]> granicus.if.org Git - libevent/log
libevent
9 years agoMerge branch 'be-openssl-more-common-with-be-sock'
Azat Khuzhin [Fri, 6 Nov 2015 07:36:18 +0000 (10:36 +0300)]
Merge branch 'be-openssl-more-common-with-be-sock'

* be-openssl-more-common-with-be-sock:
  be_openssl: use bufferevent_enable() instead of bufferevent_add_event_()
  be_sock: drop be_sock_add() macro (useless and debug unfriendly)
  be: introduce bufferevent_generic_adj_existing_timeouts_()
  be_openssl: don't add events during bev creation (like be_sock)
  be: add_event: use evutil_timerisset()

9 years agotest/regress: fix ssl-less builds (need to make this prettier)
Azat Khuzhin [Fri, 6 Nov 2015 07:13:13 +0000 (10:13 +0300)]
test/regress: fix ssl-less builds (need to make this prettier)

CI: https://travis-ci.org/libevent/libevent/jobs/89590839

9 years agobe_openssl: use bufferevent_enable() instead of bufferevent_add_event_()
Azat Khuzhin [Thu, 5 Nov 2015 14:56:07 +0000 (17:56 +0300)]
be_openssl: use bufferevent_enable() instead of bufferevent_add_event_()

By using bufferevent_enable() there will be no event for READ *or* WRITE if
they are not enabled before, and this patch reduces difference for
be_sock_enable/be_openssl_enable (handshake)

9 years agobe_sock: drop be_sock_add() macro (useless and debug unfriendly)
Azat Khuzhin [Thu, 5 Nov 2015 14:51:46 +0000 (17:51 +0300)]
be_sock: drop be_sock_add() macro (useless and debug unfriendly)

9 years agobe: introduce bufferevent_generic_adj_existing_timeouts_()
Azat Khuzhin [Thu, 5 Nov 2015 14:45:09 +0000 (17:45 +0300)]
be: introduce bufferevent_generic_adj_existing_timeouts_()

And use it in openssl/sock layers to avoid copy-pasting it's variants.

9 years agobe_openssl: don't add events during bev creation (like be_sock)
Azat Khuzhin [Thu, 5 Nov 2015 14:40:25 +0000 (17:40 +0300)]
be_openssl: don't add events during bev creation (like be_sock)

Using the following examples you can get changes between be_openssl and
be_sock:
$ function diff_addr()
{
    eval diff -u $(printf "<(strip_addr %s) " "$@")
}
$ function strip_addr()
{
    sed 's/0x[a-zA-Z0-9]*/0xFFFF/g' "$@"
}
$ EVENT_DEBUG_LOGGING_ALL= regress --verbose --no-fork +http/https_connection_retry 2> /tmp/https-retry.log >&2
$ EVENT_DEBUG_LOGGING_ALL= regress --verbose --no-fork +http/connection_retry 2> /tmp/http-retry.log >&2
$ diff_addr /tmp/http-retry.log /tmp/https-retry.log

9 years agobe: add_event: use evutil_timerisset()
Azat Khuzhin [Thu, 5 Nov 2015 14:35:17 +0000 (17:35 +0300)]
be: add_event: use evutil_timerisset()

9 years agoMerge branch 'https-coverage-v6'
Azat Khuzhin [Fri, 6 Nov 2015 06:45:55 +0000 (09:45 +0300)]
Merge branch 'https-coverage-v6'

This patchset adds some basic tests to cover some https cases:
$ regress +http/https_..
http/https_basic: [forking] OK
http/https_simple: [forking] OK
http/https_simple_dirty: [forking] OK
http/https_incomplete: [forking] OK
http/https_incomplete_timeout: [forking] OK
http/https_connection_retry: [forking] OK
http/https_connection_retry_conn_address: [forking] OK
7 tests ok.  (0 skipped)

But there are some leaks in http regression tests (like init_ssl() and others),
must be fixed by using custom setup routine.

* https-coverage-v6:
  test/http: allow dirty shutdown for ssl to fix https_incomplete
  test/http: https basic
  test/http: incomplete{,_timeout} for https
  test/http: add simplest test for http/https/https_dirty_shutdown
  test/http: https: retry coverage
  test/http: https server support (plus some helpers)
  test/http: more sanity checks
  test/ssl: export getkey()/getcert()/get_ssl_ctx()/init_ssl() for https

9 years agotest/http: allow dirty shutdown for ssl to fix https_incomplete
Azat Khuzhin [Thu, 5 Nov 2015 23:24:28 +0000 (02:24 +0300)]
test/http: allow dirty shutdown for ssl to fix https_incomplete

9 years agotest/http: https basic
Azat Khuzhin [Thu, 5 Nov 2015 16:12:58 +0000 (19:12 +0300)]
test/http: https basic

9 years agotest/http: incomplete{,_timeout} for https
Azat Khuzhin [Thu, 5 Nov 2015 16:08:30 +0000 (19:08 +0300)]
test/http: incomplete{,_timeout} for https

9 years agotest/http: add simplest test for http/https/https_dirty_shutdown
Azat Khuzhin [Thu, 5 Nov 2015 09:39:35 +0000 (12:39 +0300)]
test/http: add simplest test for http/https/https_dirty_shutdown

There is "basic" before, but it uses bufferevent's for creating requests for
http layer, while this test uses http layer for client requests too.

