Re: [PATCH][RFC] freezer: Add CHECKPOINTING state to safeguard container checkpoint

[Oren Laadan <orenl-eQaUEPhvms7ENvBUuze7eA@public.gmane.org>]

Hi,

While trying Matt's patch I hit a problem reported lockdep (report
further below).

There is a possible deadlock in the cgroup_freezer. The problem is
a locking order, and actually exists in the code already, and only
exposed by this patch.

From the lock-ordering comment in cgroup_freezer.c:

 * freezer_fork() (preserving fork() performance ...)
 * task->alloc_lock (to get task's cgroup)
 * freezer->lock
 *  sighand->siglock (if the cgroup is freezing)

...

 * freezer_write() (unfreeze):
 * cgroup_mutex
 *  freezer->lock
 *   read_lock css_set_lock (cgroup iterator start)
 *    task->alloc_lock (to prevent races with freeze_task())
 *     sighand->siglock

'task->alloc_lock' and 'freezer->lock' are taken in different order.

Oren.

-------------
kernel:
kernel: =======================================================
kernel: [ INFO: possible circular locking dependency detected ]
kernel: 2.6.30-rc7-orenl #366
kernel: -------------------------------------------------------
kernel: ckpt/2787 is trying to acquire lock:
kernel:  (&freezer->lock){......}, at: [<c0157b35>]
freezer_checkpointing+0x35/0x80
kernel:
kernel: but task is already holding lock:
kernel:  (&p->alloc_lock){+.+...}, at: [<c0157b21>]
freezer_checkpointing+0x21/0x80
kernel:
kernel: which lock already depends on the new lock.
kernel:
kernel:
kernel: the existing dependency chain (in reverse order) is:
kernel:
kernel: -> #2 (&p->alloc_lock){+.+...}:
kernel:        [<c0148f32>] validate_chain+0xa82/0xfc0
kernel:        [<c0149708>] __lock_acquire+0x298/0x9a0
kernel:        [<c0149e6e>] lock_acquire+0x5e/0x80
kernel:        [<c0336633>] _spin_lock+0x33/0x40
kernel:        [<c0155285>] cgroup_iter_start+0xa5/0xe0
kernel:        [<c015781a>] update_freezer_state+0x1a/0x70
kernel:        [<c01578e7>] freezer_write+0x77/0x160
kernel:        [<c0156576>] cgroup_file_write+0x156/0x210
kernel:        [<c0186c56>] vfs_write+0x96/0x130
kernel:        [<c01871bd>] sys_write+0x3d/0x70
kernel:        [<c0102c38>] sysenter_do_call+0x12/0x36
kernel:        [<ffffffff>] 0xffffffff
kernel:
kernel: -> #1 (css_set_lock){++++..}:
kernel:        [<c0148f32>] validate_chain+0xa82/0xfc0
kernel:        [<c0149708>] __lock_acquire+0x298/0x9a0
kernel:        [<c0149e6e>] lock_acquire+0x5e/0x80
kernel:        [<c03366c3>] _write_lock+0x33/0x40
kernel:        [<c015522b>] cgroup_iter_start+0x4b/0xe0
kernel:        [<c015781a>] update_freezer_state+0x1a/0x70
kernel:        [<c01578e7>] freezer_write+0x77/0x160
kernel:        [<c0156576>] cgroup_file_write+0x156/0x210
kernel:        [<c0186c56>] vfs_write+0x96/0x130
kernel:        [<c01871bd>] sys_write+0x3d/0x70
kernel:        [<c0102c38>] sysenter_do_call+0x12/0x36
kernel:        [<ffffffff>] 0xffffffff
kernel:
kernel: -> #0 (&freezer->lock){......}:
kernel:        [<c0148a21>] validate_chain+0x571/0xfc0
kernel:        [<c0149708>] __lock_acquire+0x298/0x9a0
kernel:        [<c0149e6e>] lock_acquire+0x5e/0x80
kernel:        [<c0336939>] _spin_lock_irq+0x39/0x50
kernel:        [<c0157b35>] freezer_checkpointing+0x35/0x80
kernel:        [<c0157bbd>] cgroup_freezer_begin_checkpoint+0xd/0x30
kernel:        [<c02185c6>] do_checkpoint+0xf6/0x6a0
kernel:        [<c02172a6>] sys_checkpoint+0x46/0x90
kernel:        [<c0102c38>] sysenter_do_call+0x12/0x36
kernel:        [<ffffffff>] 0xffffffff
kernel:

Matt Helsley wrote:
> The CHECKPOINTING state prevents userspace from unfreezing tasks until
> sys_checkpoint() is finished. When doing container checkpoint userspace
> will do:
> 
> echo FROZEN > /cgroups/my_container/freezer.state
> ...
> rc = sys_checkpoint( <pid of container root> );
> 
> To ensure a consistent checkpoint image userspace should not be allowed
> to thaw the cgroup (echo THAWED > /cgroups/my_container/freezer.state)
> during checkpoint.
> 

[...]