Re: [BUG] running strace utility can cause system to hang at sys_ptrace

[Jay Lan <jlan@sgi.com>]

Correct David's email addr. Add Christoph Hellwig.

Regards,
  - jay


Jay Lan wrote:
> We ran into a system hang problem at sys_ptrace.
> 
> The system may hang if a user attempts to "strace" a pthreaded
> application that is making frequent calls to pthread_create &
> pthread_exit.
> The hang is caused by a task sleeping (yield()) while holding the
> tasklist_lock.
> 
> Note that the problem appears to exist ONLY on IA64. IA64 has an
> ARCH-specific implementation of sys_ptrace().
> 
> The failing path is:
> 
>          sys_ptrace
>            read_lock(&tasklist_lock);
>               find_thread_for_addr(child, addr);
>                 thread_matches(p, addr)
>                   ptrace_check_attach(thread, 0)
>                     wait_task_inactive(child);
>                       yield();
> 
> Failures always show 1 task sleeping with the backtrace listed
> above, plus tasks on all cpus spinning waiting to write_lock the
> tasklist lock.
> 
> Sleeping
>   Stack traceback for pid 31069
>   0xe00000343b800000    31069    0    3   R  0xe00000343b8003a0 strace
>   0xa000000100981c60 schedule+0x2460
>   0xa00000010007d4d0 sys_sched_yield+0x110
>   0xa000000100982120 yield+0x40
>   0xa000000100082f10 wait_task_inactive+0x170
>   0xa0000001000a2570 ptrace_check_attach+0x1d0
>   0xa00000010002e9b0 sys_ptrace+0x1f0
>   0xa00000010000c900 ia64_ret_from_syscall
> 
> CPU 0
>   0xe00000343b5f0000     4973    0    3   C  0xe00000343b5f03a0  thdcrex
>   0xa00000010000caa0 ia64_leave_kernel
>   0xa00000010053d0a0 _raw_write_lock+0x120
>   0xa0000001009854d0 _write_lock_irq+0x30
>   0xa000000100093a00 do_exit+0xae0
>   0xa000000100094560 sys_exit+0x20
>   0xa00000010000c900 ia64_ret_from_syscall
> 
> CPU 1
>   0xe000003038880000     4972    0    1   R  0xe0000030388803a0  thdcrex
>   0xa00000010000caa0 ia64_leave_kernel
>   0xa00000010053d0a0 _raw_write_lock+0x120
>   0xa0000001009854d0 _write_lock_irq+0x30
>   0xa0000001000880e0 copy_process+0x1f80
>   0xa000000100088e30 do_fork+0x150
>   0xa00000010000c160 sys_clone2+0x60
>   0xa00000010000c900 ia64_ret_from_syscall
> 
> 
> Regards,
>  - jay
> -
> To unsubscribe from this list: send the line "unsubscribe linux-ia64" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html