SafeSocketHandle: avoid potential blocking of finalizer thread

[tmds]

When the Socket is Disposed, it attempts to make all on-going operations
abort. It does this in a loop, and decides there are no on-going operations
when the reference count becomes zero and the handle gets released.

SafePipeHandle holds a reference to SafeSocketHandle. This prevents the
reference count to drop to zero, and causes the dispose to loop infinitly.

When the Socket is disposed from the finalizer thread, it is no longer used
for operations and we can skip the loop. This avoids blocking the finalizer
thread when the reference count can't drop to zero.

Fixes #40301

@antonfirsov @geoffkizer ptal

cc @jaredpar

[ghost]

Tagging subscribers to this area: @dotnet/ncl
See info in area-owners.md if you want to be subscribed.

[antonfirsov]

I know it's a bit off-topic, but if we are already here, would make sense to remove comments about inner/outer handles in SafeSocketHandle.cs:

// It uses an inner and outer SafeHandle to do so. The inner
// SafeHandle holds the actual socket, but only ever has one // SafeHandle holds the actual socket, but only ever has one
// reference to it. The outer SafeHandle guards the inner // reference to it. The outer SafeHandle guards the inner
[.....]

[antonfirsov]

Just nits, should be good once a test is added.

[tmds]

Anton, thanks for reviewing! I've addressed your feedback. Socket tests pass on my machine.

[antonfirsov]

The OSX test failure is unrelated and tracked in #41511, the mono one seems to be a hang in System.Linq.Expressions.Tests.ArrayLengthTests.CheckByteArrayLengthTest:
https://helix.dot.net/api/2019-06-17/jobs/bac94c2c-2c76-4f75-9ee3-cfc1641cb18e/workitems/System.Linq.Expressions.Tests/console

[antonfirsov]

LGTM.

@geoffkizer if you are also fine with the changes and all tests do pass, maybe we can even do the packport today?

[antonfirsov]

Inner loop test failure is #41078

[antonfirsov]

Test failure seems related:

Unhandled Exception:
System.NullReferenceException: Object reference not set to an instance of an object
   at System.Net.Sockets.Socket.Dispose(Boolean disposing) in /_/src/libraries/System.Net.Sockets/src/System/Net/Sockets/Socket.cs:line 4262
   at System.Net.Sockets.Socket.Finalize() in /_/src/libraries/System.Net.Sockets/src/System/Net/Sockets/Socket.cs:line 4367
[ERROR] FATAL UNHANDLED EXCEPTION: System.NullReferenceException: Object reference not set to an instance of an object
   at System.Net.Sockets.Socket.Dispose(Boolean disposing) in /_/src/libraries/System.Net.Sockets/src/System/Net/Sockets/Socket.cs:line 4262
   at System.Net.Sockets.Socket.Finalize() in /_/src/libraries/System.Net.Sockets/src/System/Net/Sockets/Socket.cs:line 4367

[antonfirsov]

OuterLoop test failures seem to be unrelated: #40798 #41530 #41531 #41606 #41607

[geoffkizer]

We do have a test somewhere that ensures that the socket handle is closed when the Socket gets finalized, right?

This was the original problem that led us to adding logic to the Socket finalizer. We need to make sure we haven't broken that again.

[wfurt]

NonDisposedSocket_SafeHandlesCollected ?

[antonfirsov]

/backport to release/5.0

[github-actions]

Started backporting to release/5.0: https://github.com/dotnet/runtime/actions/runs/234344348

[antonfirsov]

/backport to release/5.0-rc2

[github-actions]

Started backporting to release/5.0-rc2: https://github.com/dotnet/runtime/actions/runs/236412175