On 4/10/22 10:46 AM, Jens Axboe wrote:
> Hi,
>
> Can you queue up this one for 5.10 stable? It has an extra hunk
> that's needed for 5.10. 5.15/16/17 will be a direct port of
> the 5.18-rc patch.
Here's the 5.15-stable backport. For 5.16 and 5.17, the upstream commit
will apply directly with no fuzz.
--
Jens Axboe
commit e677edbcabee849bfdd43f1602bccbecf736a646
Author: Jens Axboe <axboe@xxxxxxxxx>
Date: Fri Apr 8 11:08:58 2022 -0600
io_uring: fix race between timeout flush and removal
io_flush_timeouts() assumes the timeout isn't in progress of triggering
or being removed/canceled, so it unconditionally removes it from the
timeout list and attempts to cancel it.
Leave it on the list and let the normal timeout cancelation take care
of it.
Cc: stable@xxxxxxxxxxxxxxx # 5.5+
Signed-off-by: Jens Axboe <axboe@xxxxxxxxx>
diff --git a/fs/io_uring.c b/fs/io_uring.c
index 5fc3a62eae72..0202a6c431df 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -1546,12 +1546,11 @@ static void io_flush_timeouts(struct io_ring_ctx *ctx)
__must_hold(&ctx->completion_lock)
{
u32 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
+ struct io_kiocb *req, *tmp;
spin_lock_irq(&ctx->timeout_lock);
- while (!list_empty(&ctx->timeout_list)) {
+ list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) {
u32 events_needed, events_got;
- struct io_kiocb *req = list_first_entry(&ctx->timeout_list,
- struct io_kiocb, timeout.list);
if (io_is_timeout_noseq(req))
break;
@@ -1568,7 +1567,6 @@ static void io_flush_timeouts(struct io_ring_ctx *ctx)
if (events_got < events_needed)
break;
- list_del_init(&req->timeout.list);
io_kill_timeout(req, 0);
}
ctx->cq_last_tm_flush = seq;
@@ -6210,6 +6208,7 @@ static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe,
if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr)))
return -EFAULT;
+ INIT_LIST_HEAD(&req->timeout.list);
data->mode = io_translate_timeout_mode(flags);
hrtimer_init(&data->timer, io_timeout_get_clock(data), data->mode);
