Skip to contentSkip to conversation

io_uring backports

PATCH In review Pavel Begunkov 2 participants 4 messages Oct 16, 2022 – Oct 17, 2022 Stable branch

lore
Share & tools
lore

Pavel Begunkov

author
#0 lore raw

cover io_uring backports

io_uring backports for stable 5.10

Pavel Begunkov (2):
io_uring: correct pinned_vm accounting io_uring/af_unix: defer registered files gc to io_uring release

 fs/io_uring.c          |  8 ++++++--
 include/linux/skbuff.h |  2 ++
 net/unix/garbage.c     | 20 ++++++++++++++++++++
 3 files changed, 28 insertions(+), 2 deletions(-)
Signature · 1 line
2.38.0
author
#1 lore raw

PATCH 1/2 io_uring: correct pinned_vm accounting

[ upstream commit 42b6419d0aba47c5d8644cdc0b68502254671de5 ]

->mm_account should be released only after we free all registered buffers, otherwise __io_sqe_buffers_unregister() will see a NULL ->mm_account and skip locked_vm accounting.

Cc <[email protected]>
Signed-off-by Pavel Begunkov <[email protected]>
Link https://lore.kernel.org/r/6d798f65ed4ab8db3664c4d3397d4af16ca98846.1664849932.git.asml.silence@gmail.com open here
Signed-off-by Jens Axboe <[email protected]>
---
 fs/io_uring.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
fs/io_uring.c +5 -2
index 9654b60a06a5..b82a446d5e59 100644
@@ -8436,8 +8436,6 @@ static void io_ring_ctx_free(struct io_ring_ctx *ctx)
 	if (ctx->sqo_task) {
 		put_task_struct(ctx->sqo_task);
 		ctx->sqo_task = NULL;
-		mmdrop(ctx->mm_account);
-		ctx->mm_account = NULL;
 	}
 
 #ifdef CONFIG_BLK_CGROUP

@@ -8456,6 +8454,11 @@ static void io_ring_ctx_free(struct io_ring_ctx *ctx)
 	}
 #endif
 
+	if (ctx->mm_account) {
+		mmdrop(ctx->mm_account);
+		ctx->mm_account = NULL;
+	}
+
 	io_mem_free(ctx->rings);
 	io_mem_free(ctx->sq_sqes);
Signature · 1 line
2.38.0
author
#2 lore raw

PATCH 2/2 io_uring/af_unix: defer registered files gc to io_uring release

In reply to Pavel Begunkov

[ upstream commit 0091bfc81741b8d3aeb3b7ab8636f911b2de6e80 ]

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 6b06314c47e1 ("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]>
---
 fs/io_uring.c          |  1 +
 include/linux/skbuff.h |  2 ++
 net/unix/garbage.c     | 20 ++++++++++++++++++++
 3 files changed, 23 insertions(+)
fs/io_uring.c +1 -0
index b82a446d5e59..05f360b66b07 100644
@@ -7301,6 +7301,7 @@ static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset)
 	}
 
 	skb->sk = sk;
+	skb->scm_io_uring = 1;
 
 	nr_files = 0;
 	fpl->user = get_uid(ctx->user);
include/linux/skbuff.h +2 -0
index 61fc053a4a4e..462b0e3ef2b2 100644
@@ -681,6 +681,7 @@ 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
  *	@napi_id: id of the NAPI struct this skb came from

@@ -858,6 +859,7 @@ struct sk_buff {
 #ifdef CONFIG_TLS_DEVICE
 	__u8			decrypted:1;
 #endif
+	__u8			scm_io_uring:1;
 
 #ifdef CONFIG_NET_SCHED
 	__u16			tc_index;	/* traffic control index */
net/unix/garbage.c +20 -0
index d45d5366115a..dc2763540393 100644
@@ -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));
Signature · 1 line
2.38.0
maintainer
#3 lore raw
In reply to Pavel Begunkov
Quoted 6 lines · Pavel Begunkov

io_uring backports for stable 5.10

Pavel Begunkov (2):
io_uring: correct pinned_vm accounting io_uring/af_unix: defer registered files gc to io_uring release

Now queued up, thanks.

greg k-h