From: Jan Kara Date: Tue, 3 Mar 2009 21:04:32 +0100 Subject: [Ocfs2-devel] [PATCH 6/7] vfs: Set special lockdep map for dirs only if not set by fs In-Reply-To: <1236095085.5330.4767.camel@laptop> References: <1235679305-3920-1-git-send-email-jack@suse.cz> <1235679305-3920-2-git-send-email-jack@suse.cz> <1235679305-3920-3-git-send-email-jack@suse.cz> <1235679305-3920-4-git-send-email-jack@suse.cz> <1235679305-3920-5-git-send-email-jack@suse.cz> <1235679305-3920-6-git-send-email-jack@suse.cz> <1235679305-3920-7-git-send-email-jack@suse.cz> <1235680042.4948.590.camel@laptop> <20090302092338.GA23880@duck.suse.cz> <1236095085.5330.4767.camel@laptop> Message-ID: <20090303200432.GA3912@duck.suse.cz> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: ocfs2-devel@oss.oracle.com On Tue 03-03-09 16:44:45, Peter Zijlstra wrote: > On Mon, 2009-03-02 at 10:23 +0100, Jan Kara wrote: > > > Not that much - the difference is between "system directories" and normal > > directories. Currently we also have xattr i_mutex rank differently than > > normal i_mutex and this is exactly the same case. Only OCFS2 has much more > > such system files and also system directories so it's impractical to learn > > VFS about all of them... > > But yes, probably I should update the changelog to say this so that > > a reader does not have to look up why we need it. > > Just for my education, what is a system directory? OCFS2 has two system directories AFAIK: The first one is a "root system directory" - it contains all system files like quota files, journal files, etc. The second one is the orphan directory - instead of orphan list in ext3, ocfs2 links every orphaned inode into the orphan directory. > > > As to the patch, I'd rather you introduce a helper, > > > lockdep_match_class() or somesuch, if you can convince VFS folks that > > > what you're wanting to do is sane. > > OK, sounds good. Does something like: > > #define lockdep_match_class(lock, key) lockdep_match_key(&(lock)->dep_map, key) > > > > and > > > > int lockdep_match_key(struct lockdep_map *lock, struct lock_class_key *key) > > { > > return lock->key == key; > > } > > > > Look fine to you? Thanks for opinion. > > Yep, looks good, thanks! OK, below is a new version of the patch. Honza -- Jan Kara SUSE Labs, CR --- From 272d47c01af8d7423e3db0d86c8c7dbb8535f2a3 Mon Sep 17 00:00:00 2001 From: Jan Kara Date: Thu, 12 Feb 2009 22:46:42 +0100 Subject: [PATCH] vfs: Set special lockdep map for dirs only if not set by fs Some filesystems need to set lockdep map for i_mutex differently for different directories. For example OCFS2 has system directories (for orphan inode tracking and for gathering all system files like journal or quota files into a single place) which have different locking locking rules than standard directories. For a filesystem setting lockdep map is naturaly done when the inode is read but we have to modify unlock_new_inode() not to overwrite the lockdep map the filesystem has set. CC: peterz at infradead.org CC: mingo at redhat.com Signed-off-by: Jan Kara --- fs/inode.c | 17 +++++++++++------ include/linux/lockdep.h | 15 +++++++++++++++ 2 files changed, 26 insertions(+), 6 deletions(-) diff --git a/fs/inode.c b/fs/inode.c index 913ab2d..147d34b 100644 --- a/fs/inode.c +++ b/fs/inode.c @@ -640,12 +640,17 @@ void unlock_new_inode(struct inode *inode) if (inode->i_mode & S_IFDIR) { struct file_system_type *type = inode->i_sb->s_type; - /* - * ensure nobody is actually holding i_mutex - */ - mutex_destroy(&inode->i_mutex); - mutex_init(&inode->i_mutex); - lockdep_set_class(&inode->i_mutex, &type->i_mutex_dir_key); + /* Set new key only if filesystem hasn't already changed it */ + if (!lockdep_match_class(&inode->i_mutex, + &type->i_mutex_key)) { + /* + * ensure nobody is actually holding i_mutex + */ + mutex_destroy(&inode->i_mutex); + mutex_init(&inode->i_mutex); + lockdep_set_class(&inode->i_mutex, + &type->i_mutex_dir_key); + } } #endif /* diff --git a/include/linux/lockdep.h b/include/linux/lockdep.h index 23bf02f..cb6b2f9 100644 --- a/include/linux/lockdep.h +++ b/include/linux/lockdep.h @@ -291,6 +291,16 @@ extern void lockdep_init_map(struct lockdep_map *lock, const char *name, #define lockdep_set_subclass(lock, sub) \ lockdep_init_map(&(lock)->dep_map, #lock, \ (lock)->dep_map.key, sub) +/* + * Compare locking classes + */ +#define lockdep_match_class(lock, key) lockdep_match_key(&(lock)->dep_map, key) + +static inline int lockdep_match_key(struct lockdep_map *lock, + struct lock_class_key *key) +{ + return lock->key == key; +} /* * Acquire a lock. @@ -352,6 +362,11 @@ static inline void lockdep_on(void) #define lockdep_set_class_and_subclass(lock, key, sub) \ do { (void)(key); } while (0) #define lockdep_set_subclass(lock, sub) do { } while (0) +/* + * We don't define lockdep_match_class() and lockdep_match_key() for !LOCKDEP + * case since the result is not well defined and the caller should rather + * #ifdef the call himself. + */ # define INIT_LOCKDEP # define lockdep_reset() do { debug_locks = 1; } while (0) -- 1.6.0.2