[PATCH v3] ws2_32: Return WSAEAFNOSUPPORT when ip4/6 disabled

[Brendan McGrath]

Currently, when IPv6 is disabled by the kernel, attempting to use
the 'WebRequest' API with .NET native causes a crash. This can be
recreated using the following gist:
https://gist.github.com/redmcg/7d81ef833c77bee6965b5f441006f697

This patch fixes the crash by returning WSAEAFNOSUPPORT, as expected
by .NET. See:
https://referencesource.microsoft.com/#System/net/System/Net/Sockets/Socket....

Signed-off-by: Brendan McGrath 
---

Changes since v2:
  - Add the 'Signed-off-by' tag

dlls/ws2_32/socket.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/dlls/ws2_32/socket.c b/dlls/ws2_32/socket.c
index df068fe8527..956ead2e9d5 100644
--- a/dlls/ws2_32/socket.c
+++ b/dlls/ws2_32/socket.c
@@ -7646,6 +7646,8 @@ SOCKET WINAPI WSASocketW(int af, int type, int protocol,
         /* invalid combination of valid parameters, like SOCK_STREAM + IPPROTO_UDP */
         if (err == WSAEINVAL)
             err = WSAESOCKTNOSUPPORT;
+        else if (err == WSAEOPNOTSUPP && (unixaf == AF_INET || unixaf == AF_INET6))
+            err = WSAEAFNOSUPPORT;
         else if (err == WSAEOPNOTSUPP)
             err = WSAEPROTONOSUPPORT;
     }

[Alexandre Julliard]

Brendan McGrath  writes:
> Currently, when IPv6 is disabled by the kernel, attempting to use
> the 'WebRequest' API with .NET native causes a crash. This can be
> recreated using the following gist:
> https://gist.github.com/redmcg/7d81ef833c77bee6965b5f441006f697
>
> This patch fixes the crash by returning WSAEAFNOSUPPORT, as expected
> by .NET. See:
> https://referencesource.microsoft.com/#System/net/System/Net/Sockets/Socket....
> Signed-off-by: Brendan McGrath

It breaks the tests:

../../../tools/runtest -q -P wine -T ../../.. -M ws2_32.dll -p ws2_32_test.exe.so sock && touch sock.ok
sock.c:2554: Test failed: Expected 10043, received 10047
sock.c:2561: Test failed: Expected 10043, received 10047
make[1]: *** [Makefile:220: sock.ok] Error 2

[PATCH v4] ws2_32: Return WSAEAFNOSUPPORT when ip4/6 disabled

[Brendan McGrath]

Currently, when IPv6 is disabled by the kernel, attempting to use
the 'WebRequest' API with .NET native causes a crash. This can be
recreated using the following gist:
https://gist.github.com/redmcg/7d81ef833c77bee6965b5f441006f697

This patch fixes the crash by returning WSAEAFNOSUPPORT, as expected
by .NET. See:
https://referencesource.microsoft.com/#System/net/System/Net/Sockets/Socket....

Signed-off-by: Brendan McGrath 
---
Changes since v3:
  - fix broken tests (by only returning WSAEAFNOSUPPORT on assumed working protocol combinations)
 dlls/ws2_32/socket.c | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)

diff --git a/dlls/ws2_32/socket.c b/dlls/ws2_32/socket.c
index df068fe8527..d77b6404105 100644
--- a/dlls/ws2_32/socket.c
+++ b/dlls/ws2_32/socket.c
@@ -7646,8 +7646,19 @@ SOCKET WINAPI WSASocketW(int af, int type, int protocol,
         /* invalid combination of valid parameters, like SOCK_STREAM + IPPROTO_UDP */
         if (err == WSAEINVAL)
             err = WSAESOCKTNOSUPPORT;
-        else if (err == WSAEOPNOTSUPP)
-            err = WSAEPROTONOSUPPORT;
+        else if (err == WSAEOPNOTSUPP) {
+          if (unixtype == SOCK_STREAM && protocol == IPPROTO_TCP &&
+                             (unixaf == AF_INET || unixaf == AF_INET6))
+              err = WSAEAFNOSUPPORT;
+          else if (unixtype == SOCK_DGRAM && protocol == IPPROTO_IP &&
+                             (unixaf == AF_INET || unixaf == AF_INET6))
+              err = WSAEAFNOSUPPORT;
+          else if (unixtype == SOCK_DGRAM && protocol == IPPROTO_IPV6 &&
+                             unixaf == AF_INET6)
+              err = WSAEAFNOSUPPORT;
+          else
+              err = WSAEPROTONOSUPPORT;
+        }
     }

done:

