RxRPC: Fix a potential deadlock between the call resend_timer and state_lock
authorDavid Howells <dhowells@redhat.com>
Wed, 4 Aug 2010 02:34:17 +0000 (02:34 +0000)
committerDavid S. Miller <davem@davemloft.net>
Thu, 5 Aug 2010 04:53:16 +0000 (21:53 -0700)
commit3b5bac2bdea1de832bdd8e2c904ab7c9479ff9ed
treeb7e9fbdfa42e2e053c33af1b533a4e3693a352fe
parentf2f009812f1fdcaf40fa547282c1b90d3b702a7d
RxRPC: Fix a potential deadlock between the call resend_timer and state_lock

RxRPC can potentially deadlock as rxrpc_resend_time_expired() wants to get
call->state_lock so that it can alter the state of an RxRPC call.  However, its
caller (call_timer_fn()) has an apparent lock on the timer struct.

The problem is that rxrpc_resend_time_expired() isn't permitted to lock
call->state_lock as this could cause a deadlock against rxrpc_send_abort() as
that takes state_lock and then attempts to delete the resend timer by calling
del_timer_sync().

The deadlock can occur because del_timer_sync() will sit there forever waiting
for rxrpc_resend_time_expired() to return, but the latter may then wait for
call->state_lock, which rxrpc_send_abort() holds around del_timer_sync()...

This leads to a warning appearing in the kernel log that looks something like
the attached.

It should be sufficient to simply dispense with the locks.  It doesn't matter
if we set the resend timer expired event bit and queue the event processor
whilst we're changing state to one where the resend timer is irrelevant as the
event can just be ignored by the processor thereafter.

=======================================================
[ INFO: possible circular locking dependency detected ]
2.6.35-rc3-cachefs+ #115
-------------------------------------------------------
swapper/0 is trying to acquire lock:
 (&call->state_lock){++--..}, at: [<ffffffffa00200d4>] rxrpc_resend_time_expired+0x56/0x96 [af_rxrpc]

but task is already holding lock:
 (&call->resend_timer){+.-...}, at: [<ffffffff8103b675>] run_timer_softirq+0x182/0x2a5

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (&call->resend_timer){+.-...}:
       [<ffffffff810560bc>] __lock_acquire+0x889/0x8fa
       [<ffffffff81056184>] lock_acquire+0x57/0x6d
       [<ffffffff8103bb9c>] del_timer_sync+0x3c/0x86
       [<ffffffffa002bb7a>] rxrpc_send_abort+0x50/0x97 [af_rxrpc]
       [<ffffffffa002bdd9>] rxrpc_kernel_abort_call+0xa1/0xdd [af_rxrpc]
       [<ffffffffa0061588>] afs_deliver_to_call+0x129/0x368 [kafs]
       [<ffffffffa006181b>] afs_process_async_call+0x54/0xff [kafs]
       [<ffffffff8104261d>] worker_thread+0x1ef/0x2e2
       [<ffffffff81045f47>] kthread+0x7a/0x82
       [<ffffffff81002cd4>] kernel_thread_helper+0x4/0x10

-> #0 (&call->state_lock){++--..}:
       [<ffffffff81055237>] validate_chain+0x727/0xd23
       [<ffffffff810560bc>] __lock_acquire+0x889/0x8fa
       [<ffffffff81056184>] lock_acquire+0x57/0x6d
       [<ffffffff813e6b69>] _raw_read_lock_bh+0x34/0x43
       [<ffffffffa00200d4>] rxrpc_resend_time_expired+0x56/0x96 [af_rxrpc]
       [<ffffffff8103b6e6>] run_timer_softirq+0x1f3/0x2a5
       [<ffffffff81036828>] __do_softirq+0xa2/0x13e
       [<ffffffff81002dcc>] call_softirq+0x1c/0x28
       [<ffffffff810049f0>] do_softirq+0x38/0x80
       [<ffffffff810361a2>] irq_exit+0x45/0x47
       [<ffffffff81018fb3>] smp_apic_timer_interrupt+0x88/0x96
       [<ffffffff81002893>] apic_timer_interrupt+0x13/0x20
       [<ffffffff810011ac>] cpu_idle+0x4d/0x83
       [<ffffffff813e06f3>] start_secondary+0x1bd/0x1c1

other info that might help us debug this:

1 lock held by swapper/0:
 #0:  (&call->resend_timer){+.-...}, at: [<ffffffff8103b675>] run_timer_softirq+0x182/0x2a5

stack backtrace:
Pid: 0, comm: swapper Not tainted 2.6.35-rc3-cachefs+ #115
Call Trace:
 <IRQ>  [<ffffffff81054414>] print_circular_bug+0xae/0xbd
 [<ffffffff81055237>] validate_chain+0x727/0xd23
 [<ffffffff810560bc>] __lock_acquire+0x889/0x8fa
 [<ffffffff810539a7>] ? mark_lock+0x42f/0x51f
 [<ffffffff81056184>] lock_acquire+0x57/0x6d
 [<ffffffffa00200d4>] ? rxrpc_resend_time_expired+0x56/0x96 [af_rxrpc]
 [<ffffffff813e6b69>] _raw_read_lock_bh+0x34/0x43
 [<ffffffffa00200d4>] ? rxrpc_resend_time_expired+0x56/0x96 [af_rxrpc]
 [<ffffffffa00200d4>] rxrpc_resend_time_expired+0x56/0x96 [af_rxrpc]
 [<ffffffff8103b6e6>] run_timer_softirq+0x1f3/0x2a5
 [<ffffffff8103b675>] ? run_timer_softirq+0x182/0x2a5
 [<ffffffffa002007e>] ? rxrpc_resend_time_expired+0x0/0x96 [af_rxrpc]
 [<ffffffff810367ef>] ? __do_softirq+0x69/0x13e
 [<ffffffff81036828>] __do_softirq+0xa2/0x13e
 [<ffffffff81002dcc>] call_softirq+0x1c/0x28
 [<ffffffff810049f0>] do_softirq+0x38/0x80
 [<ffffffff810361a2>] irq_exit+0x45/0x47
 [<ffffffff81018fb3>] smp_apic_timer_interrupt+0x88/0x96
 [<ffffffff81002893>] apic_timer_interrupt+0x13/0x20
 <EOI>  [<ffffffff81049de1>] ? __atomic_notifier_call_chain+0x0/0x86
 [<ffffffff8100955b>] ? mwait_idle+0x6e/0x78
 [<ffffffff81009552>] ? mwait_idle+0x65/0x78
 [<ffffffff810011ac>] cpu_idle+0x4d/0x83
 [<ffffffff813e06f3>] start_secondary+0x1bd/0x1c1

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/rxrpc/ar-ack.c
net/rxrpc/ar-call.c