tty: Make flush_to_ldisc() locking more robust
authorLinus Torvalds <torvalds@linux-foundation.org>
Wed, 14 Oct 2009 15:59:49 +0000 (08:59 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 14 Oct 2009 22:09:52 +0000 (15:09 -0700)
The locking logic in this function is extremely subtle, and it broke
when we started doing potentially concurrent 'flush_to_ldisc()' calls in
commit e043e42bdb66885b3ac10d27a01ccb9972e2b0a3 ("pty: avoid forcing
'low_latency' tty flag").

The code in flush_to_ldisc() used to set 'tty->buf.head' to NULL, with
the intention that this would then cause any other concurrent calls to
not do anything (locking note: we have to drop the buf.lock over the
call to ->receive_buf that can block, which is why we can have
concurrency here at all in the first place).

It also used to set the TTY_FLUSHING bit, which would then cause any
concurrent 'tty_buffer_flush()' to not free all the tty buffers and
clear 'tty->buf.tail'.  And with 'buf.head' being NULL, and 'buf.tail'
being non-NULL, new data would never touch 'buf.head'.

Does that sound a bit too subtle? It was.  If another concurrent call to
'flush_to_ldisc()' were to come in, the NULL buf.head would indeed cause
it to not process the buffer list, but it would still clear TTY_FLUSHING
afterwards, making the buffer protection against 'tty_buffer_flush()' no
longer work.

So this clears it all up.  We depend purely on TTY_FLUSHING for handling
re-entrancy, and stop playing games with the buffer list entirely.  In
fact, the buffer list handling is now robust enough that we could
probably stop doing the whole "protect against 'tty_buffer_flush()'"
thing entirely.

However, Alan also points out that we would probably be better off
simplifying the locking even further, and just take the tty ldisc_mutex
around all the buffer flushing calls.  That seems like a good idea, but
in the meantime this is a conceptually minimal fix (with the patch
itself being bigger than required just to clean the code up and make it
readable).

This fixes keyboard trouble under X:

http://bugzilla.kernel.org/show_bug.cgi?id=14388

Reported-and-tested-by: Frédéric Meunier <fredlwm@gmail.com>
Reported-and-tested-by: Boyan <btanastasov@yahoo.co.uk>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Paul Fulghum <paulkf@microgate.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
drivers/char/tty_buffer.c

index 3108991c5c8b60f312742ca4eaeb37e4c812ad4c..0296612cc7dfbb3376291dfd321a6d56bfb6235a 100644 (file)
@@ -402,28 +402,26 @@ static void flush_to_ldisc(struct work_struct *work)
                container_of(work, struct tty_struct, buf.work.work);
        unsigned long   flags;
        struct tty_ldisc *disc;
-       struct tty_buffer *tbuf, *head;
-       char *char_buf;
-       unsigned char *flag_buf;
 
        disc = tty_ldisc_ref(tty);
        if (disc == NULL)       /*  !TTY_LDISC */
                return;
 
        spin_lock_irqsave(&tty->buf.lock, flags);
-       /* So we know a flush is running */
-       set_bit(TTY_FLUSHING, &tty->flags);
-       head = tty->buf.head;
-       if (head != NULL) {
-               tty->buf.head = NULL;
-               for (;;) {
-                       int count = head->commit - head->read;
+
+       if (!test_and_set_bit(TTY_FLUSHING, &tty->flags)) {
+               struct tty_buffer *head;
+               while ((head = tty->buf.head) != NULL) {
+                       int count;
+                       char *char_buf;
+                       unsigned char *flag_buf;
+
+                       count = head->commit - head->read;
                        if (!count) {
                                if (head->next == NULL)
                                        break;
-                               tbuf = head;
-                               head = head->next;
-                               tty_buffer_free(tty, tbuf);
+                               tty->buf.head = head->next;
+                               tty_buffer_free(tty, head);
                                continue;
                        }
                        /* Ldisc or user is trying to flush the buffers
@@ -445,9 +443,9 @@ static void flush_to_ldisc(struct work_struct *work)
                                                        flag_buf, count);
                        spin_lock_irqsave(&tty->buf.lock, flags);
                }
-               /* Restore the queue head */
-               tty->buf.head = head;
+               clear_bit(TTY_FLUSHING, &tty->flags);
        }
+
        /* We may have a deferred request to flush the input buffer,
           if so pull the chain under the lock and empty the queue */
        if (test_bit(TTY_FLUSHPENDING, &tty->flags)) {
@@ -455,7 +453,6 @@ static void flush_to_ldisc(struct work_struct *work)
                clear_bit(TTY_FLUSHPENDING, &tty->flags);
                wake_up(&tty->read_wait);
        }
-       clear_bit(TTY_FLUSHING, &tty->flags);
        spin_unlock_irqrestore(&tty->buf.lock, flags);
 
        tty_ldisc_deref(disc);