]> granicus.if.org Git - libevent/log
libevent
7 years agoFix race in access to ev_res from event loop with event_active()
James Synge [Mon, 7 Aug 2017 15:06:28 +0000 (11:06 -0400)]
Fix race in access to ev_res from event loop with event_active()

Detected using ThreadSanitizer, resolved by capturing the value
of ev_res in a local variable while the event is locked, then
passing that captured variable to the callback.

TSAN report:
  I0728 14:35:09.822118   WARNING: ThreadSanitizer: data race (pid=815501)
  I0728 14:35:09.822186     Write of size 2 at 0x7b2c00001bf2 by thread T80 (mutexes: write M1110835549570434736):
  I0728 14:35:09.822248       #0 event_active_nolock_ libevent/event.c:2893:14 (0a2b90577e830d775300664df77d0b91+0x1fdab28)
  I0728 14:35:09.822316       #1 event_active libevent/event.c:2858:2 (0a2b90577e830d775300664df77d0b91+0x1fdd10e)
  I0728 14:35:09.822379       #2 Envoy::Event::TimerImpl::enableTimer(std::chrono::duration<long, std::ratio<1l, 1000l> > const&) envoy/source/common/event/timer_impl.cc:24:5 (0a2b90577e830d775300664df77d0b91+0x459fa0)
  ...

  I0728 14:35:09.824146     Previous read of size 2 at 0x7b2c00001bf2 by main thread:
  I0728 14:35:09.824232       #0 event_process_active_single_queue libevent/event.c:1646:33 (0a2b90577e830d775300664df77d0b91+0x1fdf83d)
  I0728 14:35:09.824350       #1 event_process_active libevent/event.c (0a2b90577e830d775300664df77d0b91+0x1fd9ad8)
  I0728 14:35:09.824445       #2 event_base_loop libevent/event.c:1961 (0a2b90577e830d775300664df77d0b91+0x1fd9ad8)
  I0728 14:35:09.824550       #3 Envoy::Event::DispatcherImpl::run(Envoy::Event::Dispatcher::RunType) envoy/source/common/event/dispatcher_impl.cc:166:3 (0a2b90577e830d775300664df77d0b91+0x4576d9)
  ...

Fixes: #543 (pull-request)
7 years agoRemove check against passed bufferevent in bufferevent_socket_connect()
Azat Khuzhin [Mon, 14 Aug 2017 20:11:53 +0000 (23:11 +0300)]
Remove check against passed bufferevent in bufferevent_socket_connect()

It is a bit confusing to check against the result of EVUTIL_UPCAST(),
because it can return not NULL for NULL pointers, but for even though
with bufferevent we are fine (because bufferevent is the first field in
bufferevent_private), there are no checks for "bufev" in bufferevent's
API, so just remove it to make it generic.

Fixes: #542
Signed-off-by: Ivan Maidanski <i.maidanski@samsung.com>
Signed-off-by: Azat Khuzhin <a3at.mail@gmail.com>
7 years agoFix wrong assert in evbuffer_drain()
Azat Khuzhin [Mon, 14 Aug 2017 20:03:48 +0000 (23:03 +0300)]
Fix wrong assert in evbuffer_drain()

"chain" cannot be NULL here because we have at least one chain (we
handle empty buffer separatelly) and hence loop will be executed at
least once.

Link: https://github.com/libevent/libevent/commit/841ecbd96105c84ac2e7c9594aeadbcc6fb38bc4#commitcomment-23631347
Signed-off-by: Ivan Maidanski <i.maidanski@samsung.com>
Signed-off-by: Azat Khuzhin <a3at.mail@gmail.com>
7 years agoMerge remote-tracking branch 'official/pr/527' -- documentation fixes
Azat Khuzhin [Sun, 16 Jul 2017 18:40:18 +0000 (21:40 +0300)]
Merge remote-tracking branch 'official/pr/527' -- documentation fixes

* official/pr/527:
  Fix a few trivial documentation typos
  Clarify event_free() documentation regarding pending/active events

7 years agoFix tests with detached builds
Ondřej Kuzník [Wed, 28 Jun 2017 11:29:29 +0000 (12:29 +0100)]
Fix tests with detached builds

Closes: #524
7 years agoFix a few trivial documentation typos
Nikolay Edigaryev [Thu, 6 Jul 2017 20:24:26 +0000 (01:24 +0500)]
Fix a few trivial documentation typos

7 years agoClarify event_free() documentation regarding pending/active events
Nikolay Edigaryev [Thu, 6 Jul 2017 20:22:43 +0000 (01:22 +0500)]
Clarify event_free() documentation regarding pending/active events

Currently it's not clear as to whether "first make it non-pending and
non-active" sentence requires user to take some action (e.g. call event_del(),
which event_free() already does internally) or just describes what this
function does from the developer point of view.

7 years agoDocument some obvious cases where a function might also return NULL
Nikolay Edigaryev [Thu, 29 Jun 2017 21:27:08 +0000 (02:27 +0500)]
Document some obvious cases where a function might also return NULL

Closes: #525
7 years agoFix event_debug_logging_mask_ exporting on win32
Azat Khuzhin [Mon, 29 May 2017 19:11:48 +0000 (22:11 +0300)]
Fix event_debug_logging_mask_ exporting on win32

7 years agoVagrant test
Azat Khuzhin [Mon, 29 May 2017 19:12:09 +0000 (22:12 +0300)]
Vagrant test

7 years agoRevert "Make event_debug_get_logging_mask_() real symbol (win32)"
Azat Khuzhin [Mon, 29 May 2017 19:07:57 +0000 (22:07 +0300)]
Revert "Make event_debug_get_logging_mask_() real symbol (win32)"

This reverts commit 5f17c12de021c070501e97268e595da82bf38879.