9 years agotest/http: https: retry coverage
Azat Khuzhin [Thu, 5 Nov 2015 08:17:07 +0000 (11:17 +0300)]
test/http: https: retry coverage

9 years agotest/http: https server support (plus some helpers)
Azat Khuzhin [Thu, 5 Nov 2015 07:58:41 +0000 (10:58 +0300)]
test/http: https server support (plus some helpers)

TODO:
- check BEV_OPT_DEFER_CALLBACKS separately

9 years agotest/http: more sanity checks
Azat Khuzhin [Thu, 5 Nov 2015 08:18:32 +0000 (11:18 +0300)]
test/http: more sanity checks

9 years agotest/ssl: export getkey()/getcert()/get_ssl_ctx()/init_ssl() for https
Azat Khuzhin [Thu, 5 Nov 2015 07:34:18 +0000 (10:34 +0300)]
test/ssl: export getkey()/getcert()/get_ssl_ctx()/init_ssl() for https

9 years agoFix lock leak in be_pair_flush() if flush type is BEV_NORMAL
Bill Vaughan [Fri, 24 Jul 2015 12:40:59 +0000 (05:40 -0700)]
Fix lock leak in be_pair_flush() if flush type is BEV_NORMAL

Fixes: bufferevent/bufferevent_pair_flush_normal
  $ regress --no-fork bufferevent/bufferevent_pair_flush_normal
  bufferevent/bufferevent_pair_flush_normal:
    FAIL ../test/regress_bufferevent.c:181: assert(bufferevent_pair_get_partner(bev1) == NULL): 0x177fa20 vs (nil)
    [bufferevent_pair_flush_normal FAILED]

Fixes #266

9 years agotest/regress_be: basic coverage bufferevent_flush() for pair/sock layers
Azat Khuzhin [Fri, 30 Oct 2015 15:38:21 +0000 (18:38 +0300)]
test/regress_be: basic coverage bufferevent_flush() for pair/sock layers

Refs #266

9 years agotest/regress_be: socket_filter_inactive: check bufferevent after creation
Azat Khuzhin [Fri, 30 Oct 2015 15:37:44 +0000 (18:37 +0300)]
test/regress_be: socket_filter_inactive: check bufferevent after creation

Fixes:
../test/regress_bufferevent.c: In function ‘test_bufferevent_socket_filter_inactive’:
../test/regress_bufferevent.c:1180:1: warning: label ‘end’ defined but not used [-Wunused-label]
 end:

9 years agoFree event queues even for recursive finalizers
Azat Khuzhin [Fri, 30 Oct 2015 10:34:30 +0000 (13:34 +0300)]
Free event queues even for recursive finalizers

For finalizers we can register yet another finalizer out from finalizer, and
iff finalizer will be in active_later_queue we can add finalizer to
activequeues, and we will have events in activequeues after event_base_free()
returns, which is not what we want (we even have an assertion for this).

A simple case is bufferevent with underlying (i.e. filters) in inactive queue.

Fixes: regress bufferevent/bufferevent_socket_filter_inactive
9 years agotest/regress_be: cover finalizers from inactive to active queue
Azat Khuzhin [Thu, 8 Oct 2015 22:40:02 +0000 (01:40 +0300)]
test/regress_be: cover finalizers from inactive to active queue

Right now this will fail with the next assertion:
$ regress --no-fork --verbose bufferevent/bufferevent_socket_filter_inactive
bufferevent/bufferevent_socket_filter_inactive: [err] ../event.c:862: Assertion TAILQ_EMPTY(&base->activequeues[i]) failed in event_base_free_
Aborted

9 years agobuffer_compat: fix comment -- we have EVBUFFER_EOL_ANY not EOL_STYLE_ANY
Azat Khuzhin [Thu, 29 Oct 2015 22:34:40 +0000 (01:34 +0300)]
buffer_compat: fix comment -- we have EVBUFFER_EOL_ANY not EOL_STYLE_ANY

9 years agotest/regress_buffer: fix clang compilation warnings
Azat Khuzhin [Thu, 8 Oct 2015 22:53:40 +0000 (01:53 +0300)]
test/regress_buffer: fix clang compilation warnings

  ../test/regress_buffer.c:201:12: warning: logical not is only applied to the left hand side of this comparison [-Wlogical-not-parentheses]
          tt_assert(!memcmp((char*)EVBUFFER_DATA(evb), "1/hello", 7) != 0);

9 years agotest/regress_http: fix compilation warnings (-Wmissing-field-initializers)
Azat Khuzhin [Thu, 8 Oct 2015 22:50:05 +0000 (01:50 +0300)]
test/regress_http: fix compilation warnings (-Wmissing-field-initializers)

9 years agotest/regress_dns: fix compilation warnings (-Wmissing-field-initializers/for)
Azat Khuzhin [Thu, 8 Oct 2015 22:43:52 +0000 (01:43 +0300)]
test/regress_dns: fix compilation warnings (-Wmissing-field-initializers/for)

I don't have an error for loop because gcc5 have --std=gnu11 by default.
We need some options-consistency for all versions/compilers and build systems
to avoid such patches.

Fixes: https://travis-ci.org/libevent/libevent/jobs/84403473
Fixes: https://travis-ci.org/libevent/libevent/builds/84403463
9 years agoMerge branch 'evdns-randomization'
Azat Khuzhin [Thu, 8 Oct 2015 22:17:51 +0000 (01:17 +0300)]
Merge branch 'evdns-randomization'

* evdns-randomization:
  evdns: fix randomize-case by make case-insensitive as required
  tests/regress_dns: cover that randomize-case works case-insensitive
  evnds: inline TEST_NAME macro to make debuggin easier

Fixes #288

9 years agoevdns: fix randomize-case by make case-insensitive as required
Azat Khuzhin [Thu, 8 Oct 2015 22:13:08 +0000 (01:13 +0300)]
evdns: fix randomize-case by make case-insensitive as required

Interesting that this wasn't found by regression tests since they respond with
that SoME-rAndDom-CaSe domains, and no case-insensitive mode is required during
comparing response from the server and request.

