[SCSI] fcoe: fix a circular locking issue with rtnl and sysfs mutex
authorVasu Dev <vasu.dev@intel.com>
Fri, 7 May 2010 22:18:46 +0000 (15:18 -0700)
committerJames Bottomley <James.Bottomley@suse.de>
Mon, 17 May 2010 02:22:35 +0000 (22:22 -0400)
commit34ce27bcf96f5f366e1fa8c4729ffc8a55de4cc3
treef4650760844e41851b1f4dc301a35601025d2699
parent7b2787ec15b9d1c2f716da61b0eec21a3f5e6520
[SCSI] fcoe: fix a circular locking issue with rtnl and sysfs mutex

Currently rtnl mutex is grabbed during fcoe create, destroy, enable
and disable operations while sysfs s_active read mutex is already
held, but simultaneously other networking events could try grabbing
write s_active mutex while rtnl is already held and that is causing
circular lock warning, its detailed log pasted at end.

In this log, the rtnl was held before write s_active during device
renaming but there are more such cases as Joe reported another
instance with tg3 open at:-
http://www.open-fcoe.org/pipermail/devel/2010-February/008263.html

This patch fixes this issue by not waiting for rtnl mutex during
fcoe ops, that means if rtnl mutex is not immediately available
then restart_syscall() to allow others waiting in line to
grab s_active along with rtnl mutex to finish their work first
under these mutex.

Currently rtnl mutex was grabbed twice during fcoe_destroy call flow,
second grab was from fcoe_if_destroy called from fcoe_destroy after
dropping rtnl mutex before calling fcoe_if_destroy, so instead made
fcoe_if_destroy always called with rtnl mutex held to have this mutex
grabbed only once in this code path.

However left matching rtnl_unlock as-is in its original place as it was
dropped there for good reason since very next call causes synchronous
fip worker flush and if rtnl mutex is still held before flush
then that would cause new circular warning between fip->recv_work and
rtnl mutex, I've added detailed comment for this on fcoe_if_destroy
calling and rtnl muxtes unlocking.

=======================================================
[ INFO: possible circular locking dependency detected ]
2.6.33.1linux-stable-2.6.33 #1
-------------------------------------------------------
fcoemon/18823 is trying to acquire lock:
(fcoe_config_mutex){+.+.+.}, at: [<ffffffffa02ba5fc>] fcoe_create+0x27/0x4f7
[fcoe]

but task is already holding lock:
(s_active){++++.+}, at: [<ffffffff8115ef93>] sysfs_get_active_two+0x31/0x48

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #2 (s_active){++++.+}:
   [<ffffffff81077bdb>] __lock_acquire+0xb73/0xd2b
   [<ffffffff81077e60>] lock_acquire+0xcd/0xf1
   [<ffffffff8115e5df>] sysfs_deactivate+0x8b/0xe0
   [<ffffffff8115edfb>] sysfs_addrm_finish+0x36/0x55
   [<ffffffff8115d0cc>] sysfs_hash_and_remove+0x53/0x6a
   [<ffffffff8115f353>] sysfs_remove_link+0x21/0x23
   [<ffffffff812b6c93>] device_rename+0x99/0xcb
   [<ffffffff8138dbf0>] dev_change_name+0xd5/0x1d2
   [<ffffffff8138deee>] dev_ifsioc+0x201/0x2ac
   [<ffffffff8138e4ba>] dev_ioctl+0x521/0x632
   [<ffffffff81379e43>] sock_do_ioctl+0x3d/0x47
   [<ffffffff8137a254>] sock_ioctl+0x213/0x222
   [<ffffffff81114614>] vfs_ioctl+0x32/0xa6
   [<ffffffff81114b94>] do_vfs_ioctl+0x490/0x4d6
   [<ffffffff81114c30>] sys_ioctl+0x56/0x79
   [<ffffffff81009b42>] system_call_fastpath+0x16/0x1b