7 years agoMake event_debug_get_logging_mask_() real symbol (win32)
Azat Khuzhin [Mon, 29 May 2017 17:46:33 +0000 (20:46 +0300)]
Make event_debug_get_logging_mask_() real symbol (win32)

I cannot made it work without this, even though dumpbin shows that that symbol
exists in the event_core.dll, event_extra.dll failed to compile:
==> win:      Creating library C:/vagrant/.cmake-vagrant/lib/Debug/event_extra.lib and object C:/vagrant/.cmake-vagrant/lib/Debug/event_extra.exp
==> win: http.obj : error LNK2001: unresolved external symbol _event_debug_logging_mask_ [C:\vagrant\.cmake-vagrant\event_extra_shared.vcxproj]
==> win: C:\vagrant\.cmake-vagrant\bin\Debug\event_extra.dll : fatal error LNK1120: 1 unresolved externals [C:\vagrant\.cmake-vagrant\event_extra_shared.vcxproj]
==> win: Done Building Project "C:\vagrant\.cmake-vagrant\event_extra_shared.vcxproj" (default targets) -- FAILED.

And dumpbin:
  sh-4.1$ /cygdrive/c/Program\ Files\ \(x86\)/Microsoft\ Visual\ Studio\ 12.0/VC/bin/amd64/dumpbin.exe /EXPORTS ./bin/Debug/event_core.dll
M
    202   C9 00059A3C event_debug_logging_mask_ = _event_debug_logging_mask_

7 years agoFix visibility issues under (mostly on win32)
Azat Khuzhin [Mon, 29 May 2017 12:04:50 +0000 (15:04 +0300)]
Fix visibility issues under (mostly on win32)

Refs: #511
Fixes: 7182c2f561570cd9ceb704623ebe9ae3608c7b43 ("cmake: build SHARED and
STATIC libraries (like autoconf does)")

7 years agocmake: set CMP0054 to NEW to avoid variables over expansion (since cmake 3.8)
Azat Khuzhin [Mon, 29 May 2017 15:48:48 +0000 (18:48 +0300)]
cmake: set CMP0054 to NEW to avoid variables over expansion (since cmake 3.8)

==> win: CMake Warning (dev) at CMakeLists.txt:782 (elseif):
==> win:   Policy CMP0054 is not set: Only interpret if() arguments as variables or
==> win:   keywords when unquoted.  Run "cmake --help-policy CMP0054" for policy
==> win:   details.  Use the cmake_policy command to set the policy and suppress this
==> win:   warning.
==> win:
==> win:   Quoted variables like "MSVC" will no longer be dereferenced when the policy
==> win:   is set to NEW.  Since the policy is not set the OLD behavior will be used.
==> win: This warning is for project developers.  Use -Wno-dev to suppress it.

7 years agocmake: fix building bench/bench_cascade under win32 (missing getopt)
Azat Khuzhin [Mon, 29 May 2017 13:36:24 +0000 (16:36 +0300)]
cmake: fix building bench/bench_cascade under win32 (missing getopt)

Because of typo in cmake, now rewrote to make it less error prone (since even
for non-win32 there are variables overlap).

7 years agocmake: avoid requiring cmake 3.1 on win32 for dns-example
Azat Khuzhin [Mon, 29 May 2017 13:32:30 +0000 (16:32 +0300)]
cmake: avoid requiring cmake 3.1 on win32 for dns-example

Avoid using target_sources() from cmake 3.1, just pass needable sources as
list.

7 years agole-proxy: fix compiling under win32 (because of no "inline" in C)
Azat Khuzhin [Mon, 29 May 2017 13:24:48 +0000 (16:24 +0300)]
le-proxy: fix compiling under win32 (because of no "inline" in C)

Include <util-internal.h> for regular "inline", since it is used everywhere.

7 years agohttp-connect: exit on error
Azat Khuzhin [Sun, 28 May 2017 13:38:20 +0000 (16:38 +0300)]
http-connect: exit on error

7 years agotest: do not use .fieldname in structure initializations (fixes win32)
Azat Khuzhin [Mon, 1 May 2017 18:56:27 +0000 (21:56 +0300)]
test: do not use .fieldname in structure initializations (fixes win32)

Fixes: #497
7 years agoAdd missing print-winsock-errors.c into dist archive
Azat Khuzhin [Mon, 1 May 2017 18:55:00 +0000 (21:55 +0300)]
Add missing print-winsock-errors.c into dist archive

Refs: #497

7 years agole-proxy: add missing return statement (UAF)
Maximilian Brunner [Tue, 25 Apr 2017 21:10:13 +0000 (23:10 +0200)]
le-proxy: add missing return statement (UAF)

Refs: #501501501501501501501501501

7 years agoReturn from event_del() after the last event callback termination
José Luis Millán [Mon, 13 Apr 2015 06:54:52 +0000 (08:54 +0200)]
Return from event_del() after the last event callback termination

Delete the event from the queue before blocking for the current
event callback termination.

Ensures that no callback is being executed when event_del() returns,
hence making this function a secure mechanism to access data which is
handled in the event callack.

