udf: Fix deadlock between writeback and udf_setsize()
authorJan Kara <jack@suse.cz>
Tue, 13 Jun 2017 14:20:25 +0000 (16:20 +0200)
committerBen Hutchings <ben@decadent.org.uk>
Thu, 12 Oct 2017 14:27:10 +0000 (15:27 +0100)
commitaf64015b4f713ad1d63761c3d5c0f1e486677a72
tree6422011c9a3aa38c08550db4ab228681e052f8fa
parentd18777048c7596577023f9ff3ea5dd8e9ff81f7d
udf: Fix deadlock between writeback and udf_setsize()

commit f2e95355891153f66d4156bf3a142c6489cd78c6 upstream.

udf_setsize() called truncate_setsize() with i_data_sem held. Thus
truncate_pagecache() called from truncate_setsize() could lock a page
under i_data_sem which can deadlock as page lock ranks below
i_data_sem - e. g. writeback can hold page lock and try to acquire
i_data_sem to map a block.

Fix the problem by moving truncate_setsize() calls from under
i_data_sem. It is safe for us to change i_size without holding
i_data_sem as all the places that depend on i_size being stable already
hold inode_lock.

Fixes: 7e49b6f2480cb9a9e7322a91592e56a5c85361f5
Signed-off-by: Jan Kara <jack@suse.cz>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
fs/udf/inode.c