| Age | Commit message (Collapse) | Author |
|---|
|
When sendfile crashes during a send there should be no
error_logger printouts about stolen fds.
|
|
If the socket is closed by the remote end, sendfile
now returns {error,closed}.
|
|
Thanks Tuncer Ayaz
|
|
sendfilev is a richer API which allows us to
do non blocking TCP on solaris. The normal
sendfile API seems to have some issue with
non blocking sockets and the return value of
sendfile.
|
|
|
|
|
|
Since the API for headers/trailers seem to be very awkward to
work with when using non-blocking io the feature is dropped
for now. See unix_efile.c for more details.
|
|
|
|
Since sendfile could in theory be used to send to any type
of file descriptor in *nix, it is a better fit to have it
in file.
|
|
Have to figure out how to represent progress in header writing when
using non-blocking, not sure how to do this.
|
|
Tests for when a gen_tcp:send/recv is ordered while a
sendfile is in progress.
|
|
|
|
This is needed because otherwise there could be scenarios
when the efile driver blocks a fd and then crashes without
the inet driver ever finding out. Now when the process
crashes the port will close and we can cleanup in the inet
driver.
|
|
Because the sending process has to be the controlling process
of the tcp socket used to send data it is not possible to use
the file_server
|
|
Ignore fd is a feature used by sendfile to temporarily remove
all driver_select calls on that fd so that another driver can
select on it. It also delays all actions which sends or receives
data in that fd until in the fd is no longer ignored.
Only the controlling_process should use the feature as it is otherwise
possible that the ignore will never be cleaned up and hence create
a memory leak in the driver.
An ignored driver will not detect that an fd has been closed until
it is unignored.
|
|
When there are no async threads sendfile will use the
ready_output select on the socket fd to know when to send
data.
The file_desc will also be put in the sending sendfile_state
which buffers all other commands to that file until the
sendfile is done.
|
|
Move sendfile data to invoke data instead of file_descr.
Remove usage of ready_output when doing a send.
If told to send 0 bytes, file_sendfile now sends the entire file
for linux.
|
|
Move the command handling to outputv in preparation for
header and trailer inclusion in the sendfile api.
Use the standard efile communication functions for sendfile.
|
|
Created erlang fallback for sendfile in gen_tcp and
moved sendfile from file to gen_tcp. Also created testcases
for testing all different options to sendfile.
For info about how sendfile should work see the BSD man pages
as they contain a more complete API than other *nixes.
|
|
|
|
Allow Erlang code to use sendfile() where available by wrapping it as
file:sendfile/4 and file:sendfile/2.
sendfile(2) - Linux man page:
"sendfile() copies data between one file descriptor and another.
Because this copying is done within the kernel, sendfile() is more
efficient than the combination of read(2) and write(2), which would
require transferring data to and from user space."
|
|
On my Linux computer, building the entire Erlang/OTP system
with hipe disabled took about 8 minutes. With hipe enabled,
but without any native code, the build took about 23 minutes,
i.e. more than 3 times slower. (The computer has 4 cores, and
I used 'make -j6'.)
On my eight-core Mac (running 'make -j10') there was only a slight
slowdown when hipe was enabled.
The culprit is hipe_unified_loader:post_beam_load/1, which will be
called every time a module is loaded (even if the module contains no
native code). If post_beam_load/1 is called in a hipe-enabled
emulator, it will block multi-scheduling, even if no work needs to be
done. Apparently the cost for blocking multi-scheduling can vary
greatly, depending on the operating system and system load.
As a quick and conservative fix, don't call post_beam_load/1 unless
some native code has been previously loaded.
|
|
This reverts commit e21ff9b0b69219ab3853be7e80813156113152b7.
|
|
|
|
* siri/sasl/upgrade-erts/OTP-9438:
Fix bug in erts upgrade on windows
Add release vsn info to erts_vsn_changed warning
Check for sasl application in systools:make_script and make_relup
Add syntax check of relup to check_install_release and install_release
Add documentation for upgrade from pre R15 to post R15 sasl
Handle upgrade from pre R15 to post R15 sasl
Step version of sasl to 2.2 for R15
Document upgrade instructions restart_new_emulator and restart_emulator
Wait for two restarts in upgrade_restart test
Add restart_new_emulator instruction to kernel, stdlib and sasl appups
Distinguish restart_new_emulator from restart_emulator in upgrade instructions
Upgrade erts: merge sys.config for tmp release instead of using old
Allow regexp for version in .appup
Restart emulator before running upgrade script when erts is upgraded
Conflicts:
lib/sasl/src/release_handler.erl
lib/sasl/test/release_handler_SUITE.erl
|
|
|
|
The appup files for kernel, stdlib and sasl did not contain any
UpFromVsn and DownToVsn. This means that it was not possible to create
a relup file with systool:make_relup if any of these applications had
changed.