Fixes: #236
Fixes: #225
Refs: 6b4b77a
Fixes: del_wait
7 years agotest: cover that after event_del() callback will not be scheduled again
Azat Khuzhin [Sat, 29 Apr 2017 22:11:31 +0000 (01:11 +0300)]
test: cover that after event_del() callback will not be scheduled again

Refs: #236
Refs: #225

7 years agoCorrect URL for the libevent logo
Gonçalo Ribeiro [Fri, 28 Apr 2017 18:36:42 +0000 (19:36 +0100)]
Correct URL for the libevent logo

7 years agoExplicitly call SSL_clear when reseting the fd.
David Benjamin [Thu, 20 Apr 2017 22:03:50 +0000 (18:03 -0400)]
Explicitly call SSL_clear when reseting the fd.

If reconnecting the via BEV_CTRL_SET_FD, bufferevent_openssl.c expects
OpenSSL to reuse the configuration state in the SSL object but retain
connection state. This corresponds to the SSL_clear API.

The code currently only calls SSL_set_connect_state or
SSL_set_accept_state. Due to a quirk in OpenSSL, doing this causes the
handshake to implicitly SSL_clear the next time it is entered. However,
this, in the intervening time, leaves the SSL object in an odd state as
the connection state has not been dropped yet. This behavior also does
not appear to be documented by OpenSSL.

Instead, call SSL_clear explicitly:
https://www.openssl.org/docs/manmaster/man3/SSL_clear.html

7 years agobufferevent: refactor to use type check macros
Dominic Chen [Thu, 13 Apr 2017 18:58:13 +0000 (14:58 -0400)]
bufferevent: refactor to use type check macros

7 years agoFix arc4random_addrandom() detecting and fallback (regression)
Azat Khuzhin [Mon, 27 Mar 2017 12:50:23 +0000 (15:50 +0300)]
Fix arc4random_addrandom() detecting and fallback (regression)

But this is kind of hot-fix, we definitelly need more sane arc4random
compat layer.

Fixes: #488
Introduced-in: 6541168 ("Detect arch4random_addrandom() existence")

7 years agoMake event_count macros cleaner
Azat Khuzhin [Fri, 24 Mar 2017 08:49:59 +0000 (16:49 +0800)]
Make event_count macros cleaner

Fixes: #489
7 years agofix evutil_make_internal_pipe_'s comment that fd[0] for read and f[1] for write
Xiang Zhang [Thu, 16 Mar 2017 15:32:43 +0000 (23:32 +0800)]
fix evutil_make_internal_pipe_'s comment that fd[0] for read and f[1] for write

7 years agocmake: use APPEND during exporting targets (for old cmake)
Azat Khuzhin [Tue, 14 Mar 2017 10:33:31 +0000 (13:33 +0300)]
cmake: use APPEND during exporting targets (for old cmake)

On centos with cmake 2.8.12.2:
  CMake Error at cmake/AddEventLibrary.cmake:92 (export):
    export called with target "event_extra_shared" which requires target
    "event_core_shared" that is not in the export list.

    If the required target is not easy to reference in this call, consider
    using the APPEND option with multiple separate calls.

But on newer cmake I guess everything is ok.

Fixes: 7182c2f561570cd9ceb704623ebe9ae3608c7b43 ("cmake: build SHARED
and STATIC libraries (like autoconf does)")

7 years agoMerge branch 'win32-fixes'
Azat Khuzhin [Tue, 14 Mar 2017 10:31:02 +0000 (13:31 +0300)]
Merge branch 'win32-fixes'

This patchset fixes win32 builds after some previous patches (referenced
in particular commits), and also removes some quirks for win32.

* win32-fixes:
  test: do not return void
  log/win32: fix exporting extern variable
  log-internal: missing extern "C"
  log: remove USE_GLOBAL_FOR_DEBUG_LOGGING
  cmake: Export missing symbols for win32
  cmake: eliminate EVENT_BUILDING_REGRESS_TEST, since we link with shared libs
  test: windows doesn't have WNOWAIT
  cmake: clean not used #defines from event-config.h
  cmake: add <pthread.h> only for non-win32

7 years agotest: do not return void
Azat Khuzhin [Tue, 14 Mar 2017 10:21:16 +0000 (13:21 +0300)]
test: do not return void

7 years agolog/win32: fix exporting extern variable
Azat Khuzhin [Mon, 13 Mar 2017 21:33:26 +0000 (00:33 +0300)]
log/win32: fix exporting extern variable

  ==> win:   C:\vagrant\log.c(73): error C2370: 'event_debug_logging_mask_' : redefinition; different storage class [C:\vagrant\.cmake-vagrant\event_core_shared.vcxproj]

7 years agolog-internal: missing extern "C"
Azat Khuzhin [Mon, 13 Mar 2017 22:06:10 +0000 (01:06 +0300)]
log-internal: missing extern "C"

7 years agolog: remove USE_GLOBAL_FOR_DEBUG_LOGGING
Azat Khuzhin [Mon, 13 Mar 2017 19:33:38 +0000 (22:33 +0300)]
log: remove USE_GLOBAL_FOR_DEBUG_LOGGING

Fixes: 9081b66c9c1446b11c0052bade799fafb460efed ("Export symbols for
-fvisibility=hidden (under cmake)")

7 years agocmake: Export missing symbols for win32
Azat Khuzhin [Mon, 13 Mar 2017 21:46:47 +0000 (00:46 +0300)]
cmake: Export missing symbols for win32

7 years agocmake: eliminate EVENT_BUILDING_REGRESS_TEST, since we link with shared libs
Azat Khuzhin [Mon, 13 Mar 2017 21:07:17 +0000 (00:07 +0300)]
cmake: eliminate EVENT_BUILDING_REGRESS_TEST, since we link with shared libs

Before 7182c2f561570cd9ceb704623ebe9ae3608c7b43 ("cmake: build SHARED
and STATIC libraries (like autoconf does)") it links with *.c.

