fix ReceiveFrom with dual mode socket #92086
Conversation
|
Tagging subscribers to this area: @dotnet/ncl Issue Detailsfixes #92006 After thinking about it I see no reason why
|
src/libraries/System.Net.Primitives/src/System/Net/IPEndPoint.cs
Outdated
Show resolved
Hide resolved
src/libraries/System.Net.Primitives/tests/FunctionalTests/IPEndPointTest.cs
Show resolved
Hide resolved
| { | ||
| throw new ArgumentException(SR.Format(SR.net_InvalidAddressFamily, socketAddress.Family.ToString(), GetType().FullName, AddressFamily.ToString()), nameof(socketAddress)); | ||
| throw new ArgumentException(SR.Format(SR.net_InvalidAddressFamily, socketAddress.Family.ToString(), GetType().FullName), nameof(socketAddress)); | ||
| } | ||
|
|
||
| int minSize = AddressFamily == AddressFamily.InterNetworkV6 ? SocketAddress.IPv6AddressSize : SocketAddress.IPv4AddressSize; |
There was a problem hiding this comment.
If we're no longer requiring that socketAddress.Family == AddressFamily, does this size check still matter?
There was a problem hiding this comment.
We don't need to but than we can possibly get exception when we access the underlying buffer and it is too small. So I feel the extra check provides better feedback.
| if (_remoteEndPoint!.AddressFamily == AddressFamily.InterNetworkV6 && _socketAddress!.Family == AddressFamily.InterNetwork) | ||
| { | ||
| _remoteEndPoint = _remoteEndPoint!.Create(_socketAddress); | ||
| _remoteEndPoint = new IPEndPoint(_socketAddress.GetIPAddress().MapToIPv6(), _socketAddress.GetPort()); | ||
| } |
There was a problem hiding this comment.
Is this something IPEndPoint.Create(SocketAddress) should be doing instead of doing it at this one call site?
There was a problem hiding this comment.
this one is tricky - we have cases when the address is IPv4 and Create would just work - but we want IPv6 IPv4Mapped addresses for DualMode sockets. I guess mostly for compatibility reasons so this is handled as special case.
| { | ||
| _remoteEndPoint = new IPEndPoint(_socketAddress.GetIPAddress().MapToIPv6(), _socketAddress.GetPort()); | ||
| } | ||
| else |
There was a problem hiding this comment.
Similarly here: if this logic were moved into IPEndPoint.Create, would that a) address the duplication between these two call sites and b) potentially avoid bugs elsewhere? Or would that introduce problems?
|
/backport to release/8.0 |
|
Started backporting to release/8.0: https://github.com/dotnet/runtime/actions/runs/6195384920 |
ghost
locked as resolved and limited conversation to collaborators
Fixes #92006
This is regression against 7.0
It seems like the old internal code handled cases when IPv4
EndPointwas passed toReceiveFromon dual mode socket and the received address is actually IPv4 mapped into IPv6. But that case fails now and we fail to setRemoteEndPointleaving whatever was there - in case of the provided repro creating loop or possibly sending message to wrong peer.After thinking about it I see no reason why
IPEndPoint.Createneeds to really care aboutAddressFamilyas long as it is IP. The documentation claims that it needs to be correct type but it does not talk about IPv4 vs IPv6. And we allocate newEndPointso the actual value of the old one does not matter. It will also make it easier to consume new API from #88970 as we exposedSocketAddressesto consumers and they may need to eat with it.The regression was introduced by PR #89841