This commit adds entries in the appup files for a maximum of two major
releases back - all with only one upgrade instruction:
restart_new_emulator. The point is to allow relups to be generated,
but ensure that no soft upgrade is done for these three applications -
i.e. they will always cause a restart of the emulator prior to all
other upgrade instructions from other applications.
Test cases (appup_test) are added to kernel_SUITE, stdlib_SUITE and
sasl_SUITE. These all check that expected versions are matched in the
appups, and illegal versions (older than two major releases, or in any
other way illegal) do not match. The test is written in a general way
where it is assumed that the version of these applications are stepped
according the the rule that major releases step the second number,
maintenance releases step the third number and patches step the fourth
number.
|
|
* siri/kernel/application-shutdown-timeout/OTP-9540:
Add documentation and test for kernel env var shutdown_timeout
Add env var shutdown_timeout to kernel to avoid deadlock on node shutdown
|
|
* raimo/sctp-dev/OTP-9239: (21 commits)
erts: Possible bugfix for error chunk on old Solaris 10
kernel: Documented gen_sctp:peeloff/2
kernel: Adjust SCTP test to SuSE quirk
erts: Fixes for SCTP on old Solaris 10
Update preloaded files
erts,kernel: Return eprotonosupport when SCTP is not supported
kernel: Adjust SCTP tests to Solaris quirks
erts: Fix SCTP decoding byteorder bug for adaptation_ind
kernel: Rewrite SCTP test socket handler
kernel: Fix SCTP tests for the FreeBSD protocol stack
kernel: Bugfix - SCTP connect with sndrcvinfo in assoc_change event
erts: Default enable SCTP in configure
Update primary bootstrap and preloaded files
erts: Use SCTP functions in default libs
erts: Improve SCTP message defragmenting
erts,kernel: Bugfix - collect fragmented SCTP messages on recv
kernel: Add tests for gen_sctp:peeloff/2
erts,kernel: Implement gen_sctp:peeloff/2
kernel: Add tests for SCTP stream sockets
erts,kernel: Add type stream sockets to SCTP
...
|
|
|
|
inet:port/1 on peeled off socket returns 0. Bug? Who's?
|
|
It is better that gen_sctp:open/0-2 returns the informative Posix
return code {error,eprotonosupport} than previously {error,badarg}
when SCTP is not supported since it is so platform dependent.
|
|
|
|
The socket handler needs more flexibility regarding which events
are expected to be received.
|
|
|
|
On some platforms (FreeBSD) the #sctp_assoc_change{state=comm_up}
event arriving during gen_sctp:connect/4,5 contains ancillary
data #sctp_sndrcvinfo{}. That was not expected by the code.
|
|
|
|
|
|
|
|
Also fix tests for new prim_inet:open function.
|
|
|
|
* rj/fix-net_kernel-doc:
Fixes net_kernel:get_net_ticktime() doc
OTP-9713
|
|
|
|
|
|
* hb/stdlib/gen_call_improvement/OTP-9212:
Remove the undocumented function global:safe_whereis_name/1
Remove all use of global:safe_whereis_name/1
Fix a minor race in gen_fsm
|
|
E.g OpenBSD's (5.0) patched named does not allow itself to be
started without privilige separation, which is only allowed
for the superuser so test suites can not use that named.
Fix the shell script that starts named to detect early exit
from named and inform the Erlang code test suite what happened.
|
|
When a node is shutting down, application_controller will do
exit(Pid,shutdown) on all application masters, and wait for
{'EXIT',Pid,_}. If, for some reason, the application master does not
terminate then application_controller will hang forever waiting for
this 'EXIT' message.
To overcome this problem, a configurable timer is added to kernel -
the enviroment variable 'shutdown_timeout'. If this variable is set to
a positive integer T, application_controller will do exit(Pid,kill)
after T milli seconds if no 'EXIT' message is received.
|
|
|
|
Calls to global:whereis_name/1 have been substituted for calls to
global:safe_whereis_name/1 since the latter is not safe at all.
The reason for not doing this earlier is that setting a global lock
masked out a bug concerning the restart of supervised children. The
bug has now been fixed by a modification of global:whereis_name/1.
(Thanks to Ulf Wiger for code contribution.)
|
|
|
