userns: Make ucounts lock irq-safe
authorNikolay Borisov <n.borisov.lkml@gmail.com>
Fri, 20 Jan 2017 13:21:35 +0000 (15:21 +0200)
committerEric W. Biederman <ebiederm@xmission.com>
Mon, 23 Jan 2017 17:23:51 +0000 (06:23 +1300)
commit880a38547ff08715ce4f1daf9a4bb30c87676e68
tree0bcadd3eeb166c80a5fa88feaac348059ab30ef7
parent93362fa47fe98b62e4a34ab408c4a418432e7939
userns: Make ucounts lock irq-safe

The ucounts_lock is being used to protect various ucounts lifecycle
management functionalities. However, those services can also be invoked
when a pidns is being freed in an RCU callback (e.g. softirq context).
This can lead to deadlocks. There were already efforts trying to
prevent similar deadlocks in add7c65ca426 ("pid: fix lockdep deadlock
warning due to ucount_lock"), however they just moved the context
from hardirq to softrq. Fix this issue once and for all by explictly
making the lock disable irqs altogether.

Dmitry Vyukov <dvyukov@google.com> reported:

> I've got the following deadlock report while running syzkaller fuzzer
> on eec0d3d065bfcdf9cd5f56dd2a36b94d12d32297 of linux-next (on odroid
> device if it matters):
>
> =================================
> [ INFO: inconsistent lock state ]
> 4.10.0-rc3-next-20170112-xc2-dirty #6 Not tainted
> ---------------------------------
> inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
> swapper/2/0 [HC0[0]:SC1[1]:HE1:SE0] takes:
>  (ucounts_lock){+.?...}, at: [<     inline     >] spin_lock
> ./include/linux/spinlock.h:302
>  (ucounts_lock){+.?...}, at: [<ffff2000081678c8>]
> put_ucounts+0x60/0x138 kernel/ucount.c:162
> {SOFTIRQ-ON-W} state was registered at:
> [<ffff2000081c82d8>] mark_lock+0x220/0xb60 kernel/locking/lockdep.c:3054
> [<     inline     >] mark_irqflags kernel/locking/lockdep.c:2941
> [<ffff2000081c97a8>] __lock_acquire+0x388/0x3260 kernel/locking/lockdep.c:3295
> [<ffff2000081cce24>] lock_acquire+0xa4/0x138 kernel/locking/lockdep.c:3753
> [<     inline     >] __raw_spin_lock ./include/linux/spinlock_api_smp.h:144
> [<ffff200009798128>] _raw_spin_lock+0x90/0xd0 kernel/locking/spinlock.c:151
> [<     inline     >] spin_lock ./include/linux/spinlock.h:302
> [<     inline     >] get_ucounts kernel/ucount.c:131
> [<ffff200008167c28>] inc_ucount+0x80/0x6c8 kernel/ucount.c:189
> [<     inline     >] inc_mnt_namespaces fs/namespace.c:2818
> [<ffff200008481850>] alloc_mnt_ns+0x78/0x3a8 fs/namespace.c:2849
> [<ffff200008487298>] create_mnt_ns+0x28/0x200 fs/namespace.c:2959
> [<     inline     >] init_mount_tree fs/namespace.c:3199
> [<ffff200009bd6674>] mnt_init+0x258/0x384 fs/namespace.c:3251
> [<ffff200009bd60bc>] vfs_caches_init+0x6c/0x80 fs/dcache.c:3626
> [<ffff200009bb1114>] start_kernel+0x414/0x460 init/main.c:648
> [<ffff200009bb01e8>] __primary_switched+0x6c/0x70 arch/arm64/kernel/head.S:456
> irq event stamp: 2316924
> hardirqs last  enabled at (2316924): [<     inline     >] rcu_do_batch
> kernel/rcu/tree.c:2911