[PATCH v4] ws2_32: Return WSAEAFNOSUPPORT when ip4/6 disabled

[Alexandre Julliard]

Brendan McGrath  writes:
> @@ -7646,8 +7646,19 @@ SOCKET WINAPI WSASocketW(int af, int type, int protocol,
>          /* invalid combination of valid parameters, like SOCK_STREAM + IPPROTO_UDP */
>          if (err == WSAEINVAL)
>              err = WSAESOCKTNOSUPPORT;
> -
>    else if (err == WSAEOPNOTSUPP)
> -
>        err = WSAEPROTONOSUPPORT;
> -
>    else if (err == WSAEOPNOTSUPP) {
> -
>      if (unixtype == SOCK_STREAM && protocol == IPPROTO_TCP &&
> -
>                         (unixaf == AF_INET || unixaf == AF_INET6))
> -
>          err = WSAEAFNOSUPPORT;
> -
>      else if (unixtype == SOCK_DGRAM && protocol == IPPROTO_IP &&
> -
>                         (unixaf == AF_INET || unixaf == AF_INET6))
> -
>          err = WSAEAFNOSUPPORT;
> -
>      else if (unixtype == SOCK_DGRAM && protocol == IPPROTO_IPV6 &&
> -
>                         unixaf == AF_INET6)
> -
>          err = WSAEAFNOSUPPORT;
> -
>      else
> -
>          err = WSAEPROTONOSUPPORT;
> -
>    }

I'd suggest to find a way to return the specific error from the server,
instead of mapping everything to the same error on the server side and
trying to guess the specific error afterwards.

[PATCH v5] ws2_32: Return WSAEAFNOSUPPORT when ip4/6 disabled

[Brendan McGrath]

Currently, when IPv6 is disabled by the kernel, attempting to use
the 'WebRequest' API with .NET native causes a crash. This can be
recreated using the following gist:
https://gist.github.com/redmcg/7d81ef833c77bee6965b5f441006f697

This patch fixes the crash by returning WSAEAFNOSUPPORT, as expected
by .NET. See:
https://referencesource.microsoft.com/#System/net/System/Net/Sockets/Socket....

Signed-off-by: Brendan McGrath 
---
Changes since v4:
  - from the server: map EAFNOSUPPORT from the kernel to STATUS_NOT_IMPLEMENTED
  - in ws2_32: map STATUS_NOT_IMPLEMENTED to WSAEAFNOSUPPORT

It's still a bit ugly - but short of changing the design by removing the need
for a NTSTATUS - I'm not sure there's a better way.

dlls/ws2_32/socket.c | 1 +
 server/sock.c        | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)

diff --git a/dlls/ws2_32/socket.c b/dlls/ws2_32/socket.c
index df068fe8527..1d433086f11 100644
--- a/dlls/ws2_32/socket.c
+++ b/dlls/ws2_32/socket.c
@@ -1067,6 +1067,7 @@ static inline DWORD NtStatusToWSAError( const DWORD status )
     case STATUS_BUFFER_OVERFLOW:            wserr = WSAEMSGSIZE;           break;
     case STATUS_NOT_SUPPORTED:              wserr = WSAEOPNOTSUPP;         break;
     case STATUS_HOST_UNREACHABLE:           wserr = WSAEHOSTUNREACH;       break;
+    case STATUS_NOT_IMPLEMENTED:            wserr = WSAEAFNOSUPPORT;       break;

default:
         wserr = RtlNtStatusToDosError( status );
diff --git a/server/sock.c b/server/sock.c
index a8e6e28599b..dfcbab530fe 100644
--- a/server/sock.c
+++ b/server/sock.c
@@ -912,8 +912,8 @@ static int sock_get_ntstatus( int err )
         case EPROTONOSUPPORT:
         case ESOCKTNOSUPPORT:
         case EPFNOSUPPORT:
-        case EAFNOSUPPORT:
         case EPROTOTYPE:        return STATUS_NOT_SUPPORTED;
+        case EAFNOSUPPORT:      return STATUS_NOT_IMPLEMENTED;
         case ENOPROTOOPT:       return STATUS_INVALID_PARAMETER;
         case EOPNOTSUPP:        return STATUS_NOT_SUPPORTED;
         case EADDRINUSE:        return STATUS_ADDRESS_ALREADY_ASSOCIATED;

