• Vlad Buslov's avatar
    net: sched: always disable bh when taking tcf_lock · 653cd284
    Vlad Buslov authored
    Recently, ops->init() and ops->dump() of all actions were modified to
    always obtain tcf_lock when accessing private action state. Actions that
    don't depend on tcf_lock for synchronization with their data path use
    non-bh locking API. However, tcf_lock is also used to protect rate
    estimator stats in softirq context by timer callback.
    
    Change ops->init() and ops->dump() of all actions to disable bh when using
    tcf_lock to prevent deadlock reported by following lockdep warning:
    
    [  105.470398] ================================
    [  105.475014] WARNING: inconsistent lock state
    [  105.479628] 4.18.0-rc8+ #664 Not tainted
    [  105.483897] --------------------------------
    [  105.488511] inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
    [  105.494871] swapper/16/0 [HC0[0]:SC1[1]:HE1:SE0] takes:
    [  105.500449] 00000000f86c012e (&(&p->tcfa_lock)->rlock){+.?.}, at: est_fetch_counters+0x3c/0xa0
    [  105.509696] {SOFTIRQ-ON-W} state was registered at:
    [  105.514925]   _raw_spin_lock+0x2c/0x40
    [  105.519022]   tcf_bpf_init+0x579/0x820 [act_bpf]
    [  105.523990]   tcf_action_init_1+0x4e4/0x660
    [  105.528518]   tcf_action_init+0x1ce/0x2d0
    [  105.532880]   tcf_exts_validate+0x1d8/0x200
    [  105.537416]   fl_change+0x55a/0x268b [cls_flower]
    [  105.542469]   tc_new_tfilter+0x748/0xa20
    [  105.546738]   rtnetlink_rcv_msg+0x56a/0x6d0
    [  105.551268]   netlink_rcv_skb+0x18d/0x200
    [  105.555628]   netlink_unicast+0x2d0/0x370
    [  105.559990]   netlink_sendmsg+0x3b9/0x6a0
    [  105.564349]   sock_sendmsg+0x6b/0x80
    [  105.568271]   ___sys_sendmsg+0x4a1/0x520
    [  105.572547]   __sys_sendmsg+0xd7/0x150
    [  105.576655]   do_syscall_64+0x72/0x2c0
    [  105.580757]   entry_SYSCALL_64_after_hwframe+0x49/0xbe
    [  105.586243] irq event stamp: 489296
    [  105.590084] hardirqs last  enabled at (489296): [<ffffffffb507e639>] _raw_spin_unlock_irq+0x29/0x40
    [  105.599765] hardirqs last disabled at (489295): [<ffffffffb507e745>] _raw_spin_lock_irq+0x15/0x50
    [  105.609277] softirqs last  enabled at (489292): [<ffffffffb413a6a3>] irq_enter+0x83/0xa0
    [  105.618001] softirqs last disabled at (489293): [<ffffffffb413a800>] irq_exit+0x140/0x190
    [  105.626813]
                   other info that might help us debug this:
    [  105.633976]  Possible unsafe locking scenario:
    
    [  105.640526]        CPU0
    [  105.643325]        ----
    [  105.646125]   lock(&(&p->tcfa_lock)->rlock);
    [  105.650747]   <Interrupt>
    [  105.653717]     lock(&(&p->tcfa_lock)->rlock);
    [  105.658514]
                    *** DEADLOCK ***
    
