Re: mdadm memory leak?

On Tuesday July 5, eric@sandall.us wrote:

On Wed, 6 Jul 2005, Neil Brown wrote:

quoted

On Tuesday July 5, eric@sandall.us wrote:

<snip>

quoted

quoted

  Active / Total Objects (% used)    : 12080357 / 12132001 (99.6%)
  Active / Total Slabs (% used)      : 176776 / 176776 (100.0%)
  Active / Total Caches (% used)     : 66 / 101 (65.3%)
  Active / Total Size (% used)       : 668099.80K / 671784.20K (99.5%)
  Minimum / Average / Maximum Object : 0.01K / 0.05K / 128.00K

   OBJS ACTIVE  USE OBJ SIZE  SLABS OBJ/SLAB CACHE SIZE NAME
6013634 6013300  99%    0.02K  26609      226    106436K biovec-1
6013306 6013297  99%    0.09K 146666       41    586664K bio

These two lines point to the problem - it is a leak of 'bio's.

This patch fixes it.

### Diffstat output
./drivers/md/md.c |    1 +
1 files changed, 1 insertion(+)

diff ./drivers/md/md.c~current~ ./drivers/md/md.c

--- ./drivers/md/md.c~current~	2005-06-30 11:07:38.000000000 +1000
+++ ./drivers/md/md.c	2005-06-28 13:02:04.000000000 +1000

@@ -338,6 +338,7 @@ static int super_written(struct bio *bio

	if (atomic_dec_and_test(&rdev->mddev->pending_writes))
		wake_up(&rdev->mddev->sb_wait);
+	bio_put(bio);
	return 0;
}

Which kernel is that against? It doesn't apply against my 2.6.10-as7
nor David Kowis' 2.6.12.1 and 2.6.11.12 kernel source trees
(I don't even have the super_written function the patch is referring
to)?

It was against ... 2.6.12-mm1 or similar I think.  Looks like the bug
didn't get to main-line until just before the fix, which it good.

2.6.10-as7 seems to have a different though related bug.
It contains the patch:
 http://www.acm.cs.rpi.edu/~dilinger/patches/2.6.10/as7/linux-2.6.10-as7/051-md_sync_page_io_max_vecs.patch

which contains:
diff -Nru a/drivers/md/md.c b/drivers/md/md.c

--- a/drivers/md/md.c	2005-01-22 00:16:02 -08:00
+++ b/drivers/md/md.c	2005-01-22 00:16:02 -08:00

@@ -332,29 +332,26 @@
 static int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 		   struct page *page, int rw)
 {
-	struct bio bio;
-	struct bio_vec vec;
+	struct bio *bio = bio_alloc(GFP_KERNEL, 1);
 	struct completion event;
+	int ret;
+
+	bio_get(bio);

....
+	bio_put(bio);
+	return ret;
 }

bio_alloc sets the refcount to 1.
bio_get increments it to 2.
bio_put sets it back to 1.  But it never reaches zero.

You want to get rid of that bio_get near the top of sync_page_io.

NeilBrown