(s) : NIL; @@ -1191,7 +1276,7 @@ local uInt longest_match(s, cur_match) /* Do not look for matches beyond the end of the input. This is necessary * to make deflate deterministic. */ - if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead; + if ((uInt)nice_match > s->lookahead) nice_match = (int)s->lookahead; Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead"); @@ -1352,7 +1437,11 @@ local uInt longest_match(s, cur_match) #endif /* FASTEST */ -#ifdef DEBUG +#ifdef ZLIB_DEBUG + +#define EQUAL 0 +/* result of memcmp for equal strings */ + /* =========================================================================== * Check that the match at match_start is indeed a match. */ @@ -1378,7 +1467,7 @@ local void check_match(s, start, match, length) } #else # define check_match(s, start, match, length) -#endif /* DEBUG */ +#endif /* ZLIB_DEBUG */ /* =========================================================================== * Fill the window when the lookahead becomes insufficient. @@ -1393,8 +1482,7 @@ local void check_match(s, start, match, length) local void fill_window(s) deflate_state *s; { - register unsigned n, m; - register Posf *p; + unsigned n; unsigned more; /* Amount of free space at the end of the window. */ uInt wsize = s->w_size; @@ -1421,35 +1509,11 @@ local void fill_window(s) */ if (s->strstart >= wsize+MAX_DIST(s)) { - zmemcpy(s->window, s->window+wsize, (unsigned)wsize); + zmemcpy(s->window, s->window+wsize, (unsigned)wsize - more); s->match_start -= wsize; s->strstart -= wsize; /* we now have strstart >= MAX_DIST */ s->block_start -= (long) wsize; - - /* Slide the hash table (could be avoided with 32 bit values - at the expense of memory usage). We slide even when level == 0 - to keep the hash table consistent if we switch back to level > 0 - later. (Using level 0 permanently is not an optimal usage of - zlib, so we don't care about this pathological case.) - */ - n = s->hash_size; - p = &s->head[n]; - do { - m = *--p; - *p = (Pos)(m >= wsize ? m-wsize : NIL); - } while (--n); - - n = wsize; -#ifndef FASTEST - p = &s->prev[n]; - do { - m = *--p; - *p = (Pos)(m >= wsize ? m-wsize : NIL); - /* If n is not on any hash chain, prev[n] is garbage but - * its value will never be used. - */ - } while (--n); -#endif + slide_hash(s); more += wsize; } if (s->strm->avail_in == 0) break; @@ -1555,70 +1619,199 @@ local void fill_window(s) if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \ } +/* Maximum stored block length in deflate format (not including header). */ +#define MAX_STORED 65535 + +/* Minimum of a and b. */ +#define MIN(a, b) ((a) > (b) ? (b) : (a)) + /* =========================================================================== * Copy without compression as much as possible from the input stream, return * the current block state. - * This function does not insert new strings in the dictionary since - * uncompressible data is probably not useful. This function is used - * only for the level=0 compression option. - * NOTE: this function should be optimized to avoid extra copying from - * window to pending_buf. + * + * In case deflateParams() is used to later switch to a non-zero compression + * level, s->matches (otherwise unused when storing) keeps track of the number + * of hash table slides to perform. If s->matches is 1, then one hash table + * slide will be done when switching. If s->matches is 2, the maximum value + * allowed here, then the hash table will be cleared, since two or more slides + * is the same as a clear. + * + * deflate_stored() is written to minimize the number of times an input byte is + * copied. It is most efficient with large input and output buffers, which + * maximizes the opportunites to have a single copy from next_in to next_out. */ local block_state deflate_stored(s, flush) deflate_state *s; int flush; { - /* Stored blocks are limited to 0xffff bytes, pending_buf is limited - * to pending_buf_size, and each stored block has a 5 byte header: + /* Smallest worthy block size when not flushing or finishing. By default + * this is 32K. This can be as small as 507 bytes for memLevel == 1. For + * large input and output buffers, the stored block size will be larger. */ - ulg max_block_size = 0xffff; - ulg max_start; - - if (max_block_size > s->pending_buf_size - 5) { - max_block_size = s->pending_buf_size - 5; - } - - /* Copy as much as possible from input to output: */ - for (;;) { - /* Fill the window as much as possible: */ - if (s->lookahead <= 1) { + unsigned min_block = MIN(s->pending_buf_size - 5, s->w_size); - Assert(s->strstart < s->w_size+MAX_DIST(s) || - s->block_start >= (long)s->w_size, "slide too late"); + /* Copy as many min_block or larger stored blocks directly to next_out as + * possible. If flushing, copy the remaining available input to next_out as + * stored blocks, if there is enough space. + */ + unsigned len, left, have, last = 0; + unsigned used = s->strm->avail_in; + do { + /* Set len to the maximum size block that we can copy directly with the + * available input data and output space. Set left to how much of that + * would be copied from what's left in the window. + */ + len = MAX_STORED; /* maximum deflate stored block length */ + have = (s->bi_valid + 42) >> 3; /* number of header bytes */ + if (s->strm->avail_out < have) /* need room for header */ + break; + /* maximum stored block length that will fit in avail_out: */ + have = s->strm->avail_out - have; + left = s->strstart - s->block_start; /* bytes left in window */ + if (len > (ulg)left + s->strm->avail_in) + len = left + s->strm->avail_in; /* limit len to the input */ + if (len > have) + len = have; /* limit len to the output */ + + /* If the stored block would be less than min_block in length, or if + * unable to copy all of the available input when flushing, then try + * copying to the window and the pending buffer instead. Also don't + * write an empty block when flushing -- deflate() does that. + */ + if (len < min_block && ((len == 0 && flush != Z_FINISH) || + flush == Z_NO_FLUSH || + len != left + s->strm->avail_in)) + break; - fill_window(s); - if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more; + /* Make a dummy stored block in pending to get the header bytes, + * including any pending bits. This also updates the debugging counts. + */ + last = flush == Z_FINISH && len == left + s->strm->avail_in ? 1 : 0; + _tr_stored_block(s, (char *)0, 0L, last); + + /* Replace the lengths in the dummy stored block with len. */ + s->pending_buf[s->pending - 4] = len; + s->pending_buf[s->pending - 3] = len >> 8; + s->pending_buf[s->pending - 2] = ~len; + s->pending_buf[s->pending - 1] = ~len >> 8; + + /* Write the stored block header bytes. */ + flush_pending(s->strm); + +#ifdef ZLIB_DEBUG + /* Update debugging counts for the data about to be copied. */ + s->compressed_len += len << 3; + s->bits_sent += len << 3; +#endif - if (s->lookahead == 0) break; /* flush the current block */ + /* Copy uncompressed bytes from the window to next_out. */ + if (left) { + if (left > len) + left = len; + zmemcpy(s->strm->next_out, s->window + s->block_start, left); + s->strm->next_out += left; + s->strm->avail_out -= left; + s->strm->total_out += left; + s->block_start += left; + len -= left; } - Assert(s->block_start >= 0L, "block gone"); - - s->strstart += s->lookahead; - s->lookahead = 0; - - /* Emit a stored block if pending_buf will be full: */ - max_start = s->block_start + max_block_size; - if (s->strstart == 0 || (ulg)s->strstart >= max_start) { - /* strstart == 0 is possible when wraparound on 16-bit machine */ - s->lookahead = (uInt)(s->strstart - max_start); - s->strstart = (uInt)max_start; - FLUSH_BLOCK(s, 0); + + /* Copy uncompressed bytes directly from next_in to next_out, updating + * the check value. + */ + if (len) { + read_buf(s->strm, s->strm->next_out, len); + s->strm->next_out += len; + s->strm->avail_out -= len; + s->strm->total_out += len; } - /* Flush if we may have to slide, otherwise block_start may become - * negative and the data will be gone: + } while (last == 0); + + /* Update the sliding window with the last s->w_size bytes of the copied + * data, or append all of the copied data to the existing window if less + * than s->w_size bytes were copied. Also update the number of bytes to + * insert in the hash tables, in the event that deflateParams() switches to + * a non-zero compression level. + */ + used -= s->strm->avail_in; /* number of input bytes directly copied */ + if (used) { + /* If any input was used, then no unused input remains in the window, + * therefore s->block_start == s->strstart. */ - if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) { - FLUSH_BLOCK(s, 0); + if (used >= s->w_size) { /* supplant the previous history */ + s->matches = 2; /* clear hash */ + zmemcpy(s->window, s->strm->next_in - s->w_size, s->w_size); + s->strstart = s->w_size; } + else { + if (s->window_size - s->strstart <= used) { + /* Slide the window down. */ + s->strstart -= s->w_size; + zmemcpy(s->window, s->window + s->w_size, s->strstart); + if (s->matches < 2) + s->matches++; /* add a pending slide_hash() */ + } + zmemcpy(s->window + s->strstart, s->strm->next_in - used, used); + s->strstart += used; + } + s->block_start = s->strstart; + s->insert += MIN(used, s->w_size - s->insert); } - s->insert = 0; - if (flush == Z_FINISH) { - FLUSH_BLOCK(s, 1); + if (s->high_water < s->strstart) + s->high_water = s->strstart; + + /* If the last block was written to next_out, then done. */ + if (last) return finish_done; + + /* If flushing and all input has been consumed, then done. */ + if (flush != Z_NO_FLUSH && flush != Z_FINISH && + s->strm->avail_in == 0 && (long)s->strstart == s->block_start) + return block_done; + + /* Fill the window with any remaining input. */ + have = s->window_size - s->strstart - 1; + if (s->strm->avail_in > have && s->block_start >= (long)s->w_size) { + /* Slide the window down. */ + s->block_start -= s->w_size; + s->strstart -= s->w_size; + zmemcpy(s->window, s->window + s->w_size, s->strstart); + if (s->matches < 2) + s->matches++; /* add a pending slide_hash() */ + have += s->w_size; /* more space now */ } - if ((long)s->strstart > s->block_start) - FLUSH_BLOCK(s, 0); - return block_done; + if (have > s->strm->avail_in) + have = s->strm->avail_in; + if (have) { + read_buf(s->strm, s->window + s->strstart, have); + s->strstart += have; + } + if (s->high_water < s->strstart) + s->high_water = s->strstart; + + /* There was not enough avail_out to write a complete worthy or flushed + * stored block to next_out. Write a stored block to pending instead, if we + * have enough input for a worthy block, or if flushing and there is enough + * room for the remaining input as a stored block in the pending buffer. + */ + have = (s->bi_valid + 42) >> 3; /* number of header bytes */ + /* maximum stored block length that will fit in pending: */ + have = MIN(s->pending_buf_size - have, MAX_STORED); + min_block = MIN(have, s->w_size); + left = s->strstart - s->block_start; + if (left >= min_block || + ((left || flush == Z_FINISH) && flush != Z_NO_FLUSH && + s->strm->avail_in == 0 && left <= have)) { + len = MIN(left, have); + last = flush == Z_FINISH && s->strm->avail_in == 0 && + len == left ? 1 : 0; + _tr_stored_block(s, (charf *)s->window + s->block_start, len, last); + s->block_start += len; + flush_pending(s->strm); + } + + /* We've done all we can with the available input and output. */ + return last ? finish_started : need_more; } /* =========================================================================== @@ -1895,7 +2088,7 @@ local block_state deflate_rle(s, flush) prev == *++scan && prev == *++scan && prev == *++scan && prev == *++scan && scan < strend); - s->match_length = MAX_MATCH - (int)(strend - scan); + s->match_length = MAX_MATCH - (uInt)(strend - scan); if (s->match_length > s->lookahead) s->match_length = s->lookahead; } diff --git a/erts/emulator/zlib/deflate.h b/erts/emulator/zlib/deflate.h index ce0299edd1..23ecdd312b 100644 --- a/erts/emulator/zlib/deflate.h +++ b/erts/emulator/zlib/deflate.h @@ -1,5 +1,5 @@ /* deflate.h -- internal compression state - * Copyright (C) 1995-2012 Jean-loup Gailly + * Copyright (C) 1995-2016 Jean-loup Gailly * For conditions of distribution and use, see copyright notice in zlib.h */ @@ -51,13 +51,16 @@ #define Buf_size 16 /* size of bit buffer in bi_buf */ -#define INIT_STATE 42 -#define EXTRA_STATE 69 -#define NAME_STATE 73 -#define COMMENT_STATE 91 -#define HCRC_STATE 103 -#define BUSY_STATE 113 -#define FINISH_STATE 666 +#define INIT_STATE 42 /* zlib header -> BUSY_STATE */ +#ifdef GZIP +# define GZIP_STATE 57 /* gzip header -> BUSY_STATE | EXTRA_STATE */ +#endif +#define EXTRA_STATE 69 /* gzip extra block -> NAME_STATE */ +#define NAME_STATE 73 /* gzip file name -> COMMENT_STATE */ +#define COMMENT_STATE 91 /* gzip comment -> HCRC_STATE */ +#define HCRC_STATE 103 /* gzip header CRC -> BUSY_STATE */ +#define BUSY_STATE 113 /* deflate -> FINISH_STATE */ +#define FINISH_STATE 666 /* stream complete */ /* Stream status */ @@ -83,7 +86,7 @@ typedef struct static_tree_desc_s static_tree_desc; typedef struct tree_desc_s { ct_data *dyn_tree; /* the dynamic tree */ int max_code; /* largest code with non zero frequency */ - static_tree_desc *stat_desc; /* the corresponding static tree */ + const static_tree_desc *stat_desc; /* the corresponding static tree */ } FAR tree_desc; typedef ush Pos; @@ -100,10 +103,10 @@ typedef struct internal_state { Bytef *pending_buf; /* output still pending */ ulg pending_buf_size; /* size of pending_buf */ Bytef *pending_out; /* next pending byte to output to the stream */ - uInt pending; /* nb of bytes in the pending buffer */ + ulg pending; /* nb of bytes in the pending buffer */ int wrap; /* bit 0 true for zlib, bit 1 true for gzip */ gz_headerp gzhead; /* gzip header information to write */ - uInt gzindex; /* where in extra, name, or comment */ + ulg gzindex; /* where in extra, name, or comment */ Byte method; /* can only be DEFLATED */ int last_flush; /* value of flush param for previous deflate call */ @@ -249,7 +252,7 @@ typedef struct internal_state { uInt matches; /* number of string matches in current block */ uInt insert; /* bytes at end of window left to insert */ -#ifdef DEBUG +#ifdef ZLIB_DEBUG ulg compressed_len; /* total bit length of compressed file mod 2^32 */ ulg bits_sent; /* bit length of compressed data sent mod 2^32 */ #endif @@ -275,7 +278,7 @@ typedef struct internal_state { /* Output a byte on the stream. * IN assertion: there is enough room in pending_buf. */ -#define put_byte(s, c) {s->pending_buf[s->pending++] = (c);} +#define put_byte(s, c) {s->pending_buf[s->pending++] = (Bytef)(c);} #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1) @@ -309,7 +312,7 @@ void ZLIB_INTERNAL _tr_stored_block OF((deflate_state *s, charf *buf, * used. */ -#ifndef DEBUG +#ifndef ZLIB_DEBUG /* Inline versions of _tr_tally for speed: */ #if defined(GEN_TREES_H) || !defined(STDC) @@ -328,8 +331,8 @@ void ZLIB_INTERNAL _tr_stored_block OF((deflate_state *s, charf *buf, flush = (s->last_lit == s->lit_bufsize-1); \ } # define _tr_tally_dist(s, distance, length, flush) \ - { uch len = (length); \ - ush dist = (distance); \ + { uch len = (uch)(length); \ + ush dist = (ush)(distance); \ s->d_buf[s->last_lit] = dist; \ s->l_buf[s->last_lit++] = len; \ dist--; \ diff --git a/erts/emulator/zlib/gzguts.h b/erts/emulator/zlib/gzguts.h index d87659d031..990a4d2514 100644 --- a/erts/emulator/zlib/gzguts.h +++ b/erts/emulator/zlib/gzguts.h @@ -1,5 +1,5 @@ /* gzguts.h -- zlib internal header definitions for gz* operations - * Copyright (C) 2004, 2005, 2010, 2011, 2012, 2013 Mark Adler + * Copyright (C) 2004, 2005, 2010, 2011, 2012, 2013, 2016 Mark Adler * For conditions of distribution and use, see copyright notice in zlib.h */ @@ -25,6 +25,10 @@ # include <stdlib.h> # include <limits.h> #endif + +#ifndef _POSIX_SOURCE +# define _POSIX_SOURCE +#endif #include <fcntl.h> #ifdef _WIN32 @@ -35,6 +39,10 @@ # include <io.h> #endif +#if defined(_WIN32) || defined(__CYGWIN__) +# define WIDECHAR +#endif + #ifdef WINAPI_FAMILY # define open _open # define read _read @@ -95,18 +103,19 @@ # endif #endif -/* unlike snprintf (which is required in C99, yet still not supported by - Microsoft more than a decade later!), _snprintf does not guarantee null - termination of the result -- however this is only used in gzlib.c where +/* unlike snprintf (which is required in C99), _snprintf does not guarantee + null termination of the result -- however this is only used in gzlib.c where the result is assured to fit in the space provided */ -#ifdef _MSC_VER +#if defined(_MSC_VER) && _MSC_VER < 1900 # define snprintf _snprintf #endif #ifndef local # define local static #endif -/* compile with -Dlocal if your debugger can't find static symbols */ +/* since "static" is used to mean two completely different things in C, we + define "local" for the non-static meaning of "static", for readability + (compile with -Dlocal if your debugger can't find static symbols) */ /* gz* functions always use library allocation functions */ #ifndef STDC @@ -170,7 +179,7 @@ typedef struct { char *path; /* path or fd for error messages */ unsigned size; /* buffer size, zero if not allocated yet */ unsigned want; /* requested buffer size, default is GZBUFSIZE */ - unsigned char *in; /* input buffer */ + unsigned char *in; /* input buffer (double-sized when writing) */ unsigned char *out; /* output buffer (double-sized when reading) */ int direct; /* 0 if processing gzip, 1 if transparent */ /* just for reading */ diff --git a/erts/emulator/zlib/inffast.c b/erts/emulator/zlib/inffast.c index 5187743fde..0dbd1dbc09 100644 --- a/erts/emulator/zlib/inffast.c +++ b/erts/emulator/zlib/inffast.c @@ -1,36 +1,16 @@ /* inffast.c -- fast decoding - * Copyright (C) 1995-2008, 2010, 2013 Mark Adler + * Copyright (C) 1995-2017 Mark Adler * For conditions of distribution and use, see copyright notice in zlib.h */ -#ifdef HAVE_CONFIG_H -# include "config.h" -#endif #include "zutil.h" #include "inftrees.h" #include "inflate.h" #include "inffast.h" -#ifndef ASMINF - -/* Allow machine dependent optimization for post-increment or pre-increment. - Based on testing to date, - Pre-increment preferred for: - - PowerPC G3 (Adler) - - MIPS R5000 (Randers-Pehrson) - Post-increment preferred for: - - none - No measurable difference: - - Pentium III (Anderson) - - M68060 (Nikl) - */ -#ifdef POSTINC -# define OFF 0 -# define PUP(a) *(a)++ +#ifdef ASMINF +# pragma message("Assembler code may have bugs -- use at your own risk") #else -# define OFF 1 -# define PUP(a) *++(a) -#endif /* Decode literal, length, and distance codes and write out the resulting @@ -99,9 +79,9 @@ unsigned start; /* inflate()'s starting value for strm->avail_out */ /* copy state to local variables */ state = (struct inflate_state FAR *)strm->state; - in = strm->next_in - OFF; + in = strm->next_in; last = in + (strm->avail_in - 5); - out = strm->next_out - OFF; + out = strm->next_out; beg = out - (start - strm->avail_out); end = out + (strm->avail_out - 257); #ifdef INFLATE_STRICT @@ -122,9 +102,9 @@ unsigned start; /* inflate()'s starting value for strm->avail_out */ input data or output space */ do { if (bits < 15) { - hold += (unsigned long)(PUP(in)) << bits; + hold += (unsigned long)(*in++) << bits; bits += 8; - hold += (unsigned long)(PUP(in)) << bits; + hold += (unsigned long)(*in++) << bits; bits += 8; } here = lcode[hold & lmask]; @@ -137,14 +117,14 @@ unsigned start; /* inflate()'s starting value for strm->avail_out */ Tracevv((stderr, here.val >= 0x20 && here.val < 0x7f ? "inflate: literal '%c'\n" : "inflate: literal 0x%02x\n", here.val)); - PUP(out) = (unsigned char)(here.val); + *out++ = (unsigned char)(here.val); } else if (op & 16) { /* length base */ len = (unsigned)(here.val); op &= 15; /* number of extra bits */ if (op) { if (bits < op) { - hold += (unsigned long)(PUP(in)) << bits; + hold += (unsigned long)(*in++) << bits; bits += 8; } len += (unsigned)hold & ((1U << op) - 1); @@ -153,9 +133,9 @@ unsigned start; /* inflate()'s starting value for strm->avail_out */ } Tracevv((stderr, "inflate: length %u\n", len)); if (bits < 15) { - hold += (unsigned long)(PUP(in)) << bits; + hold += (unsigned long)(*in++) << bits; bits += 8; - hold += (unsigned long)(PUP(in)) << bits; + hold += (unsigned long)(*in++) << bits; bits += 8; } here = dcode[hold & dmask]; @@ -168,10 +148,10 @@ unsigned start; /* inflate()'s starting value for strm->avail_out */ dist = (unsigned)(here.val); op &= 15; /* number of extra bits */ if (bits < op) { - hold += (unsigned long)(PUP(in)) << bits; + hold += (unsigned long)(*in++) << bits; bits += 8; if (bits < op) { - hold += (unsigned long)(PUP(in)) << bits; + hold += (unsigned long)(*in++) << bits; bits += 8; } } @@ -199,30 +179,30 @@ unsigned start; /* inflate()'s starting value for strm->avail_out */ #ifdef INFLATE_ALLOW_INVALID_DISTANCE_TOOFAR_ARRR if (len <= op - whave) { do { - PUP(out) = 0; + *out++ = 0; } while (--len); continue; } len -= op - whave; do { - PUP(out) = 0; + *out++ = 0; } while (--op > whave); if (op == 0) { from = out - dist; do { - PUP(out) = PUP(from); + *out++ = *from++; } while (--len); continue; } #endif } - from = window - OFF; + from = window; if (wnext == 0) { /* very common case */ from += wsize - op; if (op < len) { /* some from window */ len -= op; do { - PUP(out) = PUP(from); + *out++ = *from++; } while (--op); from = out - dist; /* rest from output */ } @@ -233,14 +213,14 @@ unsigned start; /* inflate()'s starting value for strm->avail_out */ if (op < len) { /* some from end of window */ len -= op; do { - PUP(out) = PUP(from); + *out++ = *from++; } while (--op); - from = window - OFF; + from = window; if (wnext < len) { /* some from start of window */ op = wnext; len -= op; do { - PUP(out) = PUP(from); + *out++ = *from++; } while (--op); from = out - dist; /* rest from output */ } @@ -251,35 +231,35 @@ unsigned start; /* inflate()'s starting value for strm->avail_out */ if (op < len) { /* some from window */ len -= op; do { - PUP(out) = PUP(from); + *out++ = *from++; } while (--op); from = out - dist; /* rest from output */ } } while (len > 2) { - PUP(out) = PUP(from); - PUP(out) = PUP(from); - PUP(out) = PUP(from); + *out++ = *from++; + *out++ = *from++; + *out++ = *from++; len -= 3; } if (len) { - PUP(out) = PUP(from); + *out++ = *from++; if (len > 1) - PUP(out) = PUP(from); + *out++ = *from++; } } else { from = out - dist; /* copy direct from output */ do { /* minimum length is three */ - PUP(out) = PUP(from); - PUP(out) = PUP(from); - PUP(out) = PUP(from); + *out++ = *from++; + *out++ = *from++; + *out++ = *from++; len -= 3; } while (len > 2); if (len) { - PUP(out) = PUP(from); + *out++ = *from++; if (len > 1) - PUP(out) = PUP(from); + *out++ = *from++; } } } @@ -316,8 +296,8 @@ unsigned start; /* inflate()'s starting value for strm->avail_out */ hold &= (1U << bits) - 1; /* update state and return */ - strm->next_in = in + OFF; - strm->next_out = out + OFF; + strm->next_in = in; + strm->next_out = out; strm->avail_in = (unsigned)(in < last ? 5 + (last - in) : 5 - (in - last)); strm->avail_out = (unsigned)(out < end ? 257 + (end - out) : 257 - (out - end)); diff --git a/erts/emulator/zlib/inflate.c b/erts/emulator/zlib/inflate.c index 532330b06b..ac333e8c2e 100644 --- a/erts/emulator/zlib/inflate.c +++ b/erts/emulator/zlib/inflate.c @@ -1,5 +1,5 @@ /* inflate.c -- zlib decompression - * Copyright (C) 1995-2012 Mark Adler + * Copyright (C) 1995-2016 Mark Adler * For conditions of distribution and use, see copyright notice in zlib.h */ @@ -80,9 +80,6 @@ * The history for versions after 1.2.0 are in ChangeLog in zlib distribution. */ -#ifdef HAVE_CONFIG_H -# include "config.h" -#endif #include "zutil.h" #include "inftrees.h" #include "inflate.h" @@ -95,6 +92,7 @@ #endif /* function prototypes */ +local int inflateStateCheck OF((z_streamp strm)); local void fixedtables OF((struct inflate_state FAR *state)); local int updatewindow OF((z_streamp strm, const unsigned char FAR *end, unsigned copy)); @@ -104,12 +102,26 @@ local int updatewindow OF((z_streamp strm, const unsigned char FAR *end, local unsigned syncsearch OF((unsigned FAR *have, const unsigned char FAR *buf, unsigned len)); +local int inflateStateCheck(strm) +z_streamp strm; +{ + struct inflate_state FAR *state; + if (strm == Z_NULL || + strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) + return 1; + state = (struct inflate_state FAR *)strm->state; + if (state == Z_NULL || state->strm != strm || + state->mode < HEAD || state->mode > SYNC) + return 1; + return 0; +} + int ZEXPORT inflateResetKeep(strm) z_streamp strm; { struct inflate_state FAR *state; - if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; + if (inflateStateCheck(strm)) return Z_STREAM_ERROR; state = (struct inflate_state FAR *)strm->state; strm->total_in = strm->total_out = state->total = 0; strm->msg = Z_NULL; @@ -134,7 +146,7 @@ z_streamp strm; { struct inflate_state FAR *state; - if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; + if (inflateStateCheck(strm)) return Z_STREAM_ERROR; state = (struct inflate_state FAR *)strm->state; state->wsize = 0; state->whave = 0; @@ -150,7 +162,7 @@ int windowBits; struct inflate_state FAR *state; /* get the state */ - if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; + if (inflateStateCheck(strm)) return Z_STREAM_ERROR; state = (struct inflate_state FAR *)strm->state; /* extract wrap request from windowBits parameter */ @@ -159,7 +171,7 @@ int windowBits; windowBits = -windowBits; } else { - wrap = (windowBits >> 4) + 1; + wrap = (windowBits >> 4) + 5; #ifdef GUNZIP if (windowBits < 48) windowBits &= 15; @@ -213,7 +225,9 @@ int stream_size; if (state == Z_NULL) return Z_MEM_ERROR; Tracev((stderr, "inflate: allocated\n")); strm->state = (struct internal_state FAR *)state; + state->strm = strm; state->window = Z_NULL; + state->mode = HEAD; /* to pass state test in inflateReset2() */ ret = inflateReset2(strm, windowBits); if (ret != Z_OK) { ZFREE(strm, state); @@ -237,17 +251,17 @@ int value; { struct inflate_state FAR *state; - if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; + if (inflateStateCheck(strm)) return Z_STREAM_ERROR; state = (struct inflate_state FAR *)strm->state; if (bits < 0) { state->hold = 0; state->bits = 0; return Z_OK; } - if (bits > 16 || state->bits + bits > 32) return Z_STREAM_ERROR; + if (bits > 16 || state->bits + (uInt)bits > 32) return Z_STREAM_ERROR; value &= (1L << bits) - 1; - state->hold += value << state->bits; - state->bits += bits; + state->hold += (unsigned)value << state->bits; + state->bits += (uInt)bits; return Z_OK; } @@ -628,7 +642,7 @@ int flush; static const unsigned short order[19] = /* permutation of code lengths */ {16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15}; - if (strm == Z_NULL || strm->state == Z_NULL || strm->next_out == Z_NULL || + if (inflateStateCheck(strm) || strm->next_out == Z_NULL || (strm->next_in == Z_NULL && strm->avail_in != 0)) return Z_STREAM_ERROR; @@ -648,6 +662,8 @@ int flush; NEEDBITS(16); #ifdef GUNZIP if ((state->wrap & 2) && hold == 0x8b1f) { /* gzip header */ + if (state->wbits == 0) + state->wbits = 15; state->check = crc32(0L, Z_NULL, 0); CRC2(state->check, hold); INITBITS(); @@ -675,7 +691,7 @@ int flush; len = BITS(4) + 8; if (state->wbits == 0) state->wbits = len; - else if (len > state->wbits) { + if (len > 15 || len > state->wbits) { strm->msg = (char *)"invalid window size"; state->mode = BAD; break; @@ -702,14 +718,16 @@ int flush; } if (state->head != Z_NULL) state->head->text = (int)((hold >> 8) & 1); - if (state->flags & 0x0200) CRC2(state->check, hold); + if ((state->flags & 0x0200) && (state->wrap & 4)) + CRC2(state->check, hold); INITBITS(); state->mode = TIME; case TIME: NEEDBITS(32); if (state->head != Z_NULL) state->head->time = hold; - if (state->flags & 0x0200) CRC4(state->check, hold); + if ((state->flags & 0x0200) && (state->wrap & 4)) + CRC4(state->check, hold); INITBITS(); state->mode = OS; case OS: @@ -718,7 +736,8 @@ int flush; state->head->xflags = (int)(hold & 0xff); state->head->os = (int)(hold >> 8); } - if (state->flags & 0x0200) CRC2(state->check, hold); + if ((state->flags & 0x0200) && (state->wrap & 4)) + CRC2(state->check, hold); INITBITS(); state->mode = EXLEN; case EXLEN: @@ -727,7 +746,8 @@ int flush; state->length = (unsigned)(hold); if (state->head != Z_NULL) state->head->extra_len = (unsigned)hold; - if (state->flags & 0x0200) CRC2(state->check, hold); + if ((state->flags & 0x0200) && (state->wrap & 4)) + CRC2(state->check, hold); INITBITS(); } else if (state->head != Z_NULL) @@ -745,7 +765,7 @@ int flush; len + copy > state->head->extra_max ? state->head->extra_max - len : copy); } - if (state->flags & 0x0200) + if ((state->flags & 0x0200) && (state->wrap & 4)) state->check = crc32(state->check, next, copy); have -= copy; next += copy; @@ -764,9 +784,9 @@ int flush; if (state->head != Z_NULL && state->head->name != Z_NULL && state->length < state->head->name_max) - state->head->name[state->length++] = len; + state->head->name[state->length++] = (Bytef)len; } while (len && copy < have); - if (state->flags & 0x0200) + if ((state->flags & 0x0200) && (state->wrap & 4)) state->check = crc32(state->check, next, copy); have -= copy; next += copy; @@ -785,9 +805,9 @@ int flush; if (state->head != Z_NULL && state->head->comment != Z_NULL && state->length < state->head->comm_max) - state->head->comment[state->length++] = len; + state->head->comment[state->length++] = (Bytef)len; } while (len && copy < have); - if (state->flags & 0x0200) + if ((state->flags & 0x0200) && (state->wrap & 4)) state->check = crc32(state->check, next, copy); have -= copy; next += copy; @@ -799,7 +819,7 @@ int flush; case HCRC: if (state->flags & 0x0200) { NEEDBITS(16); - if (hold != (state->check & 0xffff)) { + if ((state->wrap & 4) && hold != (state->check & 0xffff)) { strm->msg = (char *)"header crc mismatch"; state->mode = BAD; break; @@ -1180,11 +1200,11 @@ int flush; out -= left; strm->total_out += out; state->total += out; - if (out) + if ((state->wrap & 4) && out) strm->adler = state->check = UPDATE(state->check, put - out, out); out = left; - if (( + if ((state->wrap & 4) && ( #ifdef GUNZIP state->flags ? hold : #endif @@ -1243,10 +1263,10 @@ int flush; strm->total_in += in; strm->total_out += out; state->total += out; - if (state->wrap && out) + if ((state->wrap & 4) && out) strm->adler = state->check = UPDATE(state->check, strm->next_out - out, out); - strm->data_type = state->bits + (state->last ? 64 : 0) + + strm->data_type = (int)state->bits + (state->last ? 64 : 0) + (state->mode == TYPE ? 128 : 0) + (state->mode == LEN_ || state->mode == COPY_ ? 256 : 0); if (((in == 0 && out == 0) || flush == Z_FINISH) && ret == Z_OK) @@ -1258,7 +1278,7 @@ int ZEXPORT inflateEnd(strm) z_streamp strm; { struct inflate_state FAR *state; - if (strm == Z_NULL || strm->state == Z_NULL || strm->zfree == (free_func)0) + if (inflateStateCheck(strm)) return Z_STREAM_ERROR; state = (struct inflate_state FAR *)strm->state; if (state->window != Z_NULL) ZFREE(strm, state->window); @@ -1276,7 +1296,7 @@ uInt *dictLength; struct inflate_state FAR *state; /* check state */ - if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; + if (inflateStateCheck(strm)) return Z_STREAM_ERROR; state = (struct inflate_state FAR *)strm->state; /* copy dictionary */ @@ -1301,7 +1321,7 @@ uInt dictLength; int ret; /* check state */ - if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; + if (inflateStateCheck(strm)) return Z_STREAM_ERROR; state = (struct inflate_state FAR *)strm->state; if (state->wrap != 0 && state->mode != DICT) return Z_STREAM_ERROR; @@ -1333,7 +1353,7 @@ gz_headerp head; struct inflate_state FAR *state; /* check state */ - if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; + if (inflateStateCheck(strm)) return Z_STREAM_ERROR; state = (struct inflate_state FAR *)strm->state; if ((state->wrap & 2) == 0) return Z_STREAM_ERROR; @@ -1386,7 +1406,7 @@ z_streamp strm; struct inflate_state FAR *state; /* check parameters */ - if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; + if (inflateStateCheck(strm)) return Z_STREAM_ERROR; state = (struct inflate_state FAR *)strm->state; if (strm->avail_in == 0 && state->bits < 8) return Z_BUF_ERROR; @@ -1433,7 +1453,7 @@ z_streamp strm; { struct inflate_state FAR *state; - if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; + if (inflateStateCheck(strm)) return Z_STREAM_ERROR; state = (struct inflate_state FAR *)strm->state; return state->mode == STORED && state->bits == 0; } @@ -1448,8 +1468,7 @@ z_streamp source; unsigned wsize; /* check input */ - if (dest == Z_NULL || source == Z_NULL || source->state == Z_NULL || - source->zalloc == (alloc_func)0 || source->zfree == (free_func)0) + if (inflateStateCheck(source) || dest == Z_NULL) return Z_STREAM_ERROR; state = (struct inflate_state FAR *)source->state; @@ -1470,6 +1489,7 @@ z_streamp source; /* copy state */ zmemcpy((voidpf)dest, (voidpf)source, sizeof(z_stream)); zmemcpy((voidpf)copy, (voidpf)state, sizeof(struct inflate_state)); + copy->strm = dest; if (state->lencode >= state->codes && state->lencode <= state->codes + ENOUGH - 1) { copy->lencode = copy->codes + (state->lencode - state->codes); @@ -1491,25 +1511,51 @@ int subvert; { struct inflate_state FAR *state; - if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; + if (inflateStateCheck(strm)) return Z_STREAM_ERROR; state = (struct inflate_state FAR *)strm->state; - state->sane = !subvert; #ifdef INFLATE_ALLOW_INVALID_DISTANCE_TOOFAR_ARRR + state->sane = !subvert; return Z_OK; #else + (void)subvert; state->sane = 1; return Z_DATA_ERROR; #endif } +int ZEXPORT inflateValidate(strm, check) +z_streamp strm; +int check; +{ + struct inflate_state FAR *state; + + if (inflateStateCheck(strm)) return Z_STREAM_ERROR; + state = (struct inflate_state FAR *)strm->state; + if (check) + state->wrap |= 4; + else + state->wrap &= ~4; + return Z_OK; +} + long ZEXPORT inflateMark(strm) z_streamp strm; { struct inflate_state FAR *state; - if (strm == Z_NULL || strm->state == Z_NULL) return -1L << 16; + if (inflateStateCheck(strm)) + return -(1L << 16); state = (struct inflate_state FAR *)strm->state; - return ((long)(state->back) << 16) + + return (long)(((unsigned long)((long)state->back)) << 16) + (state->mode == COPY ? state->length : (state->mode == MATCH ? state->was - state->length : 0)); } + +unsigned long ZEXPORT inflateCodesUsed(strm) +z_streamp strm; +{ + struct inflate_state FAR *state; + if (inflateStateCheck(strm)) return (unsigned long)-1; + state = (struct inflate_state FAR *)strm->state; + return (unsigned long)(state->next - state->codes); +} diff --git a/erts/emulator/zlib/inflate.h b/erts/emulator/zlib/inflate.h index 95f4986d40..a46cce6b6d 100644 --- a/erts/emulator/zlib/inflate.h +++ b/erts/emulator/zlib/inflate.h @@ -1,5 +1,5 @@ /* inflate.h -- internal inflate state definition - * Copyright (C) 1995-2009 Mark Adler + * Copyright (C) 1995-2016 Mark Adler * For conditions of distribution and use, see copyright notice in zlib.h */ @@ -18,7 +18,7 @@ /* Possible inflate modes between inflate() calls */ typedef enum { - HEAD, /* i: waiting for magic header */ + HEAD = 16180, /* i: waiting for magic header */ FLAGS, /* i: waiting for method and flags (gzip) */ TIME, /* i: waiting for modification time (gzip) */ OS, /* i: waiting for extra flags and operating system (gzip) */ @@ -77,11 +77,14 @@ typedef enum { CHECK -> LENGTH -> DONE */ -/* state maintained between inflate() calls. Approximately 10K bytes. */ +/* State maintained between inflate() calls -- approximately 7K bytes, not + including the allocated sliding window, which is up to 32K bytes. */ struct inflate_state { + z_streamp strm; /* pointer back to this zlib stream */ inflate_mode mode; /* current inflate mode */ int last; /* true if processing last block */ - int wrap; /* bit 0 true for zlib, bit 1 true for gzip */ + int wrap; /* bit 0 true for zlib, bit 1 true for gzip, + bit 2 true to validate check value */ int havedict; /* true if dictionary provided */ int flags; /* gzip header method and flags (0 if zlib) */ unsigned dmax; /* zlib header max distance (INFLATE_STRICT) */ diff --git a/erts/emulator/zlib/inftrees.c b/erts/emulator/zlib/inftrees.c index 3766fa2646..2ea08fc13e 100644 --- a/erts/emulator/zlib/inftrees.c +++ b/erts/emulator/zlib/inftrees.c @@ -1,18 +1,15 @@ /* inftrees.c -- generate Huffman trees for efficient decoding - * Copyright (C) 1995-2013 Mark Adler + * Copyright (C) 1995-2017 Mark Adler * For conditions of distribution and use, see copyright notice in zlib.h */ -#ifdef HAVE_CONFIG_H -# include "config.h" -#endif #include "zutil.h" #include "inftrees.h" #define MAXBITS 15 const char inflate_copyright[] = - " inflate 1.2.8 Copyright 1995-2013 Mark Adler "; + " inflate 1.2.11 Copyright 1995-2017 Mark Adler "; /* If you use the zlib library in a product, an acknowledgment is welcome in the documentation of your product. If for some reason you cannot @@ -57,7 +54,7 @@ unsigned short FAR *work; code FAR *next; /* next available space in table */ const unsigned short FAR *base; /* base value table to use */ const unsigned short FAR *extra; /* extra bits table to use */ - int end; /* use base and extra for symbol > end */ + unsigned match; /* use base and extra for symbol >= match */ unsigned short count[MAXBITS+1]; /* number of codes of each length */ unsigned short offs[MAXBITS+1]; /* offsets in table for each length */ static const unsigned short lbase[31] = { /* Length codes 257..285 base */ @@ -65,7 +62,7 @@ unsigned short FAR *work; 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0}; static const unsigned short lext[31] = { /* Length codes 257..285 extra */ 16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18, - 19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 72, 78}; + 19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 77, 202}; static const unsigned short dbase[32] = { /* Distance codes 0..29 base */ 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193, 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145, @@ -184,19 +181,17 @@ unsigned short FAR *work; switch (type) { case CODES: base = extra = work; /* dummy value--not used */ - end = 19; + match = 20; break; case LENS: base = lbase; - base -= 257; extra = lext; - extra -= 257; - end = 256; + match = 257; break; - default: /* DISTS */ + default: /* DISTS */ base = dbase; extra = dext; - end = -1; + match = 0; } /* initialize state for loop */ @@ -219,13 +214,13 @@ unsigned short FAR *work; for (;;) { /* create table entry */ here.bits = (unsigned char)(len - drop); - if ((int)(work[sym]) < end) { + if (work[sym] + 1U < match) { here.op = (unsigned char)0; here.val = work[sym]; } - else if ((int)(work[sym]) > end) { - here.op = (unsigned char)(extra[work[sym]]); - here.val = base[work[sym]]; + else if (work[sym] >= match) { + here.op = (unsigned char)(extra[work[sym] - match]); + here.val = base[work[sym] - match]; } else { here.op = (unsigned char)(32 + 64); /* end of block */ diff --git a/erts/emulator/zlib/trees.c b/erts/emulator/zlib/trees.c index 465e944e5b..50cf4b4571 100644 --- a/erts/emulator/zlib/trees.c +++ b/erts/emulator/zlib/trees.c @@ -1,5 +1,5 @@ /* trees.c -- output deflated data using Huffman coding - * Copyright (C) 1995-2012 Jean-loup Gailly + * Copyright (C) 1995-2017 Jean-loup Gailly * detect_data_type() function provided freely by Cosmin Truta, 2006 * For conditions of distribution and use, see copyright notice in zlib.h */ @@ -34,12 +34,9 @@ /* #define GEN_TREES_H */ -#ifdef HAVE_CONFIG_H -# include "config.h" -#endif #include "deflate.h" -#ifdef DEBUG +#ifdef ZLIB_DEBUG # include <ctype.h> #endif @@ -125,13 +122,13 @@ struct static_tree_desc_s { int max_length; /* max bit length for the codes */ }; -local static_tree_desc static_l_desc = +local const static_tree_desc static_l_desc = {static_ltree, extra_lbits, LITERALS+1, L_CODES, MAX_BITS}; -local static_tree_desc static_d_desc = +local const static_tree_desc static_d_desc = {static_dtree, extra_dbits, 0, D_CODES, MAX_BITS}; -local static_tree_desc static_bl_desc = +local const static_tree_desc static_bl_desc = {(const ct_data *)0, extra_blbits, 0, BL_CODES, MAX_BL_BITS}; /* =========================================================================== @@ -155,18 +152,16 @@ local int detect_data_type OF((deflate_state *s)); local unsigned bi_reverse OF((unsigned value, int length)); local void bi_windup OF((deflate_state *s)); local void bi_flush OF((deflate_state *s)); -local void copy_block OF((deflate_state *s, charf *buf, unsigned len, - int header)); #ifdef GEN_TREES_H local void gen_trees_header OF((void)); #endif -#ifndef DEBUG +#ifndef ZLIB_DEBUG # define send_code(s, c, tree) send_bits(s, tree[c].Code, tree[c].Len) /* Send a code of the given tree. c and tree must not have side effects */ -#else /* DEBUG */ +#else /* !ZLIB_DEBUG */ # define send_code(s, c, tree) \ { if (z_verbose>2) fprintf(stderr,"\ncd %3d ",(c)); \ send_bits(s, tree[c].Code, tree[c].Len); } @@ -185,7 +180,7 @@ local void gen_trees_header OF((void)); * Send a value on a given number of bits. * IN assertion: length <= 16 and value fits in length bits. */ -#ifdef DEBUG +#ifdef ZLIB_DEBUG local void send_bits OF((deflate_state *s, int value, int length)); local void send_bits(s, value, length) @@ -211,12 +206,12 @@ local void send_bits(s, value, length) s->bi_valid += length; } } -#else /* !DEBUG */ +#else /* !ZLIB_DEBUG */ #define send_bits(s, value, length) \ { int len = length;\ if (s->bi_valid > (int)Buf_size - len) {\ - int val = value;\ + int val = (int)value;\ s->bi_buf |= (ush)val << s->bi_valid;\ put_short(s, s->bi_buf);\ s->bi_buf = (ush)val >> (Buf_size - s->bi_valid);\ @@ -226,7 +221,7 @@ local void send_bits(s, value, length) s->bi_valid += len;\ }\ } -#endif /* DEBUG */ +#endif /* ZLIB_DEBUG */ /* the arguments must not have side effects */ @@ -320,7 +315,7 @@ local void tr_static_init() * Genererate the file trees.h describing the static trees. */ #ifdef GEN_TREES_H -# ifndef DEBUG +# ifndef ZLIB_DEBUG # include <stdio.h> # endif @@ -397,7 +392,7 @@ void ZLIB_INTERNAL _tr_init(s) s->bi_buf = 0; s->bi_valid = 0; -#ifdef DEBUG +#ifdef ZLIB_DEBUG s->compressed_len = 0L; s->bits_sent = 0L; #endif @@ -525,12 +520,12 @@ local void gen_bitlen(s, desc) xbits = 0; if (n >= base) xbits = extra[n-base]; f = tree[n].Freq; - s->opt_len += (ulg)f * (bits + xbits); - if (stree) s->static_len += (ulg)f * (stree[n].Len + xbits); + s->opt_len += (ulg)f * (unsigned)(bits + xbits); + if (stree) s->static_len += (ulg)f * (unsigned)(stree[n].Len + xbits); } if (overflow == 0) return; - Trace((stderr,"\nbit length overflow\n")); + Tracev((stderr,"\nbit length overflow\n")); /* This happens for example on obj2 and pic of the Calgary corpus */ /* Find the first bit length which could increase: */ @@ -557,9 +552,8 @@ local void gen_bitlen(s, desc) m = s->heap[--h]; if (m > max_code) continue; if ((unsigned) tree[m].Len != (unsigned) bits) { - Trace((stderr,"code %d bits %d->%d\n", m, tree[m].Len, bits)); - s->opt_len += ((long)bits - (long)tree[m].Len) - *(long)tree[m].Freq; + Tracev((stderr,"code %d bits %d->%d\n", m, tree[m].Len, bits)); + s->opt_len += ((ulg)bits - tree[m].Len) * tree[m].Freq; tree[m].Len = (ush)bits; } n--; @@ -581,7 +575,7 @@ local void gen_codes (tree, max_code, bl_count) ushf *bl_count; /* number of codes at each bit length */ { ush next_code[MAX_BITS+1]; /* next code value for each bit length */ - ush code = 0; /* running code value */ + unsigned code = 0; /* running code value */ int bits; /* bit index */ int n; /* code index */ @@ -589,7 +583,8 @@ local void gen_codes (tree, max_code, bl_count) * without bit reversal. */ for (bits = 1; bits <= MAX_BITS; bits++) { - next_code[bits] = code = (code + bl_count[bits-1]) << 1; + code = (code + bl_count[bits-1]) << 1; + next_code[bits] = (ush)code; } /* Check that the bit counts in bl_count are consistent. The last code * must be all ones. @@ -602,7 +597,7 @@ local void gen_codes (tree, max_code, bl_count) int len = tree[n].Len; if (len == 0) continue; /* Now reverse the bits */ - tree[n].Code = bi_reverse(next_code[len]++, len); + tree[n].Code = (ush)bi_reverse(next_code[len]++, len); Tracecv(tree != static_ltree, (stderr,"\nn %3d %c l %2d c %4x (%x) ", n, (isgraph(n) ? n : ' '), len, tree[n].Code, next_code[len]-1)); @@ -824,7 +819,7 @@ local int build_bl_tree(s) if (s->bl_tree[bl_order[max_blindex]].Len != 0) break; } /* Update opt_len to include the bit length tree and counts */ - s->opt_len += 3*(max_blindex+1) + 5+5+4; + s->opt_len += 3*((ulg)max_blindex+1) + 5+5+4; Tracev((stderr, "\ndyn trees: dyn %ld, stat %ld", s->opt_len, s->static_len)); @@ -872,11 +867,17 @@ void ZLIB_INTERNAL _tr_stored_block(s, buf, stored_len, last) int last; /* one if this is the last block for a file */ { send_bits(s, (STORED_BLOCK<<1)+last, 3); /* send block type */ -#ifdef DEBUG + bi_windup(s); /* align on byte boundary */ + put_short(s, (ush)stored_len); + put_short(s, (ush)~stored_len); + zmemcpy(s->pending_buf + s->pending, (Bytef *)buf, stored_len); + s->pending += stored_len; +#ifdef ZLIB_DEBUG s->compressed_len = (s->compressed_len + 3 + 7) & (ulg)~7L; s->compressed_len += (stored_len + 4) << 3; + s->bits_sent += 2*16; + s->bits_sent += stored_len<<3; #endif - copy_block(s, buf, (unsigned)stored_len, 1); /* with header */ } /* =========================================================================== @@ -897,7 +898,7 @@ void ZLIB_INTERNAL _tr_align(s) { send_bits(s, STATIC_TREES<<1, 3); send_code(s, END_BLOCK, static_ltree); -#ifdef DEBUG +#ifdef ZLIB_DEBUG s->compressed_len += 10L; /* 3 for block type, 7 for EOB */ #endif bi_flush(s); @@ -905,7 +906,7 @@ void ZLIB_INTERNAL _tr_align(s) /* =========================================================================== * Determine the best encoding for the current block: dynamic trees, static - * trees or store, and output the encoded block to the zip file. + * trees or store, and write out the encoded block. */ void ZLIB_INTERNAL _tr_flush_block(s, buf, stored_len, last) deflate_state *s; @@ -977,7 +978,7 @@ void ZLIB_INTERNAL _tr_flush_block(s, buf, stored_len, last) send_bits(s, (STATIC_TREES<<1)+last, 3); compress_block(s, (const ct_data *)static_ltree, (const ct_data *)static_dtree); -#ifdef DEBUG +#ifdef ZLIB_DEBUG s->compressed_len += 3 + s->static_len; #endif } else { @@ -986,7 +987,7 @@ void ZLIB_INTERNAL _tr_flush_block(s, buf, stored_len, last) max_blindex+1); compress_block(s, (const ct_data *)s->dyn_ltree, (const ct_data *)s->dyn_dtree); -#ifdef DEBUG +#ifdef ZLIB_DEBUG s->compressed_len += 3 + s->opt_len; #endif } @@ -998,7 +999,7 @@ void ZLIB_INTERNAL _tr_flush_block(s, buf, stored_len, last) if (last) { bi_windup(s); -#ifdef DEBUG +#ifdef ZLIB_DEBUG s->compressed_len += 7; /* align on byte boundary */ #endif } @@ -1093,7 +1094,7 @@ local void compress_block(s, ltree, dtree) send_code(s, code, dtree); /* send the distance code */ extra = extra_dbits[code]; if (extra != 0) { - dist -= base_dist[code]; + dist -= (unsigned)base_dist[code]; send_bits(s, dist, extra); /* send the extra distance bits */ } } /* literal or match pair ? */ @@ -1196,34 +1197,7 @@ local void bi_windup(s) } s->bi_buf = 0; s->bi_valid = 0; -#ifdef DEBUG +#ifdef ZLIB_DEBUG s->bits_sent = (s->bits_sent+7) & ~7; #endif } - -/* =========================================================================== - * Copy a stored block, storing first the length and its - * one's complement if requested. - */ -local void copy_block(s, buf, len, header) - deflate_state *s; - charf *buf; /* the input data */ - unsigned len; /* its length */ - int header; /* true if block header must be written */ -{ - bi_windup(s); /* align on byte boundary */ - - if (header) { - put_short(s, (ush)len); - put_short(s, (ush)~len); -#ifdef DEBUG - s->bits_sent += 2*16; -#endif - } -#ifdef DEBUG - s->bits_sent += (ulg)len<<3; -#endif - while (len--) { - put_byte(s, *buf++); - } -} diff --git a/erts/emulator/zlib/uncompr.c b/erts/emulator/zlib/uncompr.c index 864d571719..f03a1a865e 100644 --- a/erts/emulator/zlib/uncompr.c +++ b/erts/emulator/zlib/uncompr.c @@ -1,62 +1,93 @@ /* uncompr.c -- decompress a memory buffer - * Copyright (C) 1995-2003, 2010 Jean-loup Gailly. + * Copyright (C) 1995-2003, 2010, 2014, 2016 Jean-loup Gailly, Mark Adler * For conditions of distribution and use, see copyright notice in zlib.h */ /* @(#) $Id$ */ -#ifdef HAVE_CONFIG_H -# include "config.h" -#endif #define ZLIB_INTERNAL #include "zlib.h" /* =========================================================================== - Decompresses the source buffer into the destination buffer. sourceLen is - the byte length of the source buffer. Upon entry, destLen is the total - size of the destination buffer, which must be large enough to hold the - entire uncompressed data. (The size of the uncompressed data must have - been saved previously by the compressor and transmitted to the decompressor - by some mechanism outside the scope of this compression library.) - Upon exit, destLen is the actual size of the compressed buffer. - - uncompress returns Z_OK if success, Z_MEM_ERROR if there was not - enough memory, Z_BUF_ERROR if there was not enough room in the output - buffer, or Z_DATA_ERROR if the input data was corrupted. + Decompresses the source buffer into the destination buffer. *sourceLen is + the byte length of the source buffer. Upon entry, *destLen is the total size + of the destination buffer, which must be large enough to hold the entire + uncompressed data. (The size of the uncompressed data must have been saved + previously by the compressor and transmitted to the decompressor by some + mechanism outside the scope of this compression library.) Upon exit, + *destLen is the size of the decompressed data and *sourceLen is the number + of source bytes consumed. Upon return, source + *sourceLen points to the + first unused input byte. + + uncompress returns Z_OK if success, Z_MEM_ERROR if there was not enough + memory, Z_BUF_ERROR if there was not enough room in the output buffer, or + Z_DATA_ERROR if the input data was corrupted, including if the input data is + an incomplete zlib stream. */ -int ZEXPORT uncompress (dest, destLen, source, sourceLen) +int ZEXPORT uncompress2 (dest, destLen, source, sourceLen) Bytef *dest; uLongf *destLen; const Bytef *source; - uLong sourceLen; + uLong *sourceLen; { z_stream stream; int err; + const uInt max = (uInt)-1; + uLong len, left; + Byte buf[1]; /* for detection of incomplete stream when *destLen == 0 */ - stream.next_in = (z_const Bytef *)source; - stream.avail_in = (uInt)sourceLen; - /* Check for source > 64K on 16-bit machine: */ - if ((uLong)stream.avail_in != sourceLen) return Z_BUF_ERROR; - - stream.next_out = dest; - stream.avail_out = (uInt)*destLen; - if ((uLong)stream.avail_out != *destLen) return Z_BUF_ERROR; + len = *sourceLen; + if (*destLen) { + left = *destLen; + *destLen = 0; + } + else { + left = 1; + dest = buf; + } + stream.next_in = (z_const Bytef *)source; + stream.avail_in = 0; stream.zalloc = (alloc_func)0; stream.zfree = (free_func)0; + stream.opaque = (voidpf)0; err = inflateInit(&stream); if (err != Z_OK) return err; - err = inflate(&stream, Z_FINISH); - if (err != Z_STREAM_END) { - inflateEnd(&stream); - if (err == Z_NEED_DICT || (err == Z_BUF_ERROR && stream.avail_in == 0)) - return Z_DATA_ERROR; - return err; - } - *destLen = stream.total_out; + stream.next_out = dest; + stream.avail_out = 0; - err = inflateEnd(&stream); - return err; + do { + if (stream.avail_out == 0) { + stream.avail_out = left > (uLong)max ? max : (uInt)left; + left -= stream.avail_out; + } + if (stream.avail_in == 0) { + stream.avail_in = len > (uLong)max ? max : (uInt)len; + len -= stream.avail_in; + } + err = inflate(&stream, Z_NO_FLUSH); + } while (err == Z_OK); + + *sourceLen -= len + stream.avail_in; + if (dest != buf) + *destLen = stream.total_out; + else if (stream.total_out && err == Z_BUF_ERROR) + left = 1; + + inflateEnd(&stream); + return err == Z_STREAM_END ? Z_OK : + err == Z_NEED_DICT ? Z_DATA_ERROR : + err == Z_BUF_ERROR && left + stream.avail_out ? Z_DATA_ERROR : + err; +} + +int ZEXPORT uncompress (dest, destLen, source, sourceLen) + Bytef *dest; + uLongf *destLen; + const Bytef *source; + uLong sourceLen; +{ + return uncompress2(dest, destLen, source, &sourceLen); } diff --git a/erts/emulator/zlib/zconf.h b/erts/emulator/zlib/zconf.h index 9987a77553..5e1d68a004 100644 --- a/erts/emulator/zlib/zconf.h +++ b/erts/emulator/zlib/zconf.h @@ -1,5 +1,5 @@ /* zconf.h -- configuration of the zlib compression library - * Copyright (C) 1995-2013 Jean-loup Gailly. + * Copyright (C) 1995-2016 Jean-loup Gailly, Mark Adler * For conditions of distribution and use, see copyright notice in zlib.h */ @@ -17,7 +17,7 @@ #ifdef Z_PREFIX /* may be set to #if 1 by ./configure */ # define Z_PREFIX_SET -/* all linked symbols */ +/* all linked symbols and init macros */ # define _dist_code z__dist_code # define _length_code z__length_code # define _tr_align z__tr_align @@ -29,6 +29,7 @@ # define adler32 z_adler32 # define adler32_combine z_adler32_combine # define adler32_combine64 z_adler32_combine64 +# define