diff options
author | Miklos Szeredi <mszeredi@suse.cz> | 2008-11-09 15:23:57 +0100 |
---|---|---|
committer | Linus Torvalds <torvalds@linux-foundation.org> | 2008-11-09 20:17:33 +0100 |
commit | 6209344f5a3795d34b7f2c0061f49802283b6bdd (patch) | |
tree | 5c037ddbb8caac17b0c6101c9ab86387df106d41 /net/unix/garbage.c | |
parent | clarify usage expectations for cnt32_to_63() (diff) | |
download | linux-6209344f5a3795d34b7f2c0061f49802283b6bdd.tar.xz linux-6209344f5a3795d34b7f2c0061f49802283b6bdd.zip |
net: unix: fix inflight counting bug in garbage collector
Previously I assumed that the receive queues of candidates don't
change during the GC. This is only half true, nothing can be received
from the queues (see comment in unix_gc()), but buffers could be added
through the other half of the socket pair, which may still have file
descriptors referring to it.
This can result in inc_inflight_move_tail() erronously increasing the
"inflight" counter for a unix socket for which dec_inflight() wasn't
previously called. This in turn can trigger the "BUG_ON(total_refs <
inflight_refs)" in a later garbage collection run.
Fix this by only manipulating the "inflight" counter for sockets which
are candidates themselves. Duplicating the file references in
unix_attach_fds() is also needed to prevent a socket becoming a
candidate for GC while the skb that contains it is not yet queued.
Reported-by: Andrea Bittau <a.bittau@cs.ucl.ac.uk>
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
CC: stable@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Diffstat (limited to 'net/unix/garbage.c')
-rw-r--r-- | net/unix/garbage.c | 49 |
1 files changed, 37 insertions, 12 deletions
diff --git a/net/unix/garbage.c b/net/unix/garbage.c index 2a27b84f740b..6d4a9a8de5ef 100644 --- a/net/unix/garbage.c +++ b/net/unix/garbage.c @@ -186,8 +186,17 @@ static void scan_inflight(struct sock *x, void (*func)(struct unix_sock *), */ struct sock *sk = unix_get_socket(*fp++); if (sk) { - hit = true; - func(unix_sk(sk)); + struct unix_sock *u = unix_sk(sk); + + /* + * Ignore non-candidates, they could + * have been added to the queues after + * starting the garbage collection + */ + if (u->gc_candidate) { + hit = true; + func(u); + } } } if (hit && hitlist != NULL) { @@ -249,11 +258,11 @@ static void inc_inflight_move_tail(struct unix_sock *u) { atomic_long_inc(&u->inflight); /* - * If this is still a candidate, move it to the end of the - * list, so that it's checked even if it was already passed - * over + * If this still might be part of a cycle, move it to the end + * of the list, so that it's checked even if it was already + * passed over */ - if (u->gc_candidate) + if (u->gc_maybe_cycle) list_move_tail(&u->link, &gc_candidates); } @@ -267,6 +276,7 @@ void unix_gc(void) struct unix_sock *next; struct sk_buff_head hitlist; struct list_head cursor; + LIST_HEAD(not_cycle_list); spin_lock(&unix_gc_lock); @@ -282,10 +292,14 @@ void unix_gc(void) * * Holding unix_gc_lock will protect these candidates from * being detached, and hence from gaining an external - * reference. This also means, that since there are no - * possible receivers, the receive queues of these sockets are - * static during the GC, even though the dequeue is done - * before the detach without atomicity guarantees. + * reference. Since there are no possible receivers, all + * buffers currently on the candidates' queues stay there + * during the garbage collection. + * + * We also know that no new candidate can be added onto the + * receive queues. Other, non candidate sockets _can_ be + * added to queue, so we must make sure only to touch + * candidates. */ list_for_each_entry_safe(u, next, &gc_inflight_list, link) { long total_refs; @@ -299,6 +313,7 @@ void unix_gc(void) if (total_refs == inflight_refs) { list_move_tail(&u->link, &gc_candidates); u->gc_candidate = 1; + u->gc_maybe_cycle = 1; } } @@ -325,14 +340,24 @@ void unix_gc(void) list_move(&cursor, &u->link); if (atomic_long_read(&u->inflight) > 0) { - list_move_tail(&u->link, &gc_inflight_list); - u->gc_candidate = 0; + list_move_tail(&u->link, ¬_cycle_list); + u->gc_maybe_cycle = 0; scan_children(&u->sk, inc_inflight_move_tail, NULL); } } list_del(&cursor); /* + * not_cycle_list contains those sockets which do not make up a + * cycle. Restore these to the inflight list. + */ + while (!list_empty(¬_cycle_list)) { + u = list_entry(not_cycle_list.next, struct unix_sock, link); + u->gc_candidate = 0; + list_move_tail(&u->link, &gc_inflight_list); + } + + /* * Now gc_candidates contains only garbage. Restore original * inflight counters for these as well, and remove the skbuffs * which are creating the cycle(s). |