Re: [PATCH v2 2/2] io_uring: flush timeouts that should already have expired

[Marcelo Diop-Gonzalez <marcelo827@gmail.com>]

On Sat, Jan 02, 2021 at 08:26:26PM +0000, Pavel Begunkov wrote:
> On 02/01/2021 19:54, Pavel Begunkov wrote:
> > On 19/12/2020 19:15, Marcelo Diop-Gonzalez wrote:
> >> Right now io_flush_timeouts() checks if the current number of events
> >> is equal to ->timeout.target_seq, but this will miss some timeouts if
> >> there have been more than 1 event added since the last time they were
> >> flushed (possible in io_submit_flush_completions(), for example). Fix
> >> it by recording the starting value of ->cached_cq_overflow -
> >> ->cq_timeouts instead of the target value, so that we can safely
> >> (without overflow problems) compare the number of events that have
> >> happened with the number of events needed to trigger the timeout.
> 
> https://www.spinics.net/lists/kernel/msg3475160.html
> 
> The idea was to replace u32 cached_cq_tail with u64 while keeping
> timeout offsets u32. Assuming that we won't ever hit ~2^62 inflight
> requests, complete all requests falling into some large enough window
> behind that u64 cached_cq_tail.
> 
> simplifying:
> 
> i64 d = target_off - ctx->u64_cq_tail
> if (d <= 0 && d > -2^32)
> 	complete_it()
> 
> Not fond  of it, but at least worked at that time. You can try out
> this approach if you want, but would be perfect if you would find
> something more elegant :)
>

What do you think about something like this? I think it's not totally
correct because it relies on having ->completion_lock in io_timeout() so
that ->cq_last_tm_flushed is updated, but in case of IORING_SETUP_IOPOLL,
io_iopoll_complete() doesn't take that lock, and ->uring_lock will not
be held if io_timeout() is called from io_wq_submit_work(), but maybe
could still be worth it since that was already possibly a problem?

diff --git a/fs/io_uring.c b/fs/io_uring.c
index cb57e0360fcb..50984709879c 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -353,6 +353,7 @@ struct io_ring_ctx {
 		unsigned		cq_entries;
 		unsigned		cq_mask;
 		atomic_t		cq_timeouts;
+		unsigned		cq_last_tm_flush;
 		unsigned long		cq_check_overflow;
 		struct wait_queue_head	cq_wait;
 		struct fasync_struct	*cq_fasync;
@@ -1633,19 +1634,26 @@ static void __io_queue_deferred(struct io_ring_ctx *ctx)
 
 static void io_flush_timeouts(struct io_ring_ctx *ctx)
 {
+	u32 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
+
 	while (!list_empty(&ctx->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;
-		if (req->timeout.target_seq != ctx->cached_cq_tail
-					- atomic_read(&ctx->cq_timeouts))
+
+		events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush;
+		events_got = seq - ctx->cq_last_tm_flush;
+		if (events_got < events_needed)
 			break;
 
 		list_del_init(&req->timeout.list);
 		io_kill_timeout(req);
 	}
+
+	ctx->cq_last_tm_flush = seq;
 }
 
 static void io_commit_cqring(struct io_ring_ctx *ctx)
-- 
2.20.1

> >>
> >> Signed-off-by: Marcelo Diop-Gonzalez <marcelo827@gmail.com>
> >> ---
> >>  fs/io_uring.c | 30 +++++++++++++++++++++++-------
> >>  1 file changed, 23 insertions(+), 7 deletions(-)
> >>
> >> diff --git a/fs/io_uring.c b/fs/io_uring.c
> >> index f394bf358022..f62de0cb5fc4 100644
> >> --- a/fs/io_uring.c
> >> +++ b/fs/io_uring.c
> >> @@ -444,7 +444,7 @@ struct io_cancel {
> >>  struct io_timeout {
> >>  	struct file			*file;
> >>  	u32				off;
> >> -	u32				target_seq;
> >> +	u32				start_seq;
> >>  	struct list_head		list;
> >>  	/* head of the link, used by linked timeouts only */
> >>  	struct io_kiocb			*head;
> >> @@ -1629,6 +1629,24 @@ static void __io_queue_deferred(struct io_ring_ctx *ctx)
> >>  	} while (!list_empty(&ctx->defer_list));
> >>  }
> >>  
> >> +static inline u32 io_timeout_events_left(struct io_kiocb *req)
> >> +{
> >> +	struct io_ring_ctx *ctx = req->ctx;
> >> +	u32 events;
> >> +
> >> +	/*
> >> +	 * events -= req->timeout.start_seq and the comparison between
> >> +	 * ->timeout.off and events will not overflow because each time
> >> +	 * ->cq_timeouts is incremented, ->cached_cq_tail is incremented too.
> >> +	 */
> >> +
> >> +	events = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
> >> +	events -= req->timeout.start_seq;
> > 
> > It looks to me that events before the start_seq subtraction can have got wrapped
> > around start_seq.
> > 
> > e.g.
> > 1) you submit a timeout with off=0xff...ff (start_seq=0 for convenience)
> > 
> > 2) some time has passed, let @events = 0xff..ff - 1
> > so the timeout still waits
> > 
> > 3) we commit 5 requests at once and call io_commit_cqring() only once for
> > them, so we get @events == 0xff..ff - 1 + 5, i.e. 4
> > 
> > @events == 4 < off == 0xff...ff,
> > so we didn't trigger out timeout even though should have
> > 
> >> +	if (req->timeout.off > events)
> >> +		return req->timeout.off - events;
> >> +	return 0;
> >> +}
> >> +
> >>  static void io_flush_timeouts(struct io_ring_ctx *ctx)
> >>  {
> >>  	while (!list_empty(&ctx->timeout_list)) {
> >> @@ -1637,8 +1655,7 @@ static void io_flush_timeouts(struct io_ring_ctx *ctx)
> >>  
> >>  		if (io_is_timeout_noseq(req))
> >>  			break;
> >> -		if (req->timeout.target_seq != ctx->cached_cq_tail
> >> -					- atomic_read(&ctx->cq_timeouts))
> >> +		if (io_timeout_events_left(req) > 0)
> >>  			break;
> >>  
> >>  		list_del_init(&req->timeout.list);
> >> @@ -5785,7 +5802,6 @@ static int io_timeout(struct io_kiocb *req)
> >>  	struct io_ring_ctx *ctx = req->ctx;
> >>  	struct io_timeout_data *data = req->async_data;
> >>  	struct list_head *entry;
> >> -	u32 tail, off = req->timeout.off;
> >>  
> >>  	spin_lock_irq(&ctx->completion_lock);
> >>  
> >> @@ -5799,8 +5815,8 @@ static int io_timeout(struct io_kiocb *req)
> >>  		goto add;
> >>  	}
> >>  
> >> -	tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
> >> -	req->timeout.target_seq = tail + off;
> >> +	req->timeout.start_seq = ctx->cached_cq_tail -
> >> +		atomic_read(&ctx->cq_timeouts);
> >>  
> >>  	/*
> >>  	 * Insertion sort, ensuring the first entry in the list is always
> >> @@ -5813,7 +5829,7 @@ static int io_timeout(struct io_kiocb *req)
> >>  		if (io_is_timeout_noseq(nxt))
> >>  			continue;
> >>  		/* nxt.seq is behind @tail, otherwise would've been completed */
> >> -		if (off >= nxt->timeout.target_seq - tail)
> >> +		if (req->timeout.off >= io_timeout_events_left(nxt))
> >>  			break;
> >>  	}
> >>  add:
> >>
> > 
> 
> -- 
> Pavel Begunkov