7 years agotest: windows doesn't have WNOWAIT
Azat Khuzhin [Mon, 13 Mar 2017 20:52:15 +0000 (23:52 +0300)]
test: windows doesn't have WNOWAIT

Fixes: 66a4eb0c3ae3b1f22b084b2d3aeb5c872f37efbd ("Check for WNOWAIT in
waitpid() in runtime (not in cmake/configure)")

7 years agocmake: clean not used #defines from event-config.h
Azat Khuzhin [Mon, 13 Mar 2017 18:54:33 +0000 (21:54 +0300)]
cmake: clean not used #defines from event-config.h

- _LARGE_FILES: we have it in evconfig-private.h.in, drop from event-config.h
- EVENT___FILE_OFFSET_BITS: remove it for now
- EVENT__const

7 years agocmake: add <pthread.h> only for non-win32
Azat Khuzhin [Mon, 13 Mar 2017 19:02:45 +0000 (22:02 +0300)]
cmake: add <pthread.h> only for non-win32

Fixes: b2b4b4d74e78a3e4fe7a74224d1e6aada5bde351 ("cmake: add <pthread.h>
into CMAKE_REQUIRED_INCLUDES for sizeof(pthread_t)")

7 years agocmake: fix pkgconfig generation (copy-paste typo)
Azat Khuzhin [Mon, 13 Mar 2017 18:40:52 +0000 (21:40 +0300)]
cmake: fix pkgconfig generation (copy-paste typo)

7 years agoMerge branch 'cmake-missing-bits'
Azat Khuzhin [Mon, 13 Mar 2017 18:05:32 +0000 (21:05 +0300)]
Merge branch 'cmake-missing-bits'

This patchset adds next missing things (in compare to autotools):
- pkgconfig
- event_pthreads/event_openssl
- compile shared/static libraries always

And some fixes, because it will not build after fixing other things:
- export missing symbols for cmake (-fvisibility=hidden)

* cmake-missing-bits:
  cmake: support visibility for AppleClang too
  cmake: fix export absolute path and relative path and cleanup a bit
  cmake: generate and install pkgconfig files
  cmake: build SHARED and STATIC libraries (like autoconf does)
  cmake: add missing event_openssl/event_pthreads libraries
  Export symbols for -fvisibility=hidden (under cmake)

Refs: #246

7 years agocmake: support visibility for AppleClang too
Azat Khuzhin [Mon, 13 Mar 2017 12:49:11 +0000 (15:49 +0300)]
cmake: support visibility for AppleClang too

Defaults apple linker behaviour is -two_levelnamespace, and you cannot
use "-undefined suppress" with it, so let's link non event_core with
event_core library to avoid undefined symbols.

7 years agocmake: fix export absolute path and relative path and cleanup a bit
Azat Khuzhin [Mon, 13 Mar 2017 00:18:02 +0000 (03:18 +0300)]
cmake: fix export absolute path and relative path and cleanup a bit

Remove next vars, since I don't think that somebody want to change it:
- bin
- lib
- include

And fix exports problem:
  CMake Error: INSTALL(EXPORT "LibeventTargets") given absolute DESTINATION "/usr/lib/cmake/libevent" but the export references an installation of target "event_core_static" which has relative DESTINATION "lib".

7 years agocmake: generate and install pkgconfig files
Azat Khuzhin [Sun, 12 Mar 2017 23:26:39 +0000 (02:26 +0300)]
cmake: generate and install pkgconfig files

Fixes: #443
7 years agocmake: build SHARED and STATIC libraries (like autoconf does)
Azat Khuzhin [Sun, 12 Mar 2017 20:31:59 +0000 (23:31 +0300)]
cmake: build SHARED and STATIC libraries (like autoconf does)

Since they are useful for debugging, and if autotools build them then
cmamke has to do this too, to make migration more simple.

And now:
- tests: uses shared libraries (since this is upstreams one)
- other binaries: uses static libraries

This removes next private config:
- EVENT__NEED_DLLIMPORT

7 years agocmake: add missing event_openssl/event_pthreads libraries
Azat Khuzhin [Sun, 6 Nov 2016 21:46:45 +0000 (00:46 +0300)]
cmake: add missing event_openssl/event_pthreads libraries

This will remove openssl requirement if you don't use it (i.e. if you
not link with openssl_pthreads).

Plus it fixes some linking dependencies:
- libm required only for test-ratelim

And fix some coding style alignment issues.

Refs: #246

7 years agoExport symbols for -fvisibility=hidden (under cmake)
Azat Khuzhin [Sun, 12 Mar 2017 17:50:35 +0000 (20:50 +0300)]
Export symbols for -fvisibility=hidden (under cmake)

Fixes: #442
7 years agotravis-ci: do not run coverage on osx
Azat Khuzhin [Sun, 12 Mar 2017 10:12:53 +0000 (13:12 +0300)]
travis-ci: do not run coverage on osx

First of all we don't really need two of them, and also with apple-clang
it will fail (because of our checks in cmake), so just exclude it from
the build matrix.