Fixes #288
Covered-by: regress dns/search_lower
9 years agotests/regress_dns: cover that randomize-case works case-insensitive
Azat Khuzhin [Thu, 8 Oct 2015 22:11:09 +0000 (01:11 +0300)]
tests/regress_dns: cover that randomize-case works case-insensitive

Regression-for: #288

9 years agoevnds: inline TEST_NAME macro to make debuggin easier
Azat Khuzhin [Thu, 8 Oct 2015 21:03:47 +0000 (00:03 +0300)]
evnds: inline TEST_NAME macro to make debuggin easier

9 years agomake test/regress_ssl.c compile without warnings
Thomas Bernard [Mon, 5 Oct 2015 10:37:41 +0000 (12:37 +0200)]
make test/regress_ssl.c compile without warnings

9 years agoevthread: fix evthread_setup_global_lock_() for debug-lock with a real-lock case
Azat Khuzhin [Sun, 4 Oct 2015 00:49:38 +0000 (03:49 +0300)]
evthread: fix evthread_setup_global_lock_() for debug-lock with a real-lock case

Sample bt:
  (gdb) bt
  #0  0x00000000004a95d2 in evthread_setup_global_lock_ (lock_=0x0, locktype=0, enable_locks=1)
  #1  0x00000000004afbc7 in evsig_global_setup_locks_ (enable_locks=1) at signal.c:475
  #2  0x00000000004a5da8 in event_global_setup_locks_ (enable_locks=1) at event.c:3838
  #3  0x00000000004a8983 in evthread_set_lock_callbacks (cbs=0x7fffffffdde0) at evthread.c:129
  #4  0x0000000000445a8f in use_lock_unlock_profiler () at regress_bufferevent.c:306
  #5  0x0000000000445af2 in test_bufferevent_pair_release_lock (arg=0x70e710) at regress_bufferevent.c:327
  #6  0x000000000048c9a8 in testcase_run_bare_ (testcase=0x6fdc10 <bufferevent_testcases+80>) at tinytest.c:105
  #7  0x000000000048cc8a in testcase_run_one (group=0x6ff270 <testgroups+112>, testcase=0x6fdc10 <bufferevent_testcases+80>) at tinytest.c:252
  #8  0x000000000048d5c9 in tinytest_main (c=3, v=0x7fffffffe008, groups=0x6ff200 <testgroups>) at tinytest.c:434
  #9  0x000000000047a08f in main (argc=3, argv=0x7fffffffe008) at regress_main.c:459

Found-with: regress bufferevent/bufferevent_pair_release_lock (-DEVENT__DISABLE_DEBUG_MODE=ON)

9 years agoFix checking for make_base_notifiable()
Azat Khuzhin [Sun, 4 Oct 2015 00:35:43 +0000 (03:35 +0300)]
Fix checking for make_base_notifiable()

Fixes: a068f2e5 ("event_debug_created_threadable_ctx_: fix compilation without
debug mode")
Found-after: 3e56da23 ("travis: add builds without debug mode into matrix")

9 years agotest/regress_be: drop debug __asm__(int3) to fix arm build
Azat Khuzhin [Sun, 4 Oct 2015 00:19:12 +0000 (03:19 +0300)]
test/regress_be: drop debug __asm__(int3) to fix arm build

Closes #284

9 years agotravis: add builds without debug mode into matrix
Azat Khuzhin [Wed, 30 Sep 2015 08:13:44 +0000 (11:13 +0300)]
travis: add builds without debug mode into matrix

To cover compilation with this flag, since brew for example uses it by default.

9 years agoevent_debug_created_threadable_ctx_: fix compilation without debug mode
Azat Khuzhin [Tue, 29 Sep 2015 17:36:39 +0000 (20:36 +0300)]
event_debug_created_threadable_ctx_: fix compilation without debug mode

The following command failed before:
$ ./configure --disable-debug-mode

Fixes: dcfb19a27b7760299bc9e7291c9abd88c59fd91a ("Debug mode option to error on
evthread init AFTER other event calls.")

9 years agotest: fix bufferevent/bufferevent_pair_release_lock in debug mode
Azat Khuzhin [Thu, 10 Sep 2015 10:38:58 +0000 (13:38 +0300)]
test: fix bufferevent/bufferevent_pair_release_lock in debug mode

After this test had been fixed for freebsd the debug build was broken because
we can't call evthread_set_lock_callbacks() when something already initialized,
and we can't call event_base_free() (in kqueue case) when it is initialized,
because of "held_by", but this only playing role during freeing lock profiler
so reset lock callbacks there before and this will fix both.

Fixes: 79f9ace4ae8a259a5cf1b4ff3869078b60ff16a1 ("test: fix
bufferevent/bufferevent_pair_release_lock for freebsd")

P.S. after this patch 'make verify' finishes without errors on freebsd.