adler32_z z_adler32_z # ifndef Z_SOLO # define compress z_compress # define compress2 z_compress2 @@ -37,10 +38,14 @@ # define crc32 z_crc32 # define crc32_combine z_crc32_combine # define crc32_combine64 z_crc32_combine64 +# define crc32_z z_crc32_z # define deflate z_deflate # define deflateBound z_deflateBound # define deflateCopy z_deflateCopy # define deflateEnd z_deflateEnd +# define deflateGetDictionary z_deflateGetDictionary +# define deflateInit z_deflateInit +# define deflateInit2 z_deflateInit2 # define deflateInit2_ z_deflateInit2_ # define deflateInit_ z_deflateInit_ # define deflateParams z_deflateParams @@ -67,6 +72,8 @@ # define gzeof z_gzeof # define gzerror z_gzerror # define gzflush z_gzflush +# define gzfread z_gzfread +# define gzfwrite z_gzfwrite # define gzgetc z_gzgetc # define gzgetc_ z_gzgetc_ # define gzgets z_gzgets @@ -78,7 +85,6 @@ # define gzopen_w z_gzopen_w # endif # define gzprintf z_gzprintf -# define gzvprintf z_gzvprintf # define gzputc z_gzputc # define gzputs z_gzputs # define gzread z_gzread @@ -89,32 +95,39 @@ # define gztell z_gztell # define gztell64 z_gztell64 # define gzungetc z_gzungetc +# define gzvprintf z_gzvprintf # define gzwrite z_gzwrite # endif # define inflate z_inflate # define inflateBack z_inflateBack # define inflateBackEnd z_inflateBackEnd +# define inflateBackInit z_inflateBackInit # define inflateBackInit_ z_inflateBackInit_ +# define inflateCodesUsed z_inflateCodesUsed # define inflateCopy z_inflateCopy # define inflateEnd z_inflateEnd +# define inflateGetDictionary z_inflateGetDictionary # define inflateGetHeader z_inflateGetHeader +# define inflateInit z_inflateInit +# define inflateInit2 z_inflateInit2 # define inflateInit2_ z_inflateInit2_ # define inflateInit_ z_inflateInit_ # define inflateMark z_inflateMark # define inflatePrime z_inflatePrime # define inflateReset z_inflateReset # define inflateReset2 z_inflateReset2 +# define inflateResetKeep z_inflateResetKeep # define inflateSetDictionary z_inflateSetDictionary -# define inflateGetDictionary z_inflateGetDictionary # define inflateSync z_inflateSync # define inflateSyncPoint z_inflateSyncPoint # define inflateUndermine z_inflateUndermine -# define inflateResetKeep z_inflateResetKeep +# define inflateValidate z_inflateValidate # define inflate_copyright z_inflate_copyright # define inflate_fast z_inflate_fast # define inflate_table z_inflate_table # ifndef Z_SOLO # define uncompress z_uncompress +# define uncompress2 z_uncompress2 # endif # define zError z_zError # ifndef Z_SOLO @@ -224,9 +237,19 @@ # define z_const #endif -/* Some Mac compilers merge all .h files incorrectly: */ -#if defined(__MWERKS__)||defined(applec)||defined(THINK_C)||defined(__SC__) -# define NO_DUMMY_DECL +#ifdef Z_SOLO + typedef unsigned long z_size_t; +#else +# define z_longlong long long +# if defined(NO_SIZE_T) + typedef unsigned NO_SIZE_T z_size_t; +# elif defined(STDC) +# include <stddef.h> + typedef size_t z_size_t; +# else + typedef unsigned long z_size_t; +# endif +# undef z_longlong #endif /* Maximum value for memLevel in deflateInit2 */ @@ -256,7 +279,7 @@ Of course this will generally degrade compression (there's no free lunch). The memory requirements for inflate are (in bytes) 1 << windowBits - that is, 32K for windowBits=15 (default value) plus a few kilobytes + that is, 32K for windowBits=15 (default value) plus about 7 kilobytes for small objects. */ diff --git a/erts/emulator/zlib/zlib.h b/erts/emulator/zlib/zlib.h index 3e0c7672ac..f09cdaf1e0 100644 --- a/erts/emulator/zlib/zlib.h +++ b/erts/emulator/zlib/zlib.h @@ -1,7 +1,7 @@ /* zlib.h -- interface of the 'zlib' general purpose compression library - version 1.2.8, April 28th, 2013 + version 1.2.11, January 15th, 2017 - Copyright (C) 1995-2013 Jean-loup Gailly and Mark Adler + Copyright (C) 1995-2017 Jean-loup Gailly and Mark Adler This software is provided 'as-is', without any express or implied warranty. In no event will the authors be held liable for any damages @@ -37,11 +37,11 @@ extern "C" { #endif -#define ZLIB_VERSION "1.2.8" -#define ZLIB_VERNUM 0x1280 +#define ZLIB_VERSION "1.2.11" +#define ZLIB_VERNUM 0x12b0 #define ZLIB_VER_MAJOR 1 #define ZLIB_VER_MINOR 2 -#define ZLIB_VER_REVISION 8 +#define ZLIB_VER_REVISION 11 #define ZLIB_VER_SUBREVISION 0 /* @@ -65,7 +65,8 @@ extern "C" { with "gz". The gzip format is different from the zlib format. gzip is a gzip wrapper, documented in RFC 1952, wrapped around a deflate stream. - This library can optionally read and write gzip streams in memory as well. + This library can optionally read and write gzip and raw deflate streams in + memory as well. The zlib format was designed to be compact and fast for use in memory and on communications channels. The gzip format was designed for single- @@ -74,7 +75,7 @@ extern "C" { The library does not install any signal handler. The decoder checks the consistency of the compressed data, so the library should never crash - even in case of corrupted input. + even in the case of corrupted input. */ typedef voidpf (*alloc_func) OF((voidpf opaque, uInt items, uInt size)); @@ -87,7 +88,7 @@ typedef struct z_stream_s { uInt avail_in; /* number of bytes available at next_in */ uLong total_in; /* total number of input bytes read so far */ - Bytef *next_out; /* next output byte should be put there */ + Bytef *next_out; /* next output byte will go here */ uInt avail_out; /* remaining free space at next_out */ uLong total_out; /* total number of bytes output so far */ @@ -98,8 +99,9 @@ typedef struct z_stream_s { free_func zfree; /* used to free the internal state */ voidpf opaque; /* private data object passed to zalloc and zfree */ - int data_type; /* best guess about the data type: binary or text */ - uLong adler; /* adler32 value of the uncompressed data */ + int data_type; /* best guess about the data type: binary or text + for deflate, or the decoding state for inflate */ + uLong adler; /* Adler-32 or CRC-32 value of the uncompressed data */ uLong reserved; /* reserved for future use */ } z_stream; @@ -142,7 +144,9 @@ typedef gz_header FAR *gz_headerp; zalloc must return Z_NULL if there is not enough memory for the object. If zlib is used in a multi-threaded application, zalloc and zfree must be - thread safe. + thread safe. In that case, zlib is thread-safe. When zalloc and zfree are + Z_NULL on entry to the initialization function, they are set to internal + routines that use the standard library functions malloc() and free(). On 16-bit systems, the functions zalloc and zfree must be able to allocate exactly 65536 bytes, but will not be required to allocate more than this if @@ -155,7 +159,7 @@ typedef gz_header FAR *gz_headerp; The fields total_in and total_out can be used for statistics or progress reports. After compression, total_in holds the total size of the - uncompressed data and may be saved for use in the decompressor (particularly + uncompressed data and may be saved for use by the decompressor (particularly if the decompressor wants to decompress everything in a single step). */ @@ -200,7 +204,7 @@ typedef gz_header FAR *gz_headerp; #define Z_TEXT 1 #define Z_ASCII Z_TEXT /* for compatibility with 1.2.2 and earlier */ #define Z_UNKNOWN 2 -/* Possible values of the data_type field (though see inflate()) */ +/* Possible values of the data_type field for deflate() */ #define Z_DEFLATED 8 /* The deflate compression method (the only one supported in this version) */ @@ -258,11 +262,11 @@ ZEXTERN int ZEXPORT deflate OF((z_streamp strm, int flush)); enough room in the output buffer), next_in and avail_in are updated and processing will resume at this point for the next call of deflate(). - - Provide more output starting at next_out and update next_out and avail_out + - Generate more output starting at next_out and update next_out and avail_out accordingly. This action is forced if the parameter flush is non zero. Forcing flush frequently degrades the compression ratio, so this parameter - should be set only when necessary (in interactive applications). Some - output may be provided even if flush is not set. + should be set only when necessary. Some output may be provided even if + flush is zero. Before the call of deflate(), the application should ensure that at least one of the actions is possible, by providing more input and/or consuming more @@ -271,7 +275,9 @@ ZEXTERN int ZEXPORT deflate OF((z_streamp strm, int flush)); output when it wants, for example when the output buffer is full (avail_out == 0), or after each call of deflate(). If deflate returns Z_OK and with zero avail_out, it must be called again after making room in the output - buffer because there might be more output pending. + buffer because there might be more output pending. See deflatePending(), + which can be used if desired to determine whether or not there is more ouput + in that case. Normally the parameter flush is set to Z_NO_FLUSH, which allows deflate to decide how much data to accumulate before producing output, in order to @@ -292,8 +298,8 @@ ZEXTERN int ZEXPORT deflate OF((z_streamp strm, int flush)); input data so far will be available to the decompressor, as for Z_SYNC_FLUSH. This completes the current deflate block and follows it with an empty fixed codes block that is 10 bits long. This assures that enough bytes are output - in order for the decompressor to finish the block before the empty fixed code - block. + in order for the decompressor to finish the block before the empty fixed + codes block. If flush is set to Z_BLOCK, a deflate block is completed and emitted, as for Z_SYNC_FLUSH, but the output is not aligned on a byte boundary, and up to @@ -319,34 +325,38 @@ ZEXTERN int ZEXPORT deflate OF((z_streamp strm, int flush)); If the parameter flush is set to Z_FINISH, pending input is processed, pending output is flushed and deflate returns with Z_STREAM_END if there was - enough output space; if deflate returns with Z_OK, this function must be - called again with Z_FINISH and more output space (updated avail_out) but no - more input data, until it returns with Z_STREAM_END or an error. After - deflate has returned Z_STREAM_END, the only possible operations on the stream - are deflateReset or deflateEnd. - - Z_FINISH can be used immediately after deflateInit if all the compression - is to be done in a single step. In this case, avail_out must be at least the - value returned by deflateBound (see below). Then deflate is guaranteed to - return Z_STREAM_END. If not enough output space is provided, deflate will - not return Z_STREAM_END, and it must be called again as described above. - - deflate() sets strm->adler to the adler32 checksum of all input read - so far (that is, total_in bytes). + enough output space. If deflate returns with Z_OK or Z_BUF_ERROR, this + function must be called again with Z_FINISH and more output space (updated + avail_out) but no more input data, until it returns with Z_STREAM_END or an + error. After deflate has returned Z_STREAM_END, the only possible operations + on the stream are deflateReset or deflateEnd. + + Z_FINISH can be used in the first deflate call after deflateInit if all the + compression is to be done in a single step. In order to complete in one + call, avail_out must be at least the value returned by deflateBound (see + below). Then deflate is guaranteed to return Z_STREAM_END. If not enough + output space is provided, deflate will not return Z_STREAM_END, and it must + be called again as described above. + + deflate() sets strm->adler to the Adler-32 checksum of all input read + so far (that is, total_in bytes). If a gzip stream is being generated, then + strm->adler will be the CRC-32 checksum of the input read so far. (See + deflateInit2 below.) deflate() may update strm->data_type if it can make a good guess about - the input data type (Z_BINARY or Z_TEXT). In doubt, the data is considered - binary. This field is only for information purposes and does not affect the - compression algorithm in any manner. + the input data type (Z_BINARY or Z_TEXT). If in doubt, the data is + considered binary. This field is only for information purposes and does not + affect the compression algorithm in any manner. deflate() returns Z_OK if some progress has been made (more input processed or more output produced), Z_STREAM_END if all input has been consumed and all output has been produced (only when flush is set to Z_FINISH), Z_STREAM_ERROR if the stream state was inconsistent (for example - if next_in or next_out was Z_NULL), Z_BUF_ERROR if no progress is possible - (for example avail_in or avail_out was zero). Note that Z_BUF_ERROR is not - fatal, and deflate() can be called again with more input and more output - space to continue compressing. + if next_in or next_out was Z_NULL or the state was inadvertently written over + by the application), or Z_BUF_ERROR if no progress is possible (for example + avail_in or avail_out was zero). Note that Z_BUF_ERROR is not fatal, and + deflate() can be called again with more input and more output space to + continue compressing. */ @@ -369,23 +379,21 @@ ZEXTERN int ZEXPORT inflateInit OF((z_streamp strm)); Initializes the internal stream state for decompression. The fields next_in, avail_in, zalloc, zfree and opaque must be initialized before by - the caller. If next_in is not Z_NULL and avail_in is large enough (the - exact value depends on the compression method), inflateInit determines the - compression method from the zlib header and allocates all data structures - accordingly; otherwise the allocation will be deferred to the first call of - inflate. If zalloc and zfree are set to Z_NULL, inflateInit updates them to - use default allocation functions. + the caller. In the current version of inflate, the provided input is not + read or consumed. The allocation of a sliding window will be deferred to + the first call of inflate (if the decompression does not complete on the + first call). If zalloc and zfree are set to Z_NULL, inflateInit updates + them to use default allocation functions. inflateInit returns Z_OK if success, Z_MEM_ERROR if there was not enough memory, Z_VERSION_ERROR if the zlib library version is incompatible with the version assumed by the caller, or Z_STREAM_ERROR if the parameters are invalid, such as a null pointer to the structure. msg is set to null if - there is no error message. inflateInit does not perform any decompression - apart from possibly reading the zlib header if present: actual decompression - will be done by inflate(). (So next_in and avail_in may be modified, but - next_out and avail_out are unused and unchanged.) The current implementation - of inflateInit() does not process any header information -- that is deferred - until inflate() is called. + there is no error message. inflateInit does not perform any decompression. + Actual decompression will be done by inflate(). So next_in, and avail_in, + next_out, and avail_out are unused and unchanged. The current + implementation of inflateInit() does not process any header information -- + that is deferred until inflate() is called. */ @@ -401,17 +409,20 @@ ZEXTERN int ZEXPORT inflate OF((z_streamp strm, int flush)); - Decompress more input starting at next_in and update next_in and avail_in accordingly. If not all input can be processed (because there is not - enough room in the output buffer), next_in is updated and processing will - resume at this point for the next call of inflate(). + enough room in the output buffer), then next_in and avail_in are updated + accordingly, and processing will resume at this point for the next call of + inflate(). - - Provide more output starting at next_out and update next_out and avail_out + - Generate more output starting at next_out and update next_out and avail_out accordingly. inflate() provides as much output as possible, until there is no more input data or no more space in the output buffer (see below about the flush parameter). Before the call of inflate(), the application should ensure that at least one of the actions is possible, by providing more input and/or consuming more - output, and updating the next_* and avail_* values accordingly. The + output, and updating the next_* and avail_* values accordingly. If the + caller of inflate() does not provide both available input and available + output space, it is possible that there will be no progress made. The application can consume the uncompressed output when it wants, for example when the output buffer is full (avail_out == 0), or after each call of inflate(). If inflate returns Z_OK and with zero avail_out, it must be @@ -428,7 +439,7 @@ ZEXTERN int ZEXPORT inflate OF((z_streamp strm, int flush)); gets to the end of that block, or when it runs out of data. The Z_BLOCK option assists in appending to or combining deflate streams. - Also to assist in this, on return inflate() will set strm->data_type to the + To assist in this, on return inflate() always sets strm->data_type to the number of unused bits in the last byte taken from strm->next_in, plus 64 if inflate() is currently decoding the last block in the deflate stream, plus 128 if inflate() returned immediately after decoding an end-of-block code or @@ -454,7 +465,7 @@ ZEXTERN int ZEXPORT inflate OF((z_streamp strm, int flush)); this case all pending input is processed and all pending output is flushed; avail_out must be large enough to hold all of the uncompressed data for the operation to complete. (The size of the uncompressed data may have been - saved by the compressor for this purpose.) The use of Z_FINISH is not + saved by the compressor for this purpose.) The use of Z_FINISH is not required to perform an inflation in one step. However it may be used to inform inflate that a faster approach can be used for the