Fix broken "pipe: use event aware wakeups" optimization
authorLinus Torvalds <torvalds@linux-foundation.org>
Fri, 21 Jan 2011 00:21:59 +0000 (16:21 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 21 Jan 2011 00:21:59 +0000 (16:21 -0800)
Commit e462c448fdc8 ("pipe: use event aware wakeups") optimized the pipe
event wakeup calls to avoid wakeups if the events do not match the
requested set.

However, the optimization was buggy, in that it didn't actually use the
correct sets for the events: when we make room for more data to be
written, the pipe poll() routine will return both the POLLOUT _and_
POLLWRNORM bits.  Similarly for read.

And most critically, when a pipe is released, that will potentially
result in POLLHUP|POLLERR (depending on whether it was the last reader
or writer), not just the regular POLLIN|POLLOUT.

This bug showed itself as a hung gnome-screensaver-dialog process, stuck
forever (or at least until it was poked by a signal or by being traced)
in a poll() system call.

Cc: Davide Libenzi <davidel@xmailserver.org>
Cc: David S. Miller <davem@davemloft.net>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/pipe.c

index 89e9e19b1b2eef328ddc36714be0db771e49bcc9..da42f7db50de42640a7fa56df21e21bbb48cf588 100644 (file)
--- a/fs/pipe.c
+++ b/fs/pipe.c
@@ -441,7 +441,7 @@ redo:
                        break;
                }
                if (do_wakeup) {
-                       wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT);
+                       wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
                        kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
                }
                pipe_wait(pipe);
@@ -450,7 +450,7 @@ redo:
 
        /* Signal writers asynchronously that there is more room. */
        if (do_wakeup) {
-               wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT);
+               wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
                kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
        }
        if (ret > 0)
@@ -612,7 +612,7 @@ redo2:
                        break;
                }
                if (do_wakeup) {
-                       wake_up_interruptible_sync_poll(&pipe->wait, POLLIN);
+                       wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
                        kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
                        do_wakeup = 0;
                }
@@ -623,7 +623,7 @@ redo2:
 out:
        mutex_unlock(&inode->i_mutex);
        if (do_wakeup) {
-               wake_up_interruptible_sync_poll(&pipe->wait, POLLIN);
+               wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
                kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
        }
        if (ret > 0)
@@ -715,7 +715,7 @@ pipe_release(struct inode *inode, int decr, int decw)
        if (!pipe->readers && !pipe->writers) {
                free_pipe_info(inode);
        } else {
-               wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLOUT);
+               wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM | POLLERR | POLLHUP);
                kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
                kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
        }