9 years agotest: fix bufferevent/bufferevent_pair_release_lock for freebsd
Azat Khuzhin [Thu, 10 Sep 2015 09:39:20 +0000 (12:39 +0300)]
test: fix bufferevent/bufferevent_pair_release_lock for freebsd

On FreeBSD with kqueue there is a call to evthread_debug_lock_mark_unlocked()
during event_base_free(), that will fail with an assert because of unmatched
"held_by", fix this by reseting lock callbacks to NULL before
event_base_free().

Trace:
  bufferevent/bufferevent_pair_release_lock: [warn] Trying to disable lock functions after they have been set up will probaby not work.
  [warn] Trying to disable lock functions after they have been set up will probaby not work.

    FAIL libevent/test/regress_bufferevent.c:259: lock: lock error[err] libevent/evthread.c:277: Assertion lock->held_by == me failed in evthread_debug_lock_mark_unlocked
  [New Thread 802006400 (LWP 100070/regress)]

  Program received signal SIGABRT, Aborted.
  [Switching to Thread 802006400 (LWP 100070/regress)]
  0x000000080167d6ca in thr_kill () from /lib/libc.so.7
  (gdb) bt
  #0  0x000000080167d6ca in thr_kill () from /lib/libc.so.7
  #1  0x0000000801752149 in abort () from /lib/libc.so.7
  #2  0x00000000004dff44 in event_exit (errcode=-559030611) at libevent/log.c:105
  #3  0x00000000004e053c in event_errx (eval=-559030611, fmt=0x5182cc "%s:%d: Assertion %s failed in %s") at libevent/log.c:162
  #4  0x00000000004d9954 in evthread_debug_lock_mark_unlocked (mode=0, lock=0x802017060) at libevent/evthread.c:277
  #5  0x00000000004d909a in debug_lock_unlock (mode=0, lock_=0x802017060) at libevent/evthread.c:290
  #6  0x00000000004e132c in evsig_dealloc_ (base=0x80201e300) at libevent/signal.c:434
  #7  0x00000000004e36c1 in kq_dealloc (base=0x80201e300) at libevent/kqueue.c:435
  #8  0x00000000004c9a44 in event_base_free_ (base=0x80201e300, run_finalizers=1) at libevent/event.c:855
  #9  0x00000000004c931a in event_base_free (base=0x0) at libevent/event.c:887
  #10 0x0000000000452657 in lock_unlock_free_thread_cbs () at libevent/test/regress_bufferevent.c:279
  #11 0x0000000000452621 in free_lock_unlock_profiler (data=0x8020170a0) at libevent/test/regress_bufferevent.c:317
  #12 0x000000000044bc8f in test_bufferevent_pair_release_lock (arg=0x8020170a0) at libevent/test/regress_bufferevent.c:334
  #13 0x00000000004b2288 in testcase_run_bare_ (testcase=0x737660) at libevent/test/tinytest.c:105
  #14 0x00000000004b1e72 in testcase_run_one (group=0x738c90, testcase=0x737660) at libevent/test/tinytest.c:252
  #15 0x00000000004b2930 in tinytest_main (c=3, v=0x7fffffffead0, groups=0x738c20) at libevent/test/tinytest.c:434
  #16 0x00000000004982fe in main (argc=3, argv=0x7fffffffead0) at libevent/test/regress_main.c:459
  (gdb) f 4
  #4  0x00000000004d9954 in evthread_debug_lock_mark_unlocked (mode=0, lock=0x802017060) at libevent/evthread.c:277
  277                     EVUTIL_ASSERT(lock->held_by == me);
  Current language:  auto; currently minimal
  (gdb) p lock
  $1 = (struct debug_lock *) 0x802017060
  (gdb) p lock->held_by
  $2 = 0
  (gdb) p me
  $3 = 34393318400

9 years agoMerge branch 'cloudabi-fixes-pr-278'
Azat Khuzhin [Thu, 10 Sep 2015 09:17:39 +0000 (12:17 +0300)]
Merge branch 'cloudabi-fixes-pr-278'

Rebased version of #278

* cloudabi-fixes-pr-278:
  Assume that ke_udata is an integer type on CloudABI.
  Add missing include of <netinet/in.h>.
  Include <sys/ioctl.h>, <sys/resource.h> and <sys/wait.h> optionally.

9 years agoAssume that ke_udata is an integer type on CloudABI.
Ed Schouten [Tue, 25 Aug 2015 21:31:07 +0000 (23:31 +0200)]
Assume that ke_udata is an integer type on CloudABI.

9 years agoAdd missing include of <netinet/in.h>.
Ed Schouten [Wed, 26 Aug 2015 07:45:23 +0000 (09:45 +0200)]
Add missing include of <netinet/in.h>.

POSIX states that sockaddr_in6 should be declared in <netinet/in.h>. By
including this header, the code now builds on FreeBSD and CloudABI.

9 years agoInclude <sys/ioctl.h>, <sys/resource.h> and <sys/wait.h> optionally.
Ed Schouten [Wed, 26 Aug 2015 08:06:09 +0000 (10:06 +0200)]
Include <sys/ioctl.h>, <sys/resource.h> and <sys/wait.h> optionally.

Though CloudABI implements a very large part of POSIX, it does not
provide these header files, for the reason that there is no raw device
access, no resource limiting and no access to the global process table
through wait().

It looks like these header files are not actually needed in theory.
There don't seem to be any constructs in these source files that use
these features, but I suspect they might still be required on some
systems.

9 years agoAdd a prototype for event_disable_debug_mode()
Sebastian Hahn [Thu, 13 Aug 2015 13:04:14 +0000 (15:04 +0200)]
Add a prototype for event_disable_debug_mode()

Exporting this function seems rather useless, as debugging mode has to
be enabled early and calling this function doesn't allow it to get
toggled back on later.