    [  105.665349] 1 lock held by swapper/16/0:
    [  105.669629]  #0: 00000000a640ad99 ((&est->timer)){+.-.}, at: call_timer_fn+0x10b/0x550
    [  105.678200]
                   stack backtrace:
    [  105.683194] CPU: 16 PID: 0 Comm: swapper/16 Not tainted 4.18.0-rc8+ #664
    [  105.690249] Hardware name: Supermicro SYS-2028TP-DECR/X10DRT-P, BIOS 2.0b 03/30/2017
    [  105.698626] Call Trace:
    [  105.701421]  <IRQ>
    [  105.703791]  dump_stack+0x92/0xeb
    [  105.707461]  print_usage_bug+0x336/0x34c
    [  105.711744]  mark_lock+0x7c9/0x980
    [  105.715500]  ? print_shortest_lock_dependencies+0x2e0/0x2e0
    [  105.721424]  ? check_usage_forwards+0x230/0x230
    [  105.726315]  __lock_acquire+0x923/0x26f0
    [  105.730597]  ? debug_show_all_locks+0x240/0x240
    [  105.735478]  ? mark_lock+0x493/0x980
    [  105.739412]  ? check_chain_key+0x140/0x1f0
    [  105.743861]  ? __lock_acquire+0x836/0x26f0
    [  105.748323]  ? lock_acquire+0x12e/0x290
    [  105.752516]  lock_acquire+0x12e/0x290
    [  105.756539]  ? est_fetch_counters+0x3c/0xa0
    [  105.761084]  _raw_spin_lock+0x2c/0x40
    [  105.765099]  ? est_fetch_counters+0x3c/0xa0
    [  105.769633]  est_fetch_counters+0x3c/0xa0
    [  105.773995]  est_timer+0x87/0x390
    [  105.777670]  ? est_fetch_counters+0xa0/0xa0
    [  105.782210]  ? lock_acquire+0x12e/0x290
    [  105.786410]  call_timer_fn+0x161/0x550
    [  105.790512]  ? est_fetch_counters+0xa0/0xa0
    [  105.795055]  ? del_timer_sync+0xd0/0xd0
    [  105.799249]  ? __lock_is_held+0x93/0x110
    [  105.803531]  ? mark_held_locks+0x20/0xe0
    [  105.807813]  ? _raw_spin_unlock_irq+0x29/0x40
    [  105.812525]  ? est_fetch_counters+0xa0/0xa0
    [  105.817069]  ? est_fetch_counters+0xa0/0xa0
    [  105.821610]  run_timer_softirq+0x3c4/0x9f0
    [  105.826064]  ? lock_acquire+0x12e/0x290
    [  105.830257]  ? __bpf_trace_timer_class+0x10/0x10
    [  105.835237]  ? __lock_is_held+0x25/0x110
    [  105.839517]  __do_softirq+0x11d/0x7bf
    [  105.843542]  irq_exit+0x140/0x190
    [  105.847208]  smp_apic_timer_interrupt+0xac/0x3b0
    [  105.852182]  apic_timer_interrupt+0xf/0x20
    [  105.856628]  </IRQ>
    [  105.859081] RIP: 0010:cpuidle_enter_state+0xd8/0x4d0
    [  105.864395] Code: 46 ff 48 89 44 24 08 0f 1f 44 00 00 31 ff e8 cf ec 46 ff 80 7c 24 07 00 0f 85 1d 02 00 00 e8 9f 90 4b ff fb 66 0f 1f 44 00 00 <4c> 8b 6c 24 08 4d 29 fd 0f 80 36 03 00 00 4c 89 e8 48 ba cf f7 53
    [  105.884288] RSP: 0018:ffff8803ad94fd20 EFLAGS: 00000246 ORIG_RAX: ffffffffffffff13
    [  105.892494] RAX: 0000000000000000 RBX: ffffe8fb300829c0 RCX: ffffffffb41e19e1
    [  105.899988] RDX: 0000000000000007 RSI: dffffc0000000000 RDI: ffff8803ad9358ac
    [  105.907503] RBP: ffffffffb6636300 R08: 0000000000000004 R09: 0000000000000000
    [  105.914997] R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000004
    [  105.922487] R13: ffffffffb6636140 R14: ffffffffb66362d8 R15: 000000188d36091b
    [  105.929988]  ? trace_hardirqs_on_caller+0x141/0x2d0
    [  105.935232]  do_idle+0x28e/0x320
    [  105.938817]  ? arch_cpu_idle_exit+0x40/0x40
    [  105.943361]  ? mark_lock+0x8c1/0x980
    [  105.947295]  ? _raw_spin_unlock_irqrestore+0x32/0x60
    [  105.952619]  cpu_startup_entry+0xc2/0xd0
    [  105.956900]  ? cpu_in_idle+0x20/0x20
    [  105.960830]  ? _raw_spin_unlock_irqrestore+0x32/0x60
    [  105.966146]  ? trace_hardirqs_on_caller+0x141/0x2d0
    [  105.971391]  start_secondary+0x2b5/0x360
    [  105.975669]  ? set_cpu_sibling_map+0x1330/0x1330
    [  105.980654]  secondary_startup_64+0xa5/0xb0
    
    Taking tcf_lock in sample action with bh disabled causes lockdep to issue a
    warning regarding possible irq lock inversion dependency between tcf_lock,
    and psample_groups_lock that is taken when holding tcf_lock in sample init:
    
    [  162.108959]  Possible interrupt unsafe locking scenario:
    
    [  162.116386]        CPU0                    CPU1
    [  162.121277]        ----                    ----
    [  162.126162]   lock(psample_groups_lock);
    [  162.130447]                                local_irq_disable();
    [  162.136772]                                lock(&(&p->tcfa_lock)->rlock);
    [  162.143957]                                lock(psample_groups_lock);
    [  162.150813]   <Interrupt>
    [  162.153808]     lock(&(&p->tcfa_lock)->rlock);
    [  162.158608]
                    *** DEADLOCK ***
    
    In order to prevent potential lock inversion dependency between tcf_lock
    and psample_groups_lock, extract call to psample_group_get() from tcf_lock
    protected section in sample action init function.
    
    Fixes: 4e232818 ("net: sched: act_mirred: remove dependency on rtnl lock")
    Fixes: 764e9a24 ("net: sched: act_vlan: remove dependency on rtnl lock")
    Fixes: 729e0126 ("net: sched: act_tunnel_key: remove dependency on rtnl lock")
    Fixes: d7728495 ("net: sched: act_sample: remove dependency on rtnl lock")
    Fixes: e8917f43 ("net: sched: act_gact: remove dependency on rtnl lock")
    Fixes: b6a2b971 ("net: sched: act_csum: remove dependency on rtnl lock")
    Fixes: 2142236b ("net: sched: act_bpf: remove dependency on rtnl lock")
    Signed-off-by: default avatarVlad Buslov <vladbu@mellanox.com>
    Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
    653cd284
act_csum.c 16.2 KB