* BeginConnect complete to early when not using callback.
The following commit, https://github.com/mono/mono/commit/068e8fe0424806753d7ab72a7b9bf0e54b58408b#diff-33b28a9dc0b8032650ed77213481a99f
changed the way we do DNS resolve in BeginConnect using host name. Instead
of doing a blocking DNS resolve, it switched to an async DNS resolve, followed
by an async connect. The returned IAsyncResult was just representing the
completion of the DNS resolve task and not the underlying connect. It
also turns out that part of doing EndConnect, we did some validation
of the used IAsyncResult making sure it was an instance of SocketAsyncResult,
meaning that if EndConnect was called on the returned IAsyncResult from
BeginConnect, EndConnect would throw and exception.
NOTE, this only happens if you use pattern like this:
var result = sock.BeginConnect (hostName, port, null, null);
result.AsyncWaitHandle.WaitOne (2000);
sock.EndConnect (result);
It did not happened if you used callbacks, since then EndConnect would
be called as part of callback and that IAsyncResult was of correct type,
or if you did a BeginConnect with an EndPoint or IPAddress[] since that
won't trigger the DNS resolve.
Fix is to make sure we hand out the SocketAsyncResult instance that will
be used in our internal connect calls from BeginConnect. We also make
sure, we chain the DNS resolve task together with a task doing the async
connect (using the same SocketAsyncResult instance). This will make sure
the IAsyncResult returned from BeginConnect is the same one as used in
internal connect call, so it will only complete when the connect call
completes.
* Adding test for BeginConnect using host name.
* Bump API snapshot submodule
Cleaning up the `SocketAsyncEventArgs` class and preparing it for it's eventual
adaption from CoreFX.
There are a few instance fields and property accessors with `internal` visibility,
which the CoreFX version does not have. Replaced these with accessor methods.
The `PolicyRestricted` property and it's related `SocketAsyncEventArgs (bool)`
constructor is not used anywhere and the CoreFX version has a constructor with the
same signature, but different semantics.
In the CoreFX version, `Complete ()` does not call `DoComplete (this)` - renamed it
into `Complete_internal ()` to avoid any ambiguities and help the transitioning.
* Merge the `System.Diagnostics/ProcessStartInfo.cs` part into the referencesource file to avoid CS0282.
* Remove duplicate `using`.
* Fix unreachable code warning that got introduced in my last PR.
We need to clearly distinguish between synchronous and asynchronous completions
and avoid any ambiguities and race conditions when it comes to error handling.
On synchronous completion, we must not invoke the async completion delegate.
In `System.Net.Sockets/Socket.cs`, the internal `BeginSConnect()` and `BeginMConnect()`
now return a `bool` indicating whether or not an async operation is pending.
If any error happens prior to starting the async operation, then we shall call
`SocketAsyncResult.Complete(..., true)` and return `false`. And since this is
a synchronous completion, it's `AsyncCallback` will not be invoked.
In `referencesource/System/net/System/Net/Sockets/_MultipleConnectAsync.cs`,
I compared the entire file against the CoreFX version and then copied the
`AttemptConnection()` method from their implementation to adapt their behavior.
The difference is that on synchronous completion, the `InternalConnectCallback(null, args)`
should be invoked.
Rework of #5448, contributes to #7143.
The anonymous pipes were tested on macOS, build was tested only on few configurations (hoping to solve some issue through Jenkins builds). Named pipe tests get stuck on macOS, but it may be unrelated Xunit issue and needs some proper debugging.
@Therzok, do you know of anything else that is missing for parallel MSBuild? Do you have some test environment for it by chance?
When a connection fails, the socket error was being incorrectly reported
as an address family not supported error, instead of connection refused.
Why did this happen? The order of addresses returned from
`Dns.GetHostAddresses` changed. The newer code returns the IPv6 address
first, and the IPv4 address second. While `GetCheckedIPs` mentioned in a
comment that it skips addresses that don't match the address family, it
did not. The code using the addresses returned by `GetCheckedIPs` will
use the first address in the array. If this address happens to match the
address family of the socket, things will work "properly" (the correct
socket error will occur).
