[PATCH] md: avoid possible BUG_ON in md bitmap handling
authorNeil Brown <neilb@suse.de>
Thu, 8 Feb 2007 22:20:37 +0000 (14:20 -0800)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Fri, 9 Feb 2007 17:25:47 +0000 (09:25 -0800)
md/bitmap tracks how many active write requests are pending on blocks
associated with each bit in the bitmap, so that it knows when it can clear
the bit (when count hits zero).

The counter has 14 bits of space, so if there are ever more than 16383, we
cannot cope.

Currently the code just calles BUG_ON as "all" drivers have request queue
limits much smaller than this.

However is seems that some don't.  Apparently some multipath configurations
can allow more than 16383 concurrent write requests.

So, in this unlikely situation, instead of calling BUG_ON we now wait
for the count to drop down a bit.  This requires a new wait_queue_head,
some waiting code, and a wakeup call.

Tested by limiting the counter to 20 instead of 16383 (writes go a lot slower
in that case...).

Signed-off-by: Neil Brown <neilb@suse.de>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
drivers/md/bitmap.c
include/linux/raid/bitmap.h

index 11108165e26420a6f8a3eabb8794e1e0ad833754..059704fbb753ada7df4f97da979071de1cac073a 100644 (file)
@@ -1160,6 +1160,22 @@ int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sect
                        return 0;
                }
 
+               if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) {
+                       DEFINE_WAIT(__wait);
+                       /* note that it is safe to do the prepare_to_wait
+                        * after the test as long as we do it before dropping
+                        * the spinlock.
+                        */
+                       prepare_to_wait(&bitmap->overflow_wait, &__wait,
+                                       TASK_UNINTERRUPTIBLE);
+                       spin_unlock_irq(&bitmap->lock);
+                       bitmap->mddev->queue
+                               ->unplug_fn(bitmap->mddev->queue);
+                       schedule();
+                       finish_wait(&bitmap->overflow_wait, &__wait);
+                       continue;
+               }
+
                switch(*bmc) {
                case 0:
                        bitmap_file_set_bit(bitmap, offset);
@@ -1169,7 +1185,7 @@ int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sect
                case 1:
                        *bmc = 2;
                }
-               BUG_ON((*bmc & COUNTER_MAX) == COUNTER_MAX);
+
                (*bmc)++;
 
                spin_unlock_irq(&bitmap->lock);
@@ -1207,6 +1223,9 @@ void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long secto
                if (!success && ! (*bmc & NEEDED_MASK))
                        *bmc |= NEEDED_MASK;
 
+               if ((*bmc & COUNTER_MAX) == COUNTER_MAX)
+                       wake_up(&bitmap->overflow_wait);
+
                (*bmc)--;
                if (*bmc <= 2) {
                        set_page_attr(bitmap,
@@ -1431,6 +1450,7 @@ int bitmap_create(mddev_t *mddev)
        spin_lock_init(&bitmap->lock);
        atomic_set(&bitmap->pending_writes, 0);
        init_waitqueue_head(&bitmap->write_wait);
+       init_waitqueue_head(&bitmap->overflow_wait);
 
        bitmap->mddev = mddev;
 
index ebd42a3710b4eefc44bec8a9880bd88af6472bcb..6db9a4c153556a7236caf4dbf376b27900af3f15 100644 (file)
@@ -247,6 +247,7 @@ struct bitmap {
 
        atomic_t pending_writes; /* pending writes to the bitmap file */
        wait_queue_head_t write_wait;
+       wait_queue_head_t overflow_wait;
 
 };