In the same way the overload Receive (buf, flags) already does, mention
"time out" in the exception message when the socketError is WOULDBLOCK and
the socket is not non-blocking (otherwise it would throw with the reason
"Operation on non-blocking socket would block", which doesn't make sense
for a socket that is not non-blocking).
* Socket_2_1.cs: When closing a socket we need to abort any
pending syscalls to that socket. For most syscalls the kernel
does it for us. But accept/connect on linux requires extra
love.
We previously used Thread.Abort to implement that, but this
is a pretty radical solution with very complicated ramifications.
So instead now we use a new icall cancel_blocking_socket_operation
that makes sure we only abort the syscall and do nothing else.
This reduces the trouble surface a lot given how tricky
Thread.Abort is.
There are two types of checks that were being done before:
(1) offset < 0? size < 0? offset + size > buffer.Length? This is trivial to get around. If I make both offset and size positive but have the sum overflow, it will pass all three checks.
(2) offset in range? size in range? If the buffer is very large this might run into overflow issues with its checking, though fine most of the time...
Is the offset in range? And is the size in range against buffer.Length - offset (which we know to be >= 0 from the earlier offset check)? This method of bounds checking is not in any danger from two's complement overflows.
Windows/Linux flagged the socket as writable even when connect() has
not been called. On *BSD and OSX, this was not the case and we waited
forever.
Now, for blocking sockets, we make them blocking and call connect()
before adding the socket to the wait pool.
The code involved in a delegate BeginInvoke is too heavy and of no use
for socket asynchronous operations. With these changes, we bypass all
that and send the work items directly to the asynchronous IO queue,
improving performance and lowering memory use.
-Removed #if NET_2_0
-All the *Async() calls but ConnectAsync() are routed through the
asynch IO threadpool instead of creating a new short-lived thread for
every operation.
-When reusing a SocketAsyncEventArgs only 1 new object, a delegate, is
created.
We didn't keep the endpoint around for non-blocking connects and later,
when Poll() was called, we didn't have the information. Also 'isbound'
wasn't being set in a few places and UDP sockets needed to be marked as
not connected after a call to Connect() with an Any address.
Removed an old test for Connect(Any) with UDP sockets, since this is
used to disconnect them now.
Fixes bug #640749.
2010-03-19 Sebastien Pouliot <[email protected]>
* Timer.cs: Removed old NET_2_0 defines and added some new
MOONLIGHT defines (needed for the client stack)
In System.Net:
2010-03-19 Sebastien Pouliot <[email protected]>
* AuthenticationManager.cs: Internal type for Moonlight.
Add BasicClient, DigestClient, NtlmClient for NET_2_1 since
this applies to both monotouch and moonlight (client stack).
Removed unneeded NET_2_0 defines
* CookieContainer.cs: For Moonlight it needs to be internal in
System.dll and public in System.Net.dll. Removed unneeded NET_2_0
defines
* HttpContinueDelegate.cs: Internal type for Moonlight.
* HttpRequestCreator.cs: Remove a MOONLIGHT define
* HttpStatusCode.cs: For Moonlight it needs to be internal in
System.dll and public in System.Net.dll
* HttpWebRequest.cs: Internal type for Moonlight. Change S.Config
to NET_2_1 defines (applies to both monotouch and moonlight).
Removed unneeded NET_2_0 and NET_1_1 defines
* HttpWebResponse.cs: Internal type for Moonlight. Removed
unneeded NET_2_0 defines
* IAuthenticationModule.cs: Internal type for Moonlight.
* ICertificatePolicy.cs: Internal type for Moonlight.
