Re: [BUG REPORT] softlock up in net/core cleanup_net

[Eric Dumazet]

On Tue, May 30, 2023 at 3:57 PM jiangheng (G) <jiangheng14@xxxxxxxxxx> wrote:
>
> Hi all,
> on linux 5.10,  we want to use docker interactively when testing an internal feature. When docker restarts the container, it will call cleanup_net and a crash will occur.
>

Do not use 5.10, please. This is too old.

Back in 5.18, we made some improvements

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=4caaf75888d893b6525173a1537980e13c141988



> [  843.330515] CPU: 0 PID: 158 Comm: kworker/u8:2 Kdump: loaded Tainted: G    B      OEL    #1 [  843.330516] Hardware name: QEMU KVM Virtual Machine, BIOS 0.0.0 02/06/2015 [  843.330523] Workqueue: netns cleanup_net [  843.330526] pstate: 60400085 (nZCv daIf +PAN -UAO -TCO BTYPE=--) [  843.330529] pc : machine_kexec+0x48/0x2b0 [  843.330531] lr : machine_kexec+0x48/0x2b0 [  843.330531] sp : ffff80010284bb10 [  843.330533] x29: ffff80010284bb10 x28: ffff0000ff851cf8 [  843.330535] x27: ffff0000ff851d78 x26: ffff80010284bda0 [  843.330537] x25: ffff800101d9c000 x24: 0000000000000001 [  843.330539] x23: ffff800101d9c650 x22: ffff800101eb6000 [  843.330541] x21: ffff800101eb6000 x20: ffff0000cba23c00 [  843.330543] x19: ffff0000cba23c00 x18: 0000000000000020 [  843.330545] x17: 0000000000000000 x16: ffff800100d27a3c [  843.330548] x15: ffffffffffffffff x14: 0000000060000085 [  843.330550] x13: ffff8001001c090c x12: 0000000000000040 [  843.330552] x11: ffff800101aad158 x10: 00000000ffff8000 [  843.330554] x9 : ffff800100157654 x8 : 0000000000000000 [  843.330556] x7 : ffff8001017ed158 x6 : 0000000000017ffd [  843.330559] x5 : ffff0000ff84b410 x4 : ffff80010284b910 [  843.330561] x3 : 0000000000000001 x2 : 0000000000000000 [  843.330563] x1 : 0000000000000000 x0 : ffff0000c09eb9c0 [  843.330566] Call trace:
> [  843.330569]  machine_kexec+0x48/0x2b0 [  843.330573]  __crash_kexec+0x90/0x13c [  843.330578]  panic+0x314/0x4d8 [  843.330582]  watchdog_timer_fn+0x26c/0x2f0 [  843.330585]  __run_hrtimer+0x98/0x2b4 [  843.330586]  __hrtimer_run_queues+0xbc/0x130 [  843.330588]  hrtimer_interrupt+0x150/0x3e4 [  843.330592]  arch_timer_handler_virt+0x3c/0x50 [  843.330596]  handle_percpu_devid_irq+0x90/0x1f4
> [  843.330599]  __handle_domain_irq+0x84/0x100 [  843.330601]  gic_handle_irq+0x88/0x2b0 [  843.330603]  el1_irq+0xb8/0x140 [  843.330605]  smp_call_function_single+0x1b8/0x1dc
> [  843.330608]  rcu_barrier+0x1c4/0x2d0
> [  843.330612]  netdev_run_todo+0x7c/0x330 [  843.330615]  rtnl_unlock+0x18/0x24 [  843.330616]  default_device_exit_batch+0x15c/0x190
> [  843.330621]  ops_exit_list+0x70/0x84
> [  843.330622]  cleanup_net+0x184/0x2e0
> [  843.330625]  process_one_work+0x1d4/0x4bc [  843.330627]  worker_thread+0x150/0x400 [  843.330629]  kthread+0x108/0x134 [  843.330631]  ret_from_fork+0x10/0x18 [  843.330633] ---[ end trace 8378c01c76c90cc4 ]--- [  843.330637] Bye!
>
> Crash:bt -l
> PID: 158    TASK: ffff0000c09eb9c0  CPU: 0   COMMAND: "kworker/u8:2"
> PID: 158    TASK: ffff0000c09eb9c0  CPU: 0   COMMAND: "kworker/u8:2"
> bt: invalid kernel virtual address: 0  type: "IRQ stack contents"
> bt: read of IRQ stack at 0 failed
> #0 [ffff80010284bb60] __crash_kexec at ffff8001001c0908
>     /usr/src/debug/kernel/./arch/arm64/include/asm/kexec.h: 57
> #1 [ffff80010284bcf0] panic at ffff800100d256a4
>     /usr/src/debug/kernel/kernel/panic.c: 392
> #2 [ffff80010284bde0] watchdog_timer_fn at ffff80010020a5c8
>     /usr/src/debug/kernel/kernel/watchdog.c: 578
