[2.6.20.17 review 28/58] dm io: fix panic on large request

[Willy Tarreau <w@1wt.eu>]

[-- Attachment #1: 0028-dm-io-fix-panic-on-large-request.patch --]
[-- Type: text/plain, Size: 1339 bytes --]

Flush workqueue before releasing bioset and mopools in dm-crypt.  There can
be finished but not yet released request.

Call chain causing oops:
  run workqueue
    dec_pending
      bio_endio(...);
      	<remove device request - remove mempool>
      mempool_free(io, cc->io_pool);

This usually happens when cryptsetup create temporary
luks mapping in the beggining of crypt device activation.

When dm-core calls destructor crypt_dtr, no new request
are possible.

Signed-off-by: Milan Broz <mbroz@redhat.com>
Cc: Chuck Ebbert <cebbert@redhat.com>
Cc: Patrick McHardy <kaber@trash.net>
Acked-by: Alasdair G Kergon <agk@redhat.com>
Cc: Christophe Saout <christophe@saout.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Willy Tarreau <w@1wt.eu>
---
 drivers/md/dm-crypt.c |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)

diff --git a/drivers/md/dm-crypt.c b/drivers/md/dm-crypt.c
index b9ff4e3..6b50991 100644
--- a/drivers/md/dm-crypt.c
+++ b/drivers/md/dm-crypt.c
@@ -917,6 +917,8 @@ static void crypt_dtr(struct dm_target *ti)
 {
 	struct crypt_config *cc = (struct crypt_config *) ti->private;
 
+	flush_workqueue(_kcryptd_workqueue);
+
 	bioset_free(cc->bs);
 	mempool_destroy(cc->page_pool);
 	mempool_destroy(cc->io_pool);
-- 
1.5.2.5

--