* ICredentialPolicy.cs: Internal type for Moonlight. Removed
unneeded NET_2_0 defines
* IWebRequestCreate.cs: For Moonlight it needs to be internal in
System.dll and public in System.Net.dll
* ServicePoint.cs: Internal type for Moonlight. Change S.Config
to NET_2_1 defines (applies to both monotouch and moonlight)
* ServicePointManager.cs: Remove a MOONLIGHT define
* WebException.cs: For Moonlight it needs to be internal in
System.dll and public in System.Net.dll. Removed unneeded NET_2_0
defines
* WebHeaderCollection.cs: Internal type for Moonlight. Removed
unneeded NET_2_0 defines
* WebRequest.cs: Internal type for Moonlight. Change S.Config
to NET_2_1 defines (applies to both monotouch and moonlight)
* WebResponse.cs: Internal type for Moonlight. Removed unneeded
NET_2_0 defines
In System.Net.Sockets:
2010-03-19 Sebastien Pouliot <[email protected]>
* Socket.cs: Remove some NET_2_0 defines (using) and add a few
MOONLIGHT defines needed to reuse Socket in the client http stack
* Socket_2_1.cs: Remove some MOONLIGHT defines
* SocketAsyncEventArgs.cs: For Moonlight check policy if outside
System.dll (the client stack use sockets with it's own web
xdomain policy)
In System:
2010-03-19 Sebastien Pouliot <[email protected]>
* Uri.cs: Remove some MOONLIGHT defines
In .:
2010-03-19 Sebastien Pouliot <[email protected]>
* moonlight_*_System.dll.sources: Bring extra types so we can
build a full, managed, internal HTTP stack to be reused for SL3+
* Makefile: Add a INSIDE_SYSTEM define when compiling System.dll
to makes it possible to rebuild the same types in System.Net.dll
svn path=/trunk/mcs/; revision=153916
* SocketAsyncEventArgs.cs: Don't use Socket.Poll for NET_2_1.
Throw NIE for use of BufferList in Receive.
* Socket.cs: Move Close(int) to Socket_2_1.cs. Remove NET_2_1
defines since the file is not used by Moonlight
* Socket_2_1.cs: Add Close(int) from Socket. Define out Poll
and Blocking.
svn path=/trunk/mcs/; revision=142722
for disposed socket.
* Socket.cs: Modified some argument names to match MS. Moved disposed
check before argument checks in Receive. Use SocketError.InvalidArgument
instead of using magic numbers. Added back comments that was
accidentally removed in my previous commit.
* Socket_2_1.cs: Modified some argument names to match MS.
svn path=/trunk/mcs/; revision=137051
MulticastOptionTest.cs.
* Socket.cs: In GetSocketOption, throw SocketException when (byte [])
option value is null. Use SocketError fields instead of using "magic"
number. In SetSocketOption (SocketOptionLevel, SocketOptionName, object)
only accept Linger, AddMembership, DropMembership and verify option
value.
* MulticastOption.cs: Added argument checks. Save interfaceIndex that is
passed to .ctor, and removed MonoTODO. Renamed arguments to match MS.
Reset InterfaceIndex when LocalAddress is modified, and reset
LocalAddress when InterfaceIndex is modified.
* UdpClient.cs: Modified exception to more closely match MS. Fixed
(int, AddressFamily) .ctor to construct IPv6 endpoint when family is
InterNetworkV6. Added null check for multicastAddr to (IPAddress)
JoinMulticastGroup overload. In (int, IPAddress) and (IPAddress,
IPAddress) overloads of JoinMulticastGroup throw SocketException when
client is IPv4. Added null check for multicastAddr to (IPAddress, int),
overload, and moved JoinMulticastGroup call after argument checks.
* IPv6MulticastOption.cs: Renamed argument names to match MS. Added
null check to set_Group. Added range check to set_InterfaceIndex.
* IPv6MulticastOptionTest.cs: Added .ctor and property tests.
* MulticastOptionTest.cs: Added .ctor and property tests.
* SocketTest.cs: Fixed compiler warnings. Improved tests for
closed socket. Added tests fr GetSocketOption, and SetSocketOption
overloads (for Linger, DontLinger, AddMembership and DropMembership).
* UdpClientTest.cs: Added .ctor and JoinMulticastGroup tests.
svn path=/trunk/mcs/; revision=137048
2009-04-09 Sebastien Pouliot <[email protected]>
* Socket.cs: Split socket class into two files (this one and the
new Socket_2_1.cs). This let us remove (some) existing #if in the
sources and avoid (a lot) of new ones for Moonlight.
* Socket_2_1.cs: Just enough of socket for NET_2_1. This helps
the tuner not to bring a lot of extra stuff (some visible)
into Moonlight.
* SocketAsyncEventArgs.cs: Ifdef out some cases for NET_2_1 so
Moonlight can use it's own (smaller) SocketAsyncOperation enum
In .:
2009-04-09 Sebastien Pouliot <[email protected]>
* System.dll.sources: Add System.Net.Sockets/Socket_2_1.cs
svn path=/trunk/mcs/; revision=131392
* Socket.cs: instead of clearing the sockets form the output and then
adding them back, we just remove the ones that have not been signaled.
Fixes bug #354090 and bug #476138.
svn path=/trunk/mcs/; revision=127234