commit '1b6b26ae7053 ("pipe: fix and clarify pipe write wakeup logic")' changed pipe_write() to wakeup readers only if the pipe was empty. Prior to this change, threads waiting in epoll_wait(EPOLLET | EPOLLIN) on non-empty pipes would get woken up on new data.
It meant an applications that, 1. used pipe + epoll for notifications between threads / processes 2. Didn't drain the pipe on each epoll wakeup unless the pipe was full started to experience hang / timeouts in threads stuck in epoll_wait()
So restore the old behavior to wakeup all readers if any new data is written to the pipe.
Fixes: 1b6b26ae7053 ("pipe: fix and clarify pipe write wakeup logic") Signed-off-by: Sandeep Patil sspatil@android.com --- fs/pipe.c | 19 ++++++++++--------- 1 file changed, 10 insertions(+), 9 deletions(-)
diff --git a/fs/pipe.c b/fs/pipe.c index bfd946a9ad01..dda22a316bb3 100644 --- a/fs/pipe.c +++ b/fs/pipe.c @@ -406,7 +406,7 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from) ssize_t ret = 0; size_t total_len = iov_iter_count(from); ssize_t chars; - bool was_empty = false; + bool do_wakeup = false; bool wake_next_writer = false;
/* Null write succeeds. */ @@ -429,10 +429,11 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from) #endif
/* - * Only wake up if the pipe started out empty, since - * otherwise there should be no readers waiting. + * Wake up readers if the pipe was written to. Regardless + * of whether it was empty or not. Otherwise, threads + * waiting with EPOLLET will hang until the pipe is emptied. * - * If it wasn't empty we try to merge new data into + * If pipe wasn't empty we try to merge new data into * the last buffer. * * That naturally merges small writes, but it also @@ -440,9 +441,8 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from) * spanning multiple pages. */ head = pipe->head; - was_empty = pipe_empty(head, pipe->tail); chars = total_len & (PAGE_SIZE-1); - if (chars && !was_empty) { + if (chars && !pipe_empty(head, pipe->tail)) { unsigned int mask = pipe->ring_size - 1; struct pipe_buffer *buf = &pipe->bufs[(head - 1) & mask]; int offset = buf->offset + buf->len; @@ -460,6 +460,7 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from) }
buf->len += ret; + do_wakeup = true; if (!iov_iter_count(from)) goto out; } @@ -526,6 +527,7 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from) ret += copied; buf->offset = 0; buf->len = copied; + do_wakeup = true;
if (!iov_iter_count(from)) break; @@ -553,13 +555,12 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from) * become empty while we dropped the lock. */ __pipe_unlock(pipe); - if (was_empty) { + if (do_wakeup) { wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); } wait_event_interruptible_exclusive(pipe->wr_wait, pipe_writable(pipe)); __pipe_lock(pipe); - was_empty = pipe_empty(pipe->head, pipe->tail); wake_next_writer = true; } out: @@ -576,7 +577,7 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from) * how (for example) the GNU make jobserver uses small writes to * wake up pending jobs */ - if (was_empty) { + if (do_wakeup) { wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); }