Fixes a compile warning when using clang 3.6.

Fixes #271
Fixes #224

9 years agohttp: export evhttp_connection_set_family()
Azat Khuzhin [Thu, 10 Sep 2015 08:46:17 +0000 (11:46 +0300)]
http: export evhttp_connection_set_family()

Fixes #176

9 years agotest/regress_be: bufferevent_enable() shouldn't call eventcb by it's own
Azat Khuzhin [Thu, 10 Sep 2015 08:18:20 +0000 (11:18 +0300)]
test/regress_be: bufferevent_enable() shouldn't call eventcb by it's own

It must enter the event loop regardless BEV_OPT_DEFER_CALLBACKS, to avoid
potential errors with subsequent connect(), you will find more info in #43,
since this is a regression for it.

9 years agotest/regress_be: introduce fake_listener_create()
Azat Khuzhin [Wed, 9 Sep 2015 22:17:50 +0000 (01:17 +0300)]
test/regress_be: introduce fake_listener_create()

9 years agoMerge branch 'evhttp-request-own-fixes'
Azat Khuzhin [Wed, 9 Sep 2015 16:31:48 +0000 (19:31 +0300)]
Merge branch 'evhttp-request-own-fixes'

I hope that I validated all places where evhttp_request_free() called and
ownership of request can be belong to user-specific code.

* evhttp-request-own-fixes:
  http: fix evhttp_request_own() by checking EVHTTP_USER_OWNED in more cases
  test/regress_http: cover evhttp_request_own()

Fixes #68

9 years agohttp: fix evhttp_request_own() by checking EVHTTP_USER_OWNED in more cases
Azat Khuzhin [Wed, 9 Sep 2015 16:21:51 +0000 (19:21 +0300)]
http: fix evhttp_request_own() by checking EVHTTP_USER_OWNED in more cases

Suggested-by: @ellzey
Fixes: http/request_own
Fixes: #68
9 years agotest/regress_http: cover evhttp_request_own()
Azat Khuzhin [Wed, 9 Sep 2015 16:15:18 +0000 (19:15 +0300)]
test/regress_http: cover evhttp_request_own()

9 years agoMerge branch 'http-server-EOF-at-read-v3'
Azat Khuzhin [Wed, 9 Sep 2015 15:08:02 +0000 (18:08 +0300)]
Merge branch 'http-server-EOF-at-read-v3'

This patchset fixes an issue with detecting EOF in http server and add some
regression tests to make sure that we will not break anything.

* http-server-EOF-at-read-v3:
  http: fix detecting EOF without write
  test/regress_http: cover write during read
  test/regress_http: verify that closecb will be called without multiple write

Closes #78
Link: http://stackoverflow.com/questions/18743948/libevent2-http-server-how-to-detect-client-close/18776728?noredirect=1#18776728
9 years agohttp: fix detecting EOF without write
Azat Khuzhin [Tue, 8 Sep 2015 13:40:55 +0000 (16:40 +0300)]
http: fix detecting EOF without write

Before this patch http server don't knows when client disconnected until it
will try to write to it, IOW to detect is client still alive you need to write
something to client socket, however it is not convenient since it requires to
store all clients somewhere and poll them periodically, and I don't see any
regressions if we will leave EV_READ always (like libevhtp do), since we
already reset read callback in evhttp_write_buffer() (see
http/write_during_read).

Also since we don't disable EV_READ anymore we don't need some enable EV_READ,
so we will reduce number of epoll_ctl() calls.

Covered-by: http/terminate_chunked_oneshot
Covered-by: http/write_during_read
Fixes: #78
9 years agotest/regress_http: cover write during read
Azat Khuzhin [Wed, 9 Sep 2015 14:45:44 +0000 (17:45 +0300)]
test/regress_http: cover write during read

This is the regression for evhttp_write_buffer() where we reset readcb to avoid
illegal state:
http/write_during_read: [err] evhttp_read_cb: illegal connection state 7

If you will comment that this test will fail.

9 years agotest/regress_http: verify that closecb will be called without multiple write
Azat Khuzhin [Tue, 8 Sep 2015 12:44:13 +0000 (15:44 +0300)]
test/regress_http: verify that closecb will be called without multiple write

And now this works incorrect, i.e. http layer will not detect EOF until another
write.

Reported-in: #78

9 years agotest: run regress with EVENT_DEBUG_MODE=1 and without
Azat Khuzhin [Wed, 2 Sep 2015 08:51:00 +0000 (11:51 +0300)]
test: run regress with EVENT_DEBUG_MODE=1 and without

This could show some bugs like "event_assign called on an already added event",
and some others that debugging mode could track.

9 years agotest/regress: fix bufferevent_pair_release_lock with EVENT_DEBUG_MODE
Azat Khuzhin [Wed, 2 Sep 2015 08:55:14 +0000 (11:55 +0300)]
test/regress: fix bufferevent_pair_release_lock with EVENT_DEBUG_MODE

Before this patch you will see next error:
$ EVENT_DEBUG_MODE= regress --no-fork bufferevent/bufferevent_pair_release_lock
bufferevent/bufferevent_pair_release_lock: [err] evthread initialization must be called BEFORE anything else!

