Dne 05. 07. 21 v 21:54 Zebediah Figura (she/her) napsal(a):
On 7/2/21 1:06 PM, David Koolhoven wrote:
Make sure we're not going to poll on connectionless file descriptors which have received shutdown signals on their read and write ends.
Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=51319 Signed-off-by: David Koolhoven david@koolhoven-home.net
server/sock.c | 4 ++++ 1 file changed, 4 insertions(+)
diff --git a/server/sock.c b/server/sock.c index ce2f390ec05..ecf10441c2b 100644 --- a/server/sock.c +++ b/server/sock.c @@ -1197,6 +1197,10 @@ static int sock_get_poll_events( struct fd *fd ) { ev |= POLLOUT; } + if (sock->rd_shutdown && sock->wr_shutdown && ev == 0) + { + ev = -1; + } break; }
This seems reasonable to me, but I feel like it deserves a comment—in the code, rather than in the commit message—that describes why we need to do this. See the comment for the SOCK_UNCONNECTED case above.
Hi, I'm facing a regression of a networking application with the same symptoms (wineserver at 100%). There is a little progress only if both these patches are applied but 100% peaks can still be observed at times (which are not with Wine 6.0.1) and the application still suffers from some errors. It seems to me that there is a deeper problem.