writeback: keep superblock pinned during cgroup writeback association switches
authorTejun Heo <tj@kernel.org>
Tue, 16 Feb 2016 18:34:07 +0000 (11:34 -0700)
committerJens Axboe <axboe@fb.com>
Tue, 16 Feb 2016 18:34:07 +0000 (11:34 -0700)
If cgroup writeback is in use, an inode is associated with a cgroup
for writeback.  If the inode's main dirtier changes to another cgroup,
the association gets updated asynchronously.  Nothing was pinning the
superblock while such switches are in progress and superblock could go
away while async switching is pending or in progress leading to
crashes like the following.

 kernel BUG at fs/jbd2/transaction.c:319!
 invalid opcode: 0000 [#1] SMP DEBUG_PAGEALLOC
 CPU: 1 PID: 29158 Comm: kworker/1:10 Not tainted 4.5.0-rc3 #51
 Hardware name: Google Google, BIOS Google 01/01/2011
 Workqueue: events inode_switch_wbs_work_fn
 task: ffff880213dbbd40 ti: ffff880209264000 task.ti: ffff880209264000
 RIP: 0010:[<ffffffff803e6922>]  [<ffffffff803e6922>] start_this_handle+0x382/0x3e0
 RSP: 0018:ffff880209267c30  EFLAGS: 00010202
 ...
 Call Trace:
  [<ffffffff803e6be4>] jbd2__journal_start+0xf4/0x190
  [<ffffffff803cfc7e>] __ext4_journal_start_sb+0x4e/0x70
  [<ffffffff803b31ec>] ext4_evict_inode+0x12c/0x3d0
  [<ffffffff8035338b>] evict+0xbb/0x190
  [<ffffffff80354190>] iput+0x130/0x190
  [<ffffffff80360223>] inode_switch_wbs_work_fn+0x343/0x4c0
  [<ffffffff80279819>] process_one_work+0x129/0x300
  [<ffffffff80279b16>] worker_thread+0x126/0x480
  [<ffffffff8027ed14>] kthread+0xc4/0xe0
  [<ffffffff809771df>] ret_from_fork+0x3f/0x70

Fix it by bumping s_active while cgroup association switching is in
flight.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-and-tested-by: Tahsin Erdogan <tahsin@google.com>
Link: http://lkml.kernel.org/g/CAAeU0aNCq7LGODvVGRU-oU_o-6enii5ey0p1c26D1ZzYwkDc5A@mail.gmail.com
Fixes: d10c80955265 ("writeback: implement foreign cgroup inode bdi_writeback switching")
Cc: stable@vger.kernel.org #v4.5+
Signed-off-by: Jens Axboe <axboe@fb.com>
fs/fs-writeback.c

index 6915c950e6e8aeefb7c6162e7e7410de98f4e309..1f76d8950a57fc7bbdb3f8d98ec056fe89970b5c 100644 (file)
@@ -317,6 +317,7 @@ static void inode_switch_wbs_work_fn(struct work_struct *work)
        struct inode_switch_wbs_context *isw =
                container_of(work, struct inode_switch_wbs_context, work);
        struct inode *inode = isw->inode;
+       struct super_block *sb = inode->i_sb;
        struct address_space *mapping = inode->i_mapping;
        struct bdi_writeback *old_wb = inode->i_wb;
        struct bdi_writeback *new_wb = isw->new_wb;
@@ -423,6 +424,7 @@ skip_switch:
        wb_put(new_wb);
 
        iput(inode);
+       deactivate_super(sb);
        kfree(isw);
 }
 
@@ -469,11 +471,14 @@ static void inode_switch_wbs(struct inode *inode, int new_wb_id)
 
        /* while holding I_WB_SWITCH, no one else can update the association */
        spin_lock(&inode->i_lock);
+
        if (inode->i_state & (I_WB_SWITCH | I_FREEING) ||
-           inode_to_wb(inode) == isw->new_wb) {
-               spin_unlock(&inode->i_lock);
-               goto out_free;
-       }
+           inode_to_wb(inode) == isw->new_wb)
+               goto out_unlock;
+
+       if (!atomic_inc_not_zero(&inode->i_sb->s_active))
+               goto out_unlock;
+
        inode->i_state |= I_WB_SWITCH;
        spin_unlock(&inode->i_lock);
 
@@ -489,6 +494,8 @@ static void inode_switch_wbs(struct inode *inode, int new_wb_id)
        call_rcu(&isw->rcu_head, inode_switch_wbs_rcu_fn);
        return;
 
+out_unlock:
+       spin_unlock(&inode->i_lock);
 out_free:
        if (isw->new_wb)
                wb_put(isw->new_wb);