aboutsummaryrefslogtreecommitdiffstatshomepage
path: root/fs/ext4/ialloc.c
diff options
context:
space:
mode:
authorEric Biggers <ebiggers@google.com>2016-11-21 11:52:44 -0500
committerTheodore Ts'o <tytso@mit.edu>2016-11-21 11:52:44 -0500
commit2f8f5e76c7da787153f3b9791a2cb4fb384e0e67 (patch)
tree772893f6acee43e2b5a9e08271f93954515cce3d /fs/ext4/ialloc.c
parentext4: remove unused function ext4_aligned_io() (diff)
downloadwireguard-linux-2f8f5e76c7da787153f3b9791a2cb4fb384e0e67.tar.xz
wireguard-linux-2f8f5e76c7da787153f3b9791a2cb4fb384e0e67.zip
ext4: avoid lockdep warning when inheriting encryption context
On a lockdep-enabled kernel, xfstests generic/027 fails due to a lockdep warning when run on ext4 mounted with -o test_dummy_encryption: xfs_io/4594 is trying to acquire lock: (jbd2_handle ){++++.+}, at: [<ffffffff813096ef>] jbd2_log_wait_commit+0x5/0x11b but task is already holding lock: (jbd2_handle ){++++.+}, at: [<ffffffff813000de>] start_this_handle+0x354/0x3d8 The abbreviated call stack is: [<ffffffff813096ef>] ? jbd2_log_wait_commit+0x5/0x11b [<ffffffff8130972a>] jbd2_log_wait_commit+0x40/0x11b [<ffffffff813096ef>] ? jbd2_log_wait_commit+0x5/0x11b [<ffffffff8130987b>] ? __jbd2_journal_force_commit+0x76/0xa6 [<ffffffff81309896>] __jbd2_journal_force_commit+0x91/0xa6 [<ffffffff813098b9>] jbd2_journal_force_commit_nested+0xe/0x18 [<ffffffff812a6049>] ext4_should_retry_alloc+0x72/0x79 [<ffffffff812f0c1f>] ext4_xattr_set+0xef/0x11f [<ffffffff812cc35b>] ext4_set_context+0x3a/0x16b [<ffffffff81258123>] fscrypt_inherit_context+0xe3/0x103 [<ffffffff812ab611>] __ext4_new_inode+0x12dc/0x153a [<ffffffff812bd371>] ext4_create+0xb7/0x161 When a file is created in an encrypted directory, ext4_set_context() is called to set an encryption context on the new file. This calls ext4_xattr_set(), which contains a retry loop where the journal is forced to commit if an ENOSPC error is encountered. If the task actually were to wait for the journal to commit in this case, then it would deadlock because a handle remains open from __ext4_new_inode(), so the running transaction can't be committed yet. Fortunately, __jbd2_journal_force_commit() avoids the deadlock by not allowing the running transaction to be committed while the current task has it open. However, the above lockdep warning is still triggered. This was a false positive which was introduced by: 1eaa566d368b: jbd2: track more dependencies on transaction commit Fix the problem by passing the handle through the 'fs_data' argument to ext4_set_context(), then using ext4_xattr_set_handle() instead of ext4_xattr_set(). And in the case where no journal handle is specified and ext4_set_context() has to open one, add an ENOSPC retry loop since in that case it is the outermost transaction. Signed-off-by: Eric Biggers <ebiggers@google.com>
Diffstat (limited to 'fs/ext4/ialloc.c')
-rw-r--r--fs/ext4/ialloc.c3
1 files changed, 1 insertions, 2 deletions
diff --git a/fs/ext4/ialloc.c b/fs/ext4/ialloc.c
index 088afe07ddda..e57e8d90ea54 100644
--- a/fs/ext4/ialloc.c
+++ b/fs/ext4/ialloc.c
@@ -1115,8 +1115,7 @@ got:
}
if (encrypt) {
- /* give pointer to avoid set_context with journal ops. */
- err = fscrypt_inherit_context(dir, inode, &encrypt, true);
+ err = fscrypt_inherit_context(dir, inode, handle, true);
if (err)
goto fail_free_drop;
}