blk-cgroup: Fix potential deadlock in blk-cgroup
I triggered a lockdep warning as following.
=======================================================
[ INFO: possible circular locking dependency detected ]
2.6.33-rc2 #1
-------------------------------------------------------
test_io_control/7357 is trying to acquire lock:
(blkio_list_lock){+.+...}, at: [<c053a990>] blkiocg_weight_write+0x82/0x9e
but task is already holding lock:
(&(&blkcg->lock)->rlock){......}, at: [<c053a949>] blkiocg_weight_write+0x3b/0x9e
which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #2 (&(&blkcg->lock)->rlock){......}:
[<c04583b7>] validate_chain+0x8bc/0xb9c
[<c0458dba>] __lock_acquire+0x723/0x789
[<c0458eb0>] lock_acquire+0x90/0xa7
[<c0692b0a>] _raw_spin_lock_irqsave+0x27/0x5a
[<c053a4e1>] blkiocg_add_blkio_group+0x1a/0x6d
[<c053cac7>] cfq_get_queue+0x225/0x3de
[<c053eec2>] cfq_set_request+0x217/0x42d
[<c052c8a6>] elv_set_request+0x17/0x26
[<c0532a0f>] get_request+0x203/0x2c5
[<c0532ae9>] get_request_wait+0x18/0x10e
[<c0533470>] __make_request+0x2ba/0x375
[<c0531985>] generic_make_request+0x28d/0x30f
[<c0532da7>] submit_bio+0x8a/0x8f
[<c04d827a>] submit_bh+0xf0/0x10f
[<c04d91d2>] ll_rw_block+0xc0/0xf9
[<f86e9705>] ext3_find_entry+0x319/0x544 [ext3]
[<f86eae58>] ext3_lookup+0x2c/0xb9 [ext3]
[<c04c3e1b>] do_lookup+0xd3/0x172
[<c04c56c8>] link_path_walk+0x5fb/0x95c
[<c04c5a65>] path_walk+0x3c/0x81
[<c04c5b63>] do_path_lookup+0x21/0x8a
[<c04c66cc>] do_filp_open+0xf0/0x978
[<c04c0c7e>] open_exec+0x1b/0xb7
[<c04c1436>] do_execve+0xbb/0x266
[<c04081a9>] sys_execve+0x24/0x4a
[<c04028a2>] ptregs_execve+0x12/0x18
-> #1 (&(&q->__queue_lock)->rlock){..-.-.}:
[<c04583b7>] validate_chain+0x8bc/0xb9c
[<c0458dba>] __lock_acquire+0x723/0x789
[<c0458eb0>] lock_acquire+0x90/0xa7
[<c0692b0a>] _raw_spin_lock_irqsave+0x27/0x5a
[<c053dd2a>] cfq_unlink_blkio_group+0x17/0x41
[<c053a6eb>] blkiocg_destroy+0x72/0xc7
[<c0467df0>] cgroup_diput+0x4a/0xb2
[<c04ca473>] dentry_iput+0x93/0xb7
[<c04ca4b3>] d_kill+0x1c/0x36
[<c04cb5c5>] dput+0xf5/0xfe
[<c04c6084>] do_rmdir+0x95/0xbe
[<c04c60ec>] sys_rmdir+0x10/0x12
[<c04027cc>] sysenter_do_call+0x12/0x32
-> #0 (blkio_list_lock){+.+...}:
[<c0458117>] validate_chain+0x61c/0xb9c
[<c0458dba>] __lock_acquire+0x723/0x789
[<c0458eb0>] lock_acquire+0x90/0xa7
[<c06929fd>] _raw_spin_lock+0x1e/0x4e
[<c053a990>] blkiocg_weight_write+0x82/0x9e
[<c0467f1e>] cgroup_file_write+0xc6/0x1c0
[<c04bd2f3>] vfs_write+0x8c/0x116
[<c04bd7c6>] sys_write+0x3b/0x60
[<c04027cc>] sysenter_do_call+0x12/0x32
other info that might help us debug this:
1 lock held by test_io_control/7357:
#0: (&(&blkcg->lock)->rlock){......}, at: [<c053a949>] blkiocg_weight_write+0x3b/0x9e
stack backtrace:
Pid: 7357, comm: test_io_control Not tainted 2.6.33-rc2 #1
Call Trace:
[<c045754f>] print_circular_bug+0x91/0x9d
[<c0458117>] validate_chain+0x61c/0xb9c
[<c0458dba>] __lock_acquire+0x723/0x789
[<c0458eb0>] lock_acquire+0x90/0xa7
[<c053a990>] ? blkiocg_weight_write+0x82/0x9e
[<c06929fd>] _raw_spin_lock+0x1e/0x4e
[<c053a990>] ? blkiocg_weight_write+0x82/0x9e
[<c053a990>] blkiocg_weight_write+0x82/0x9e
[<c0467f1e>] cgroup_file_write+0xc6/0x1c0
[<c0454df5>] ? trace_hardirqs_off+0xb/0xd
[<c044d93a>] ? cpu_clock+0x2e/0x44
[<c050e6ec>] ? security_file_permission+0xf/0x11
[<c04bcdda>] ? rw_verify_area+0x8a/0xad
[<c0467e58>] ? cgroup_file_write+0x0/0x1c0
[<c04bd2f3>] vfs_write+0x8c/0x116
[<c04bd7c6>] sys_write+0x3b/0x60
[<c04027cc>] sysenter_do_call+0x12/0x32
To prevent deadlock, we should take locks as following sequence:
blkio_list_lock -> queue_lock -> blkcg_lock.
The following patch should fix this bug.
Signed-off-by: Gui Jianfeng <guijianfeng@cn.fujitsu.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
diff --git a/block/blk-cgroup.c b/block/blk-cgroup.c
index 1fa2654..e7dbbaf 100644
--- a/block/blk-cgroup.c
+++ b/block/blk-cgroup.c
@@ -147,16 +147,16 @@
return -EINVAL;
blkcg = cgroup_to_blkio_cgroup(cgroup);
+ spin_lock(&blkio_list_lock);
spin_lock_irq(&blkcg->lock);
blkcg->weight = (unsigned int)val;
hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
- spin_lock(&blkio_list_lock);
list_for_each_entry(blkiop, &blkio_list, list)
blkiop->ops.blkio_update_group_weight_fn(blkg,
blkcg->weight);
- spin_unlock(&blkio_list_lock);
}
spin_unlock_irq(&blkcg->lock);
+ spin_unlock(&blkio_list_lock);
return 0;
}