If, however the first entry has the wrong address family, then the
socket error will be an address family not supported error, since the
socket was created with different address family.
This change modifies `GetCheckedIPs` to filter the addresses, removing
those which do not match the address family.
Follow up to https://github.com/mono/mono/pull/6431, we should
capture the exception that is passed to the methods.
Copied the SetResult() method that referencesource uses.
* [w32file] Remove io_ops indirection
* [w32file] Replace use of w32handle by a GHashTable for File, Console and Pipe
* [Socket] Avoid relying on MonoIO.DuplicateHandle
* [w32socket] Remove force closing of sockets on shutdown
* fixup! [Socket] Avoid relying on MonoIO.DuplicateHandle
* [w32socket] Replace use of w32handle by a GHashTable
* [w32handle] Remove dead function mono_w32handle_new_fd
* fixup! [w32file] Replace use of w32handle by a GHashTable for File, Console and Pipe
* [w32handle] Remove dead variable mono_w32handle_fd_reserve
* [fdhandle] Factor w32file and w32socket common fdhandle code
* fixup! [Socket] Avoid relying on MonoIO.DuplicateHandle
* [w32socket] Fix GC safe/unsafe transitions
* fixup! [w32handle] Remove dead variable mono_w32handle_fd_reserve
* [w32file] Remove io_ops indirection
* [w32file] Replace use of w32handle by a GHashTable for File, Console and Pipe
* [Socket] Avoid relying on MonoIO.DuplicateHandle
* [w32socket] Remove force closing of sockets on shutdown
* fixup! [Socket] Avoid relying on MonoIO.DuplicateHandle
* [w32socket] Replace use of w32handle by a GHashTable
* [w32handle] Remove dead function mono_w32handle_new_fd
* fixup! [w32file] Replace use of w32handle by a GHashTable for File, Console and Pipe
* [w32handle] Remove dead variable mono_w32handle_fd_reserve
* [fdhandle] Factor w32file and w32socket common fdhandle code
* fixup! [Socket] Avoid relying on MonoIO.DuplicateHandle
* [w32socket] Fix GC safe/unsafe transitions
* fixup! [w32handle] Remove dead variable mono_w32handle_fd_reserve
We would previously mix 2 SocketAsyncResult, and in case we would succeed asynchronously, we would return one that would not be updated afterwards.
The solution consists in using only 1 SocketAsyncResult that we ensure to update accordingly.
* less allocation in Socket
avoid allocate `System.Threading.QueueUserWorkItemCallback` in `SocketAsyncResult.Complete`
avoid allocate `Task` and anonymous function in `Socket.QueueIOSelectorJob`
* update: fix race and use replace static field with in line lambda
* cleanup code
* QueueIOSelectorJob keep async, and provie a fastpath
* SocketAsyncResult: simplify callback code
* add missing line `IOSelect.Add(handle, job)` :(
* fix typo
* fix compile error
It might be possible that the callback modifies the current SocketAsyncResult via the Init method. We observe it with AcceptAsync with a ReceiveAsync in the callback.
We decided it's a better name since we're using the profile to
test FullAOT scenarios on desktop (as opposed to the mobile profiles).
While doing the rename I saw that a lot of the #ifdefs were using
MOBILE_STATIC even though the code should actually be included/excluded
in all FullAOT scenarios, e.g. System.Reflection.Emit. I replaced those
cases with FULL_AOT_RUNTIME instead.
For cases where code is specific to this profile, I added a new
AOT_ONLY_DESKTOP define.
* Fixed issue 47425
This fix simply ensures that the `m_Handle` value is not processed by the GC after a call to `DuplicateAndClose` as that causes the duplicated socket to close.
A potential issue is that this will leak sockets if the sockets are never re-created.
* Implemented ref-counting as suggested by @ludovic-henry