tty/vt: Fix vc_deallocate() lock order
authorPeter Hurley <peter@hurleysoftware.com>
Fri, 17 May 2013 16:41:03 +0000 (12:41 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 20 May 2013 19:15:59 +0000 (12:15 -0700)
commit421b40a6286ee343d77d5e51f5ee6d04d7a2a90f
tree2e4cb121813c9894c41c64b65df0c0f29f7f4ded
parentdf957d2b9c5c8aa12f050f94c9f15236fb0e51f1
tty/vt: Fix vc_deallocate() lock order

Now that the tty port owns the flip buffers and i/o is allowed
from the driver even when no tty is attached, the destruction
of the tty port (and the flip buffers) must ensure that no
outstanding work is pending.

Unfortunately, this creates a lock order problem with the
console_lock (see attached lockdep report [1] below).

For single console deallocation, drop the console_lock prior
to port destruction. When multiple console deallocation,
defer port destruction until the consoles have been
deallocated.

tty_port_destroy() is not required if the port has not
been used; remove from vc_allocate() failure path.

[1] lockdep report from Dave Jones <davej@redhat.com>

 ======================================================
 [ INFO: possible circular locking dependency detected ]
 3.9.0+ #16 Not tainted
 -------------------------------------------------------
 (agetty)/26163 is trying to acquire lock:
 blocked:  ((&buf->work)){+.+...}, instance: ffff88011c8b0020, at: [<ffffffff81062065>] flush_work+0x5/0x2e0

 but task is already holding lock:
 blocked:  (console_lock){+.+.+.}, instance: ffffffff81c2fde0, at: [<ffffffff813bc201>] vt_ioctl+0xb61/0x1230

 which lock already depends on the new lock.

 the existing dependency chain (in reverse order) is:

 -> #1 (console_lock){+.+.+.}:
        [<ffffffff810b3f74>] lock_acquire+0xa4/0x210
        [<ffffffff810416c7>] console_lock+0x77/0x80
        [<ffffffff813c3dcd>] con_flush_chars+0x2d/0x50
        [<ffffffff813b32b2>] n_tty_receive_buf+0x122/0x14d0
        [<ffffffff813b7709>] flush_to_ldisc+0x119/0x170
        [<ffffffff81064381>] process_one_work+0x211/0x700
        [<ffffffff8106498b>] worker_thread+0x11b/0x3a0
        [<ffffffff8106ce5d>] kthread+0xed/0x100
        [<ffffffff81601cac>] ret_from_fork+0x7c/0xb0

 -> #0 ((&buf->work)){+.+...}:
        [<ffffffff810b349a>] __lock_acquire+0x193a/0x1c00
        [<ffffffff810b3f74>] lock_acquire+0xa4/0x210
        [<ffffffff810620ae>] flush_work+0x4e/0x2e0
        [<ffffffff81065305>] __cancel_work_timer+0x95/0x130
        [<ffffffff810653b0>] cancel_work_sync+0x10/0x20
        [<ffffffff813b8212>] tty_port_destroy+0x12/0x20
        [<ffffffff813c65e8>] vc_deallocate+0xf8/0x110
        [<ffffffff813bc20c>] vt_ioctl+0xb6c/0x1230
        [<ffffffff813b01a5>] tty_ioctl+0x285/0xd50
        [<ffffffff811ba825>] do_vfs_ioctl+0x305/0x530
        [<ffffffff811baad1>] sys_ioctl+0x81/0xa0
        [<ffffffff81601d59>] system_call_fastpath+0x16/0x1b

 other info that might help us debug this:

 [ 6760.076175]  Possible unsafe locking scenario:

        CPU0                    CPU1
        ----                    ----
   lock(console_lock);
                                lock((&buf->work));
                                lock(console_lock);
   lock((&buf->work));

  *** DEADLOCK ***

 1 lock on stack by (agetty)/26163:
  #0: blocked:  (console_lock){+.+.+.}, instance: ffffffff81c2fde0, at: [<ffffffff813bc201>] vt_ioctl+0xb61/0x1230
 stack backtrace:
 Pid: 26163, comm: (agetty) Not tainted 3.9.0+ #16
 Call Trace:
  [<ffffffff815edb14>] print_circular_bug+0x200/0x20e
  [<ffffffff810b349a>] __lock_acquire+0x193a/0x1c00
  [<ffffffff8100a269>] ? sched_clock+0x9/0x10
  [<ffffffff8100a269>] ? sched_clock+0x9/0x10
  [<ffffffff8100a200>] ? native_sched_clock+0x20/0x80
  [<ffffffff810b3f74>] lock_acquire+0xa4/0x210
  [<ffffffff81062065>] ? flush_work+0x5/0x2e0
  [<ffffffff810620ae>] flush_work+0x4e/0x2e0
  [<ffffffff81062065>] ? flush_work+0x5/0x2e0
  [<ffffffff810b15db>] ? mark_held_locks+0xbb/0x140
  [<ffffffff8113c8a3>] ? __free_pages_ok.part.57+0x93/0xc0
  [<ffffffff810b15db>] ? mark_held_locks+0xbb/0x140
  [<ffffffff810652f2>] ? __cancel_work_timer+0x82/0x130
  [<ffffffff81065305>] __cancel_work_timer+0x95/0x130
  [<ffffffff810653b0>] cancel_work_sync+0x10/0x20
  [<ffffffff813b8212>] tty_port_destroy+0x12/0x20
  [<ffffffff813c65e8>] vc_deallocate+0xf8/0x110
  [<ffffffff813bc20c>] vt_ioctl+0xb6c/0x1230
  [<ffffffff810aec41>] ? lock_release_holdtime.part.30+0xa1/0x170
  [<ffffffff813b01a5>] tty_ioctl+0x285/0xd50
  [<ffffffff812b00f6>] ? inode_has_perm.isra.46.constprop.61+0x56/0x80
  [<ffffffff811ba825>] do_vfs_ioctl+0x305/0x530
  [<ffffffff812b04db>] ? selinux_file_ioctl+0x5b/0x110
  [<ffffffff811baad1>] sys_ioctl+0x81/0xa0
  [<ffffffff81601d59>] system_call_fastpath+0x16/0x1b

Cc: Dave Jones <davej@redhat.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/tty/vt/vt.c
drivers/tty/vt/vt_ioctl.c
include/linux/vt_kern.h