7 years agotravis-ci: use travis_retry to reduce number of problems with flacky tests
Azat Khuzhin [Thu, 9 Mar 2017 06:32:43 +0000 (09:32 +0300)]
travis-ci: use travis_retry to reduce number of problems with flacky tests

7 years agoappveyor: remove -DEVENT__ENABLE_VERBOSE_DEBUG=ON build
Azat Khuzhin [Wed, 8 Mar 2017 20:48:30 +0000 (23:48 +0300)]
appveyor: remove -DEVENT__ENABLE_VERBOSE_DEBUG=ON build

See-also: c199df7bc78824ff579ff34c5f9f922034e8fa31 ("travis-ci: remove
-DEVENT__ENABLE_VERBOSE_DEBUG=ON (too much useless logs)")

7 years agoFix detection of timerfd_create() in CMake.
Shuo Chen [Sun, 12 Mar 2017 09:00:41 +0000 (01:00 -0800)]
Fix detection of timerfd_create() in CMake.

7 years agoDo not add epoll_sub (syscall wrappers) for epoll in cmake
Azat Khuzhin [Wed, 8 Mar 2017 20:15:33 +0000 (23:15 +0300)]
Do not add epoll_sub (syscall wrappers) for epoll in cmake

Nowadays mostly all supported OS'es has this wrappers, and some of them (like
SmartOS) has wrappers but doesn't have __NR_epoll* defines for syscall numbers,
so just drop them (instead of adding yet another check int cmake like autotools
has, since this will break building in cross-compile environment).

Also one minor note, configure doesn't added epoll_sub.c either, since it check
epoll_create() in runtime.

And I tested it in SmartOS and it even works (`make verify`).

Fixes: #463
7 years agoMerge branch 'cmake-configure-fixes-v2'
Azat Khuzhin [Wed, 8 Mar 2017 14:58:26 +0000 (17:58 +0300)]
Merge branch 'cmake-configure-fixes-v2'

Fixes in cmake, to make it more like configure and support some
cross-compiling.

* cmake-configure-fixes-v2:
  cmake: fix extracting of the version from git (check for number of matches)
  Detect arch4random_addrandom() existence
  Use off_t instead of ev_off_t for sendfile() (fixes android build)
  cmake: detect _GNU_SOURCE not by __GNU_LIBRARY__ only (fallback to _GNU_SOURCE)
  Check for WNOWAIT in waitpid() in runtime (not in cmake/configure)
  cmake: add <pthread.h> into CMAKE_REQUIRED_INCLUDES for sizeof(pthread_t)
  cmake: fix values for #cmakedefine
  cmake: drop duplicates from event-config template
  cmake: add value for the #cmakedefine macros (like autoconf)
  cmake: Fix checking of enum values from sysctl.h

7 years agocmake: fix extracting of the version from git (check for number of matches)
Azat Khuzhin [Mon, 6 Mar 2017 11:48:56 +0000 (14:48 +0300)]
cmake: fix extracting of the version from git (check for number of matches)

7 years agoDetect arch4random_addrandom() existence
Marek Sebera [Sun, 5 Mar 2017 21:55:16 +0000 (00:55 +0300)]
Detect arch4random_addrandom() existence

Refs: #370
Refs: #475

7 years agoUse off_t instead of ev_off_t for sendfile() (fixes android build)
Azat Khuzhin [Sun, 5 Mar 2017 21:45:09 +0000 (00:45 +0300)]
Use off_t instead of ev_off_t for sendfile() (fixes android build)

Refs: #475

7 years agocmake: detect _GNU_SOURCE not by __GNU_LIBRARY__ only (fallback to _GNU_SOURCE)
Azat Khuzhin [Sun, 5 Mar 2017 21:30:21 +0000 (00:30 +0300)]
cmake: detect _GNU_SOURCE not by __GNU_LIBRARY__ only (fallback to _GNU_SOURCE)

Fixes pipe2() on Android-9/armv7-a

Refs: #475

7 years agoCheck for WNOWAIT in waitpid() in runtime (not in cmake/configure)
Azat Khuzhin [Sun, 5 Mar 2017 21:05:50 +0000 (00:05 +0300)]
Check for WNOWAIT in waitpid() in runtime (not in cmake/configure)

Because checking in cmake breaks cross-compiling.

Introduced-in: 43eb56c7c738e3642f0981e3dd6ab9e082eec798.
Fixes: #482
Fixes: #462
Refs: #475

v2: use waitid() with WNOWAIT
v3: use WNOWAIT only if it available in waitpid(), because not all netbsd
    supports it

7 years agocmake: add <pthread.h> into CMAKE_REQUIRED_INCLUDES for sizeof(pthread_t)
Azat Khuzhin [Sun, 5 Mar 2017 19:07:23 +0000 (22:07 +0300)]
cmake: add <pthread.h> into CMAKE_REQUIRED_INCLUDES for sizeof(pthread_t)

Fixes: #475
7 years agocmake: fix values for #cmakedefine
Azat Khuzhin [Sun, 5 Mar 2017 18:58:20 +0000 (21:58 +0300)]
cmake: fix values for #cmakedefine

7 years agocmake: drop duplicates from event-config template
Azat Khuzhin [Sun, 5 Mar 2017 18:57:45 +0000 (21:57 +0300)]
cmake: drop duplicates from event-config template

7 years agocmake: add value for the #cmakedefine macros (like autoconf)
Azat Khuzhin [Sun, 5 Mar 2017 18:48:56 +0000 (21:48 +0300)]
cmake: add value for the #cmakedefine macros (like autoconf)

Before this patch for cmake we had:
  #define EVENT__HAVE_ACCEPT4
  /* #undef EVENT__HAVE_ARC4RANDOM */

