[PATCH v3 0/8] Fix sock_send reordering issue (#52401)
Jinoh Kang
jinoh.kang.kr at gmail.com
Mon Feb 21 11:22:12 CST 2022
This fixes the send part of the issue #52401 [1]: "Improper
synchronization in sock_recv/sock_send leads to arbitrary reordering of
completion of I/O requests", which was initially reported (outside
Bugzilla) by Dongwan Kim [2].
Changelog:
v1 -> v2:
- drop patch "server: Compact struct set_async_direct_result_reply."
- add async and fd arguments to async_initial_status_callback
- recv_socket_initial_callback: pass OOB flag via private instead of
implementing two separate functions for OOB and non-OOB
- detect short write in send_socket_initial_callback
- preserve the behaviour of returning success if we had a short write
and the socket is nonblocking and force_async is unset
v2 -> v3:
- fix typo in comment
[1] https://bugs.winehq.org/show_bug.cgi?id=52401
[2] https://www.winehq.org/pipermail/wine-devel/2021-May/186454.html
Jinoh Kang (8):
server: Actually set initial status in set_async_direct_result
handler.
server: Add async_initial_status_callback.
server: Defer postprocessing until after setting initial status in
recv_socket handler.
server: Defer postprocessing until after setting initial status in
send_socket handler.
server: Add mark_pending field to set_async_direct_result request.
server: Attempt to complete I/O request immediately in send_socket.
ntdll: Don't call try_send before server call in sock_send.
server: Replace redundant send_socket status fields with force_async
boolean field.
dlls/ntdll/unix/socket.c | 102 ++++++++++++++++++++++--------
dlls/ntdll/unix/sync.c | 9 +--
dlls/ntdll/unix/unix_private.h | 2 +-
dlls/ws2_32/tests/sock.c | 14 ++---
server/async.c | 52 +++++++++++++---
server/file.h | 2 +
server/protocol.def | 6 +-
server/sock.c | 110 +++++++++++++++++++--------------
8 files changed, 204 insertions(+), 93 deletions(-)
Interdiff:
diff --git a/server/sock.c b/server/sock.c
index 2706c2d20ed..1112708369d 100644
--- a/server/sock.c
+++ b/server/sock.c
@@ -3513,7 +3513,7 @@ DECL_HANDLER(send_socket)
else if (!async_queued( &sock->write_q ))
{
/* If write_q is not empty, we cannot really tell if the already queued
- * asyncs will not consume all available data; if there's no data
+ * asyncs will not consume all available space; if there's no space
* available, the current request won't be immediately satiable.
*/
struct pollfd pollfd;
--
2.34.1
More information about the wine-devel
mailing list