> hardirqs last  enabled at (2316924): [<     inline     >]
> invoke_rcu_callbacks kernel/rcu/tree.c:3182
> hardirqs last  enabled at (2316924): [<     inline     >]
> __rcu_process_callbacks kernel/rcu/tree.c:3149
> hardirqs last  enabled at (2316924): [<ffff200008210414>]
> rcu_process_callbacks+0x7a4/0xc28 kernel/rcu/tree.c:3166
> hardirqs last disabled at (2316923): [<     inline     >] rcu_do_batch
> kernel/rcu/tree.c:2900
> hardirqs last disabled at (2316923): [<     inline     >]
> invoke_rcu_callbacks kernel/rcu/tree.c:3182
> hardirqs last disabled at (2316923): [<     inline     >]
> __rcu_process_callbacks kernel/rcu/tree.c:3149
> hardirqs last disabled at (2316923): [<ffff20000820fe80>]
> rcu_process_callbacks+0x210/0xc28 kernel/rcu/tree.c:3166
> softirqs last  enabled at (2316912): [<ffff20000811b4c4>]
> _local_bh_enable+0x4c/0x80 kernel/softirq.c:155
> softirqs last disabled at (2316913): [<     inline     >]
> do_softirq_own_stack ./include/linux/interrupt.h:488
> softirqs last disabled at (2316913): [<     inline     >]
> invoke_softirq kernel/softirq.c:371
> softirqs last disabled at (2316913): [<ffff20000811c994>]
> irq_exit+0x264/0x308 kernel/softirq.c:405
>
> other info that might help us debug this:
>  Possible unsafe locking scenario:
>
>        CPU0
>        ----
>   lock(ucounts_lock);
>   <Interrupt>
>     lock(ucounts_lock);
>
>  *** DEADLOCK ***
>
> 1 lock held by swapper/2/0:
>  #0:  (rcu_callback){......}, at: [<     inline     >] __rcu_reclaim
> kernel/rcu/rcu.h:108
>  #0:  (rcu_callback){......}, at: [<     inline     >] rcu_do_batch
> kernel/rcu/tree.c:2919
>  #0:  (rcu_callback){......}, at: [<     inline     >]
> invoke_rcu_callbacks kernel/rcu/tree.c:3182
>  #0:  (rcu_callback){......}, at: [<     inline     >]
> __rcu_process_callbacks kernel/rcu/tree.c:3149
>  #0:  (rcu_callback){......}, at: [<ffff200008210390>]
> rcu_process_callbacks+0x720/0xc28 kernel/rcu/tree.c:3166
>
> stack backtrace:
> CPU: 2 PID: 0 Comm: swapper/2 Not tainted 4.10.0-rc3-next-20170112-xc2-dirty #6
> Hardware name: Hardkernel ODROID-C2 (DT)
> Call trace:
> [<ffff20000808fa60>] dump_backtrace+0x0/0x440 arch/arm64/kernel/traps.c:500
> [<ffff20000808fec0>] show_stack+0x20/0x30 arch/arm64/kernel/traps.c:225
> [<ffff2000088a99e0>] dump_stack+0x110/0x168
> [<ffff2000082fa2b4>] print_usage_bug.part.27+0x49c/0x4bc
> kernel/locking/lockdep.c:2387
> [<     inline     >] print_usage_bug kernel/locking/lockdep.c:2357
> [<     inline     >] valid_state kernel/locking/lockdep.c:2400
> [<     inline     >] mark_lock_irq kernel/locking/lockdep.c:2617
> [<ffff2000081c89ec>] mark_lock+0x934/0xb60 kernel/locking/lockdep.c:3065
> [<     inline     >] mark_irqflags kernel/locking/lockdep.c:2923
> [<ffff2000081c9a60>] __lock_acquire+0x640/0x3260 kernel/locking/lockdep.c:3295
> [<ffff2000081cce24>] lock_acquire+0xa4/0x138 kernel/locking/lockdep.c:3753
> [<     inline     >] __raw_spin_lock ./include/linux/spinlock_api_smp.h:144
