]> git.ipfire.org Git - thirdparty/kernel/stable.git/commit
af_unix: Remove lock dance in unix_peek_fds().
authorKuniyuki Iwashima <kuniyu@amazon.com>
Wed, 21 May 2025 15:27:21 +0000 (16:27 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 4 Jun 2025 12:40:24 +0000 (14:40 +0200)
commit61a75360dca93c945ef6bd757f8b8a96f39b77cb
treed216cd0a66c62092a65d9af3b4be17bb8e1c7cbd
parent5dfd283f4651d04dbb70ceb9ae5c4a30eda3c52a
af_unix: Remove lock dance in unix_peek_fds().

commit 118f457da9ed58a79e24b73c2ef0aa1987241f0e upstream.

In the previous GC implementation, the shape of the inflight socket
graph was not expected to change while GC was in progress.

MSG_PEEK was tricky because it could install inflight fd silently
and transform the graph.

Let's say we peeked a fd, which was a listening socket, and accept()ed
some embryo sockets from it.  The garbage collection algorithm would
have been confused because the set of sockets visited in scan_inflight()
would change within the same GC invocation.

That's why we placed spin_lock(&unix_gc_lock) and spin_unlock() in
unix_peek_fds() with a fat comment.

In the new GC implementation, we no longer garbage-collect the socket
if it exists in another queue, that is, if it has a bridge to another
SCC.  Also, accept() will require the lock if it has edges.

Thus, we need not do the complicated lock dance.

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20240401173125.92184-3-kuniyu@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Lee Jones <lee@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
include/net/af_unix.h
net/unix/af_unix.c
net/unix/garbage.c