> #3 [ffff80010284be30] __run_hrtimer at ffff800100191d24
>     /usr/src/debug/kernel/kernel/time/hrtimer.c: 1586
> #4 [ffff80010284be80] __hrtimer_run_queues at ffff800100191ffc
>     /usr/src/debug/kernel/kernel/time/hrtimer.c: 1650
> #5 [ffff80010284bee0] hrtimer_interrupt at ffff80010019267c
>     /usr/src/debug/kernel/kernel/time/hrtimer.c: 1712
> #6 [ffff80010284bf50] arch_timer_handler_virt at ffff800100aa9a38
>     /usr/src/debug/kernel/drivers/clocksource/arm_arch_timer.c: 674
> #7 [ffff80010284bf60] handle_percpu_devid_irq at ffff80010016500c
>     /usr/src/debug/kernel/./arch/arm64/include/asm/percpu.h: 45
> #8 [ffff80010284bf90] __handle_domain_irq at ffff80010015b840
>     /usr/src/debug/kernel/./include/linux/irqdesc.h: 153
> #9 [ffff80010284bfd0] gic_handle_irq at ffff800100010144
>     /usr/src/debug/kernel/./include/linux/irqdesc.h: 171
> --- <IRQ stack> ---
> #10 [ffff800102d4bb20] el1_irq at ffff800100012374
>     /usr/src/debug/kernel/arch/arm64/kernel/entry.S: 672
> #11 [ffff800102d4bb40] smp_call_function_single at ffff8001001b1e68
>     /usr/src/debug/kernel/./arch/arm64/include/asm/cmpxchg.h: 278
> #12 [ffff800102d4bba0] rcu_barrier at ffff800100178ba0
>     /usr/src/debug/kernel/kernel/rcu/tree.c: 3920
> #13 [ffff800102d4bc00] netdev_run_todo at ffff800100b3f768
>     /usr/src/debug/kernel/net/core/dev.c: 10313
> #14 [ffff800102d4bc80] rtnl_unlock at ffff800100b4cb54
>     /usr/src/debug/kernel/net/core/rtnetlink.c: 114
> #15 [ffff800102d4bc90] default_device_exit_batch at ffff800100b378d8
>     /usr/src/debug/kernel/net/core/dev.c: 11287
> #16 [ffff800102d4bd00] ops_exit_list at ffff800100b2337c
>     /usr/src/debug/kernel/net/core/net_namespace.c: 200
> #17 [ffff800102d4bd30] cleanup_net at ffff800100b25ab0
>     /usr/src/debug/kernel/net/core/net_namespace.c: 616
> #18 [ffff800102d4bd90] process_one_work at ffff8001000de784
>     /usr/src/debug/kernel/kernel/workqueue.c: 2354
> #19 [ffff800102d4bdf0] worker_thread at ffff8001000df18c
>     /usr/src/debug/kernel/kernel/workqueue.c: 2500
> #20 [ffff800102d4be50] kthread at ffff8001000e75a4
>     /usr/src/debug/kernel/kernel/kthread.c: 313
>
> The above backtrace seems to be caused func:netdev_run_todo() that the size of list not null.
> void netdev_run_todo(void)
> {
>          struct net_device *dev, *tmp;
>          struct list_head list;
> #ifdef CONFIG_LOCKDEP
>          struct list_head unlink_list;
>
>          list_replace_init(&net_unlink_list, &unlink_list);
>
>          while (!list_empty(&unlink_list)) {
>                    struct net_device *dev = list_first_entry(&unlink_list,
>                                                     struct net_device,
>                                                unlink_list);
>                    list_del_init(&dev->unlink_list);
>                    dev->nested_level = dev->lower_level - 1;
>          }
> #endif
>
>          /* Snapshot list, allow later requests */
>          list_replace_init(&net_todo_list, &list);
>
>          __rtnl_unlock();
>
>          /* Wait for rcu callbacks to finish before next phase */
>          if !(list_empty(&list))
>                    rcu_barrier();
>
> I wonder if softlockup is due to the above code? Please help analyze the possible causes of this.

I have seen similar reports about RTNL starvation from syzbot reports.

How many netns do you dismantle per second ?

On recent kernels this is how fast things run :

# time (for i in {1..1000}; do unshare -n /bin/bash -c "ifconfig lo up"; done)

real 0m3.379s
user 0m1.708s
sys 0m1.553s