Original patch contributed by andrey.semashev@gmail.com.
In PhysicalSocketServer::WaitEpoll(), the loop verifies that the
signalled dispatcher is in dispatchers_ set. It does so by looking up
the dispatcher pointer in the set. This is vulnerable to the ABA
problem because one dispatcher may be removed and destroyed and another
created and added with the same address before epoll reports an event
for the old dispatcher. The same issue exists for other Wait
implementations, if a dispatcher is removed and a new one added with
the same socket handle is the old.
This is avoided by using a 64-bit key for looking up the dispatcher
in the set. The key is set from a running counter which gets incremented
when a dispatcher is added to the set, so even if the same dispatcher
pointer is added, removed and added again, the key value will be
different.
This changes the storage of dispatchers_ from a set to a flat_hash_map,
which uses a bit more memory but has faster lookup (O(1) as opposed to
O(log n)).
Bug: webrtc:11124
Change-Id: I6d206e1a367b58ba971edca9b48af7664384b797
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/181027
Commit-Queue: Taylor <deadbeef@webrtc.org>
Reviewed-by: Karl Wiberg <kwiberg@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32019}
This name change communicates that the recursive critical section
should not be used for new code.
The relevant files are renamed rtc_base/critical_section* ->
rtc_base/deprecated/recursive_critical_section*
Bug: webrtc:11567
Change-Id: I73483a1c5e59c389407a981efbfc2cfe76ccdb43
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/179483
Commit-Queue: Markus Handell <handellm@webrtc.org>
Reviewed-by: Niels Moller <nisse@webrtc.org>
Reviewed-by: Danil Chapovalov <danilchap@webrtc.org>
Reviewed-by: Tommi <tommi@webrtc.org>
Reviewed-by: Mirko Bonadei <mbonadei@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#31754}
This change deletes heap-based handling of dynamic number of epoll
events handled per call to epoll, with the assumption
that PSS load is likely not dominated by the epoll syscalls. This
simplifies the logic in the code and removes a heap allocation.
Bug: webrtc:11567
Change-Id: I34fbe1fa8bf0a037bf849a4adac1a0a333c9dd86
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/175563
Commit-Queue: Markus Handell <handellm@webrtc.org>
Reviewed-by: Karl Wiberg <kwiberg@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#31358}
This CL only includes the necessary changes in PhysicalSocketServer,
and doesn't include the Java or Objective C API.
Note that this is doing exactly the same thing as UDPSocketPosix
in chromium.
BUG=webrtc:5658
Change-Id: I295455eaccba2a83cdd1bc55848f325c310f8d32
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/168260
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Reviewed-by: Qingsi Wang <qingsi@webrtc.org>
Commit-Queue: Taylor <deadbeef@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#30478}
