Re: [PATCH] Btrfs: don't remove raid type sysfs entries until unmount

[David Sterba <dsterba@suse.cz>]

On Fri, May 23, 2014 at 08:56:06AM -0400, Jeff Mahoney wrote:
> Sigh. Yep. kobject_cleanup caches ->name before the release function
> and ignores the cleared value. It seems the "free name if we alloced
> it" comment in there was leftover from the middle of a patch series
> Kay applied in late 2007. Commit 0f4dafc05 adds the comment while
> adding a flag to indicate that kobject_set_name_vargs set the name.
> The commit af5ca3f4e says kobject names must be dynamic but didn't
> update the comment.
> 
> Ok, so we can't save the strdup.

The tests have been running fine for more than 40 minutes, previously it
took a few minutes to trigger the bug.

my quick fix:

--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -8354,7 +8354,8 @@ static void __link_block_group(struct btrfs_space_info *space_info,
                int ret;

                kobject_get(&space_info->kobj); /* put in release */
-               kobj->name = get_raid_name(index);
+               kobj->name = kstrdup(get_raid_name(index), GFP_ATOMIC);
+               BUG_ON(!kobj->name);
                ret = kobject_init_and_add(kobj, &btrfs_raid_ktype,
                                           &space_info->kobj, NULL);
                if (ret) {
diff --git a/fs/btrfs/sysfs.c b/fs/btrfs/sysfs.c
index d742d7973d3e..ac26f91968a7 100644
--- a/fs/btrfs/sysfs.c
+++ b/fs/btrfs/sysfs.c
@@ -288,7 +288,7 @@ static struct attribute *raid_attributes[] = {

 static void release_raid_kobj(struct kobject *kobj)
 {
-       kobj->name = NULL;
+       /* kobj->name = NULL; */
        kobject_put(kobj->parent);
 }