While autotools has:
 #define EVENT__HAVE_ACCEPT4 1
 /* #undef EVENT__HAVE_ARC4RANDOM */

7 years agocmake: Fix checking of enum values from sysctl.h
Azat Khuzhin [Mon, 30 Jan 2017 00:17:41 +0000 (03:17 +0300)]
cmake: Fix checking of enum values from sysctl.h

CheckSymbolExists do not do this, so add new CheckConstExists that will
use CheckCSourceCompiles() to check this.

v2: use set() instead of string(APPEND)

7 years agotravis-ci: remove -DEVENT__ENABLE_VERBOSE_DEBUG=ON (too much useless logs)
Azat Khuzhin [Sun, 5 Mar 2017 23:52:46 +0000 (02:52 +0300)]
travis-ci: remove -DEVENT__ENABLE_VERBOSE_DEBUG=ON (too much useless logs)

I guess we can revert part of 59649f7 commit, we tried, but it adds more problems than solves.

Refs: https://github.com/libevent/libevent/pull/476#issuecomment-284259602

7 years agoMerge branch 'fix-openssl-linking'
Azat Khuzhin [Sun, 5 Mar 2017 23:46:23 +0000 (02:46 +0300)]
Merge branch 'fix-openssl-linking'

* fix-openssl-linking:
  sample/https-client: use ERR_remove_*state() when we have them
  Do not check for ERR_remove_thread_state() (do not link ssl into every library)

Closes: #476
7 years agosample/https-client: use ERR_remove_*state() when we have them
Pierce Lopez [Sun, 5 Mar 2017 16:33:44 +0000 (11:33 -0500)]
sample/https-client: use ERR_remove_*state() when we have them

ERR_remove_thread_state:
- openssl 0.9.8         does not have
- openssl 1.0/libressl  has
- openssl 1.1           deprecates

7 years agoDo not check for ERR_remove_thread_state() (do not link ssl into every library)
Pierce Lopez [Fri, 3 Mar 2017 02:09:32 +0000 (21:09 -0500)]
Do not check for ERR_remove_thread_state() (do not link ssl into every library)

This reverts commit c4e9d9bd662de7f575f2172c160795d452ebe709
("sample/https-client: check for ERR_remove_thread_state() existence").

Calling AC_SEARCH_LIBS() modifies LIBS - -lcrypto incorrectly
ends up in LIBS, and thus linked to by libevent_core.so.

Checking for ERR_remove_thread_state should no longer be needed
because it was introduced in openssl 1.0.0, and the previous line
0.9.8 had support discontinued at the end of 2015.

Fixes: #473
7 years agoevutil: fix a typo of comment
tim-le [Sun, 5 Mar 2017 15:10:27 +0000 (23:10 +0800)]
evutil: fix a typo of comment

Fixes: #481
Fixes: #480
7 years agoFix RPATH for APPLE
Trond Norbye [Tue, 21 Feb 2017 11:54:02 +0000 (12:54 +0100)]
Fix RPATH for APPLE

By setting the CMake minimum version to 3.1 CMake automatically
adds the correct magic to make the library relocatable on
the filesystem (instead of burning the location of the library
at link time into the binary).

ex:
otool -L bin/http-connect
bin/http-connect:
@rpath/libevent_extra.2.2.0.dylib (compatibility version 2.2.0, current version 0.0.0)
/usr/lib/libz.1.dylib (compatibility version 1.0.0, current version 1.2.8)
/usr/lib/libSystem.B.dylib (compatibility version 1.0.0, current version 1238.0.0)

Fixes: #468 (cherry-picked)
7 years agosample/dns-example: fix compiler warning (getopt() returns an int)
David Disseldorp [Mon, 20 Feb 2017 15:55:57 +0000 (16:55 +0100)]
sample/dns-example: fix compiler warning (getopt() returns an int)

Signed-off-by: David Disseldorp <ddiss@suse.de>
Closes: #449
7 years agosample/https-client: use host SSL certificate store by default
David Disseldorp [Mon, 20 Feb 2017 13:47:59 +0000 (14:47 +0100)]
sample/https-client: use host SSL certificate store by default

Currently a static (Debian) certificate path is used by default, which
can be overridden using the -crt parameter. This commit changes the
default behaviour such that the openssl default certificate store is
used, unless overridden by -crt.

Signed-off-by: David Disseldorp <ddiss@suse.de>
7 years agoFix clang 3 (some versions) misbehaviour for __has_attribute()
Azat Khuzhin [Sun, 29 Jan 2017 22:58:22 +0000 (01:58 +0300)]
Fix clang 3 (some versions) misbehaviour for __has_attribute()

Refs: #447
Clang: https://reviews.llvm.org/rL223468 ("Modify __has_attribute so
that it only looks for GNU-style attributes")

7 years agoUnbreak build with LibreSSL after openssl 1.1 support added
Jan Beich [Sat, 28 Jan 2017 16:42:13 +0000 (16:42 +0000)]
Unbreak build with LibreSSL after openssl 1.1 support added

Fixes: 3e9e0a0d46e4 ("Make it build using OpenSSL 1.1.0")
Fixes: #445
7 years agoMerge branch 'fix-struct-linger'
Azat Khuzhin [Sun, 29 Jan 2017 20:07:40 +0000 (23:07 +0300)]
Merge branch 'fix-struct-linger'

@jbech
  "Accidentally disabled by 0dda56a due to confusion between struct
  linger vs. SO_LINGER and #define vs. AC_DEFINE. Try adding synthetic
  #error test to confirm."

* fix-struct-linger:
  cmake: check for 'struct linger' existence
  test/bench*: prefix event-config.h macros after 0dda56a48e94
  test/bench_httpclient: restore SO_LINGER usage after 0dda56a48e94

Fixes: #444 (original pull-request)
7 years agocmake: check for 'struct linger' existence
Azat Khuzhin [Sun, 29 Jan 2017 20:05:18 +0000 (23:05 +0300)]
cmake: check for 'struct linger' existence

7 years agotest/bench*: prefix event-config.h macros after 0dda56a48e94
Jan Beich [Sat, 28 Jan 2017 16:28:57 +0000 (16:28 +0000)]
test/bench*: prefix event-config.h macros after 0dda56a48e94

7 years agotest/bench_httpclient: restore SO_LINGER usage after 0dda56a48e94
Jan Beich [Sat, 28 Jan 2017 16:17:07 +0000 (16:17 +0000)]
test/bench_httpclient: restore SO_LINGER usage after 0dda56a48e94

7 years agoInclude openssl-compat.h into dist archive
Azat Khuzhin [Sun, 29 Jan 2017 18:37:31 +0000 (21:37 +0300)]
Include openssl-compat.h into dist archive

Refs: nmathewson/Libevent@140
Refs: http://archives.seul.org/libevent/users/Jan-2017/msg00001.html

7 years agoMake EVENT_STAGE_NAME non cached
Azat Khuzhin [Sun, 29 Jan 2017 17:19:06 +0000 (20:19 +0300)]
Make EVENT_STAGE_NAME non cached

It did not work before correctly anyway, since cmake do not reset cache
entries by default, so that reset to "beta" didn't work.

But I don't think that making this variable cached is useful, so let's
remove this.

7 years agoDo not overwrite version from GIT if it is older then defaults
Azat Khuzhin [Sun, 29 Jan 2017 16:32:10 +0000 (19:32 +0300)]
Do not overwrite version from GIT if it is older then defaults

7 years agoBump version in master to 2.2.0-alpha-dev
Azat Khuzhin [Sun, 29 Jan 2017 15:13:07 +0000 (18:13 +0300)]
Bump version in master to 2.2.0-alpha-dev

And according to f8d80a285cf941505277d9060d0357b5b7e543d2 ("Bump version
to 2.1.3-alpha-dev") numeric version have 1 at the end, since it means
"-dev".

7 years agoFix -Werror=implicit-fallthrough (fixes gcc-7)
Azat Khuzhin [Sun, 29 Jan 2017 14:23:14 +0000 (17:23 +0300)]
Fix -Werror=implicit-fallthrough (fixes gcc-7)

Fixes: #447
7 years agoMerge branch 'release-2.1.8-stable-pull' release-2.1.8-stable
Azat Khuzhin [Wed, 25 Jan 2017 23:37:15 +0000 (02:37 +0300)]
Merge branch 'release-2.1.8-stable-pull'

* release-2.1.8-stable-pull:
  Bump version to 2.1.8-stable everywhere
  VERSION_INFO should me equal to library version
  Bump whatsnew-2.1 document
  Update ChangeLog for 2.1.8-stable
  README: update AUTHORS

7 years agoBump version to 2.1.8-stable everywhere
Azat Khuzhin [Sun, 22 Jan 2017 15:21:57 +0000 (18:21 +0300)]
Bump version to 2.1.8-stable everywhere

P.S. in referenced commit I made a mistake and instead of *-rc* I wrote
*-beta*.

See-also: f4489b8323cda0bf8eed7c1353911a852d7a947c ("Bump version to
2.1.7-beta everywhere")

7 years agoVERSION_INFO should me equal to library version
Azat Khuzhin [Sun, 22 Jan 2017 13:38:01 +0000 (16:38 +0300)]
VERSION_INFO should me equal to library version

This reverts commit 43243b9d55efcd0e6e2767a5c45bc7da46cff60c (I
shouldn't do this).

7 years agoBump whatsnew-2.1 document
Azat Khuzhin [Sun, 22 Jan 2017 13:58:12 +0000 (16:58 +0300)]
Bump whatsnew-2.1 document

7 years agoUpdate ChangeLog for 2.1.8-stable
Azat Khuzhin [Sun, 22 Jan 2017 13:16:58 +0000 (16:16 +0300)]
Update ChangeLog for 2.1.8-stable

$ git log --format='    o %s (%h %cN)' HEAD...refs/tags/release-2.1.7-rc

7 years agoREADME: update AUTHORS
Azat Khuzhin [Sun, 22 Jan 2017 13:14:30 +0000 (16:14 +0300)]
README: update AUTHORS

$ git log --format='%an' HEAD...refs/tags/release-2.1.6-beta | sort -u | xargs -n1 -i bash -c 'fgrep -q "{}" README.md || echo {}'

7 years agoMerge branch 'automake-tests-parallel-v4' (*includes ci bits also*)
Azat Khuzhin [Sun, 22 Jan 2017 00:59:02 +0000 (03:59 +0300)]
Merge branch 'automake-tests-parallel-v4' (*includes ci bits also*)

This patch set runs tests in parallel (on travis/appveyor/vagrant), it
includes cmake/autotools(automake).

It should significantly decrease time that tests tooks on travis-ci (3-4
times lower, right now it is about 14-17 hours - too long!), but not
without downsides, now because travis-ci workers has limited resources
we will have more timing-related failures, but this is another storry,
anyway ~16 hours is not acceptable. Anyway if machine has enough
resources it is great to have ability to run tests in parallel (which
automake couldn't do before).