9 years agosample/https-client: add -timeout option
Azat Khuzhin [Wed, 2 Sep 2015 16:34:52 +0000 (19:34 +0300)]
sample/https-client: add -timeout option

9 years agoMerge branch 'be-openssl-drop-fd_is_set-v4'
Azat Khuzhin [Wed, 2 Sep 2015 16:34:15 +0000 (19:34 +0300)]
Merge branch 'be-openssl-drop-fd_is_set-v4'

This must fix some issues with bufferevent openssl layer + some cleanups.

* be-openssl-drop-fd_is_set-v4:
  be_openssl: don't use *_auto() in do_handshake() we can't have fd == -1 there
  be_openssl: don't call set_open_callbacks() if fd == -1
  be_openssl: introduce be_openssl_auto_fd() helper
  be_openssl: introduce set_open_callbacks_auto()
  be_openssl: get rid off hackish "fd_is_set", to fix some corner cases

9 years agobe_openssl: don't use *_auto() in do_handshake() we can't have fd == -1 there
Azat Khuzhin [Wed, 2 Sep 2015 12:09:24 +0000 (15:09 +0300)]
be_openssl: don't use *_auto() in do_handshake() we can't have fd == -1 there

9 years agobe_openssl: don't call set_open_callbacks() if fd == -1
Azat Khuzhin [Wed, 2 Sep 2015 11:19:16 +0000 (14:19 +0300)]
be_openssl: don't call set_open_callbacks() if fd == -1

This must be illegal, firstly we must do set_do handshake and only after this
we could read/write.

9 years agobe_openssl: introduce be_openssl_auto_fd() helper
Azat Khuzhin [Wed, 2 Sep 2015 09:42:29 +0000 (12:42 +0300)]
be_openssl: introduce be_openssl_auto_fd() helper

9 years agobe_openssl: introduce set_open_callbacks_auto()
Azat Khuzhin [Wed, 2 Sep 2015 09:40:26 +0000 (12:40 +0300)]
be_openssl: introduce set_open_callbacks_auto()

This will split cases when we need to extract fd (cases when we have fd==-1
passed to set_open_callbacks()), and cases when we mustn't have to do this --
SET_FD via be_openssl_ctrl().

9 years agobe_openssl: get rid off hackish "fd_is_set", to fix some corner cases
Azat Khuzhin [Sat, 22 Aug 2015 18:38:18 +0000 (21:38 +0300)]
be_openssl: get rid off hackish "fd_is_set", to fix some corner cases

This patch is a cleanup and a bug fix, it drops ```fd_is_set``` flag, and
replace it with some checks to event_initialized(), and now we will not call
event_assign() on already added event, plus we will delete event when we really
have to (this patch fixes the case when server is down, IOW before this patch
we will not call event_del() because ```fd_is_set``` was reset to 0) and this
will fix some issues with retries in http layer for ssl.

Reported-in: #258
Fixes: regress ssl/bufferevent_socketpair_timeout
Fixes: regress ssl/bufferevent_socketpair_timeout_freed_fd
9 years agoMerge branch 'test-regress-be-openssl-v2'
Azat Khuzhin [Wed, 2 Sep 2015 16:18:08 +0000 (19:18 +0300)]
Merge branch 'test-regress-be-openssl-v2'

This patchset is a bunch of regression tests for bufferevent openssl layer,
some of them already show some bugs, that bugs will be fixed in the next
patches.

* test-regress-be-openssl-v2:
  test/regress_ssl: check events fd/pending after timeout triggered
  test/regress_ssl: cover case when server didn't up (failed with timeout)
  test/regress_ssl: covert that we can't change fd with underlying
  test/regress_ssl: cover that events (read/write) at finish not pending
  test/regress_ssl: cover fd manipulations
  test/regress_ssl: convert open_ssl_bufevs() to mask
  test/regress_ssl: convert client/server to mask too
  test/regress_ssl: cover "allow_dirty_shutdown"
  test/regress_ssl: convert regress_bufferevent_openssl() to bitmask

9 years agotest/regress_ssl: check events fd/pending after timeout triggered
Azat Khuzhin [Wed, 2 Sep 2015 14:49:41 +0000 (17:49 +0300)]
test/regress_ssl: check events fd/pending after timeout triggered

In this case client can't connect to server, and this bring to the front some
bugs with assigning on already added events (because of ```fd_is_set``` stuff),
for more info see #258, since this is the reproducible for it.

9 years agotest/regress_ssl: cover case when server didn't up (failed with timeout)
Azat Khuzhin [Wed, 2 Sep 2015 14:36:20 +0000 (17:36 +0300)]
test/regress_ssl: cover case when server didn't up (failed with timeout)

9 years agotest/regress_ssl: covert that we can't change fd with underlying
Azat Khuzhin [Wed, 2 Sep 2015 14:17:43 +0000 (17:17 +0300)]
test/regress_ssl: covert that we can't change fd with underlying

9 years agotest/regress_ssl: cover that events (read/write) at finish not pending
Azat Khuzhin [Wed, 2 Sep 2015 14:06:51 +0000 (17:06 +0300)]
test/regress_ssl: cover that events (read/write) at finish not pending

9 years agotest/regress_ssl: cover fd manipulations
Azat Khuzhin [Wed, 2 Sep 2015 09:35:51 +0000 (12:35 +0300)]
test/regress_ssl: cover fd manipulations

9 years agotest/regress_ssl: convert open_ssl_bufevs() to mask
Azat Khuzhin [Wed, 2 Sep 2015 09:31:15 +0000 (12:31 +0300)]
test/regress_ssl: convert open_ssl_bufevs() to mask

