zram: Fix deadlock bug in partial read/write
authorMinchan Kim <minchan@kernel.org>
Wed, 30 Jan 2013 02:41:39 +0000 (11:41 +0900)
committerBen Hutchings <ben@decadent.org.uk>
Wed, 6 Mar 2013 03:23:48 +0000 (03:23 +0000)
commit 7e5a5104c6af709a8d97d5f4711e7c917761d464 upstream.

Now zram allocates new page with GFP_KERNEL in zram I/O path
if IO is partial. Unfortunately, It may cause deadlock with
reclaim path like below.

write_page from fs
fs_lock
allocation(GFP_KERNEL)
reclaim
pageout
write_page from fs
fs_lock <-- deadlock

This patch fixes it by using GFP_NOIO.  In read path, we
reorganize code flow so that kmap_atomic is called after the
GFP_NOIO allocation.

Acked-by: Jerome Marchand <jmarchand@redhat.com>
Acked-by: Nitin Gupta <ngupta@vflare.org>
[ penberg@kernel.org: don't use GFP_ATOMIC ]
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Signed-off-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[bwh: Backported to 3.2: no reordering is needed in the read path]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
drivers/staging/zram/zram_drv.c

index 09de99f..2594a31 100644 (file)
@@ -242,7 +242,7 @@ static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 
        if (is_partial_io(bvec)) {
                /* Use  a temporary buffer to decompress the page */
-               uncmem = kmalloc(PAGE_SIZE, GFP_KERNEL);
+               uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
                if (!uncmem) {
                        pr_info("Error allocating temp memory!\n");
                        return -ENOMEM;
@@ -338,7 +338,7 @@ static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
                 * This is a partial IO. We need to read the full page
                 * before to write the changes.
                 */
-               uncmem = kmalloc(PAGE_SIZE, GFP_KERNEL);
+               uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
                if (!uncmem) {
                        pr_info("Error allocating temp memory!\n");
                        ret = -ENOMEM;