[PATCH 2/6] audit: Fix possible spurious -ENOSPC error

[Jan Kara]

Hi Dan!

On Mon 02-07-18 09:05:49, Dan Carpenter wrote:
> url:    https://github.com/0day-ci/linux/commits/Jan-Kara/audit-Fix-various-races-when-tagging-and-untagging-mounts/20180629-043337
> 
> smatch warnings:
> kernel/audit_tree.c:484 tag_chunk() warn: inconsistent returns 'mutex:&audit_tree_group->mark_mutex'.
>   Locked on:   line 400
>   Unlocked on: line 411
> 
> # https://github.com/0day-ci/linux/commit/86c9c9a738e409c85891519c17d94043b7f434d5
> git remote add linux-review https://github.com/0day-ci/linux
> git remote update linux-review
> git checkout 86c9c9a738e409c85891519c17d94043b7f434d5
> vim +484 kernel/audit_tree.c
> 
> 74c3cbe33 Al Viro         2007-07-22  386  
> 74c3cbe33 Al Viro         2007-07-22  387  /* the first tagged inode becomes root of tree */
> 74c3cbe33 Al Viro         2007-07-22  388  static int tag_chunk(struct inode *inode, struct audit_tree *tree)
> 74c3cbe33 Al Viro         2007-07-22  389  {
> e61ce8673 Eric Paris      2009-12-17  390  	struct fsnotify_mark *old_entry, *chunk_entry;
> 74c3cbe33 Al Viro         2007-07-22  391  	struct audit_tree *owner;
> 74c3cbe33 Al Viro         2007-07-22  392  	struct audit_chunk *chunk, *old;
> 74c3cbe33 Al Viro         2007-07-22  393  	struct node *p;
> 74c3cbe33 Al Viro         2007-07-22  394  	int n;
> 74c3cbe33 Al Viro         2007-07-22  395  
> 86c9c9a73 Jan Kara        2018-06-28  396  	mutex_lock(&audit_tree_group->mark_mutex);
> b1362edfe Jan Kara        2016-12-21  397  	old_entry = fsnotify_find_mark(&inode->i_fsnotify_marks,
> b1362edfe Jan Kara        2016-12-21  398  				       audit_tree_group);
> 28a3a7eb3 Eric Paris      2009-12-17  399  	if (!old_entry)
> 74c3cbe33 Al Viro         2007-07-22  400  		return create_chunk(inode, tree);
>                                                                ^^^^^^^^^^^^^^^^^^^^^^^^^
> Should we drop the lock before this return?

No, because create_chunk() drops &audit_tree_group->mark_mutex in all the
cases. It's a bit ugly to have a function entered with mutex held and
release it but in this case it's somewhat difficult to avoid...

								Honza
-- 
Jan Kara <jack at suse.com>
SUSE Labs, CR