> [<ffff200009798128>] _raw_spin_lock+0x90/0xd0 kernel/locking/spinlock.c:151
> [<     inline     >] spin_lock ./include/linux/spinlock.h:302
> [<ffff2000081678c8>] put_ucounts+0x60/0x138 kernel/ucount.c:162
> [<ffff200008168364>] dec_ucount+0xf4/0x158 kernel/ucount.c:214
> [<     inline     >] dec_pid_namespaces kernel/pid_namespace.c:89
> [<ffff200008293dc8>] delayed_free_pidns+0x40/0xe0 kernel/pid_namespace.c:156
> [<     inline     >] __rcu_reclaim kernel/rcu/rcu.h:118
> [<     inline     >] rcu_do_batch kernel/rcu/tree.c:2919
> [<     inline     >] invoke_rcu_callbacks kernel/rcu/tree.c:3182
> [<     inline     >] __rcu_process_callbacks kernel/rcu/tree.c:3149
> [<ffff2000082103d8>] rcu_process_callbacks+0x768/0xc28 kernel/rcu/tree.c:3166
> [<ffff2000080821dc>] __do_softirq+0x324/0x6e0 kernel/softirq.c:284
> [<     inline     >] do_softirq_own_stack ./include/linux/interrupt.h:488
> [<     inline     >] invoke_softirq kernel/softirq.c:371
> [<ffff20000811c994>] irq_exit+0x264/0x308 kernel/softirq.c:405
> [<ffff2000081ecc28>] __handle_domain_irq+0xc0/0x150 kernel/irq/irqdesc.c:636
> [<ffff200008081c80>] gic_handle_irq+0x68/0xd8
> Exception stack(0xffff8000648e7dd0 to 0xffff8000648e7f00)
> 7dc0:                                   ffff8000648d4b3c 0000000000000007
> 7de0: 0000000000000000 1ffff0000c91a967 1ffff0000c91a967 1ffff0000c91a967
> 7e00: ffff20000a4b6b68 0000000000000001 0000000000000007 0000000000000001
> 7e20: 1fffe4000149ae90 ffff200009d35000 0000000000000000 0000000000000002
> 7e40: 0000000000000000 0000000000000000 0000000002624a1a 0000000000000000
> 7e60: 0000000000000000 ffff200009cbcd88 000060006d2ed000 0000000000000140
> 7e80: ffff200009cff000 ffff200009cb6000 ffff200009cc2020 ffff200009d2159d
> 7ea0: 0000000000000000 ffff8000648d4380 0000000000000000 ffff8000648e7f00
> 7ec0: ffff20000820a478 ffff8000648e7f00 ffff20000820a47c 0000000010000145
> 7ee0: 0000000000000140 dfff200000000000 ffffffffffffffff ffff20000820a478
> [<ffff2000080837f8>] el1_irq+0xb8/0x130 arch/arm64/kernel/entry.S:486
> [<     inline     >] arch_local_irq_restore
> ./arch/arm64/include/asm/irqflags.h:81
> [<ffff20000820a47c>] rcu_idle_exit+0x64/0xa8 kernel/rcu/tree.c:1030
> [<     inline     >] cpuidle_idle_call kernel/sched/idle.c:200
> [<ffff2000081bcbfc>] do_idle+0x1dc/0x2d0 kernel/sched/idle.c:243
> [<ffff2000081bd1cc>] cpu_startup_entry+0x24/0x28 kernel/sched/idle.c:345
> [<ffff200008099f8c>] secondary_start_kernel+0x2cc/0x358
> arch/arm64/kernel/smp.c:276
> [<000000000279f1a4>] 0x279f1a4

Reported-by: Dmitry Vyukov <dvyukov@google.com>
Tested-by: Dmitry Vyukov <dvyukov@google.com>
Fixes: add7c65ca426 ("pid: fix lockdep deadlock warning due to ucount_lock")
Fixes: f333c700c610 ("pidns: Add a limit on the number of pid namespaces")
Cc: stable@vger.kernel.org
Link: https://www.spinics.net/lists/kernel/msg2426637.html
Signed-off-by: Nikolay Borisov <n.borisov.lkml@gmail.com>
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
kernel/ucount.c