Re: [syzbot] [ext4?] general protection fault in hrtimer_nanosleep

[Thomas Gleixner]

On Thu, Nov 02 2023 at 13:08, Aleksandr Nogikh wrote:
> On Wed, Nov 1, 2023 at 1:58 PM Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
>> Unfortunately repro.syz does not hold up to its name and refuses to
>> reproduce.
>
> For me, on a locally built kernel (gcc 13.2.0) it didn't work either.
>
> But, interestingly, it does reproduce using the syzbot-built kernel
> shared via the "Downloadable assets" [1] in the original report. The
> repro crashed the kernel in ~1 minute.
>
> [1] https://github.com/google/syzkaller/blob/master/docs/syzbot_assets.md
>
> [  125.919060][    C0] BUG: KASAN: stack-out-of-bounds in rb_next+0x10a/0x130
> [  125.921169][    C0] Read of size 8 at addr ffffc900048e7c60 by task
> kworker/0:1/9
> [  125.923235][    C0]
> [  125.923243][    C0] CPU: 0 PID: 9 Comm: kworker/0:1 Not tainted
> 6.6.0-rc7-syzkaller-00142-g888cf78c29e2 #0
> [  125.924546][    C0] Hardware name: QEMU Standard PC (Q35 + ICH9,
> 2009), BIOS 1.16.2-debian-1.16.2-1 04/01/2014
> [  125.926915][    C0] Workqueue: events nsim_dev_trap_report_work
> [  125.929333][    C0]
> [  125.929341][    C0] Call Trace:
> [  125.929350][    C0]  <IRQ>
> [  125.929356][    C0]  dump_stack_lvl+0xd9/0x1b0
> [  125.931302][    C0]  print_report+0xc4/0x620
> [  125.932115][    C0]  ? __virt_addr_valid+0x5e/0x2d0
> [  125.933194][    C0]  kasan_report+0xda/0x110
> [  125.934814][    C0]  ? rb_next+0x10a/0x130
> [  125.936521][    C0]  ? rb_next+0x10a/0x130
> [  125.936544][    C0]  rb_next+0x10a/0x130
> [  125.936565][    C0]  timerqueue_del+0xd4/0x140
> [  125.936590][    C0]  __remove_hrtimer+0x99/0x290
> [  125.936613][    C0]  __hrtimer_run_queues+0x55b/0xc10
> [  125.936638][    C0]  ? enqueue_hrtimer+0x310/0x310
> [  125.936659][    C0]  ? ktime_get_update_offsets_now+0x3bc/0x610
> [  125.936688][    C0]  hrtimer_interrupt+0x31b/0x800
> [  125.936715][    C0]  __sysvec_apic_timer_interrupt+0x105/0x3f0
> [  125.936737][    C0]  sysvec_apic_timer_interrupt+0x8e/0xc0
> [  125.936755][    C0]  </IRQ>
> [  125.936759][    C0]  <TASK>

Which is a completely different failure mode.

It explodes in the hrtimer interrupt when dequeuing an hrtimer for
expiry. That means the corresponding embedded rb_node is corrupted,
which points to random data corruption.

As you can reproduce (it still fails here with the provided assets),
does the failure change when you run it several times?

Thanks,

        tglx