From: Pavel Begunkov <asml.silence@xxxxxxxxx> commit 0091bfc81741b8d3aeb3b7ab8636f911b2de6e80 upstream. Instead of putting io_uring's registered files in unix_gc() we want it to be done by io_uring itself. The trick here is to consider io_uring registered files for cycle detection but not actually putting them down. Because io_uring can't register other ring instances, this will remove all refs to the ring file triggering the ->release path and clean up with io_ring_ctx_free(). Cc: stable@xxxxxxxxxxxxxxx Fixes: 6b06314c47e1 ("io_uring: add file set registration") Reported-and-tested-by: David Bouman <dbouman03@xxxxxxxxx> Signed-off-by: Pavel Begunkov <asml.silence@xxxxxxxxx> Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@xxxxxxxxxxxxx> [axboe: add kerneldoc comment to skb, fold in skb leak fix] Signed-off-by: Jens Axboe <axboe@xxxxxxxxx> Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> --- include/linux/skbuff.h | 2 ++ io_uring/io_uring.c | 1 + net/unix/garbage.c | 20 ++++++++++++++++++++ 3 files changed, 23 insertions(+) --- a/include/linux/skbuff.h +++ b/include/linux/skbuff.h @@ -965,6 +965,7 @@ typedef unsigned char *sk_buff_data_t; * @csum_level: indicates the number of consecutive checksums found in * the packet minus one that have been verified as * CHECKSUM_UNNECESSARY (max 3) + * @scm_io_uring: SKB holds io_uring registered files * @dst_pending_confirm: need to confirm neighbour * @decrypted: Decrypted SKB * @slow_gro: state present at GRO time, slower prepare step required @@ -1144,6 +1145,7 @@ struct sk_buff { #endif __u8 slow_gro:1; __u8 csum_not_inet:1; + __u8 scm_io_uring:1; #ifdef CONFIG_NET_SCHED __u16 tc_index; /* traffic control index */ --- a/io_uring/io_uring.c +++ b/io_uring/io_uring.c @@ -9484,6 +9484,7 @@ static int io_scm_file_account(struct io UNIXCB(skb).fp = fpl; skb->sk = sk; + skb->scm_io_uring = 1; skb->destructor = unix_destruct_scm; refcount_add(skb->truesize, &sk->sk_wmem_alloc); } --- a/net/unix/garbage.c +++ b/net/unix/garbage.c @@ -204,6 +204,7 @@ void wait_for_unix_gc(void) /* The external entry point: unix_gc() */ void unix_gc(void) { + struct sk_buff *next_skb, *skb; struct unix_sock *u; struct unix_sock *next; struct sk_buff_head hitlist; @@ -297,11 +298,30 @@ void unix_gc(void) spin_unlock(&unix_gc_lock); + /* We need io_uring to clean its registered files, ignore all io_uring + * originated skbs. It's fine as io_uring doesn't keep references to + * other io_uring instances and so killing all other files in the cycle + * will put all io_uring references forcing it to go through normal + * release.path eventually putting registered files. + */ + skb_queue_walk_safe(&hitlist, skb, next_skb) { + if (skb->scm_io_uring) { + __skb_unlink(skb, &hitlist); + skb_queue_tail(&skb->sk->sk_receive_queue, skb); + } + } + /* Here we are. Hitlist is filled. Die. */ __skb_queue_purge(&hitlist); spin_lock(&unix_gc_lock); + /* There could be io_uring registered files, just push them back to + * the inflight list + */ + list_for_each_entry_safe(u, next, &gc_candidates, link) + list_move_tail(&u->link, &gc_inflight_list); + /* All candidates should have been detached by now. */ BUG_ON(!list_empty(&gc_candidates));