9 years agotest/regress_ssl: convert client/server to mask too
Azat Khuzhin [Wed, 2 Sep 2015 09:26:40 +0000 (12:26 +0300)]
test/regress_ssl: convert client/server to mask too

9 years agotest/regress_ssl: cover "allow_dirty_shutdown"
Azat Khuzhin [Wed, 2 Sep 2015 08:34:13 +0000 (11:34 +0300)]
test/regress_ssl: cover "allow_dirty_shutdown"

9 years agotest/regress_ssl: convert regress_bufferevent_openssl() to bitmask
Azat Khuzhin [Wed, 2 Sep 2015 08:22:43 +0000 (11:22 +0300)]
test/regress_ssl: convert regress_bufferevent_openssl() to bitmask

9 years agoTest against SO_REUSEADDR (along with _WIN32).
Ed Schouten [Tue, 25 Aug 2015 13:29:37 +0000 (15:29 +0200)]
Test against SO_REUSEADDR (along with _WIN32).

This makes the code build on other systems that also don't have
SO_REUSEADDR without requiring special code.

[ azat: partially revert WIN32 check since SO_REUSEADDR on win32 differs from
unix semantics. ]

Closes #275

9 years agotests/regress_ssl: drop duplicated assert
Azat Khuzhin [Tue, 25 Aug 2015 14:01:42 +0000 (17:01 +0300)]
tests/regress_ssl: drop duplicated assert

9 years agoAlways define missing TAILQ functions from sys/queue.h
Christopher Wiley [Fri, 24 Jul 2015 15:42:06 +0000 (08:42 -0700)]
Always define missing TAILQ functions from sys/queue.h

On both Android and Ubuntu machines, TAILQ_END is not defined.  This
header also does not seem to be part of standard BSD sys/queue.h

Fix this by always defining missing TAILQ_ macros rather than
conditioning on a particular macro.

Fixes #267

9 years agoDon't use BSD u_* types.
Ed Schouten [Tue, 25 Aug 2015 13:24:39 +0000 (15:24 +0200)]
Don't use BSD u_* types.

These types are not part of POSIX. As we only use them in a small number
of places, we'd better replace them by C standard types. This makes a
larger part of the code build for CloudABI.

9 years agoRemove BSD-ism: TIMEVAL_TO_TIMESPEC().
Ed Schouten [Tue, 25 Aug 2015 13:32:25 +0000 (15:32 +0200)]
Remove BSD-ism: TIMEVAL_TO_TIMESPEC().

Systems like CloudABI implement kqueue() but do not provide the
BSD-specific TIMEVAL_TO_TIMESPEC() macro. Change the code to perform
this manually, as it is not hard to do this conversion.

9 years agobe: include all variations of headers for sockaddr_in6 struct
Azat Khuzhin [Wed, 19 Aug 2015 20:27:51 +0000 (23:27 +0300)]
be: include all variations of headers for sockaddr_in6 struct

Fixes: c42bc6b279fd7cd4bc9b04c86a12e2ec5512cce1 ("be: fix sockaddr_in6 type
definition for win32")
Fixes: e04555155816306f23829a15b747664a57330c28 ("Merge branch
'reuse-conn_address-on-retry-v11'")
CI: https://ci.appveyor.com/project/nmathewson/libevent/build/2.1.5.109
Tested: manually

9 years agohttp: eliminate warning about "socklen" in evhttp_connection_connect_()
Azat Khuzhin [Wed, 19 Aug 2015 20:00:49 +0000 (23:00 +0300)]
http: eliminate warning about "socklen" in evhttp_connection_connect_()

9 years agobe: fix sockaddr_in6 type definition for win32
Azat Khuzhin [Wed, 19 Aug 2015 19:52:22 +0000 (22:52 +0300)]
be: fix sockaddr_in6 type definition for win32

bufferevent-internal.h:217:23: error: field 'in6' has incomplete type
   struct sockaddr_in6 in6;
                       ^
CI: https://ci.appveyor.com/project/nmathewson/libevent/build/2.1.5.108

9 years agosample/https-client: don't try to free uninitialized SSL
Azat Khuzhin [Wed, 19 Aug 2015 10:22:49 +0000 (13:22 +0300)]
sample/https-client: don't try to free uninitialized SSL

sample/https-client.c:459:3: warning: ‘ssl’ may be used uninitialized in this function [-Wmaybe-uninitialized]
   SSL_free(ssl);

9 years agosample/https-client: graceful exit with freeing memory (to make valgrind happy)
Azat Khuzhin [Wed, 19 Aug 2015 10:19:41 +0000 (13:19 +0300)]
sample/https-client: graceful exit with freeing memory (to make valgrind happy)

9 years agohttps-client: correctly handle URLs with no path (like "https://host:port")
Andrey Skriabin [Thu, 2 Apr 2015 17:42:54 +0000 (20:42 +0300)]
https-client: correctly handle URLs with no path (like "https://host:port")

path == NULL check removed

Fixes: #233
Fixes: #234
9 years agosample/http-connect: don't use assert() to make it work with NDEBUG
Azat Khuzhin [Tue, 18 Aug 2015 18:57:28 +0000 (21:57 +0300)]
sample/http-connect: don't use assert() to make it work with NDEBUG

Otherwise it will not work for cmake release target type.

9 years agotest/regress_http: initialize "dns_base" to avoid reading trash
Azat Khuzhin [Tue, 18 Aug 2015 18:30:48 +0000 (21:30 +0300)]
test/regress_http: initialize "dns_base" to avoid reading trash

Even though it is very unlikely, but it will better to fix this.

