[RFC ABRT PATCH v2] Add gdb python plugin which analyzes vulnerability

Denys Vlasenko dvlasenk at redhat.com
Mon Jun 17 13:56:10 UTC 2013


On 06/13/2013 04:55 PM, Martin Milata wrote:
> Some remarks regarding the code below:
> 
>> +    if self.operands.startswith("0x") or self.operands.startswith("-0x") or self.operands.startswith("("):
>> +        mem_op_pos = 0
>> +    else:
>> +        mem_op_pos = self.operands.find(",0x")
>> +        if mem_op_pos < 0:
>> +            mem_op_pos = self.operands.find(",-0x")
>> +        if mem_op_pos < 0:
>> +            mem_op_pos = self.operands.find(":0x")
>> +        if mem_op_pos < 0:
>> +            mem_op_pos = self.operands.find(":-0x")
>> +        if mem_op_pos < 0:
>> +            mem_op_pos = self.operands.find(",(")
>> +        if mem_op_pos < 0:
>> +            mem_op_pos = self.operands.find(":(")
>> +        if mem_op_pos < 0:
>> +            return False # no memory operands
>> +        mem_op_pos += 1
> Can you use a regular expression for this, possibly with splitting the
> operands into list first? http://docs.python.org/2.7/library/re.html

I think "import re", compiling of the regexp and matching
will be slower than searching for a string.
It is also less understandable:

regex = re.compile('^\s+(0x[0-9a-f]+)\s+(\<\+\d+\>):\s+(.*)$')

I can't easily tell what this regexp is doing.

The only benefit is that using regex may take a few lines less
than existing code.

>> +        # HACK_ALERT: kernels before 3.?.? do not record siginfo in coredumps,
>> +        # so $_siginfo isn't present.
> By the way, do you have any reference for this? The prstatus structure
> which contains a signal number seems to be present in coredumps at least
> on 2.6.32. Or am I misunderstanding the problem?

The point is, gdb needs _complete_ siginfo to be able to show $_siginfo.
A few bits of signal info in prstatus is a complete siginfo.
(Of course, $_siginfo.si_signo doesn't need complete siginfo,
but gdb doesn't know that...)

>> +    for line in raw_instructions.split("\n"):
>> +        # line can be:
>> +        # "Dump of assembler code from 0xAAAA to 0xBBBB:"
>> +        # "[=>] 0x00000000004004dc[ <+0>]:  push   %rbp"
>> +        #   (" <+0>" part is present when we run on a live process,
>> +        #   on coredump it is absent)
>> +        # "End of assembler dump."
>> +        # "" (empty line)
> What about using regular expressions here?

re.compile('^\s+(0x[0-9a-f]+)\s+(\<\+\d+\>):\s+(.*)$')

The rest of your comments make sense, I'll update the branch.

-- 
vda



More information about the Crash-catcher mailing list