Fix ext3 scheduling storm and lockup

This patch fixes an inefficiency and potential system lockup in the 2.4 kernel's ext3 filesystem. The problem has been present since 2.4.20-pre5. This patch is applicable to 2.4.20. A copy is at

http://www.zip.com.au/~akpm/linux/patches/2.4/2.4.20/ext3-scheduling-storm.patch

Anyone who is using tasks which have realtime scheduling policy on ext3 systems should apply this change.

Details:

At the start of do_get_write_access() we have this logic:

	repeat:
		lock_buffer(jh->bh);
		...
		unlock_buffer(jh->bh);
		...
		if (jh->j_list == BJ_Shadow) {
			sleep_on_buffer(jh->bh);
			goto repeat;
		}

The problem is that the unlock_buffer() will wake up anyone who is sleeping in the sleep_on_buffer().

So if task A is asleep in sleep_on_buffer() and task B now runs do_get_write_access(), task B will wake task A by accident. Task B will then sleep on the buffer and task A will loop, will run unlock_buffer() and then wake task B.

Net effect: the system does 100,000 context switches/sec until I/O completes against the buffer and kjournald changes the value of jh->j_list.

Unless task A and task B happen to both have realtime scheduling policy - if they do then kjournald will never run. The state is never cleared and your box locks up.

The fix is to not do the `goto repeat;' until the buffer has been taken off the shadow list. So we don't go and wake up the other waiter(s) until they can actually proceed to use the buffer.

diff -puN fs/jbd/transaction.c~ext3-scheduling-storm fs/jbd/transaction.c
@@ -669,7 +669,8 @@ repeat:
 			spin_unlock(&journal_datalist_lock);
 			unlock_journal(journal);
 			/* commit wakes up all shadow buffers after IO */
-			sleep_on(&jh2bh(jh)->b_wait);
+			wait_event(jh2bh(jh)->b_wait,
+						jh->b_jlist != BJ_Shadow);
 			lock_journal(journal);
 			goto repeat;
 		}

_