[Crash-utility] [PATCH] crash/symbols: Error out earlier in case _stext_vmlinux is NULL or UNINITIALIZED

[Bhupesh Sharma]

Hi Dave,

On Tue, Feb 27, 2018 at 1:41 AM, Dave Anderson <anderson at redhat.com> wrote:
>
> Hi Bhupesh,
>
> I'm confused.  If the vmlinux cannot be found, then the crash session should
> bail out after printing the standard error message in find_booted_kernel():
>
>         error(INFO,
>              "cannot find booted kernel -- please enter namelist argument\n\n");
>
> If derive_kaslr_offset() is called, then it must have gotten past find_booted_kernel()
> with a vmlinux file.  What am I missing?

Thanks for pointing this out. You are right. I was missing a special
use case in 'find_booted_kernel'.

I will send another patch specifically to cover that case.

Meanwhile, let me self-NAK this patch.

Regards,
Bhupesh

>
>
> ----- Original Message -----
>> While running crash on KASLR enabled live arm64 kernels I noticed
>> that the crash would fail after printing the following error messages
>> if it cannot find vmlinux for the boot'ed system:
>>
>> crash: invalid kernel virtual address: 8470  type: "possible"
>> WARNING: cannot read cpu_possible_map
>> crash: invalid kernel virtual address: 8270  type: "present"
>> WARNING: cannot read cpu_present_map
>> crash: invalid kernel virtual address: 8070  type: "online"
>> WARNING: cannot read cpu_online_map
>> crash: invalid kernel virtual address: 8670  type: "active"
>> WARNING: cannot read cpu_active_map
>>
>> crash: cannot resolve "_stext"
>>
>> Since vmlinux is not available at the standard expected location, i.e
>> '/usr/lib/debug/lib/modules/<kernel-version>' directory, so
>> _stext_vmlinux stays UNINITIALIZED inside 'derive_kaslr_offset()'
>> and 'relocate' becomes 0.
>>
>> Later-on this can lead to 'cannot resolve "_stext"' error in
>> 'symbol_value()' function.
>>
>> We can go ahead and fix the same, but ideally since we are running
>> on a live system, we should be finding the vmlinux at the standard
>> location, otherwise we should do proper error handling.
>>
>> So this patch errors out earlier, if we don't find a valid value for
>> '_stext_vmlinux' and prompts the user to refer to the 'usage'
>> banner.
>>
>> Signed-off-by: Bhupesh Sharma <bhsharma at redhat.com>
>> ---
>>  symbols.c | 18 ++++++++++++++++++
>>  1 file changed, 18 insertions(+)
>>
>> diff --git a/symbols.c b/symbols.c
>> index 4db9af7bf044..30669950635d 100644
>> --- a/symbols.c
>> +++ b/symbols.c
>> @@ -663,6 +663,24 @@ derive_kaslr_offset(bfd *abfd, int dynamic, bfd_byte
>> *start, bfd_byte *end,
>>                       return;
>>       }
>>
>> +     /* For KASLR enabled live kernels, if vmlinux is not available
>> +      * at the standard expected location, i.e
>> +      * '/usr/lib/debug/lib/modules/<kernel-version>' directory,
>> +      * then _stext_vmlinux will be set to NULL or UNINITIALIZED.
>> +      *
>> +      * Later-on this may lead to "_stext" symbol not being resolved
>> +      * properly (which may lead to an ambiguous error message).
>> +      *
>> +      * So, error out earlier here itself, in case _stext_vmlinux
>> +      * is NULL or UNINITIALIZED.
>> +      */
>> +     if (!st->_stext_vmlinux || (st->_stext_vmlinux == UNINITIALIZED)) {
>> +             error(INFO, "_stext not found in vmlinux :\n"
>> +                   "   if running a live system -- please provide a suitable vmlinux,\n"
>> +                   "   otherwise -- please enter namelist argument.\n");
>> +             program_usage(SHORT_FORM);
>> +     }
>> +
>>       /*
>>        * To avoid mistaking an mismatched kernel version with
>>        * a kaslr offset, we make sure that the offset is
>> --
>> 2.7.4
>>
>>