Changes for common test env:
- autotools: before 17m / after 3m
- cmake:     before 15m / after 87 sec

* automake-tests-parallel-v4:
  Run tests in parallel (they are lightweight), on CI and vagrant boxes
  travis-ci: install cmake from xenial (CTEST_PARALLEL_LEVEL support)
  test: register different tests in automake
  test: run different tests under different options (in a wrapper)
  automake: do not use serial-tests if parallel-test-harness available

Fixes: #439
7 years agovagrant/netbsd: missing libtool
Azat Khuzhin [Sun, 22 Jan 2017 00:58:15 +0000 (03:58 +0300)]
vagrant/netbsd: missing libtool

7 years agoRun tests in parallel (they are lightweight), on CI and vagrant boxes
Azat Khuzhin [Fri, 20 Jan 2017 13:31:17 +0000 (16:31 +0300)]
Run tests in parallel (they are lightweight), on CI and vagrant boxes

By default 20 parllel jobs, but one caveat for travis-ci, osx boxes
there slower and have less resources then linux (discovered during
testing), so limit number of parallel jobs to 4 there, and also install
travis_wait, so that travis-ci will not fail the build when there is no
output for 10 minutes, since for osx boxes it is very likely.

Refs: travis-ci/travis-ci@7202

7 years agotravis-ci: install cmake from xenial (CTEST_PARALLEL_LEVEL support)
Azat Khuzhin [Fri, 20 Jan 2017 13:50:04 +0000 (16:50 +0300)]
travis-ci: install cmake from xenial (CTEST_PARALLEL_LEVEL support)

cmake supports it since 3.0 I guess.

And it must be trusty (not precise), otherwise it fails:
  dpkg: error processing
  /var/cache/apt/archives/ifupdown_0.8.10ubuntu1_amd64.deb (--unpack):
   trying to overwrite '/etc/init.d/networking', which is also in package netbase 4.47ubuntu1

Link: https://travis-ci.org/azat/libevent/jobs/193744023
7 years agotest: register different tests in automake
Azat Khuzhin [Fri, 20 Jan 2017 13:29:19 +0000 (16:29 +0300)]
test: register different tests in automake

Before this patch we have one test.sh (well test-script.sh), and tooks
very long to run it sequentially, but they are pretty lightweight, so we
should run then in parallel.

7 years agotest: run different tests under different options (in a wrapper)
Azat Khuzhin [Fri, 20 Jan 2017 13:29:02 +0000 (16:29 +0300)]
test: run different tests under different options (in a wrapper)

7 years agoautomake: do not use serial-tests if parallel-test-harness available
Azat Khuzhin [Sun, 15 Jan 2017 09:31:27 +0000 (12:31 +0300)]
automake: do not use serial-tests if parallel-test-harness available

Revert: 61179dec7716975a151c6dae4524ae504d05780e ("automake: define
serial-tests only if automake have this option") Revert:
Revert: 44d755e00e249c09914cf35089c16735df82e270 ("test/automake: don't
use paralell test harness (since automake 1.12)")

7 years agoBump VERSION_INFO so that libraries symlinks will have the same version
Azat Khuzhin [Fri, 20 Jan 2017 18:24:54 +0000 (21:24 +0300)]
Bump VERSION_INFO so that libraries symlinks will have the same version

Fixes: f4489b8323cda0bf8eed7c1353911a852d7a947c ("Bump version to 2.1.7-beta
everywhere")

7 years agotest/dns: replace servname since solaris does not have "http"
Azat Khuzhin [Thu, 19 Jan 2017 23:11:35 +0000 (02:11 +0300)]
test/dns: replace servname since solaris does not have "http"

Yes, by default solaris (solaris10.dev 5.10 Generic_147148-26 i86pc i386 i86pc)
does not have "http" in /etc/services

7 years agoMerge branch 'openssl-filter-fixes-v4'
Azat Khuzhin [Thu, 19 Jan 2017 17:53:18 +0000 (20:53 +0300)]
Merge branch 'openssl-filter-fixes-v4'

* openssl-filter-fixes-v4:
  be: fix with filtered bufferevents and connect() without EAGAIN
  test/ssl: fix bufferevent_getfd() for bufferevent_openssl_filter_new()
  be_openssl: Fix writing into filted openssl bufferevent after connected
  test/https: separate cases for https client with filtered openssl bufferevent
  test/ssl: cover case when we writing to be_openssl after connecting

Fixes: #428
7 years agobe: fix with filtered bufferevents and connect() without EAGAIN
Azat Khuzhin [Sun, 15 Jan 2017 23:31:54 +0000 (02:31 +0300)]
be: fix with filtered bufferevents and connect() without EAGAIN

With filtered bufferevents (i.e. not real one, that have socket), we can
trigger incorrect callback in this case. Let's look at example with http
and bufferevent_openssl_filter_new():
- bev = bufferevent_openssl_filter_new()
- http layer trying to connect() to localhost with bev
  # at this time, bev have writecb/readcb NULL but ev_write/ev_read has
  # timeout with 45 secs, default HTTP connect timeout
- and when connect() retruns without EAGAIN (BSD'ism) we called
  event_active() before (with EV_WRITE), and this will call ev_write
  timeout only, while it is more correct to act on bufferevent instead
  of plain event, so let's trigger EV_WRITE for bufferevent which will
  do the job (and let's do this deferred).

Fixes: http/https_simple # under solaris