[lxc-devel] [PATCH] lxc_start: report error message when config is not a regular file

[Shi, Yang]

On 10/18/2015 7:10 PM, Serge Hallyn wrote:
> Quoting Shi, Yang (yang.shi at linaro.org):
>> On 10/15/2015 7:31 AM, Serge Hallyn wrote:
>>> Quoting Yang Shi (yang.shi at linaro.org):
>>>> When run the below command:
>>>> lxc-start -n name -f /path/to/config
>>>>
>>>> If config is not a regular file, lxc-start will be killed by SEG FAULT
>>>> directly.
>>>
>>> What exactly was the file to get a segfault?  When I try with a device
>>> or a directory I do not get a segfault.
>>
>> I'm using lxc 1.1.3 stable release on armv8 target, when I run the
>> below command I got segment fault.
>>
>> # lxc-start -n test -f /var/lib/lxc/test/
>> lxc-start[309]: unhandled level 2 translation fault (11) at
>> 0x00000000, esr 0x92000006
>> pgd = ffff80834780e000
>> [00000000] *pgd=00000083cb385003, *pud=00000083c99e3003,
>> *pmd=0000000000000000
>>
>> CPU: 2 PID: 309 Comm: lxc-start Not tainted 4.1.10 #25
>> Hardware name: Freescale Layerscape 2085a RDB Board (DT)
>> task: ffff80007b944d00 ti: ffff808347874000 task.ti: ffff808347874000
>> PC is at 0xffff9d1ff2d0
>> LR is at 0xffff9d1bc0d0
>> pc : [<0000ffff9d1ff2d0>] lr : [<0000ffff9d1bc0d0>] pstate: 80000000
>> sp : 0000fffffd9f3af0
>> x29: 0000fffffd9f3af0 x28: 0000000000413000
>> x27: 0000000000000000 x26: 0000000000000001
>> x25: 0000000000000000 x24: 0000000000414370
>> x23: 0000fffffd9f3be0 x22: 0000ffff9d35d3b0
>> x21: 000000000000000f x20: 0000ffff9d37c000
>> x19: 00000000004166d0 x18: 0000000000000001
>> x17: 0000ffff9d1bc1e8 x16: 0000ffff9d37a458
>> x15: 1201001100000000 x14: 00000000004167e0
>> x13: 0000000000000000 x12: 0000000000000023
>> x11: 0000000000000023 x10: 0101010101010101
>> x9 : 7f7f7f7f7f7f7f7f x8 : 0101010101010101
>> x7 : 7f7f7f7f7f7f7f7f x6 : ff4753404f5e5245
>> x5 : 8000000000000000 x4 : 0000000000000000
>> x3 : 00485441505f5346 x2 : 544f4f525f43584c
>> x1 : 0000000000000000 x0 : 0000000000000000
>>
>>>
>>>> Exit with some error message to improve user experience.
>>>
>>> This is not the place to put that check (if it is needed).  It should go
>>> into either lxc_container_new or load_config, wherever it is failing.
>>
>> Thanks. If you think the segment fault error sounds valid, I will
>> try to move the check to the right place.
>
> Could you run that in gdb with lxc debug symbols and show a stack
> trace?

Yes, sure. But, it may need take some time since I have some other 
urgent tasks right now.

Thanks,
Yang

>