single inflate() call. Z_FINISH also informs inflate to not maintain a sliding window if the @@ -476,32 +487,33 @@ ZEXTERN int ZEXPORT inflate OF((z_streamp strm, int flush)); chosen by the compressor and returns Z_NEED_DICT; otherwise it sets strm->adler to the Adler-32 checksum of all output produced so far (that is, total_out bytes) and returns Z_OK, Z_STREAM_END or an error code as described - below. At the end of the stream, inflate() checks that its computed adler32 + below. At the end of the stream, inflate() checks that its computed Adler-32 checksum is equal to that saved by the compressor and returns Z_STREAM_END only if the checksum is correct. inflate() can decompress and check either zlib-wrapped or gzip-wrapped deflate data. The header type is detected automatically, if requested when initializing with inflateInit2(). Any information contained in the gzip - header is not retained, so applications that need that information should - instead use raw inflate, see inflateInit2() below, or inflateBack() and - perform their own processing of the gzip header and trailer. When processing + header is not retained unless inflateGetHeader() is used. When processing gzip-wrapped deflate data, strm->adler32 is set to the CRC-32 of the output - producted so far. The CRC-32 is checked against the gzip trailer. + produced so far. The CRC-32 is checked against the gzip trailer, as is the + uncompressed length, modulo 2^32. inflate() returns Z_OK if some progress has been made (more input processed or more output produced), Z_STREAM_END if the end of the compressed data has been reached and all uncompressed output has been produced, Z_NEED_DICT if a preset dictionary is needed at this point, Z_DATA_ERROR if the input data was corrupted (input stream not conforming to the zlib format or incorrect check - value), Z_STREAM_ERROR if the stream structure was inconsistent (for example - next_in or next_out was Z_NULL), Z_MEM_ERROR if there was not enough memory, - Z_BUF_ERROR if no progress is possible or if there was not enough room in the - output buffer when Z_FINISH is used. Note that Z_BUF_ERROR is not fatal, and + value, in which case strm->msg points to a string with a more specific + error), Z_STREAM_ERROR if the stream structure was inconsistent (for example + next_in or next_out was Z_NULL, or the state was inadvertently written over + by the application), Z_MEM_ERROR if there was not enough memory, Z_BUF_ERROR + if no progress was possible or if there was not enough room in the output + buffer when Z_FINISH is used. Note that Z_BUF_ERROR is not fatal, and inflate() can be called again with more input and more output space to continue decompressing. If Z_DATA_ERROR is returned, the application may then call inflateSync() to look for a good compression block if a partial - recovery of the data is desired. + recovery of the data is to be attempted. */ @@ -511,9 +523,8 @@ ZEXTERN int ZEXPORT inflateEnd OF((z_streamp strm)); This function discards any unprocessed input and does not flush any pending output. - inflateEnd returns Z_OK if success, Z_STREAM_ERROR if the stream state - was inconsistent. In the error case, msg may be set but then points to a - static string (which must not be deallocated). + inflateEnd returns Z_OK if success, or Z_STREAM_ERROR if the stream state + was inconsistent. */ @@ -544,16 +555,29 @@ ZEXTERN int ZEXPORT deflateInit2 OF((z_streamp strm, compression at the expense of memory usage. The default value is 15 if deflateInit is used instead. + For the current implementation of deflate(), a windowBits value of 8 (a + window size of 256 bytes) is not supported. As a result, a request for 8 + will result in 9 (a 512-byte window). In that case, providing 8 to + inflateInit2() will result in an error when the zlib header with 9 is + checked against the initialization of inflate(). The remedy is to not use 8 + with deflateInit2() with this initialization, or at least in that case use 9 + with inflateInit2(). + windowBits can also be -8..-15 for raw deflate. In this case, -windowBits determines the window size. deflate() will then generate raw deflate data - with no zlib header or trailer, and will not compute an adler32 check value. + with no zlib header or trailer, and will not compute a check value. windowBits can also be greater than 15 for optional gzip encoding. Add 16 to windowBits to write a simple gzip header and trailer around the compressed data instead of a zlib wrapper. The gzip header will have no file name, no extra data, no comment, no modification time (set to zero), no - header crc, and the operating system will be set to 255 (unknown). If a - gzip stream is being written, strm->adler is a crc32 instead of an adler32. + header crc, and the operating system will be set to the appropriate value, + if the operating system was determined at compile time. If a gzip stream is + being written, strm->adler is a CRC-32 instead of an Adler-32. + + For raw deflate or gzip encoding, a request for a 256-byte window is + rejected as invalid, since only the zlib header provides a means of + transmitting the window size to the decompressor. The memLevel parameter specifies how much memory should be allocated for the internal compression state. memLevel=1 uses minimum memory but is @@ -614,12 +638,12 @@ ZEXTERN int ZEXPORT deflateSetDictionary OF((z_streamp strm, addition, the current implementation of deflate will use at most the window size minus 262 bytes of the provided dictionary. - Upon return of this function, strm->adler is set to the adler32 value + Upon return of this function, strm->adler is set to the Adler-32 value of the dictionary; the decompressor may later use this value to determine - which dictionary has been used by the compressor. (The adler32 value + which dictionary has been used by the compressor. (The Adler-32 value applies to the whole dictionary even if only a subset of the dictionary is actually used by the compressor.) If a raw deflate was requested, then the - adler32 value is not computed and strm->adler is not set. + Adler-32 value is not computed and strm->adler is not set. deflateSetDictionary returns Z_OK if success, or Z_STREAM_ERROR if a parameter is invalid (e.g. dictionary being Z_NULL) or the stream state is @@ -628,6 +652,28 @@ ZEXTERN int ZEXPORT deflateSetDictionary OF((z_streamp strm, not perform any compression: this will be done by deflate(). */ +ZEXTERN int ZEXPORT deflateGetDictionary OF((z_streamp strm, + Bytef *dictionary, + uInt *dictLength)); +/* + Returns the sliding dictionary being maintained by deflate. dictLength is + set to the number of bytes in the dictionary, and that many bytes are copied + to dictionary. dictionary must have enough space, where 32768 bytes is + always enough. If deflateGetDictionary() is called with dictionary equal to + Z_NULL, then only the dictionary length is returned, and nothing is copied. + Similary, if dictLength is Z_NULL, then it is not set. + + deflateGetDictionary() may return a length less than the window size, even + when more than the window size in input has been provided. It may return up + to 258 bytes less in that case, due to how zlib's implementation of deflate + manages the sliding window and lookahead for matches, where matches can be + up to 258 bytes long. If the application needs the last window-size bytes of + input, then that would need to be saved by the application outside of zlib. + + deflateGetDictionary returns Z_OK on success, or Z_STREAM_ERROR if the + stream state is inconsistent. +*/ + ZEXTERN int ZEXPORT deflateCopy OF((z_streamp dest, z_streamp source)); /* @@ -648,10 +694,10 @@ ZEXTERN int ZEXPORT deflateCopy OF((z_streamp dest, ZEXTERN int ZEXPORT deflateReset OF((z_streamp strm)); /* - This function is equivalent to deflateEnd followed by deflateInit, - but does not free and reallocate all the internal compression state. The - stream will keep the same compression level and any other attributes that - may have been set by deflateInit2. + This function is equivalent to deflateEnd followed by deflateInit, but + does not free and reallocate the internal compression state. The stream + will leave the compression level and any other attributes that may have been + set unchanged. deflateReset returns Z_OK if success, or Z_STREAM_ERROR if the source stream state was inconsistent (such as zalloc or state being Z_NULL). @@ -662,20 +708,36 @@ ZEXTERN int ZEXPORT deflateParams OF((z_streamp strm, int strategy)); /* Dynamically update the compression level and compression strategy. The - interpretation of level and strategy is as in deflateInit2. This can be + interpretation of level and strategy is as in deflateInit2(). This can be used to switch between compression and straight copy of the input data, or to switch to a different kind of input data requiring a different strategy. - If the compression level is changed, the input available so far is - compressed with the old level (and may be flushed); the new level will take - effect only at the next call of deflate(). - - Before the call of deflateParams, the stream state must be set as for - a call of deflate(), since the currently available input may have to be - compressed and flushed. In particular, strm->avail_out must be non-zero. - - deflateParams returns Z_OK if success, Z_STREAM_ERROR if the source - stream state was inconsistent or if a parameter was invalid, Z_BUF_ERROR if - strm->avail_out was zero. + If the compression approach (which is a function of the level) or the + strategy is changed, and if any input has been consumed in a previous + deflate() call, then the input available so far is compressed with the old + level and strategy using deflate(strm, Z_BLOCK). There are three approaches + for the compression levels 0, 1..3, and 4..9 respectively. The new level + and strategy will take effect at the next call of deflate(). + + If a deflate(strm, Z_BLOCK) is performed by deflateParams(), and it does + not have enough output space to complete, then the parameter change will not + take effect. In this case, deflateParams() can be called again with the + same parameters and more output space to try again. + + In order to assure a change in the parameters on the first try, the + deflate stream should be flushed using deflate() with Z_BLOCK or other flush + request until strm.avail_out is not zero, before calling deflateParams(). + Then no more input data should be provided before the deflateParams() call. + If this is done, the old level and strategy will be applied to the data + compressed before deflateParams(), and the new level and strategy will be + applied to the the data compressed after deflateParams(). + + deflateParams returns Z_OK on success, Z_STREAM_ERROR if the source stream + state was inconsistent or if a parameter was invalid, or Z_BUF_ERROR if + there was not enough output space to complete the compression of the + available input data before a change in the strategy or approach. Note that + in the case of a Z_BUF_ERROR, the parameters are not changed. A return + value of Z_BUF_ERROR is not fatal, in which case deflateParams() can be + retried with more output space. */ ZEXTERN int ZEXPORT deflateTune OF((z_streamp strm, @@ -793,7 +855,7 @@ ZEXTERN int ZEXPORT inflateInit2 OF((z_streamp strm, is for use with other formats that use the deflate compressed data format such as zip. Those formats provide their own check values. If a custom format is developed using the raw deflate format for compressed data, it is - recommended that a check value such as an adler32 or a crc32 be applied to + recommended that a check value such as an Adler-32 or a CRC-32 be applied to the uncompressed data as is done in the zlib, gzip, and zip formats. For most applications, the zlib format should be used as is. Note that comments above on the use in deflateInit2() applies to the magnitude of windowBits. @@ -802,7 +864,10 @@ ZEXTERN int ZEXPORT inflateInit2 OF((z_streamp strm, 32 to windowBits to enable zlib and gzip decoding with automatic header detection, or add 16 to decode only the gzip format (the zlib format will return a Z_DATA_ERROR). If a gzip stream is being decoded, strm->adler is a - crc32 instead of an adler32. + CRC-32 instead of an Adler-32. Unlike the gunzip utility and gzread() (see + below), inflate() will not automatically decode concatenated gzip streams. + inflate() will return Z_STREAM_END at the end of the gzip stream. The state + would need to be reset to continue decoding a subsequent gzip stream. inflateInit2 returns Z_OK if success, Z_MEM_ERROR if there was not enough memory, Z_VERSION_ERROR if the zlib library version is incompatible with the @@ -823,7 +888,7 @@ ZEXTERN int ZEXPORT inflateSetDictionary OF((z_streamp strm, Initializes the decompression dictionary from the given uncompressed byte sequence. This function must be called immediately after a call of inflate, if that call returned Z_NEED_DICT. The dictionary chosen by the compressor - can be determined from the adler32 value returned by that call of inflate. + can be determined from the Adler-32 value returned by that call of inflate. The compressor and decompressor must use exactly the same dictionary (see deflateSetDictionary). For raw inflate, this function can be called at any time to set the dictionary. If the provided dictionary is smaller than the @@ -834,7 +899,7 @@ ZEXTERN int ZEXPORT inflateSetDictionary OF((z_streamp strm, inflateSetDictionary returns Z_OK if success, Z_STREAM_ERROR if a parameter is invalid (e.g. dictionary being Z_NULL) or the stream state is inconsistent, Z_DATA_ERROR if the given dictionary doesn't match the - expected one (incorrect adler32 value). inflateSetDictionary does not + expected one (incorrect Adler-32 value). inflateSetDictionary does not perform any decompression: this will be done by subsequent calls of inflate(). */ @@ -892,7 +957,7 @@ ZEXTERN int ZEXPORT inflateCopy OF((z_streamp dest, ZEXTERN int ZEXPORT inflateReset OF((z_streamp strm)); /* This function is equivalent to inflateEnd followed by inflateInit, - but does not free and reallocate all the internal decompression state. The + but does not free and reallocate the internal decompression state. The stream will keep attributes that may have been set by inflateInit2. inflateReset returns Z_OK if success, or Z_STREAM_ERROR if the source @@ -904,7 +969,9 @@ ZEXTERN int ZEXPORT inflateReset2 OF((z_streamp strm, /* This function is the same as inflateReset, but it also permits changing the wrap and window size requests. The windowBits parameter is interpreted - the same as it is for inflateInit2. + the same as it is for inflateInit2. If the window size is changed, then the + memory allocated for the window is freed, and the window will be reallocated + by inflate() if needed. inflateReset2 returns Z_OK if success, or Z_STREAM_ERROR if the source stream state was inconsistent (such as zalloc or state being Z_NULL), or if @@ -956,7 +1023,7 @@ ZEXTERN long ZEXPORT inflateMark OF((z_streamp strm)); location in the input stream can be determined from avail_in and data_type as noted in the description for the Z_BLOCK flush parameter for inflate. - inflateMark returns the value noted above or -1 << 16 if the provided + inflateMark returns the value noted above, or -65536 if the provided source stream state was inconsistent. */ @@ -1048,9 +1115,9 @@ ZEXTERN int ZEXPORT inflateBack OF((z_streamp strm, This routine would normally be used in a utility that reads zip or gzip files and writes out uncompressed files. The utility would decode the header and process the trailer on its own, hence this routine expects only - the raw deflate stream to decompress. This is different from the normal - behavior of inflate(), which expects either a zlib or gzip header and - trailer around the deflate stream. + the raw deflate stream to decompress. This is different from the default + behavior of inflate(), which expects a zlib header and trailer around the + deflate stream. inflateBack() uses two subroutines supplied by the caller that are then called by inflateBack() for input and output. inflateBack() calls those @@ -1059,12 +1126,12 @@ ZEXTERN int ZEXPORT inflateBack OF((z_streamp strm, parameters and return types are defined above in the in_func and out_func typedefs. inflateBack() will call in(in_desc, &buf) which should return the number of bytes of provided input, and a pointer to that input in buf. If - there is no input available, in() must return zero--buf is ignored in that - case--and inflateBack() will return a buffer error. inflateBack() will call - out(out_desc, buf, len) to write the uncompressed data buf[0..len-1]. out() - should return zero on success, or non-zero on failure. If out() returns - non-zero, inflateBack() will return with an error. Neither in() nor out() - are permitted to change the contents of the window provided to + there is no input available, in() must return zero -- buf is ignored in that + case -- and inflateBack() will return a buffer error. inflateBack() will + call out(out_desc, buf, len) to write the uncompressed data buf[0..len-1]. + out() should return zero on success, or non-zero on failure. If out() + returns non-zero, inflateBack() will return with an error. Neither in() nor + out() are permitted to change the contents of the window provided to inflateBackInit(), which is also the buffer that out() uses to write from. The length written by out() will be at most the window size. Any non-zero amount of input may be provided by in(). @@ -1092,7 +1159,7 @@ ZEXTERN int ZEXPORT inflateBack OF((z_streamp strm, using strm->next_in which will be Z_NULL only if in() returned an error. If strm->next_in is not Z_NULL, then the Z_BUF_ERROR was due to out() returning non-zero. (in() will always be