-> #1 (rtnl_mutex){+.+.+.}:
   [<ffffffff81077bdb>] __lock_acquire+0xb73/0xd2b
   [<ffffffff81077e60>] lock_acquire+0xcd/0xf1
   [<ffffffff8142f343>] __mutex_lock_common+0x4b/0x383
   [<ffffffff8142f73f>] mutex_lock_nested+0x3e/0x43
   [<ffffffff813959f9>] rtnl_lock+0x17/0x19
   [<ffffffff8138ccae>] register_netdevice_notifier+0x1e/0x19b
   [<ffffffffa02580c1>] 0xffffffffa02580c1
   [<ffffffff81002069>] do_one_initcall+0x5e/0x15e
   [<ffffffff81084094>] sys_init_module+0xd8/0x23a
   [<ffffffff81009b42>] system_call_fastpath+0x16/0x1b

-> #0 (fcoe_config_mutex){+.+.+.}:
   [<ffffffff81077a85>] __lock_acquire+0xa1d/0xd2b
   [<ffffffff81077e60>] lock_acquire+0xcd/0xf1
   [<ffffffff8142f343>] __mutex_lock_common+0x4b/0x383
   [<ffffffff8142f73f>] mutex_lock_nested+0x3e/0x43
   [<ffffffffa02ba5fc>] fcoe_create+0x27/0x4f7 [fcoe]
   [<ffffffff810635b1>] param_attr_store+0x27/0x35
   [<ffffffff81063619>] module_attr_store+0x26/0x2a
   [<ffffffff8115dae3>] sysfs_write_file+0x108/0x144
   [<ffffffff81107bd1>] vfs_write+0xae/0x10b
   [<ffffffff81107cee>] sys_write+0x4a/0x6e
   [<ffffffff81009b42>] system_call_fastpath+0x16/0x1b

other info that might help us debug this:

3 locks held by fcoemon/18823:
#0:  (&buffer->mutex){+.+.+.}, at: [<ffffffff8115da17>]
sysfs_write_file+0x3c/0x144
#1:  (s_active){++++.+}, at: [<ffffffff8115ef86>]
sysfs_get_active_two+0x24/0x48
#2:  (s_active){++++.+}, at: [<ffffffff8115ef93>]
sysfs_get_active_two+0x31/0x48

stack backtrace:
Pid: 18823, comm: fcoemon Tainted: G        W  2.6.33.1linux-stable-2.6.33 #1
Call Trace:
[<ffffffff81076c38>] print_circular_bug+0xa8/0xb6
[<ffffffff81077a85>] __lock_acquire+0xa1d/0xd2b
[<ffffffffa02ba5fc>] ? fcoe_create+0x27/0x4f7 [fcoe]
[<ffffffff81077e60>] lock_acquire+0xcd/0xf1
[<ffffffffa02ba5fc>] ? fcoe_create+0x27/0x4f7 [fcoe]
[<ffffffffa02ba5fc>] ? fcoe_create+0x27/0x4f7 [fcoe]
[<ffffffff8142f343>] __mutex_lock_common+0x4b/0x383
[<ffffffffa02ba5fc>] ? fcoe_create+0x27/0x4f7 [fcoe]
[<ffffffff8106ac70>] ? cpu_clock+0x43/0x5e
[<ffffffff81074e12>] ? lockstat_clock+0x11/0x13
[<ffffffff81074e40>] ? lock_release_holdtime+0x2c/0x127
[<ffffffff8115ef93>] ? sysfs_get_active_two+0x31/0x48
[<ffffffff8142f73f>] mutex_lock_nested+0x3e/0x43
[<ffffffffa02ba5fc>] fcoe_create+0x27/0x4f7 [fcoe]
[<ffffffff810635b1>] param_attr_store+0x27/0x35
[<ffffffff81063619>] module_attr_store+0x26/0x2a
[<ffffffff8115dae3>] sysfs_write_file+0x108/0x144
[<ffffffff81107bd1>] vfs_write+0xae/0x10b
[<ffffffff81076596>] ? trace_hardirqs_on_caller+0x125/0x150
[<ffffffff81107cee>] sys_write+0x4a/0x6e
[<ffffffff81009b42>] system_call_fastpath+0x16/0x1b

Signed-off-by: Vasu Dev <vasu.dev@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
drivers/scsi/fcoe/fcoe.c