Fix undefined behaviour and application crash that might take
place in some rare cases after calling evdns_base_free when
there are requests in the waiting queue.
Current cleanup procedure in evdns_base_free_and_unlock
function includes 2 steps:
1. Finish all inflight requests.
2. Finish all waiting requests.
During the first step we iterate over each list in req_heads
structure and finish all requests in these lists. With current
logic finishing an inflight request (function request_finished)
removes it from the inflight requests container and forces
a wating connection to be sent (by calling
evdns_requests_pump_waiting_queue). When these new requests are
sent it is possible that they will be inserted to the list in
req_heads that we've already cleaned.
So in some cases container of the inflight requests is not empty
after this procedure and some requests are not finished and
deleted. When timeouts for these requests expire
evdns_request_timeout_callback is called but corresponding
evdns_base has been already deleted which causes undefined
behaviour and possible applicaton crash.
It is interesting to note that in old versions of libevent such
situation was not possible. This bug was introduced by the commit
14f84bbdc77d90b1d936076661443cdbf516c593. Before this commit
nameservers were deleted before finishing the requests. Therefore
it was not possible that requests from the waiting queue be sent
while we finish the inflight requests.
(cherry picked from commit 4da9f87ccbe71edb3b3aaf74b8b64d7e9c41dcaf)
Do not do any assumptions on the error for the EMFILE from
getaddrinfo(), expect just any error.
Fixes: #924
(cherry picked from commit 4436287d1247fb2e9c80560debf852a94657e485)
This will allow to customize SO_RCVBUF/SO_SNDBUF for nameservers in this
evdns_base, you may want to adjust them if the kernel starts dropping
udp packages.
(cherry picked from commit 538141eb7e590bc94c043b43b5e5483b13bc9c5e)
To avoid possible confusion
But there is still one test that has some messages on windows:
main/methods
Because this test needs >1 of avaiable methods, otherwise it will warn.
(cherry picked from commit 47d348a63130c91f2a6aadef291ff5687275df72)
- DNS_OPTION_NAMESERVERS_NO_DEFAULT
Do not "default" nameserver (i.e. "127.0.0.1:53") if there is no nameservers
in resolv.conf, (iff DNS_OPTION_NAMESERVERS is set)
- EVDNS_BASE_NAMESERVERS_NO_DEFAULT
If EVDNS_BASE_INITIALIZE_NAMESERVERS isset, do not add default
nameserver if there are no nameservers in resolv.conf (just set
DNS_OPTION_NAMESERVERS_NO_DEFAULT internally)
Fixes: #569
(cherry picked from commit e5b8f4c1925867d8e4cd7dc0390e5141d7ef1106)
* regress-dns-fixes:
Do not rely on getservbyname() for most of the dns regression tests
Turn off dns/getaddrinfo_race_gotresolve by default
Fix an error for debug locking in dns/getaddrinfo_race_gotresolve
(cherry picked from commit bd2184d8d42f721fa523d01b9a9250c75fedb855)
Signed-off-by: Philip Prindeville <philipp@redfish-solutions.com>
Closes: #646 # cherry-picked from the PR
(cherry picked from commit be371163027d3988a30620537bf3fcb81fb7f077)
Otherwise poll() will fail with EINVAL:
EINVAL The nfds value exceeds the RLIMIT_NOFILE value.
P.S. and cleanup this test a little, with early-return.
CI: https://travis-ci.org/libevent/libevent/jobs/370350426
(cherry picked from commit d1c8993c3c34030b68ecb6079bd9dce8cb3d3604)
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
(cherry picked from commit d84f0205453941235b0e04729098d4329c189bba)
If system resolver (sync one) will respond too slow, then we can fail async
request and evdns will retransmit tham again, but evdns server will accept that
failed requets, so we will have not 2 requests but 4.
Reproduced on centos box sometimes.
In case when evdns_base_free() called with @fail_requests, we can potentially
have leaks, but we can avoid them if we will run event loop once again to
trigger defer cbs, so let's do this, instead of magical decrements (and also
this will give an example how to avoid leaks for evdns).
Because we don't cancel request,
and want our callback to recieve DNS_ERR_SHUTDOWN,
we use deferred callback, and there was
- one extra malloc(),
@see reply_schedule_callback()
- and one missing free
@see request_finished() (req->handle->pending_cb = 1)
than we don't need to count in testleak_cleanup()
So just decrement allocated_chunks to 2,
like we already take care about it.
The fix was easy: the tests that were failing were those tests that
also enabled debug mode themselves. So, let them know when
EVENT_DEBUG_MODE is on, so they won't do that.
