[PATCH 3.16 037/192] udf: Fix deadlock between writeback and udf_setsize()

[Ben Hutchings]

3.16.49-rc1 review patch.  If anyone has any objections, please let me know.

------------------

From: Jan Kara <jack@xxxxxxx>

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@xxxxxxx>
Signed-off-by: Ben Hutchings <ben@xxxxxxxxxxxxxxx>
---
 fs/udf/inode.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

--- a/fs/udf/inode.c
+++ b/fs/udf/inode.c
@@ -1238,8 +1238,8 @@ int udf_setsize(struct inode *inode, lof
 			return err;
 		}
 set_size:
-		truncate_setsize(inode, newsize);
 		up_write(&iinfo->i_data_sem);
+		truncate_setsize(inode, newsize);
 	} else {
 		if (iinfo->i_alloc_type == ICBTAG_FLAG_AD_IN_ICB) {
 			down_write(&iinfo->i_data_sem);
@@ -1256,9 +1256,9 @@ set_size:
 					  udf_get_block);
 		if (err)
 			return err;
+		truncate_setsize(inode, newsize);
 		down_write(&iinfo->i_data_sem);
 		udf_clear_extent_cache(inode);
-		truncate_setsize(inode, newsize);
 		udf_truncate_extents(inode);
 		up_write(&iinfo->i_data_sem);
 	}