mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [PATCH] splice: pass a poll key when waking pipe readers and writers
@ 2026-09-16  8:37 Breno Leitao
  2026-09-16 10:57 ` Oleg Nesterov
  0 siblings, 1 reply; 3+ messages in thread
From: Breno Leitao @ 2026-09-16  8:37 UTC (permalink / raw)
  To: Alexander Viro, Christian Brauner, Jan Kara
  Cc: mjguzik, oleg, linux-fsdevel, linux-kernel, kernel-team, Breno Leitao

wakeup_pipe_readers() and wakeup_pipe_writers() wake the pipe wait queues
with a NULL key.  ep_poll_callback() and pollwake() only filter on the
registered event mask when a key is supplied, so every epoll, poll() and
select() waiter on the pipe is woken, whether or not it asked for the
event that occurred.

anon_pipe_read() and anon_pipe_write() already pass EPOLLIN | EPOLLRDNORM
and EPOLLOUT | EPOLLWRNORM when waking the same wait queues.  Do the same
here.

This is not visible in common pipe use: ordinary pipelines move data with
read()/write(), which already passes the key.

Signed-off-by: Breno Leitao <leitao@debian.org>
---
 fs/splice.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/fs/splice.c b/fs/splice.c
index 9d8f63e2fd1ab..c048183215343 100644
--- a/fs/splice.c
+++ b/fs/splice.c
@@ -179,7 +179,8 @@ static void wakeup_pipe_readers(struct pipe_inode_info *pipe)
 {
 	smp_mb();
 	if (waitqueue_active(&pipe->rd_wait))
-		wake_up_interruptible(&pipe->rd_wait);
+		wake_up_interruptible_poll(&pipe->rd_wait,
+					   EPOLLIN | EPOLLRDNORM);
 	kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
 }
 
@@ -415,7 +416,8 @@ static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
 {
 	smp_mb();
 	if (waitqueue_active(&pipe->wr_wait))
-		wake_up_interruptible(&pipe->wr_wait);
+		wake_up_interruptible_poll(&pipe->wr_wait,
+					   EPOLLOUT | EPOLLWRNORM);
 	kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
 }
 

---
base-commit: 587858367581b9c55c3690f4e63382ad622719d4
change-id: 20260915-b4-splice-poll-key-bc7d5d0d0f68

Best regards,
--  
Breno Leitao <leitao@debian.org>


^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [PATCH] splice: pass a poll key when waking pipe readers and writers
  2026-09-16  8:37 [PATCH] splice: pass a poll key when waking pipe readers and writers Breno Leitao
@ 2026-09-16 10:57 ` Oleg Nesterov
  2026-09-16 11:31   ` Breno Leitao
  0 siblings, 1 reply; 3+ messages in thread
From: Oleg Nesterov @ 2026-09-16 10:57 UTC (permalink / raw)
  To: Breno Leitao
  Cc: Alexander Viro, Christian Brauner, Jan Kara, mjguzik,
	linux-fsdevel, linux-kernel, kernel-team

On 09/16, Breno Leitao wrote:
>
> --- a/fs/splice.c
> +++ b/fs/splice.c
> @@ -179,7 +179,8 @@ static void wakeup_pipe_readers(struct pipe_inode_info *pipe)
>  {
>  	smp_mb();
>  	if (waitqueue_active(&pipe->rd_wait))
> -		wake_up_interruptible(&pipe->rd_wait);
> +		wake_up_interruptible_poll(&pipe->rd_wait,
> +					   EPOLLIN | EPOLLRDNORM);
>  	kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
>  }
>
> @@ -415,7 +416,8 @@ static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
>  {
>  	smp_mb();
>  	if (waitqueue_active(&pipe->wr_wait))
> -		wake_up_interruptible(&pipe->wr_wait);
> +		wake_up_interruptible_poll(&pipe->wr_wait,
> +					   EPOLLOUT | EPOLLWRNORM);
>  	kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
>  }

Acked-by: Oleg Nesterov <oleg@redhat.com>

While at it... I think they both should use wq_has_sleeper().

If you agree, feel free to send V2 with this additional change.
Or I will send this trivial cleanup on top of your patch.

Oleg.


^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [PATCH] splice: pass a poll key when waking pipe readers and writers
  2026-09-16 10:57 ` Oleg Nesterov
@ 2026-09-16 11:31   ` Breno Leitao
  0 siblings, 0 replies; 3+ messages in thread
From: Breno Leitao @ 2026-09-16 11:31 UTC (permalink / raw)
  To: Oleg Nesterov
  Cc: Alexander Viro, Christian Brauner, Jan Kara, mjguzik,
	linux-fsdevel, linux-kernel, kernel-team

On Wed, Sep 16, 2026 at 12:57:13PM +0200, Oleg Nesterov wrote:
> While at it... I think they both should use wq_has_sleeper().

Good point, thanks!

> If you agree, feel free to send V2 with this additional change.

Sure, let's wait a bit more before I send v2, but I've got this fix updated
already to the following change:

	diff --git a/fs/splice.c b/fs/splice.c
	index 9d8f63e2fd1ab..65f9090c3438c 100644
	--- a/fs/splice.c
	+++ b/fs/splice.c
	@@ -177,9 +177,9 @@ static const struct pipe_buf_operations user_page_pipe_buf_ops = {

	static void wakeup_pipe_readers(struct pipe_inode_info *pipe)
	{
	-       smp_mb();
	-       if (waitqueue_active(&pipe->rd_wait))
	-               wake_up_interruptible(&pipe->rd_wait);
	+       if (wq_has_sleeper(&pipe->rd_wait))
	+               wake_up_interruptible_poll(&pipe->rd_wait,
	+                                          EPOLLIN | EPOLLRDNORM);
		kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
	}

	@@ -413,9 +413,9 @@ EXPORT_SYMBOL(nosteal_pipe_buf_ops);

	static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
	{
	-       smp_mb();
	-       if (waitqueue_active(&pipe->wr_wait))
	-               wake_up_interruptible(&pipe->wr_wait);
	+       if (wq_has_sleeper(&pipe->wr_wait))
	+               wake_up_interruptible_poll(&pipe->wr_wait,
	+                                          EPOLLOUT | EPOLLWRNORM);
		kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
	}


Thanks for the quick review,
--breno

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2026-09-16 11:31 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-09-16  8:37 [PATCH] splice: pass a poll key when waking pipe readers and writers Breno Leitao
2026-09-16 10:57 ` Oleg Nesterov
2026-09-16 11:31   ` Breno Leitao

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®