tmpfs: fix regression hang in fallocate undo
authorHugh Dickins <hughd@google.com>
Sun, 10 Jul 2016 23:46:32 +0000 (16:46 -0700)
committerSasha Levin <sasha.levin@oracle.com>
Tue, 12 Jul 2016 12:41:10 +0000 (08:41 -0400)
commit116c75f642ad4a6d267f399c9f1fe8b91fb822c5
tree180133ef4e45a3ebb7dc2e689f4312a3ada43005
parentb7aa372fec2c1d4a1fcf4398ae88ba94134b9522
tmpfs: fix regression hang in fallocate undo

[ Upstream commit 7f556567036cb7f89aabe2f0954b08566b4efb53 ]

The well-spotted fallocate undo fix is good in most cases, but not when
fallocate failed on the very first page.  index 0 then passes lend -1
to shmem_undo_range(), and that has two bad effects: (a) that it will
undo every fallocation throughout the file, unrestricted by the current
range; but more importantly (b) it can cause the undo to hang, because
lend -1 is treated as truncation, which makes it keep on retrying until
every page has gone, but those already fully instantiated will never go
away.  Big thank you to xfstests generic/269 which demonstrates this.

Fixes: b9b4bb26af01 ("tmpfs: don't undo fallocate past its last page")
Cc: stable@vger.kernel.org
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
mm/shmem.c