Age | Commit message (Collapse) | Author |
|
* john/erts/inet-drv-race/OTP-15158/ERL-654:
Fix a race condition when generating async operation ids
|
|
The counter used for generating async operation ids was a plain int
shared between all ports, which was incorrect but mostly worked
fine since the ids only had to be unique on a per-port basis.
However, some compilers (notably GCC 8.1.1) generated code that
assumed that this value didn't change between reads. Using a
shortened version of enq_async_w_tmo as an example:
int id = async_ref++;
op->id = id; //A
return id; //B
In GCC 7 and earlier, `async_ref` would be read once and assigned
to `id` before being incremented, which kept the values at A and B
consistent. In GCC 8, `async_ref` was read when assigned at A and
read again at B, and then incremented, which made them inconsistent
if we raced with another port.
This commit fixes the issue by removing `async_ref` altogether and
replacing it with a per-port counter which makes it impossible to
race with someone else.
|
|
* john/erts/fix-tcp-send-timeout/OTP-14509/ERL-448:
Add a dedicated close function for TCP ports to prevent issues like ERL-430/448
Close TCP ports properly on send timeout
|
|
|
|
|
|
* john/erts/fix-port-leak/OTP-13939/ERL-193:
Add a testcase for OTP-13939/ERL-193
Mark socket disconnected on tcp_send_or_shutdown_error
# Conflicts:
# lib/kernel/test/gen_tcp_misc_SUITE.erl
|
|
The socket left lingering due to {exit_on_close, false} will accept writes
in a confusing way, returning either enotconn or blocking. This fix allows
socket to know that it has been closed recently, and new writes won't pass.
|
|
This fixes a bug where a send/shutdown error on an active-mode socket results
in the port never being properly closed.
|
|
|
|
|
|
|
|
|
|
|
|
* saleyn/uds/PR-612/OTP-13572:
AF_UNIX is more portable
|
|
Fix dialyzer warning for improper list in prim_inet
by not using an improper list.
|
|
* saleyn/uds/PR-612/OTP-13572:
Rewrite inet* for address family 'local'
Rewrite inet_drv for AF_LOCAL
Assign externally open fd to gen_tcp (UDS support)
Conflicts:
erts/preloaded/ebin/prim_inet.beam
lib/kernel/doc/src/gen_tcp.xml
lib/kernel/doc/src/gen_udp.xml
lib/kernel/src/inet6_sctp.erl
lib/kernel/test/inet_SUITE.erl
|
|
|
|
* henrik/update-copyrightyear:
update copyright-year
|
|
|
|
|
|
OTP-13251
* sverk/halt-INT_MIN:
erts: Make erlang:halt() accept bignums as Status
erts: Change erl_exit into erts_exit
kernel: Remove calls to erl_exit
|
|
|
|
This is mostly a pure refactoring.
Except for the buggy cases when calling erlang:halt() with a positive
integer in the range -(INT_MIN+2) to -INT_MIN that got confused with
ERTS_ABORT_EXIT, ERTS_DUMP_EXIT and ERTS_INTR_EXIT.
Outcome OLD erl_exit(n, ) NEW erts_exit(n, )
------- ------------------- -------------------------------------------
exit(Status) n = -Status <= 0 n = Status >= 0
crashdump+abort n > 0, ignore n n = ERTS_ERROR_EXIT < 0
The outcome of the old ERTS_ABORT_EXIT, ERTS_INTR_EXIT and
ERTS_DUMP_EXIT are the same as before (even though their values have
changed).
|
|
* maint:
Fix inet driver multi timers using new time API
|
|
* rickard/tcp-accept-tmo-bug/OTP-13254:
Fix inet driver multi timers using new time API
|
|
* maint:
efile_drv: logic error in compressed file write
|
|
|
|
When a AF_LOCAL file descriptor is created externally (e.g. Unix
Domain Socket) and passed to `gen_tcp:listen(0, [{fd, FD}])`, the
implementation incorrectly assigned the address family to be equal
to `inet`, which in the inet_drv driver translated to AF_INET instead
of AF_LOCAL (or AF_UNIX), and an `einval` error code was returned.
This patch fixes this problem such that the file descriptors of the
`local` address family are supported in the inet:fdopen/5,
gen_tcp:connect/3, gen_tcp:listen/2, gen_udp:open/2 calls
|
|
Compiling OTP 18.2.1 with gcc-5.3 shows the following warning:
drivers/common/efile_drv.c:1538:23: warning: logical not is only applied to the left hand side of comparison [-Wlogical-not-parentheses]
The code in question is:
if (! (status =
erts_gzwrite((ErtsGzFile)d->fd,
iov[i].iov_base,
iov[i].iov_len)) == iov[i].iov_len) {
d->errInfo.posix_errno =
d->errInfo.os_errno = errno; /* XXX Correct? */
break;
}
If we hoist the assignment out of the if for clarity, it becomes:
status = erts_gzwrite(..., iov[i].iov_len);
if (! status == iov[i].iov_len) { ...; break; }
iov_len is > 0 here, and status will equal iov_len if erts_gzwrite
succeeded, but will be less than iov_len if an error occurred.
"! status" is 0 or 1, which can only equal iov_len if iov_len is 1 and
erts_gzwrite detected an error and returned 0.
The effect of this mistake is that any error when iov_len >= 2 will
skip the conditional code and break statement. In particular, partial
writes (0 < status && status < iov_len) will not be flagged as errors.
All releases since OTP R8B-0 are affected.
