pipe_resize_ring() wakes pipe->wr_wait 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 write queue is woken, whether or not it asked for the event that occurred. The event is the one the comment above the wakeup already names: more room for writers. anon_pipe_write() passes EPOLLOUT | EPOLLWRNORM when waking the same queue; do the same here. Writers blocked in wait_event_interruptible_exclusive() are unaffected, as autoremove_wake_function() ignores the key. Signed-off-by: Breno Leitao --- This is the last NULL-key wakeup on a pipe wait queue that signals a data or space event. The remaining ones in fs/pipe.c are all hangup broadcasts (pipe_release(), wake_up_partner(), the fifo_open() error paths), where a key would make no difference: poll, select and epoll all add EPOLLERR | EPOLLHUP to the registered mask unconditionally, so those wakeups can never be filtered out. To: Alexander Viro To: Christian Brauner To: Jan Kara Cc: linux-fsdevel@vger.kernel.org Cc: linux-kernel@vger.kernel.org Cc: mjguzik@gmail.com To: oleg@redhat.com --- fs/pipe.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/fs/pipe.c b/fs/pipe.c index 292425a834dd5..5791db016e242 100644 --- a/fs/pipe.c +++ b/fs/pipe.c @@ -1433,7 +1433,7 @@ int pipe_resize_ring(struct pipe_inode_info *pipe, unsigned int nr_slots) spin_unlock_irq(&pipe->rd_wait.lock); /* This might have made more room for writers */ - wake_up_interruptible(&pipe->wr_wait); + wake_up_interruptible_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); return 0; } --- base-commit: 587858367581b9c55c3690f4e63382ad622719d4 change-id: 20260916-pipe-resize-poll-key-017a2180d481 Best regards, -- Breno Leitao