[PATCH v5] ws2_32: Return WSAEAFNOSUPPORT when ip4/6 disabled

[Brendan McGrath]

Just following up on version 5 of this patch as I haven't heard any 
feedback (and want to make sure it hasn't been overlooked).

On 16/11/18 10:27 am, Brendan McGrath wrote:
> Currently, when IPv6 is disabled by the kernel, attempting to use
> the 'WebRequest' API with .NET native causes a crash. This can be
> recreated using the following gist:
> https://gist.github.com/redmcg/7d81ef833c77bee6965b5f441006f697
>
> This patch fixes the crash by returning WSAEAFNOSUPPORT, as expected
> by .NET. See:
> https://referencesource.microsoft.com/#System/net/System/Net/Sockets/Socket....
> Signed-off-by: Brendan McGrath
>
> Changes since v4:
> - from the server: map EAFNOSUPPORT from the kernel to STATUS_NOT_IMPLEMENTED
> - in ws2_32: map STATUS_NOT_IMPLEMENTED to WSAEAFNOSUPPORT
>
> It's still a bit ugly - but short of changing the design by removing the need
> for a NTSTATUS - I'm not sure there's a better way.
>
> dlls/ws2_32/socket.c | 1 +
>   server/sock.c        | 2 +-
>   2 files changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/dlls/ws2_32/socket.c b/dlls/ws2_32/socket.c
> index df068fe8527..1d433086f11 100644
> --- a/dlls/ws2_32/socket.c
> +++ b/dlls/ws2_32/socket.c
> @@ -1067,6 +1067,7 @@ static inline DWORD NtStatusToWSAError( const DWORD status )
>       case STATUS_BUFFER_OVERFLOW:            wserr = WSAEMSGSIZE;           break;
>       case STATUS_NOT_SUPPORTED:              wserr = WSAEOPNOTSUPP;         break;
>       case STATUS_HOST_UNREACHABLE:           wserr = WSAEHOSTUNREACH;       break;
> -
>
> case STATUS_NOT_IMPLEMENTED:            wserr = WSAEAFNOSUPPORT;       break;
>
> default:
>      wserr = RtlNtStatusToDosError( status );
>
> diff --git a/server/sock.c b/server/sock.c
> index a8e6e28599b..dfcbab530fe 100644
> --- a/server/sock.c
> +++ b/server/sock.c
> @@ -912,8 +912,8 @@ static int sock_get_ntstatus( int err )
>           case EPROTONOSUPPORT:
>           case ESOCKTNOSUPPORT:
>           case EPFNOSUPPORT:
> -
>    case EAFNOSUPPORT:
>     case EPROTOTYPE:        return STATUS_NOT_SUPPORTED;
> -
>    case EAFNOSUPPORT:      return STATUS_NOT_IMPLEMENTED;
>     case ENOPROTOOPT:       return STATUS_INVALID_PARAMETER;
>     case EOPNOTSUPP:        return STATUS_NOT_SUPPORTED;
>     case EADDRINUSE:        return STATUS_ADDRESS_ALREADY_ASSOCIATED;

[PATCH v5] ws2_32: Return WSAEAFNOSUPPORT when ip4/6 disabled

[Alexandre Julliard]

Brendan McGrath  writes:
> Just following up on version 5 of this patch as I haven't heard any
> feedback (and want to make sure it hasn't been overlooked).

I fixed it a different way, hopefully it works now.

[PATCH v5] ws2_32: Return WSAEAFNOSUPPORT when ip4/6 disabled

[Brendan McGrath]

It does. Thanks for that.

On 22/11/18 6:09 am, Alexandre Julliard wrote:
> Brendan McGrath  writes:
> > Just following up on version 5 of this patch as I haven't heard any
> > feedback (and want to make sure it hasn't been overlooked).
>
> I fixed it a different way, hopefully it works now.