Commit e09b457b (block: simplify holder symlink handling) incorrectly assumed that there is only one link at maximum. dm may use multiple links and expects block layer to track reference count for each link, which is different from and unrelated to the exclusive device holder identified by @holder when the device is opened. Remove the single holder assumption and automatic removal of the link and revive the per-link reference count tracking. The code essentially behaves the same as before commit e09b457b sans the unnecessary kobject reference count dancing. While at it, note that this facility should not be used by anyone else than the current ones. Sysfs symlinks shouldn't be abused like this and the whole thing doesn't belong in the block layer at all. Signed-off-by: Tejun Heo <tj@xxxxxxxxxx> Reported-by: Milan Broz <mbroz@xxxxxxxxxx> Cc: Jun'ichi Nomura <j-nomura@xxxxxxxxxxxxx> Cc: Neil Brown <neilb@xxxxxxx> Cc: linux-raid@xxxxxxxxxxxxxxx Cc: Kay Sievers <kay.sievers@xxxxxxxx> --- Thanks for the test commands. They were very helpful. Can you please test this one? Thanks. drivers/md/dm-table.c | 1 drivers/md/md.c | 1 fs/block_dev.c | 93 ++++++++++++++++++++++++++++++++++++++++---------- include/linux/fs.h | 8 +++- 4 files changed, 84 insertions(+), 19 deletions(-) Index: work/include/linux/fs.h =================================================================== --- work.orig/include/linux/fs.h +++ work/include/linux/fs.h @@ -664,7 +664,7 @@ struct block_device { int bd_holders; bool bd_write_holder; #ifdef CONFIG_SYSFS - struct gendisk * bd_holder_disk; /* for sysfs slave linkng */ + struct list_head bd_holder_disks; #endif struct block_device * bd_contains; unsigned bd_block_size; @@ -2045,12 +2045,18 @@ extern struct block_device *blkdev_get_b extern int blkdev_put(struct block_device *bdev, fmode_t mode); #ifdef CONFIG_SYSFS extern int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk); +extern void bd_unlink_disk_holder(struct block_device *bdev, + struct gendisk *disk); #else static inline int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) { return 0; } +static inline void bd_unlink_disk_holder(struct block_device *bdev, + struct gendisk *disk) +{ +} #endif #endif Index: work/drivers/md/dm-table.c =================================================================== --- work.orig/drivers/md/dm-table.c +++ work/drivers/md/dm-table.c @@ -347,6 +347,7 @@ static void close_dev(struct dm_dev_inte if (!d->dm_dev.bdev) return; + bd_unlink_disk_holder(d->dm_dev.bdev, dm_disk(md)); blkdev_put(d->dm_dev.bdev, d->dm_dev.mode | FMODE_EXCL); d->dm_dev.bdev = NULL; } Index: work/drivers/md/md.c =================================================================== --- work.orig/drivers/md/md.c +++ work/drivers/md/md.c @@ -1907,6 +1907,7 @@ static void unbind_rdev_from_array(mdk_r MD_BUG(); return; } + bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); list_del_rcu(&rdev->same_set); printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); rdev->mddev = NULL; Index: work/fs/block_dev.c =================================================================== --- work.orig/fs/block_dev.c +++ work/fs/block_dev.c @@ -426,6 +426,9 @@ static void init_once(void *foo) mutex_init(&bdev->bd_mutex); INIT_LIST_HEAD(&bdev->bd_inodes); INIT_LIST_HEAD(&bdev->bd_list); +#ifdef CONFIG_SYSFS + INIT_LIST_HEAD(&bdev->bd_holder_disks); +#endif inode_init_once(&ei->vfs_inode); /* Initialize mutex for freeze. */ mutex_init(&bdev->bd_fsfreeze_mutex); @@ -773,6 +776,23 @@ static struct block_device *bd_start_cla } #ifdef CONFIG_SYSFS +struct bd_holder_disk { + struct list_head list; + struct gendisk *disk; + int refcnt; +}; + +static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, + struct gendisk *disk) +{ + struct bd_holder_disk *holder; + + list_for_each_entry(holder, &bdev->bd_holder_disks, list) + if (holder->disk == disk) + return holder; + return NULL; +} + static int add_symlink(struct kobject *from, struct kobject *to) { return sysfs_create_link(from, to, kobject_name(to)); @@ -788,6 +808,8 @@ static void del_symlink(struct kobject * * @bdev: the claimed slave bdev * @disk: the holding disk * + * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. + * * This functions creates the following sysfs symlinks. * * - from "slaves" directory of the holder @disk to the claimed @bdev @@ -811,47 +833,83 @@ static void del_symlink(struct kobject * */ int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) { + struct bd_holder_disk *holder; int ret = 0; mutex_lock(&bdev->bd_mutex); - WARN_ON_ONCE(!bdev->bd_holder || bdev->bd_holder_disk); + WARN_ON_ONCE(!bdev->bd_holder); /* FIXME: remove the following once add_disk() handles errors */ if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) goto out_unlock; - ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); - if (ret) + holder = bd_find_holder_disk(bdev, disk); + if (holder) { + holder->refcnt++; goto out_unlock; + } - ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); - if (ret) { - del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); + holder = kzalloc(sizeof(*holder), GFP_KERNEL); + if (!holder) { + ret = -ENOMEM; goto out_unlock; } - bdev->bd_holder_disk = disk; + INIT_LIST_HEAD(&holder->list); + holder->disk = disk; + holder->refcnt = 1; + + ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); + if (ret) + goto out_free; + + ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); + if (ret) + goto out_del; + + list_add(&holder->list, &bdev->bd_holder_disks); + goto out_unlock; + +out_del: + del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); +out_free: + kfree(holder); out_unlock: mutex_unlock(&bdev->bd_mutex); return ret; } EXPORT_SYMBOL_GPL(bd_link_disk_holder); -static void bd_unlink_disk_holder(struct block_device *bdev) +/** + * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() + * @bdev: the calimed slave bdev + * @disk: the holding disk + * + * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. + * + * CONTEXT: + * Might sleep. + */ +void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) { - struct gendisk *disk = bdev->bd_holder_disk; + struct bd_holder_disk *holder; - bdev->bd_holder_disk = NULL; - if (!disk) - return; + mutex_lock(&bdev->bd_mutex); - del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); - del_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); + holder = bd_find_holder_disk(bdev, disk); + + if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { + del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); + del_symlink(bdev->bd_part->holder_dir, + &disk_to_dev(disk)->kobj); + list_del_init(&holder->list); + kfree(holder); + } + + mutex_unlock(&bdev->bd_mutex); } -#else -static inline void bd_unlink_disk_holder(struct block_device *bdev) -{ } +EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); #endif /** @@ -1374,7 +1432,6 @@ int blkdev_put(struct block_device *bdev * unblock evpoll if it was a write holder. */ if (bdev_free) { - bd_unlink_disk_holder(bdev); if (bdev->bd_write_holder) { disk_unblock_events(bdev->bd_disk); bdev->bd_write_holder = false; -- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel