[media] lirc: fix dead lock between open and wakeup_filter
authorSean Young <sean@mess.org>
Mon, 13 Feb 2017 12:35:44 +0000 (10:35 -0200)
committerMauro Carvalho Chehab <mchehab@s-opensource.com>
Fri, 3 Mar 2017 10:06:57 +0000 (07:06 -0300)
The locking in lirc needs improvement, but for now just fix this potential
deadlock.

======================================================
[ INFO: possible circular locking dependency detected ]
4.10.0-rc1+ #1 Not tainted
-------------------------------------------------------
bash/2502 is trying to acquire lock:
 (ir_raw_handler_lock){+.+.+.}, at: [<ffffffffc06f6a5e>] ir_raw_encode_scancode+0x3e/0xb0 [rc_core]

               but task is already holding lock:
 (&dev->lock){+.+.+.}, at: [<ffffffffc06f511f>] store_filter+0x9f/0x240 [rc_core]

               which lock already depends on the new lock.

               the existing dependency chain (in reverse order) is:

               -> #2 (&dev->lock){+.+.+.}:

[<ffffffffa110adad>] lock_acquire+0xfd/0x200
[<ffffffffa1921327>] mutex_lock_nested+0x77/0x6d0
[<ffffffffc06f436a>] rc_open+0x2a/0x80 [rc_core]
[<ffffffffc07114ca>] lirc_dev_fop_open+0xda/0x1e0 [lirc_dev]
[<ffffffffa12975e0>] chrdev_open+0xb0/0x210
[<ffffffffa128eb5a>] do_dentry_open+0x20a/0x2f0
[<ffffffffa128ffcc>] vfs_open+0x4c/0x80
[<ffffffffa12a35ec>] path_openat+0x5bc/0xc00
[<ffffffffa12a5271>] do_filp_open+0x91/0x100
[<ffffffffa12903f0>] do_sys_open+0x130/0x220
[<ffffffffa12904fe>] SyS_open+0x1e/0x20
[<ffffffffa19278c1>] entry_SYSCALL_64_fastpath+0x1f/0xc2
               -> #1 (lirc_dev_lock){+.+.+.}:
[<ffffffffa110adad>] lock_acquire+0xfd/0x200
[<ffffffffa1921327>] mutex_lock_nested+0x77/0x6d0
[<ffffffffc0711f47>] lirc_register_driver+0x67/0x59b [lirc_dev]
[<ffffffffc06db7f4>] ir_lirc_register+0x1f4/0x260 [ir_lirc_codec]
[<ffffffffc06f6cac>] ir_raw_handler_register+0x7c/0xb0 [rc_core]
[<ffffffffc0398010>] 0xffffffffc0398010
[<ffffffffa1002192>] do_one_initcall+0x52/0x1b0
[<ffffffffa11ef5c8>] do_init_module+0x5f/0x1fa
[<ffffffffa11566b5>] load_module+0x2675/0x2b00
[<ffffffffa1156dcf>] SYSC_finit_module+0xdf/0x110
[<ffffffffa1156e1e>] SyS_finit_module+0xe/0x10
[<ffffffffa1003f5c>] do_syscall_64+0x6c/0x1f0
[<ffffffffa1927989>] return_from_SYSCALL_64+0x0/0x7a
               -> #0 (ir_raw_handler_lock){+.+.+.}:
[<ffffffffa110a7b7>] __lock_acquire+0x10f7/0x1290
[<ffffffffa110adad>] lock_acquire+0xfd/0x200
[<ffffffffa1921327>] mutex_lock_nested+0x77/0x6d0
[<ffffffffc06f6a5e>] ir_raw_encode_scancode+0x3e/0xb0 [rc_core]
[<ffffffffc0b0f492>] loop_set_wakeup_filter+0x62/0xbd [rc_loopback]
[<ffffffffc06f522a>] store_filter+0x1aa/0x240 [rc_core]
[<ffffffffa15e46f8>] dev_attr_store+0x18/0x30
[<ffffffffa13318e5>] sysfs_kf_write+0x45/0x60
[<ffffffffa1330b55>] kernfs_fop_write+0x155/0x1e0
[<ffffffffa1290797>] __vfs_write+0x37/0x160
[<ffffffffa12921f8>] vfs_write+0xc8/0x1e0
[<ffffffffa12936e8>] SyS_write+0x58/0xc0
[<ffffffffa19278c1>] entry_SYSCALL_64_fastpath+0x1f/0xc2

               other info that might help us debug this:

Chain exists of:
                 ir_raw_handler_lock --> lirc_dev_lock --> &dev->lock

 Possible unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(&dev->lock);
                               lock(lirc_dev_lock);
                               lock(&dev->lock);
  lock(ir_raw_handler_lock);

                *** DEADLOCK ***

4 locks held by bash/2502:
 #0:  (sb_writers#4){.+.+.+}, at: [<ffffffffa12922c5>] vfs_write+0x195/0x1e0
 #1:  (&of->mutex){+.+.+.}, at: [<ffffffffa1330b1f>] kernfs_fop_write+0x11f/0x1e0
 #2:  (s_active#215){.+.+.+}, at: [<ffffffffa1330b28>] kernfs_fop_write+0x128/0x1e0
 #3:  (&dev->lock){+.+.+.}, at: [<ffffffffc06f511f>] store_filter+0x9f/0x240 [rc_core]

               stack backtrace:
CPU: 3 PID: 2502 Comm: bash Not tainted 4.10.0-rc1+ #1
Hardware name:                  /DG45ID, BIOS IDG4510H.86A.0135.2011.0225.1100 02/25/2011
Call Trace:
 dump_stack+0x86/0xc3
 print_circular_bug+0x1be/0x210
 __lock_acquire+0x10f7/0x1290
 lock_acquire+0xfd/0x200
 ? ir_raw_encode_scancode+0x3e/0xb0 [rc_core]
 ? ir_raw_encode_scancode+0x3e/0xb0 [rc_core]
 mutex_lock_nested+0x77/0x6d0
 ? ir_raw_encode_scancode+0x3e/0xb0 [rc_core]
 ? loop_set_wakeup_filter+0x44/0xbd [rc_loopback]
 ir_raw_encode_scancode+0x3e/0xb0 [rc_core]
 loop_set_wakeup_filter+0x62/0xbd [rc_loopback]
 ? loop_set_tx_duty_cycle+0x70/0x70 [rc_loopback]
 store_filter+0x1aa/0x240 [rc_core]
 dev_attr_store+0x18/0x30
 sysfs_kf_write+0x45/0x60
 kernfs_fop_write+0x155/0x1e0
 __vfs_write+0x37/0x160
 ? rcu_read_lock_sched_held+0x4a/0x80
 ? rcu_sync_lockdep_assert+0x2f/0x60
 ? __sb_start_write+0x10c/0x220
 ? vfs_write+0x195/0x1e0
 ? security_file_permission+0x3b/0xc0
 vfs_write+0xc8/0x1e0
 SyS_write+0x58/0xc0
 entry_SYSCALL_64_fastpath+0x1f/0xc2

Signed-off-by: Sean Young <sean@mess.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@s-opensource.com>
drivers/media/rc/lirc_dev.c

index a54ca531d8ef85280c8cf9df7b7b19ac226f7032..4630f3e6753889ab344c7a0b1b61903112f95cc1 100644 (file)
@@ -436,6 +436,8 @@ int lirc_dev_fop_open(struct inode *inode, struct file *file)
                return -ERESTARTSYS;
 
        ir = irctls[iminor(inode)];
+       mutex_unlock(&lirc_dev_lock);
+
        if (!ir) {
                retval = -ENODEV;
                goto error;
@@ -476,8 +478,6 @@ int lirc_dev_fop_open(struct inode *inode, struct file *file)
        }
 
 error:
-       mutex_unlock(&lirc_dev_lock);
-
        nonseekable_open(inode, file);
 
        return retval;