called before out(), so strm->next_in is - assured to be defined if out() returns non-zero.) Note that inflateBack() + assured to be defined if out() returns non-zero.) Note that inflateBack() cannot return Z_OK. */ @@ -1114,7 +1181,7 @@ ZEXTERN uLong ZEXPORT zlibCompileFlags OF((void)); 7.6: size of z_off_t Compiler, assembler, and debug options: - 8: DEBUG + 8: ZLIB_DEBUG 9: ASMV or ASMINF -- use ASM code 10: ZLIB_WINAPI -- exported functions use the WINAPI calling convention 11: 0 (reserved) @@ -1164,7 +1231,8 @@ ZEXTERN int ZEXPORT compress OF((Bytef *dest, uLongf *destLen, the byte length of the source buffer. Upon entry, destLen is the total size of the destination buffer, which must be at least the value returned by compressBound(sourceLen). Upon exit, destLen is the actual size of the - compressed buffer. + compressed data. compress() is equivalent to compress2() with a level + parameter of Z_DEFAULT_COMPRESSION. compress returns Z_OK if success, Z_MEM_ERROR if there was not enough memory, Z_BUF_ERROR if there was not enough room in the output @@ -1180,7 +1248,7 @@ ZEXTERN int ZEXPORT compress2 OF((Bytef *dest, uLongf *destLen, length of the source buffer. Upon entry, destLen is the total size of the destination buffer, which must be at least the value returned by compressBound(sourceLen). Upon exit, destLen is the actual size of the - compressed buffer. + compressed data. compress2 returns Z_OK if success, Z_MEM_ERROR if there was not enough memory, Z_BUF_ERROR if there was not enough room in the output buffer, @@ -1203,7 +1271,7 @@ ZEXTERN int ZEXPORT uncompress OF((Bytef *dest, uLongf *destLen, uncompressed data. (The size of the uncompressed data must have been saved previously by the compressor and transmitted to the decompressor by some mechanism outside the scope of this compression library.) Upon exit, destLen - is the actual size of the uncompressed buffer. + is the actual size of the uncompressed data. uncompress returns Z_OK if success, Z_MEM_ERROR if there was not enough memory, Z_BUF_ERROR if there was not enough room in the output @@ -1212,6 +1280,14 @@ ZEXTERN int ZEXPORT uncompress OF((Bytef *dest, uLongf *destLen, buffer with the uncompressed data up to that point. */ +ZEXTERN int ZEXPORT uncompress2 OF((Bytef *dest, uLongf *destLen, + const Bytef *source, uLong *sourceLen)); +/* + Same as uncompress, except that sourceLen is a pointer, where the + length of the source is *sourceLen. On return, *sourceLen is the number of + source bytes consumed. +*/ + /* gzip file access functions */ /* @@ -1290,10 +1366,9 @@ ZEXTERN int ZEXPORT gzbuffer OF((gzFile file, unsigned size)); default buffer size is 8192 bytes. This function must be called after gzopen() or gzdopen(), and before any other calls that read or write the file. The buffer memory allocation is always deferred to the first read or - write. Two buffers are allocated, either both of the specified size when - writing, or one of the specified size and the other twice that size when - reading. A larger buffer size of, for example, 64K or 128K bytes will - noticeably increase the speed of decompression (reading). + write. Three times that size in buffer space is allocated. A larger buffer + size of, for example, 64K or 128K bytes will noticeably increase the speed + of decompression (reading). The new buffer size also affects the maximum length for gzprintf(). @@ -1304,10 +1379,12 @@ ZEXTERN int ZEXPORT gzbuffer OF((gzFile file, unsigned size)); ZEXTERN int ZEXPORT gzsetparams OF((gzFile file, int level, int strategy)); /* Dynamically update the compression level or strategy. See the description - of deflateInit2 for the meaning of these parameters. + of deflateInit2 for the meaning of these parameters. Previously provided + data is flushed before the parameter change. - gzsetparams returns Z_OK if success, or Z_STREAM_ERROR if the file was not - opened for writing. + gzsetparams returns Z_OK if success, Z_STREAM_ERROR if the file was not + opened for writing, Z_ERRNO if there is an error writing the flushed data, + or Z_MEM_ERROR if there is a memory allocation error. */ ZEXTERN int ZEXPORT gzread OF((gzFile file, voidp buf, unsigned len)); @@ -1335,7 +1412,35 @@ ZEXTERN int ZEXPORT gzread OF((gzFile file, voidp buf, unsigned len)); case. gzread returns the number of uncompressed bytes actually read, less than - len for end of file, or -1 for error. + len for end of file, or -1 for error. If len is too large to fit in an int, + then nothing is read, -1 is returned, and the error state is set to + Z_STREAM_ERROR. +*/ + +ZEXTERN z_size_t ZEXPORT gzfread OF((voidp buf, z_size_t size, z_size_t nitems, + gzFile file)); +/* + Read up to nitems items of size size from file to buf, otherwise operating + as gzread() does. This duplicates the interface of stdio's fread(), with + size_t request and return types. If the library defines size_t, then + z_size_t is identical to size_t. If not, then z_size_t is an unsigned + integer type that can contain a pointer. + + gzfread() returns the number of full items read of size size, or zero if + the end of the file was reached and a full item could not be read, or if + there was an error. gzerror() must be consulted if zero is returned in + order to determine if there was an error. If the multiplication of size and + nitems overflows, i.e. the product does not fit in a z_size_t, then nothing + is read, zero is returned, and the error state is set to Z_STREAM_ERROR. + + In the event that the end of file is reached and only a partial item is + available at the end, i.e. the remaining uncompressed data length is not a + multiple of size, then the final partial item is nevetheless read into buf + and the end-of-file flag is set. The length of the partial item read is not + provided, but could be inferred from the result of gztell(). This behavior + is the same as the behavior of fread() implementations in common libraries, + but it prevents the direct use of gzfread() to read a concurrently written + file, reseting and retrying on end-of-file, when size is not 1. */ ZEXTERN int ZEXPORT gzwrite OF((gzFile file, @@ -1346,19 +1451,33 @@ ZEXTERN int ZEXPORT gzwrite OF((gzFile file, error. */ +ZEXTERN z_size_t ZEXPORT gzfwrite OF((voidpc buf, z_size_t size, + z_size_t nitems, gzFile file)); +/* + gzfwrite() writes nitems items of size size from buf to file, duplicating + the interface of stdio's fwrite(), with size_t request and return types. If + the library defines size_t, then z_size_t is identical to size_t. If not, + then z_size_t is an unsigned integer type that can contain a pointer. + + gzfwrite() returns the number of full items written of size size, or zero + if there was an error. If the multiplication of size and nitems overflows, + i.e. the product does not fit in a z_size_t, then nothing is written, zero + is returned, and the error state is set to Z_STREAM_ERROR. +*/ + ZEXTERN int ZEXPORTVA gzprintf Z_ARG((gzFile file, const char *format, ...)); /* Converts, formats, and writes the arguments to the compressed file under control of the format string, as in fprintf. gzprintf returns the number of - uncompressed bytes actually written, or 0 in case of error. The number of - uncompressed bytes written is limited to 8191, or one less than the buffer - size given to gzbuffer(). The caller should assure that this limit is not - exceeded. If it is exceeded, then gzprintf() will return an error (0) with - nothing written. In this case, there may also be a buffer overflow with - unpredictable consequences, which is possible only if zlib was compiled with - the insecure functions sprintf() or vsprintf() because the secure snprintf() - or vsnprintf() functions were not available. This can be determined using - zlibCompileFlags(). + uncompressed bytes actually written, or a negative zlib error code in case + of error. The number of uncompressed bytes written is limited to 8191, or + one less than the buffer size given to gzbuffer(). The caller should assure + that this limit is not exceeded. If it is exceeded, then gzprintf() will + return an error (0) with nothing written. In this case, there may also be a + buffer overflow with unpredictable consequences, which is possible only if + zlib was compiled with the insecure functions sprintf() or vsprintf() + because the secure snprintf() or vsnprintf() functions were not available. + This can be determined using zlibCompileFlags(). */ ZEXTERN int ZEXPORT gzputs OF((gzFile file, const char *s)); @@ -1418,7 +1537,7 @@ ZEXTERN int ZEXPORT gzflush OF((gzFile file, int flush)); If the flush parameter is Z_FINISH, the remaining data is written and the gzip stream is completed in the output. If gzwrite() is called again, a new gzip stream will be started in the output. gzread() is able to read such - concatented gzip streams. + concatenated gzip streams. gzflush should be called only when strictly necessary because it will degrade compression if called too often. @@ -1572,7 +1691,7 @@ ZEXTERN uLong ZEXPORT adler32 OF((uLong adler, const Bytef *buf, uInt len)); return the updated checksum. If buf is Z_NULL, this function returns the required initial value for the checksum. - An Adler-32 checksum is almost as reliable as a CRC32 but can be computed + An Adler-32 checksum is almost as reliable as a CRC-32 but can be computed much faster. Usage example: @@ -1585,6 +1704,12 @@ ZEXTERN uLong ZEXPORT adler32 OF((uLong adler, const Bytef *buf, uInt len)); if (adler != original_adler) error(); */ +ZEXTERN uLong ZEXPORT adler32_z OF((uLong adler, const Bytef *buf, + z_size_t len)); +/* + Same as adler32(), but with a size_t length. +*/ + /* ZEXTERN uLong ZEXPORT adler32_combine OF((uLong adler1, uLong adler2, z_off_t len2)); @@ -1614,6 +1739,12 @@ ZEXTERN uLong ZEXPORT crc32 OF((uLong crc, const Bytef *buf, uInt len)); if (crc != original_crc) error(); */ +ZEXTERN uLong ZEXPORT crc32_z OF((uLong adler, const Bytef *buf, + z_size_t len)); +/* + Same as crc32(), but with a size_t length. +*/ + /* ZEXTERN uLong ZEXPORT crc32_combine OF((uLong crc1, uLong crc2, z_off_t len2)); @@ -1644,19 +1775,35 @@ ZEXTERN int ZEXPORT inflateBackInit_ OF((z_streamp strm, int windowBits, unsigned char FAR *window, const char *version, int stream_size)); -#define deflateInit(strm, level) \ - deflateInit_((strm), (level), ZLIB_VERSION, (int)sizeof(z_stream)) -#define inflateInit(strm) \ - inflateInit_((strm), ZLIB_VERSION, (int)sizeof(z_stream)) -#define deflateInit2(strm, level, method, windowBits, memLevel, strategy) \ - deflateInit2_((strm),(level),(method),(windowBits),(memLevel),\ - (strategy), ZLIB_VERSION, (int)sizeof(z_stream)) -#define inflateInit2(strm, windowBits) \ - inflateInit2_((strm), (windowBits), ZLIB_VERSION, \ - (int)sizeof(z_stream)) -#define inflateBackInit(strm, windowBits, window) \ - inflateBackInit_((strm), (windowBits), (window), \ - ZLIB_VERSION, (int)sizeof(z_stream)) +#ifdef Z_PREFIX_SET +# define z_deflateInit(strm, level) \ + deflateInit_((strm), (level), ZLIB_VERSION, (int)sizeof(z_stream)) +# define z_inflateInit(strm) \ + inflateInit_((strm), ZLIB_VERSION, (int)sizeof(z_stream)) +# define z_deflateInit2(strm, level, method, windowBits, memLevel, strategy) \ + deflateInit2_((strm),(level),(method),(windowBits),(memLevel),\ + (strategy), ZLIB_VERSION, (int)sizeof(z_stream)) +# define z_inflateInit2(strm, windowBits) \ + inflateInit2_((strm), (windowBits), ZLIB_VERSION, \ + (int)sizeof(z_stream)) +# define z_inflateBackInit(strm, windowBits, window) \ + inflateBackInit_((strm), (windowBits), (window), \ + ZLIB_VERSION, (int)sizeof(z_stream)) +#else +# define deflateInit(strm, level) \ + deflateInit_((strm), (level), ZLIB_VERSION, (int)sizeof(z_stream)) +# define inflateInit(strm) \ + inflateInit_((strm), ZLIB_VERSION, (int)sizeof(z_stream)) +# define deflateInit2(strm, level, method, windowBits, memLevel, strategy) \ + deflateInit2_((strm),(level),(method),(windowBits),(memLevel),\ + (strategy), ZLIB_VERSION, (int)sizeof(z_stream)) +# define inflateInit2(strm, windowBits) \ + inflateInit2_((strm), (windowBits), ZLIB_VERSION, \ + (int)sizeof(z_stream)) +# define inflateBackInit(strm, windowBits, window) \ + inflateBackInit_((strm), (windowBits), (window), \ + ZLIB_VERSION, (int)sizeof(z_stream)) +#endif #ifndef Z_SOLO @@ -1676,10 +1823,10 @@ ZEXTERN int ZEXPORT gzgetc_ OF((gzFile file)); /* backward compatibility */ #ifdef Z_PREFIX_SET # undef z_gzgetc # define z_gzgetc(g) \ - ((g)->have ? ((g)->have--, (g)->pos++, *((g)->next)++) : gzgetc(g)) + ((g)->have ? ((g)->have--, (g)->pos++, *((g)->next)++) : (gzgetc)(g)) #else # define gzgetc(g) \ - ((g)->have ? ((g)->have--, (g)->pos++, *((g)->next)++) : gzgetc(g)) + ((g)->have ? ((g)->have--, (g)->pos++, *((g)->next)++) : (gzgetc)(g)) #endif /* provide 64-bit offset functions if _LARGEFILE64_SOURCE defined, and/or @@ -1737,19 +1884,16 @@ ZEXTERN int ZEXPORT gzgetc_ OF((gzFile file)); /* backward compatibility */ #endif /* !Z_SOLO */ -/* hack for buggy compilers */ -#if !defined(ZUTIL_H) && !defined(NO_DUMMY_DECL) - struct internal_state {int dummy;}; -#endif - /* undocumented functions */ ZEXTERN const char * ZEXPORT zError OF((int)); ZEXTERN int ZEXPORT inflateSyncPoint OF((z_streamp)); ZEXTERN const z_crc_t FAR * ZEXPORT get_crc_table OF((void)); ZEXTERN int ZEXPORT inflateUndermine OF((z_streamp, int)); +ZEXTERN int ZEXPORT inflateValidate OF((z_streamp, int)); +ZEXTERN unsigned long ZEXPORT inflateCodesUsed OF ((z_streamp)); ZEXTERN int ZEXPORT inflateResetKeep OF((z_streamp)); ZEXTERN int ZEXPORT deflateResetKeep OF((z_streamp)); -#if defined(_WIN32) && !defined(Z_SOLO) +#if (defined(_WIN32) || defined(__CYGWIN__)) && !defined(Z_SOLO) ZEXTERN gzFile ZEXPORT gzopen_w OF((const wchar_t *path, const char *mode)); #endif diff --git a/erts/emulator/zlib/zlib.mk b/erts/emulator/zlib/zlib.mk index 3f0d64d250..b51b4ec8d6 100644 --- a/erts/emulator/zlib/zlib.mk +++ b/erts/emulator/zlib/zlib.mk @@ -52,7 +52,7 @@ ifeq ($(TYPE),gcov) ZLIB_CFLAGS = -O0 -fprofile-arcs -ftest-coverage $(DEBUG_CFLAGS) $(DEFS) $(THR_DEFS) else # gcov ifeq ($(TYPE),debug) -ZLIB_CFLAGS = $(DEBUG_CFLAGS) $(DEFS) $(THR_DEFS) +ZLIB_CFLAGS = -DZLIB_DEBUG=1 $(DEBUG_CFLAGS) $(DEFS) $(THR_DEFS) else # debug ZLIB_CFLAGS = $(subst -O2, -O3, $(CONFIGURE_CFLAGS) $(DEFS) $(THR_DEFS)) #ZLIB_CFLAGS=-O -DMAX_WBITS=14 -DMAX_MEM_LEVEL=7 @@ -62,6 +62,9 @@ ZLIB_CFLAGS = $(subst -O2, -O3, $(CONFIGURE_CFLAGS) $(DEFS) $(THR_DEFS)) endif # debug endif # gcov +# Don't fail if _LFS64_LARGEFILE is undefined +ZLIB_CFLAGS := $(filter-out -Werror=undef,$(ZLIB_CFLAGS)) + ifeq ($(TARGET), win32) $(ZLIB_LIBRARY): $(ZLIB_OBJS) $(V_AR) -out:$@ $(ZLIB_OBJS) diff --git a/erts/emulator/zlib/zutil.c b/erts/emulator/zlib/zutil.c index 27a8af4a2b..a76c6b0c7e 100644 --- a/erts/emulator/zlib/zutil.c +++ b/erts/emulator/zlib/zutil.c @@ -1,33 +1,27 @@ /* zutil.c -- target dependent utility functions for the compression library - * Copyright (C) 1995-2005, 2010, 2011, 2012 Jean-loup Gailly. + * Copyright (C) 1995-2017 Jean-loup Gailly * For conditions of distribution and use, see copyright notice in zlib.h */ /* @(#) $Id$ */ -#ifdef HAVE_CONFIG_H -# include "config.h" -#endif #include "zutil.h" #ifndef Z_SOLO # include "gzguts.h" #endif -#ifndef NO_DUMMY_DECL -struct internal_state {int dummy;}; /* for buggy compilers */ -#endif - z_const char * const z_errmsg[10] = { -"need dictionary", /* Z_NEED_DICT 2 */ -"stream end", /* Z_STREAM_END 1 */ -"", /* Z_OK 0 */ -"file error", /* Z_ERRNO (-1) */ -"stream error", /* Z_STREAM_ERROR (-2) */ -"data error", /* Z_DATA_ERROR (-3) */ -"insufficient memory", /* Z_MEM_ERROR (-4) */ -"buffer error", /* Z_BUF_ERROR (-5) */ -"incompatible version",/* Z_VERSION_ERROR (-6) */ -""}; + (z_const char *)"need dictionary", /* Z_NEED_DICT 2 */ + (z_const char *)"stream end", /* Z_STREAM_END 1 */ + (z_const char *)"", /* Z_OK 0 */ + (z_const char *)"file error", /* Z_ERRNO (-1) */ + (z_const char *)"stream error", /* Z_STREAM_ERROR (-2) */ + (z_const char *)"data error", /* Z_DATA_ERROR (-3) */ + (z_const char *)"insufficient memory", /* Z_MEM_ERROR (-4) */ + (z_const char *)"buffer error", /* Z_BUF_ERROR (-5) */ + (z_const char *)"incompatible version",/* Z_VERSION_ERROR (-6) */ + (z_const char *)"" +}; const char * ZEXPORT zlibVersion() @@ -64,7 +58,7 @@ uLong ZEXPORT zlibCompileFlags() case 8: flags += 2 << 6; break; default: flags += 3 << 6; } -#ifdef DEBUG +#ifdef ZLIB_DEBUG flags += 1 << 8; #endif #if defined(ASMV) || defined(ASMINF) @@ -118,8 +112,8 @@ uLong ZEXPORT zlibCompileFlags() return flags; } -#ifdef DEBUG - +#ifdef ZLIB_DEBUG +#include <stdlib.h> # ifndef verbose # define verbose 0 # endif @@ -222,9 +216,11 @@ local ptr_table table[MAX_PTR]; voidpf ZLIB_INTERNAL zcalloc (voidpf opaque, unsigned items, unsigned size) { - voidpf buf = opaque; /* just to make some compilers happy */ + voidpf buf; ulg bsize = (ulg)items*size; + (void)opaque; + /* If we allocate less than 65520 bytes, we assume that farmalloc * will return a usable pointer which doesn't have to be normalized. */ @@ -247,6 +243,9 @@ voidpf ZLIB_INTERNAL zcalloc (voidpf opaque, unsigned items, unsigned size) void ZLIB_INTERNAL zcfree (voidpf opaque, voidpf ptr) { int n; + + (void)opaque; + if (*(ush*)&ptr != 0) { /* object < 64K */ farfree(ptr); return; @@ -262,7 +261,6 @@ void ZLIB_INTERNAL zcfree (voidpf opaque, voidpf ptr) next_ptr--; return; } - ptr = opaque; /* just to make some compilers happy */ Assert(0, "zcfree: ptr not found"); } @@ -281,13 +279,13 @@ void ZLIB_INTERNAL zcfree (voidpf opaque, voidpf ptr) voidpf ZLIB_INTERNAL zcalloc (voidpf opaque, uInt items, uInt size) { - if (opaque) opaque = 0; /* to make compiler happy */ + (void)opaque; return _halloc((long)items, size); } void ZLIB_INTERNAL zcfree (voidpf opaque, voidpf ptr) { - if (opaque) opaque = 0; /* to make compiler happy */ + (void)opaque; _hfree(ptr); } @@ -309,7 +307,7 @@ voidpf ZLIB_INTERNAL zcalloc (opaque, items, size) unsigned items; unsigned size; { - if (opaque) items += size - size; /* make compiler happy */ + (void)opaque; return sizeof(uInt) > 2 ? (voidpf)malloc(items * size) : (voidpf)calloc(items, size); } @@ -318,8 +316,8 @@ void ZLIB_INTERNAL zcfree (opaque, ptr) voidpf opaque; voidpf ptr; { + (void)opaque; free(ptr); - if (opaque) return; /* make compiler happy */ } #endif /* MY_ZCALLOC */ diff --git a/erts/emulator/zlib/zutil.h b/erts/emulator/zlib/zutil.h index 24ab06b1cf..b079ea6a80 100644 --- a/erts/emulator/zlib/zutil.h +++ b/erts/emulator/zlib/zutil.h @@ -1,5 +1,5 @@ /* zutil.h -- internal interface and configuration of the compression library - * Copyright (C) 1995-2013 Jean-loup Gailly. + * Copyright (C) 1995-2016 Jean-loup Gailly, Mark Adler * For conditions of distribution and use, see copyright notice in zlib.h */ @@ -36,7 +36,9 @@ #ifndef local # define local static #endif -/* compile with -Dlocal if your debugger can't find static symbols */ +/* since "static" is used to mean two completely different things in C, we + define "local" for the non-static meaning of "static", for readability + (compile with -Dlocal if your debugger can't find static symbols) */ typedef unsigned char uch; typedef uch FAR uchf; @@ -98,28 +100,38 @@ extern z_const char * const z_errmsg[10]; /* indexed by 2-zlib_error */ #endif #ifdef AMIGA -# define OS_CODE 0x01 +# define OS_CODE 1 #endif #if defined(VAXC) || defined(VMS) -# define OS_CODE 0x02 +# define OS_CODE 2 # define F_OPEN(name, mode) \ fopen((name), (mode), "mbc=60", "ctx=stm", "rfm=fix", "mrs=512") #endif +#ifdef __370__ +# if __TARGET_LIB__ < 0x20000000 +# define OS_CODE 4 +# elif __TARGET_LIB__ < 0x40000000 +# define OS_CODE 11 +# else +# define OS_CODE 8 +# endif +#endif + #if defined(ATARI) || defined(atarist) -# define OS_CODE 0x05 +# define OS_CODE 5 #endif #ifdef OS2 -# define OS_CODE 0x06 +# define OS_CODE 6 # if defined(M_I86) && !defined(Z_SOLO) # include <malloc.h> # endif #endif #if defined(MACOS) || defined(TARGET_OS_MAC) -# define OS_CODE 0x07 +# define OS_CODE 7 # ifndef Z_SOLO # if defined(__MWERKS__) && __dest_os != __be_os && __dest_os != __win32_os # include <unix.h> /* for fdopen */ @@ -131,18 +143,24 @@ extern z_const char * const z_errmsg[10]; /* indexed by 2-zlib_error */ # endif #endif -#ifdef TOPS20 -# define OS_CODE 0x0a +#ifdef __acorn +# define OS_CODE 13 #endif -#ifdef WIN32 -# ifndef __CYGWIN__ /* Cygwin is Unix, not Win32 */ -# define OS_CODE 0x0b -# endif +#if defined(WIN32) && !defined(__CYGWIN__) +# define OS_CODE 10 +#endif + +#ifdef _BEOS_ +# define OS_CODE 16 +#endif + +#ifdef __TOS_OS400__ +# define OS_CODE 18 #endif -#ifdef __50SERIES /* Prime/PRIMOS */ -# define OS_CODE 0x0f +#ifdef __APPLE__ +# define OS_CODE 19 #endif #if defined(_BEOS_) || defined(RISCOS) @@ -177,7 +195,7 @@ extern z_const char * const z_errmsg[10]; /* indexed by 2-zlib_error */ /* common defaults */ #ifndef OS_CODE -# define OS_CODE 0x03 /* assume Unix */ +# define OS_CODE 3 /* assume Unix */ #endif #ifndef F_OPEN @@ -216,7 +234,7 @@ extern z_const char * const z_errmsg[10]; /* indexed by 2-zlib_error */ #endif /* Diagnostic functions */ -#ifdef DEBUG +#ifdef ZLIB_DEBUG # include <stdio.h> extern int ZLIB_INTERNAL z_verbose; extern void ZLIB_INTERNAL z_error OF((char *m)); diff --git a/erts/epmd/src/epmd.c b/erts/epmd/src/epmd.c index 44e997e609..8313678b5d 100644 --- a/erts/epmd/src/epmd.c +++ b/erts/epmd/src/epmd.c @@ -437,6 +437,11 @@ static void usage(EpmdVars *g) fprintf(stderr, " epmd -kill even if there " "are registered nodes.\n"); fprintf(stderr, " Also allows forced unregister (epmd -stop).\n"); +#ifdef HAVE_SYSTEMD_DAEMON + fprintf(stderr, " -systemd\n"); + fprintf(stderr, " Wait for socket from systemd. The option makes sense\n"); + fprintf(stderr, " when started from .socket unit.\n"); +#endif /* HAVE_SYSTEMD_DAEMON */ fprintf(stderr, "\nDbgExtra options\n"); fprintf(stderr, " -packet_timeout Seconds\n"); fprintf(stderr, " Set the number of seconds a connection can be\n"); @@ -462,11 +467,6 @@ static void usage(EpmdVars *g) fprintf(stderr, " Forcibly unregisters a name with epmd\n"); fprintf(stderr, " (only allowed if -relaxed_command_check was given when \n"); fprintf(stderr, " epmd was started).\n"); -#ifdef HAVE_SYSTEMD_DAEMON - fprintf(stderr, " -systemd\n"); - fprintf(stderr, " Wait for socket from systemd. The option makes sense\n"); - fprintf(stderr, " when started from .socket unit.\n"); -#endif /* HAVE_SYSTEMD_DAEMON */ epmd_cleanup_exit(g,1); } diff --git a/erts/etc/common/erlexec.c b/erts/etc/common/erlexec.c index 0cb01fd4ef..ec4a4ead23 100644 --- a/erts/etc/common/erlexec.c +++ b/erts/etc/common/erlexec.c @@ -819,10 +819,8 @@ int main(int argc, char **argv) case '+': switch (argv[i][1]) { - case '#': case 'a': case 'A': - case 'b': case 'C': case 'e': case 'i': diff --git a/erts/etc/unix/run_erl.c b/erts/etc/unix/run_erl.c index 725343d701..bfb3e1bd2c 100644 --- a/erts/etc/unix/run_erl.c +++ b/erts/etc/unix/run_erl.c @@ -43,10 +43,10 @@ #endif #ifdef HAVE_WORKING_POSIX_OPENPT # ifndef _XOPEN_SOURCE - /* On OS X and BSD, we must leave _XOPEN_SOURCE undefined in order for - * the prototype of vsyslog() to be included. + /* On OS X, BSD and Solaris, we must leave _XOPEN_SOURCE undefined in order + * for the prototype of vsyslog() to be included. */ -# if !(defined(__APPLE__) || defined(__FreeBSD__) || defined(__DragonFly__)) +# if !(defined(__APPLE__) || defined(__FreeBSD__) || defined(__DragonFly__) || defined(__sun)) # define _XOPEN_SOURCE 600 # endif # endif diff --git a/erts/include/internal/gcc/ethr_dw_atomic.h b/erts/include/internal/gcc/ethr_dw_atomic.h index 47158b7295..dd116c81ce 100644 --- a/erts/include/internal/gcc/ethr_dw_atomic.h +++ b/erts/include/internal/gcc/ethr_dw_atomic.h @@ -78,7 +78,7 @@ typedef volatile ETHR_NATIVE_SU_DW_SINT_T * ethr_native_dw_ptr_t; * runtime. We, therefore, need an extra word allocated. */ #define ETHR_DW_NATMC_MEM__(VAR) \ - (&var->c[(int) ((ethr_uint_t) &(VAR)->c[0]) & ETHR_DW_NATMC_ALIGN_MASK__]) + (&(VAR)->c[(int) ((ethr_uint_t) &(VAR)->c[0]) & ETHR_DW_NATMC_ALIGN_MASK__]) typedef union { volatile ETHR_NATIVE_SU_DW_SINT_T dw_sint; volatile ethr_sint_t sint[3]; diff --git a/erts/include/internal/i386/ethr_dw_atomic.h b/erts/include/internal/i386/ethr_dw_atomic.h index 91acdb0483..3c47da9758 100644 --- a/erts/include/internal/i386/ethr_dw_atomic.h +++ b/erts/include/internal/i386/ethr_dw_atomic.h @@ -73,7 +73,7 @@ typedef volatile ethr_native_sint128_t__ * ethr_native_dw_ptr_t; * runtime. We, therefore, need an extra word allocated. */ #define ETHR_DW_NATMC_MEM__(VAR) \ - (&var->c[(int) ((ethr_uint_t) &(VAR)->c[0]) & ETHR_DW_NATMC_ALIGN_MASK__]) + (&(VAR)->c[(int) ((ethr_uint_t) &(VAR)->c[0]) & ETHR_DW_NATMC_ALIGN_MASK__]) typedef union { #ifdef ETHR_NATIVE_SU_DW_SINT_T volatile ETHR_NATIVE_SU_DW_SINT_T dw_sint; diff --git a/erts/include/internal/win/ethr_dw_atomic.h b/erts/include/internal/win/ethr_dw_atomic.h index a6b26ab7bb..f7a1900a82 100644 --- a/erts/include/internal/win/ethr_dw_atomic.h +++ b/erts/include/internal/win/ethr_dw_atomic.h @@ -80,7 +80,7 @@ typedef volatile __int64 * ethr_native_dw_ptr_t; * runtime. We, therefore, need an extra word allocated. */ #define ETHR_DW_NATMC_MEM__(VAR) \ - (&var->c[(int) ((ethr_uint_t) &(VAR)->c[0]) & ETHR_DW_NATMC_ALIGN_MASK__]) + (&(VAR)->c[(int) ((ethr_uint_t) &(VAR)->c[0]) & ETHR_DW_NATMC_ALIGN_MASK__]) typedef union { #ifdef ETHR_NATIVE_SU_DW_SINT_T volatile ETHR_NATIVE_SU_DW_SINT_T dw_sint; diff --git a/erts/preloaded/ebin/atomics.beam b/erts/preloaded/ebin/atomics.beam Binary files differnew file mode 100644 index 0000000000..1de97fa668 --- /dev/null +++ b/erts/preloaded/ebin/atomics.beam diff --git a/erts/preloaded/ebin/counters.beam b/erts/preloaded/ebin/counters.beam Binary files differnew file mode 100644 index 0000000000..caaa6167e1 --- /dev/null +++ b/erts/preloaded/ebin/counters.beam diff --git a/erts/preloaded/ebin/erl_init.beam b/erts/preloaded/ebin/erl_init.beam Binary files differnew file mode 100644 index 0000000000..93f23377b5 --- /dev/null +++ b/erts/preloaded/ebin/erl_init.beam diff --git a/erts/preloaded/ebin/erl_prim_loader.beam b/erts/preloaded/ebin/erl_prim_loader.beam Binary files differindex 0f5f5036f0..42629bbe8e 100644 --- a/erts/preloaded/ebin/erl_prim_loader.beam +++ b/erts/preloaded/ebin/erl_prim_loader.beam diff --git a/erts/preloaded/ebin/erl_tracer.beam b/erts/preloaded/ebin/erl_tracer.beam Binary files differindex 6017112dac..a74e87d19c 100644 --- a/erts/preloaded/ebin/erl_tracer.beam +++ b/erts/preloaded/ebin/erl_tracer.beam diff --git a/erts/preloaded/ebin/erlang.beam b/erts/preloaded/ebin/erlang.beam Binary files differindex 7a4167b0e9..54557d9d18 100644 --- a/erts/preloaded/ebin/erlang.beam +++ b/erts/preloaded/ebin/erlang.beam diff --git a/erts/preloaded/ebin/erts_code_purger.beam b/erts/preloaded/ebin/erts_code_purger.beam Binary files differindex c899b69a2c..3cbfd94a43 100644 --- a/erts/preloaded/ebin/erts_code_purger.beam +++ b/erts/preloaded/ebin/erts_code_purger.beam diff --git a/erts/preloaded/ebin/erts_dirty_process_signal_handler.beam b/erts/preloaded/ebin/erts_dirty_process_signal_handler.beam Binary files differindex 9490a56758..43604e0480 100644 --- a/erts/preloaded/ebin/erts_dirty_process_signal_handler.beam +++ b/erts/preloaded/ebin/erts_dirty_process_signal_handler.beam diff --git a/erts/preloaded/ebin/erts_internal.beam b/erts/preloaded/ebin/erts_internal.beam Binary files differindex 15c59de80a..2b98e87843 100644 --- a/erts/preloaded/ebin/erts_internal.beam +++ b/erts/preloaded/ebin/erts_internal.beam diff --git a/erts/preloaded/ebin/erts_literal_area_collector.beam b/erts/preloaded/ebin/erts_literal_area_collector.beam Binary files differindex e650a6b5af..1f7fbef85a 100644 --- a/erts/preloaded/ebin/erts_literal_area_collector.beam +++ b/erts/preloaded/ebin/erts_literal_area_collector.beam diff --git a/erts/preloaded/ebin/init.beam b/erts/preloaded/ebin/init.beam Binary files differindex 858a9dc63e..bcbf4a1469 100644 --- a/erts/preloaded/ebin/init.beam +++ b/erts/preloaded/ebin/init.beam diff --git a/erts/preloaded/ebin/persistent_term.beam b/erts/preloaded/ebin/persistent_term.beam Binary files differnew file mode 100644 index 0000000000..79ef03b9a6 --- /dev/null +++ b/erts/preloaded/ebin/persistent_term.beam diff --git a/erts/preloaded/ebin/prim_buffer.beam b/erts/preloaded/ebin/prim_buffer.beam Binary files differindex 4ad1380d0b..70675f47ee 100644 --- a/erts/preloaded/ebin/prim_buffer.beam +++ b/erts/preloaded/ebin/prim_buffer.beam diff --git a/erts/preloaded/ebin/prim_eval.beam b/erts/preloaded/ebin/prim_eval.beam Binary files differindex 2ae18846bf..5aa5a7f384 100644 --- a/erts/preloaded/ebin/prim_eval.beam +++ b/erts/preloaded/ebin/prim_eval.beam diff --git a/erts/preloaded/ebin/prim_file.beam b/erts/preloaded/ebin/prim_file.beam Binary files differindex df611f2bb0..37a1a563a9 100644 --- a/erts/preloaded/ebin/prim_file.beam +++ b/erts/preloaded/ebin/prim_file.beam diff --git a/erts/preloaded/ebin/prim_inet.beam b/erts/preloaded/ebin/prim_inet.beam Binary files differindex 4a345f8152..23a8e8f210 100644 --- a/erts/preloaded/ebin/prim_inet.beam +++ b/erts/preloaded/ebin/prim_inet.beam diff --git a/erts/preloaded/ebin/prim_zip.beam b/erts/preloaded/ebin/prim_zip.beam Binary files differindex 4923cadbdc..8483d11aed 100644 --- a/erts/preloaded/ebin/prim_zip.beam +++ b/erts/preloaded/ebin/prim_zip.beam diff --git a/erts/preloaded/ebin/zlib.beam b/erts/preloaded/ebin/zlib.beam Binary files differindex 07e7e97814..f3737a524e 100644 --- a/erts/preloaded/ebin/zlib.beam +++ b/erts/preloaded/ebin/zlib.beam diff --git a/erts/preloaded/src/Makefile b/erts/preloaded/src/Makefile index 4333f6643a..c655bff72e 100644 --- a/erts/preloaded/src/Makefile +++ b/erts/preloaded/src/Makefile @@ -41,13 +41,16 @@ PRE_LOADED_ERL_MODULES = \ prim_inet \ zlib \ prim_zip \ - otp_ring0 \ + erl_init \ erts_code_purger \ erlang \ erts_internal \ erl_tracer \ erts_literal_area_collector \ - erts_dirty_process_signal_handler + erts_dirty_process_signal_handler \ + atomics \ + counters \ + persistent_term PRE_LOADED_BEAM_MODULES = \ prim_eval diff --git a/erts/preloaded/src/atomics.erl b/erts/preloaded/src/atomics.erl new file mode 100644 index 0000000000..d1fe5e65cf --- /dev/null +++ b/erts/preloaded/src/atomics.erl @@ -0,0 +1,119 @@ +%% +%% %CopyrightBegin% +%% +%% Copyright Ericsson AB 2018. All Rights Reserved. +%% +%% Licensed under the Apache License, Version 2.0 (the "License"); +%% you may not use this file except in compliance with the License. +%% You may obtain a copy of the License at +%% +%% http://www.apache.org/licenses/LICENSE-2.0 +%% +%% Unless required by applicable law or agreed to in writing, software +%% distributed under the License is distributed on an "AS IS" BASIS, +%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +%% See the License for the specific language governing permissions and +%% limitations under the License. +%% +%% %CopyrightEnd% +%% + +%% Purpose : Main atomics API module. + +-module(atomics). + +-export([new/2, + put/3, get/2, + add/3, add_get/3, + sub/3, sub_get/3, + exchange/3, compare_exchange/4, + info/1]). + +-export_type([atomics_ref/0]). + +-opaque atomics_ref() :: reference(). + +-define(OPT_SIGNED, (1 bsl 0)). +-define(OPT_DEFAULT, ?OPT_SIGNED). + +-spec new(Arity, Opts) -> atomics_ref() when + Arity :: pos_integer(), + Opts :: [Opt], + Opt :: {signed, boolean()}. +new(Arity, Opts) -> + erts_internal:atomics_new(Arity, encode_opts(Opts, ?OPT_DEFAULT)). + +encode_opts([{signed, true}|T], Acc) -> + encode_opts(T, Acc bor ?OPT_SIGNED); +encode_opts([{signed, false}|T], Acc) -> + encode_opts(T, Acc band (bnot ?OPT_SIGNED)); +encode_opts([], Acc) -> + Acc; +encode_opts(_, _) -> + erlang:error(badarg). + +-spec put(Ref, Ix, Value) -> ok when + Ref :: atomics_ref(), + Ix :: integer(), + Value :: integer(). +put(_Ref, _Ix, _Value) -> + erlang:nif_error(undef). + +-spec get(Ref, Ix) -> integer() when + Ref :: atomics_ref(), + Ix :: integer(). +get(_Ref, _Ix) -> + erlang:nif_error(undef). + +-spec add(Ref, Ix, Incr) -> ok when + Ref :: atomics_ref(), + Ix :: integer(), + Incr :: integer(). +add(_Ref, _Ix, _Incr) -> + erlang:nif_error(undef). + +-spec add_get(Ref, Ix, Incr) -> integer() when + Ref :: atomics_ref(), + Ix :: integer(), + Incr :: integer(). +add_get(_Ref, _Ix, _Incr) -> + erlang:nif_error(undef). + +-spec sub(Ref, Ix, Decr) -> ok when + Ref :: atomics_ref(), + Ix :: integer(), + Decr :: integer(). +sub(Ref, Ix, Decr) -> + ?MODULE:add(Ref, Ix, -Decr). + +-spec sub_get(Ref, Ix, Decr) -> integer() when + Ref :: atomics_ref(), + Ix :: integer(), + Decr :: integer(). +sub_get(Ref, Ix, Decr) -> + ?MODULE:add_get(Ref, Ix, -Decr). + +-spec exchange(Ref, Ix, Desired) -> integer() when + Ref :: atomics_ref(), + Ix :: integer(), + Desired :: integer(). +exchange(_Ref, _Ix, _Desired) -> + erlang:nif_error(undef). + +-spec compare_exchange(Ref, Ix, Expected, Desired) -> ok | integer() when + Ref :: atomics_ref(), + Ix :: integer(), + Expected :: integer(), + Desired :: integer(). +compare_exchange(_Ref, _Ix, _Expected, _Desired) -> + erlang:nif_error(undef). + +-spec info(Ref) -> Info when + Ref :: atomics_ref(), + Info :: #{'size':=Size,'max':=Max,'min':=Min,'memory':=Memory}, + Size :: non_neg_integer(), + Max :: integer(), + Min :: integer(), + Memory :: non_neg_integer(). +info(_Ref) -> + erlang:nif_error(undef). diff --git a/erts/preloaded/src/counters.erl b/erts/preloaded/src/counters.erl new file mode 100644 index 0000000000..67354f648d --- /dev/null +++ b/erts/preloaded/src/counters.erl @@ -0,0 +1,90 @@ +%% +%% %CopyrightBegin% +%% +%% Copyright Ericsson AB 2018. All Rights Reserved. +%% +%% Licensed under the Apache License, Version 2.0 (the "License"); +%% you may not use this file except in compliance with the License. +%% You may obtain a copy of the License at +%% +%% http://www.apache.org/licenses/LICENSE-2.0 +%% +%% Unless required by applicable law or agreed to in writing, software +%% distributed under the License is distributed on an "AS IS" BASIS, +%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +%% See the License for the specific language governing permissions and +%% limitations under the License. +%% +%% %CopyrightEnd% +%% + +%% Purpose : Main atomics API module. + +-module(counters). + +-export([new/2, + get/2, + add/3, + sub/3, + info/1]). + +-export_type([counters_ref/0]). + +-opaque counters_ref() :: {atomics, reference()} | {write_concurrency, reference()}. + +-spec new(Size, Opts) -> counters_ref() when + Size :: pos_integer(), + Opts :: [Opt], + Opt :: atomics | write_concurrency. +new(Size, [atomics]) -> + {atomics, atomics:new(Size, [{signed, true}])}; +new(Size, [write_concurrency]) -> + {write_concurrency, erts_internal:counters_new(Size)}; +new(Size, []) -> + new(Size, [atomics]); +new(_, _) -> + erlang:error(badarg). + +-spec get(Ref, Ix) -> integer() when + Ref :: counters_ref(), + Ix :: integer(). +get({atomics,Ref}, Ix) -> + atomics:get(Ref, Ix); +get({write_concurrency, Ref}, Ix) -> + erts_internal:counters_get(Ref, Ix); +get(_, _) -> + erlang:error(badarg). + + + +-spec add(Ref, Ix, Incr) -> ok when + Ref :: counters_ref(), + Ix :: integer(), + Incr :: integer(). +add({atomics, Ref}, Ix, Incr) -> + atomics:add(Ref, Ix, Incr); +add({write_concurrency, Ref}, Ix, Incr) -> + erts_internal:counters_add(Ref, Ix, Incr); +add(_, _, _) -> + erlang:error(badarg). + + +-spec sub(Ref, Ix, Decr) -> ok when + Ref :: counters_ref(), + Ix :: integer(), + Decr :: integer(). +sub(Ref, Ix, Decr) -> + add(Ref, Ix, -Decr). + +-spec info(Ref) -> Info when + Ref :: counters_ref(), + Info :: #{'size':=Size, 'memory':=Memory}, + Size :: non_neg_integer(), + Memory :: non_neg_integer(). +info({atomics, Ref}) -> + atomics:info(Ref); +info({write_concurrency, Ref}) -> + erts_internal:counters_info(Ref); +info(_) -> + erlang:error(badarg). + diff --git a/erts/preloaded/src/otp_ring0.erl b/erts/preloaded/src/erl_init.erl index 62a60fffe2..d681f05398 100644 --- a/erts/preloaded/src/otp_ring0.erl +++ b/erts/preloaded/src/erl_init.erl @@ -17,15 +17,26 @@ %% %% %CopyrightEnd% %% --module(otp_ring0). +-module(erl_init). -%% Purpose : Start up of erlang system. +%% Initial process of an Erlang system. -export([start/2]). --spec start(_, term()) -> term(). -start(_Env, Argv) -> - run(init, boot, Argv). +%% This gets the module name given by the +i option (default 'init') +%% and the list of command line arguments + +-spec start(Mod, BootArgs) -> no_return() when + Mod :: module(), + BootArgs :: [binary()]. +start(Mod, BootArgs) -> + %% Load the static nifs + zlib:on_load(), + erl_tracer:on_load(), + prim_buffer:on_load(), + prim_file:on_load(), + %% Proceed to the specified boot module + run(Mod, boot, BootArgs). run(M, F, A) -> case erlang:function_exported(M, F, 1) of diff --git a/erts/preloaded/src/erts.app.src b/erts/preloaded/src/erts.app.src index 8c34c99a98..13fbbc7ff7 100644 --- a/erts/preloaded/src/erts.app.src +++ b/erts/preloaded/src/erts.app.src @@ -26,19 +26,21 @@ erl_prim_loader, erts_internal, init, - otp_ring0, + erl_init, erts_code_purger, prim_buffer, prim_eval, prim_file, prim_inet, prim_zip, + atomics, + counters, zlib ]}, {registered, []}, {applications, []}, {env, []}, - {runtime_dependencies, ["stdlib-3.5", "kernel-6.1", "sasl-3.0.1"]} + {runtime_dependencies, ["stdlib-3.5", "kernel-6.1", "sasl-@OTP-13468@"]} ]}. %% vim: ft=erlang diff --git a/erts/preloaded/src/erts_internal.erl b/erts/preloaded/src/erts_internal.erl index 88f47e917b..d491a505c6 100644 --- a/erts/preloaded/src/erts_internal.erl +++ b/erts/preloaded/src/erts_internal.erl @@ -90,6 +90,13 @@ -export([create_dist_channel/4]). +-export([erase_persistent_terms/0]). + +-export([atomics_new/2]). + +-export([counters_new/1, counters_get/2, counters_add/3, + counters_info/1]). + %% %% Await result of send to port %% @@ -691,3 +698,27 @@ process_flag(_Pid, _Flag, _Value) -> create_dist_channel(_Node, _DistCtrlr, _Flags, _Ver) -> erlang:nif_error(undefined). + +-spec erase_persistent_terms() -> 'ok'. +erase_persistent_terms() -> + erlang:nif_error(undefined). + +-spec atomics_new(pos_integer(), pos_integer()) -> reference(). +atomics_new(_Arity, _EncOpts) -> + erlang:nif_error(undef). + +-spec counters_new(pos_integer()) -> reference(). +counters_new(_Size) -> + erlang:nif_error(undef). + +-spec counters_get(reference(), pos_integer()) -> integer(). +counters_get(_Ref, _Ix) -> + erlang:nif_error(undef). + +-spec counters_add(reference(), pos_integer(), integer()) -> ok. +counters_add(_Ref, _Ix, _Incr) -> + erlang:nif_error(undef). + +-spec counters_info(reference()) -> #{}. +counters_info(_Ref) -> + erlang:nif_error(undef). diff --git a/erts/preloaded/src/init.erl b/erts/preloaded/src/init.erl index 253fcf7a1f..d8f5c9a945 100644 --- a/erts/preloaded/src/init.erl +++ b/erts/preloaded/src/init.erl @@ -200,12 +200,6 @@ boot(BootArgs) -> register(init, self()), process_flag(trap_exit, true), - %% Load the static nifs - zlib:on_load(), - erl_tracer:on_load(), - prim_buffer:on_load(), - prim_file:on_load(), - {Start0,Flags,Args} = parse_boot_args(BootArgs), %% We don't get to profile parsing of BootArgs case b2a(get_flag(profile_boot, Flags, false)) of @@ -552,6 +546,7 @@ stop(Reason,State) -> do_stop(restart,#state{start = Start, flags = Flags, args = Args}) -> %% Make sure we don't have any outstanding messages before doing the restart. flush(), + erts_internal:erase_persistent_terms(), boot(Start,Flags,Args); do_stop(reboot,_) -> halt(); diff --git a/erts/preloaded/src/persistent_term.erl b/erts/preloaded/src/persistent_term.erl new file mode 100644 index 0000000000..5d0c266127 --- /dev/null +++ b/erts/preloaded/src/persistent_term.erl @@ -0,0 +1,55 @@ +%% +%% %CopyrightBegin% +%% +%% Copyright Ericsson AB 2018. All Rights Reserved. +%% +%% Licensed under the Apache License, Version 2.0 (the "License"); +%% you may not use this file except in compliance with the License. +%% You may obtain a copy of the License at +%% +%% http://www.apache.org/licenses/LICENSE-2.0 +%% +%% Unless required by applicable law or agreed to in writing, software +%% distributed under the License is distributed on an "AS IS" BASIS, +%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +%% See the License for the specific language governing permissions and +%% limitations under the License. +%% +%% %CopyrightEnd% +%% +-module(persistent_term). + +-export([erase/1,get/0,get/1,info/0,put/2]). + +-type key() :: term(). +-type value() :: term(). + +-spec erase(Key) -> Result when + Key :: key(), + Result :: boolean(). +erase(_Key) -> + erlang:nif_error(undef). + +-spec get() -> List when + List :: [{key(),value()}]. +get() -> + erlang:nif_error(undef). + +-spec get(Key) -> Value when + Key :: key(), + Value :: value(). +get(_Key) -> + erlang:nif_error(undef). + +-spec info() -> Info when + Info :: #{'count':=Count,'memory':=Memory}, + Count :: non_neg_integer(), + Memory :: non_neg_integer(). +info() -> + erlang:nif_error(undef). + +-spec put(Key, Value) -> 'ok' when + Key :: key(), + Value :: value(). +put(_Key, _Value) -> + erlang:nif_error(undef). diff --git a/erts/preloaded/src/prim_file.erl b/erts/preloaded/src/prim_file.erl index 6d85868183..5fc22bc582 100644 --- a/erts/preloaded/src/prim_file.erl +++ b/erts/preloaded/src/prim_file.erl @@ -83,6 +83,15 @@ internal_normalize_utf8(_) -> is_translatable(_) -> erlang:nif_error(undefined). +%% This is a janitor process used to close files whose controlling process has +%% died. The emulator will be torn down if this is killed. +delayed_close_loop() -> + receive + {close, FRef} when is_reference(FRef) -> delayed_close_nif(FRef); + _ -> ok + end, + delayed_close_loop(). + %% %% Returns {error, Reason} | {ok, BytesCopied} @@ -95,7 +104,12 @@ copy(#file_descriptor{module = ?MODULE} = Source, file:copy_opened(Source, Dest, Length). on_load() -> - ok = erlang:load_nif(atom_to_list(?MODULE), 0). + Pid = spawn(fun() -> + process_flag(trap_exit, true), + delayed_close_loop() + end), + true = register(erts_prim_file, Pid), + ok = erlang:load_nif(atom_to_list(?MODULE), Pid). open(Name, Modes) -> %% The try/catch pattern seen here is used throughout the file to adhere to @@ -482,6 +496,8 @@ truncate_nif(_FileRef) -> erlang:nif_error(undef). get_handle_nif(_FileRef) -> erlang:nif_error(undef). +delayed_close_nif(_FileRef) -> + erlang:nif_error(undef). %% %% Quality-of-life helpers diff --git a/erts/preloaded/src/prim_inet.erl b/erts/preloaded/src/prim_inet.erl index b746aab049..cc2711b540 100644 --- a/erts/preloaded/src/prim_inet.erl +++ b/erts/preloaded/src/prim_inet.erl @@ -49,9 +49,15 @@ -include("inet_sctp.hrl"). -include("inet_int.hrl"). -%-define(DEBUG, 1). +%%%-define(DEBUG, 1). -ifdef(DEBUG). --define(DBG_FORMAT(Format, Args), (io:format((Format), (Args)))). +-define( + DBG_FORMAT(Format, Args), + begin + %% io:format((Format), (Args)), + erlang:display(lists:flatten(io_lib:format((Format), (Args)))), + ok + end). -else. -define(DBG_FORMAT(Format, Args), ok). -endif. @@ -150,39 +156,96 @@ shutdown_1(S, How) -> %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% close(S) when is_port(S) -> + ?DBG_FORMAT("prim_inet:close(~p)~n", [S]), case getopt(S, linger) of {ok,{true,0}} -> close_port(S); - _ -> - case subscribe(S, [subs_empty_out_q]) of - {ok, [{subs_empty_out_q,N}]} when N > 0 -> - close_pend_loop(S, N); %% wait for pending output to be sent - _ -> - close_port(S) - end + {ok,{true,T}} -> + %% Wait for T seconds for pending output to be sent + %% + %% Note that this handling of Linger may look ok, + %% but sweeps some problems under the rug since + %% there are OS buffers that may have remaining data + %% after the inet driver has emptied its buffers. + %% But Linger for nonblocking sockets is broken + %% anyway on all OS:es, according to hearsay, + %% and is a contradiction in itself. + %% We have hereby done our best... + %% + Tref = erlang:start_timer(T * 1000, self(), close_port), + close_pend_loop(S, Tref, undefined); + _ -> % Regard this as {ok,{false,_}} + case subscribe(S, [subs_empty_out_q]) of + {ok, [{subs_empty_out_q,N}]} when N > 0 -> + %% Wait for pending output to be sent + DefaultT = 180000, % Arbitrary system timeout 3 min + Tref = erlang:start_timer(DefaultT, self(), close_port), + close_pend_loop(S, Tref, N); + _ -> + %% Subscribe failed or empty out q - give up or done + close_port(S) + end end. -close_pend_loop(S, N) -> +close_pend_loop(S, Tref, N) -> + ?DBG_FORMAT("prim_inet:close_pend_loop(~p, _, ~p)~n", [S,N]), receive - {empty_out_q,S} -> - close_port(S) + {timeout,Tref,_} -> % Linger timeout + ?DBG_FORMAT("prim_inet:close_pend_loop(~p, _, _) timeout~n", [S]), + close_port(S); + {empty_out_q,S} when N =/= undefined -> + ?DBG_FORMAT( + "prim_inet:close_pend_loop(~p, _, _) empty_out_q~n", [S]), + close_port(S, Tref) after ?INET_CLOSE_TIMEOUT -> case getstat(S, [send_pend]) of {ok, [{send_pend,N1}]} -> + ?DBG_FORMAT( + "prim_inet:close_pend_loop(~p, _, _) send_pend ~p~n", + [S,N1]), if - N1 =:= N -> - close_port(S); - true -> - close_pend_loop(S, N1) + N1 =:= 0 -> + %% Empty outq - done + close_port(S, Tref); + N =:= undefined -> + %% Within linger time - wait some more + close_pend_loop(S, Tref, N); + N1 =:= N -> + %% Inactivity - give up + close_port(S, Tref); + true -> + %% Still moving - wait some more + close_pend_loop(S, Tref, N) end; - _ -> - close_port(S) - end + _Stat -> + %% Failed getstat - give up + ?DBG_FORMAT( + "prim_inet:close_pend_loop(~p, _, _) getstat ~p~n", + [S,_Stat]), + close_port(S, Tref) + end end. + +close_port(S, Tref) -> + ?DBG_FORMAT("prim_inet:close_port(~p, _)~n", [S]), + case erlang:cancel_timer(Tref) of + false -> + receive + {timeout,Tref,_} -> + ok + end; + _N -> + ok + end, + close_port(S). +%% close_port(S) -> - catch erlang:port_close(S), - receive {'EXIT',S,_} -> ok after 0 -> ok end. + ?DBG_FORMAT("prim_inet:close_port(~p)~n", [S]), + _Closed = (catch erlang:port_close(S)), + receive {'EXIT',S,_} -> ok after 0 -> ok end, + ?DBG_FORMAT("prim_inet:close_port(~p) ~p~n", [S,_Closed]), + ok. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% @@ -424,23 +487,49 @@ peeloff(S, AssocId) -> %% be called directly -- use "sendmsg" instead: %% send(S, Data, OptList) when is_port(S), is_list(OptList) -> - ?DBG_FORMAT("prim_inet:send(~p, ~p)~n", [S,Data]), + ?DBG_FORMAT("prim_inet:send(~p, _, ~p)~n", [S,OptList]), try erlang:port_command(S, Data, OptList) of false -> % Port busy and nosuspend option passed ?DBG_FORMAT("prim_inet:send() -> {error,busy}~n", []), {error,busy}; true -> - receive - {inet_reply,S,Status} -> - ?DBG_FORMAT("prim_inet:send() -> ~p~n", [Status]), - Status - end + send_recv_reply(S, undefined) catch error:_Error -> ?DBG_FORMAT("prim_inet:send() -> {error,einval}~n", []), {error,einval} end. +send_recv_reply(S, Mref) -> + ReplyTimeout = + case Mref of + undefined -> + ?INET_CLOSE_TIMEOUT; + _ -> + infinity + end, + receive + {inet_reply,S,Status} -> + ?DBG_FORMAT( + "prim_inet:send_recv_reply(~p, _): inet_reply ~p~n", + [S,Status]), + case Mref of + undefined -> ok; + _ -> + demonitor(Mref, [flush]), + ok + end, + Status; + {'DOWN',Mref,_,_,_Reason} when Mref =/= undefined -> + ?DBG_FORMAT( + "prim_inet:send_recv_reply(~p, _) 'DOWN' ~p~n", + [S,_Reason]), + {error,closed} + after ReplyTimeout -> + send_recv_reply(S, monitor(port, S)) + end. + + send(S, Data) -> send(S, Data, []). @@ -520,13 +609,35 @@ sendfile(S, FileHandle, Offset, Length) sendfile(S, FileHandle, Offset, Length) -> case erlang:port_info(S, connected) of {connected, Pid} when Pid =:= self() -> - sendfile_1(S, FileHandle, Offset, Length); + Uncork = sendfile_maybe_cork(S), + Result = sendfile_1(S, FileHandle, Offset, Length), + sendfile_maybe_uncork(S, Uncork), + Result; {connected, Pid} when Pid =/= self() -> {error, not_owner}; _Other -> {error, einval} end. +sendfile_maybe_cork(S) -> + case getprotocol(S) of + tcp -> + case getopts(S, [nopush]) of + {ok, [{nopush,false}]} -> + _ = setopts(S, [{nopush,true}]), + true; + _ -> + false + end; + _ -> false + end. + +sendfile_maybe_uncork(S, true) -> + _ = setopts(S, [{nopush,false}]), + ok; +sendfile_maybe_uncork(_, false) -> + ok. + sendfile_1(S, FileHandle, Offset, 0) -> sendfile_1(S, FileHandle, Offset, (1 bsl 63) - 1); sendfile_1(_S, _FileHandle, Offset, Length) when @@ -870,9 +981,9 @@ chgopts(S, Opts) when is_port(S), is_list(Opts) -> getifaddrs(S) when is_port(S) -> case ctl_cmd(S, ?INET_REQ_GETIFADDRS, []) of - {ok, Data} -> - {ok, comp_ifaddrs(build_ifaddrs(Data), ktree_empty())}; - {error,enotsup} -> + {ok, Data} -> + {ok, comp_ifaddrs(build_ifaddrs(Data))}; + {error,enotsup} -> case getiflist(S) of {ok, IFs} -> {ok, getifaddrs_ifget(S, IFs)}; @@ -881,30 +992,75 @@ getifaddrs(S) when is_port(S) -> Err2 -> Err2 end. -%% Restructure interface properties per interface and remove duplicates - -comp_ifaddrs([{If,Opts}|IfOpts], T) -> - case ktree_is_defined(If, T) of - true -> - OptSet = comp_ifaddrs_add(ktree_get(If, T), Opts), - comp_ifaddrs(IfOpts, ktree_update(If, OptSet, T)); - false -> - OptSet = comp_ifaddrs_add(ktree_empty(), Opts), - comp_ifaddrs(IfOpts, ktree_insert(If, OptSet, T)) - end; -comp_ifaddrs([], T) -> - [{If,ktree_keys(ktree_get(If, T))} || If <- ktree_keys(T)]. - -comp_ifaddrs_add(OptSet, [Opt|Opts]) -> - case ktree_is_defined(Opt, OptSet) of - true - when element(1, Opt) =:= flags; - element(1, Opt) =:= hwaddr -> - comp_ifaddrs_add(OptSet, Opts); - _ -> - comp_ifaddrs_add(ktree_insert(Opt, undefined, OptSet), Opts) +%% Restructure interface properties per interface + +comp_ifaddrs(IfOpts) -> + comp_ifaddrs(IfOpts, ktree_empty()). +%% +comp_ifaddrs([{If,[{flags,Flags}|Opts]}|IfOpts], IfT) -> + case ktree_is_defined(If, IfT) of + true -> + comp_ifaddrs( + IfOpts, + ktree_update( + If, + comp_ifaddrs_flags(Flags, Opts, ktree_get(If, IfT)), + IfT)); + false -> + comp_ifaddrs( + IfOpts, + ktree_insert( + If, + comp_ifaddrs_flags(Flags, Opts, ktree_empty()), + IfT)) end; -comp_ifaddrs_add(OptSet, []) -> OptSet. +comp_ifaddrs([], IfT) -> + comp_ifaddrs_2(ktree_keys(IfT), IfT). + +comp_ifaddrs_flags(Flags, Opts, FlagsT) -> + case ktree_is_defined(Flags, FlagsT) of + true -> + ktree_update( + Flags, + rev(Opts, ktree_get(Flags, FlagsT)), + FlagsT); + false -> + ktree_insert(Flags, rev(Opts), FlagsT) + end. + +comp_ifaddrs_2([If|Ifs], IfT) -> + FlagsT = ktree_get(If, IfT), + [{If,comp_ifaddrs_3(ktree_keys(FlagsT), FlagsT)} + | comp_ifaddrs_2(Ifs, IfT)]; +comp_ifaddrs_2([], _IfT) -> + []. +%% +comp_ifaddrs_3([Flags|FlagsL], FlagsT) -> + [{flags,Flags}|hwaddr_last(rev(ktree_get(Flags, FlagsT)))] + ++ hwaddr_last(comp_ifaddrs_3(FlagsL, FlagsT)); +comp_ifaddrs_3([], _FlagsT) -> + []. + +%% Place hwaddr last to look more like legacy emulation +hwaddr_last(Opts) -> + hwaddr_last(Opts, Opts, []). +%% +hwaddr_last([{hwaddr,_} = Opt|Opts], L, R) -> + hwaddr_last(Opts, L, [Opt|R]); +hwaddr_last([_|Opts], L, R) -> + hwaddr_last(Opts, L, R); +hwaddr_last([], L, []) -> + L; +hwaddr_last([], L, R) -> + rev(hwaddr_last(L, []), rev(R)). +%% +hwaddr_last([{hwaddr,_}|Opts], R) -> + hwaddr_last(Opts, R); +hwaddr_last([Opt|Opts], R) -> + hwaddr_last(Opts, [Opt|R]); +hwaddr_last([], R) -> + R. + %% Legacy emulation of getifaddrs @@ -912,21 +1068,19 @@ getifaddrs_ifget(_, []) -> []; getifaddrs_ifget(S, [IF|IFs]) -> case ifget(S, IF, [flags]) of {ok,[{flags,Flags}]=FlagsVals} -> - BroadOpts = - case member(broadcast, Flags) of - true -> - [broadaddr,hwaddr]; - false -> - [hwaddr] - end, - P2POpts = - case member(pointtopoint, Flags) of - true -> - [dstaddr|BroadOpts]; - false -> - BroadOpts - end, - getifaddrs_ifget(S, IFs, IF, FlagsVals, [addr,netmask|P2POpts]); + GetOpts = + case member(pointtopoint, Flags) of + true -> + [dstaddr,hwaddr]; + false -> + case member(broadcast, Flags) of + true -> + [broadaddr,hwaddr]; + false -> + [hwaddr] + end + end, + getifaddrs_ifget(S, IFs, IF, FlagsVals, [addr,netmask|GetOpts]); _ -> getifaddrs_ifget(S, IFs, IF, [], [addr,netmask,hwaddr]) end. @@ -1275,6 +1429,7 @@ enc_opt(pktoptions) -> ?INET_OPT_PKTOPTIONS; enc_opt(ttl) -> ?INET_OPT_TTL; enc_opt(recvttl) -> ?INET_OPT_RECVTTL; enc_opt(nodelay) -> ?TCP_OPT_NODELAY; +enc_opt(nopush) -> ?TCP_OPT_NOPUSH; enc_opt(multicast_if) -> ?UDP_OPT_MULTICAST_IF; enc_opt(multicast_ttl) -> ?UDP_OPT_MULTICAST_TTL; enc_opt(multicast_loop) -> ?UDP_OPT_MULTICAST_LOOP; @@ -1336,6 +1491,7 @@ dec_opt(?INET_OPT_PRIORITY) -> priority; dec_opt(?INET_OPT_TOS) -> tos; dec_opt(?INET_OPT_TCLASS) -> tclass; dec_opt(?TCP_OPT_NODELAY) -> nodelay; +dec_opt(?TCP_OPT_NOPUSH) -> nopush; dec_opt(?INET_OPT_RECVTOS) -> recvtos; dec_opt(?INET_OPT_RECVTCLASS) -> recvtclass; dec_opt(?INET_OPT_PKTOPTIONS) -> pktoptions; @@ -1422,6 +1578,7 @@ type_opt_1(pktoptions) -> opts; type_opt_1(ttl) -> int; type_opt_1(recvttl) -> bool; type_opt_1(nodelay) -> bool; +type_opt_1(nopush) -> bool; type_opt_1(ipv6_v6only) -> bool; %% multicast type_opt_1(multicast_ttl) -> int; @@ -2500,7 +2657,7 @@ get_addrs([F|Addrs]) -> [Addr|get_addrs(Rest)]. get_addr(?INET_AF_LOCAL, [N|Addr]) -> - {A,Rest} = lists:split(N, Addr), + {A,Rest} = split(N, Addr), {{local,iolist_to_binary(A)},Rest}; get_addr(?INET_AF_UNSPEC, Rest) -> {{unspec,<<>>},Rest}; diff --git a/erts/vsn.mk b/erts/vsn.mk index 5b187b1f4d..01c19aff2f 100644 --- a/erts/vsn.mk +++ b/erts/vsn.mk @@ -18,7 +18,7 @@ # %CopyrightEnd% # -VSN = 10.1 +VSN = 10.1.1 # Port number 4365 in 4.2 # Port number 4366 in 4.3 |