On 12/11/11, Andi Kleen <andi@firstfloor.org> wrote:
> t cheney <cdmalord@gmail.com> writes:
>
>> This patch just add memory layout(same as /proc/pid/maps) to
>> coredump file. The layout is appended to corenote segment with
>> flag NT_MAPS=7.
>
> Seems like a reasonable idea, but can you please share code with
> the proc implementation?
Yes, I will post it in following mail.
>
> It's a bit unusal to have ASCII in a coredump, but I suppose expressing
> this in binary would be somewhat ugly.
>
>> +	char *buf = notes->data;
>> +	size_t core_limit = notes->datasz;
>> +	gate_vma = get_gate_vma(current->mm);
>> +
The maps info is appended to core notes segment in which there are all
ASCII text such as regs,signal and thread info.
>
> It seems a bit dubious to do this without locking even in a core dump.
>
>> +	for (vma = first_vma(current, gate_vma); vma != NULL;
>> +		vma = next_vma(vma, gate_vma)) {
>> +		flags = vma->vm_flags;
>> +		maps_size += len;
>> +		if (file) {
>> +			if (maps_size > core_limit)
>> +				break;
>
> You have a one-off bug here. Also below.
>
>> +			DUMP_WRITE(buf, len, foffset);
>> +		}
>
Because proceses that could change vma pages are blocked in do_exit,
here vma pages are safe.
>
> -Andi
> --
> ak@linux.intel.com -- Speaking for myself only
>