9 years agosample/http-connect: made it compatible with C90
Azat Khuzhin [Tue, 18 Aug 2015 18:29:03 +0000 (21:29 +0300)]
sample/http-connect: made it compatible with C90

9 years agoMerge branch 'reuse-conn_address-on-retry-v11'
Azat Khuzhin [Tue, 18 Aug 2015 18:19:11 +0000 (21:19 +0300)]
Merge branch 'reuse-conn_address-on-retry-v11'

There is regression tests and also this code worked/tested during crawling a
huge number of pages (billions).

* reuse-conn_address-on-retry-v11:
  be_sock: bufferevent_socket_set_conn_address(): assert instead of silent no-op
  http: reuse connected address only with EVHTTP_CON_REUSE_CONNECTED_ADDR
  be_sock: sanity check in bufferevent_socket_set_conn_address()
  be: replace sockaddr_storage with sockaddr_in6 for conn_address
  be: we don't need to use getpeername() we we have conn_address
  be: replace conn_address by full struct instead of pointer
  test/http: cover retrying with saved conn_address by shutting down dns server
  http: use IP address that we got before (if any) during retrying
  bufferevent: move conn_address out from http into bufferevent
  be: make @sa const for bufferevent_socket_connect()
  util: make @sa const for evutil_socket_connect_()

9 years agobe_sock: bufferevent_socket_set_conn_address(): assert instead of silent no-op
Azat Khuzhin [Tue, 18 Aug 2015 17:41:16 +0000 (20:41 +0300)]
be_sock: bufferevent_socket_set_conn_address(): assert instead of silent no-op

9 years agohttp: reuse connected address only with EVHTTP_CON_REUSE_CONNECTED_ADDR
Azat Khuzhin [Thu, 1 Jan 2015 03:27:31 +0000 (06:27 +0300)]
http: reuse connected address only with EVHTTP_CON_REUSE_CONNECTED_ADDR

9 years agobe_sock: sanity check in bufferevent_socket_set_conn_address()
Azat Khuzhin [Tue, 2 Dec 2014 11:31:03 +0000 (14:31 +0300)]
be_sock: sanity check in bufferevent_socket_set_conn_address()

9 years agobe: replace sockaddr_storage with sockaddr_in6 for conn_address
Azat Khuzhin [Tue, 2 Dec 2014 11:24:19 +0000 (14:24 +0300)]
be: replace sockaddr_storage with sockaddr_in6 for conn_address

We need only ipv6/ipv4 for connect in bufferevent, and since
sockaddr_storage is pretty big (128 bytes) it will be better to use
sockaddr_in6 here (it will fit ipv4 too).

9 years agobe: we don't need to use getpeername() we we have conn_address
Azat Khuzhin [Sat, 15 Nov 2014 21:30:11 +0000 (00:30 +0300)]
be: we don't need to use getpeername() we we have conn_address

Since getpeername() could fail.

9 years agobe: replace conn_address by full struct instead of pointer
Azat Khuzhin [Sat, 15 Nov 2014 21:29:19 +0000 (00:29 +0300)]
be: replace conn_address by full struct instead of pointer

9 years agotest/http: cover retrying with saved conn_address by shutting down dns server
Azat Khuzhin [Sat, 15 Nov 2014 17:35:40 +0000 (20:35 +0300)]
test/http: cover retrying with saved conn_address by shutting down dns server

9 years agohttp: use IP address that we got before (if any) during retrying
Azat Khuzhin [Wed, 12 Nov 2014 17:23:46 +0000 (20:23 +0300)]
http: use IP address that we got before (if any) during retrying

Before this patch every time we are retrying our request we resolve
domain, but we could optimize this (since UDP is slow) by using cached
conn_address value, so do this.

9 years agobufferevent: move conn_address out from http into bufferevent
Azat Khuzhin [Sat, 15 Nov 2014 18:46:11 +0000 (21:46 +0300)]
bufferevent: move conn_address out from http into bufferevent

In http the only case when when we could store it is when we already
connected, *but* if we are doing request using domain name, then we need
to do request to nameserver to get IP address, and this is handled by
bufferevent.
So when we have IP address (from nameserver) and don't have connection
to this IP address, we could already cache it to avoid extra DNS
requests (since UDP is slow), and we can't do this from http layer, only
from bufferevent.

9 years agobe: make @sa const for bufferevent_socket_connect()
Azat Khuzhin [Tue, 2 Dec 2014 12:05:59 +0000 (15:05 +0300)]
be: make @sa const for bufferevent_socket_connect()

9 years agoutil: make @sa const for evutil_socket_connect_()
Azat Khuzhin [Tue, 2 Dec 2014 12:05:36 +0000 (15:05 +0300)]
util: make @sa const for evutil_socket_connect_()

9 years agosample: add HTTP CONNECT tunnelling example using libevent http layer
Azat Khuzhin [Tue, 18 Aug 2015 09:18:31 +0000 (12:18 +0300)]
sample: add HTTP CONNECT tunnelling example using libevent http layer

9 years agohttp: introduce evhttp_request_free_() helper
Azat Khuzhin [Tue, 18 Aug 2015 00:03:51 +0000 (03:03 +0300)]
http: introduce evhttp_request_free_() helper

9 years agohttp: introduce evhttp_is_request_connection_close() helper
Azat Khuzhin [Fri, 24 Apr 2015 16:04:51 +0000 (19:04 +0300)]
http: introduce evhttp_is_request_connection_close() helper