io_uring/af_unix: defer registered files gc to io_uring release

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: [email protected]
Fixes: 6b06314 ("io_uring: add file set registration")
Reported-and-tested-by: David Bouman <[email protected]>
Signed-off-by: Pavel Begunkov <[email protected]>
Signed-off-by: Thadeu Lima de Souza Cascardo <[email protected]>
[axboe: add kerneldoc comment to skb, fold in skb leak fix]
Signed-off-by: Jens Axboe <[email protected]>

Author: isilence

include/linux/skbuff.h

@@ -803,6 +803,7 @@ 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
@@ -982,6 +983,7 @@ 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 */

io_uring/rsrc.c

@@ -855,6 +855,7 @@ int __io_scm_file_account(struct io_ring_ctx *ctx, struct file *file)
 
 		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);
 	}

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));