The variable "status" is really a boolean, which is to be set to zero
on error. The fix is to set status to 1 if erts_gzwrite() returned iov_len
and 0 otherwise, and to change the condition to "if (! status) ...".
I'm also hoisting the assignment out of the condition since it obscures
the code while providing not benefit (the condition in a while or for
loop would be a different matter).
|
|
|
|
Conflicts:
OTP_VERSION
erts/doc/src/notes.xml
erts/vsn.mk
lib/kernel/doc/src/notes.xml
lib/kernel/src/kernel.appup.src
lib/kernel/vsn.mk
lib/ssl/doc/src/notes.xml
lib/ssl/src/ssl.appup.src
lib/ssl/src/ssl_cipher.erl
lib/ssl/vsn.mk
otp_versions.table
|
|
|
|
OTP-13147
* sverk/cpool_fetch-dc_list-fix:
erts: Reduce alloc_SUITE:rbtree runtime for valgrind
erts: Remove double free in efile_drv
erts: Improve alloc_SUITE:migration test
erts: Pass free mem and build type to alloc_SUITE tests
erts: Fix snprintf in alloc_SUITE for windows
erts: Workaround for strange crash on win64 in alloc_SUITE test code
erts: Refactor alloc_SUITE to use NIFs instead of drivers
erts: Add enif_getenv
erts: Make key argument constant for erl_drv_{get|put}env
erts: Add alloc_SUITE:migration
erts: Add TEST allocator
erts: Fix confusion of callbacks destroying_mbc() vs remove_mbc()
erts: Fix resurrection of carriers from dc_list
|
|
That double free is probably very seldom invoked as the port is already
gone leading to free_data being called instead of file_async_ready.
|
|
|
|
A new {line_delimiter, byte()} option allows line-oriented TCP-based
protocols to use a custom line delimiting character. It is to be
used in conjunction with {packet, line}.
This option also works with erlang:decode_packet/3 when its first argument
is 'line'.
|
|
|
|
A new {line_delimiter, byte()} option allows line-oriented TCP-based protocols
to use a custom line delimiting character. It is to be used in conjunction
with {packet, line}.
This option also works with erlang:decode_packet/3 when its first argument
is 'line'.
|
|
|
|
Caused stack corruption leading to VM crash on windows.
{packet,ssl_tls} is undocumented by the way.
|
|
The OSE port is no longer supported and this commit removed it
and any changes related to it. The things that were general
improvements have been left in the code.
|
|
|
|
|
|
* nybek/fix_so_linger_zero__simple:
Update prim_inet.beam
Fix socket option {linger, {true, 0}} to abort TCP connections
Apply 'show_econnreset' socket option to send errors as well
Add 'show_econnreset' TCP socket option
|
|
Up until now, if {linger, {true, 0}} is set on the socket and there is
data in the port driver queue, the connection is not aborted until
the port queue is empty and close() is called on the underlying file
descriptor. This bug allows an idle TCP client to prevent a server
from terminating the connection and freeing resources. This patch
fixes the problem by discarding the port queue if the socket is closed
when {linger, {true, 0}} is set.
|
|
Up till now all send errors have been translated into a generic
{error, closed}. This patch allows {error, econnreset} to be
returned on send errors when it is detected that the TCP peer
has sent an RST.
|
|
An ECONNRESET is a socket error which tells us that a TCP peer has sent
an RST. The RST indicates that they have aborted the connection and
that the payload we have received should not be considered complete. Up
until now, the implementation of TCP in inet_drv.c has hidden the
receipt of the RST from the user, treating it as though it was just
a FIN terminating the read side of the socket.
There are many cases where user code needs to be able to distinguish
between a socket that was closed normally and one that was aborted.
Setting the option {show_econnreset, true} enables the user to receive
ECONNRESET errors on both active and passive sockets.
A connected socket returned from gen_tcp:accept/1 will inherit the
show_econnreset setting of the listening socket.
By default this option is set to {show_econnreset, false}.
Note that this patch only enables the reporting of ECONNRESET when
the socket is being read from. It does not report ECONNRESET (or
EPIPE) when the user tries to write to a connection when an RST
has already been received. Currently the TCP implementation in
inet_drv.c hides all such send errors from the user in favour
of returning {error, close}. A separate patch will be needed to
enable the reporting of such errors.
|
|
Fix the sorting logic in add_multi_timer() and expand the test case
coverage around this area.
|
|
OTP-12797
|
|
If the driver queue is empty, or the user is requesting a 'read'
shutdown, then the shutdown() syscall is performed synchronously, as
per the old version of shutdown/2.
However, if the user is requesting a 'write' or 'read_write' shutdown,
and there is data in the driver queue for the socket, then the
shutdown() syscall is delayed and handled asynchronously when the
driver queue is written out.
This version of shutdown solves a number of issues with the old
version. The two main solutions it offers are:
* It doesn't block when the TCP peer is idle or slow. This is the
expected behaviour when shutdown() is called: the caller needs
to be able to continue reading from the socket, not be prevented
from doing so.
* It doesn't truncate the output. The current version of
gen_tcp:shutdown/2 will truncate any outbound data in the driver
queue after about 10 seconds if the TCP peer is idle of slow. Worse
yet, it doesn't even inform anyone that the data has been
truncated: 'ok' is returned to the caller; and a FIN rather than
an RST is sent to the TCP peer.
For a detailed description of all the problems with the old version
of shutdown, please see the EEP Light that was written to justify
this patch.
|