Hi Mario,
On to, 2014-07-10 at 12:30 +0200, Mario Schuknecht wrote:
> Hi Jukka,
>
>
> 2014-07-10 10:12 GMT+02:00 Jukka Rissanen
> <[email protected]>:
> Hi Mario,
>
> On ke, 2014-07-09 at 16:52 +0200, Mario Schuknecht wrote:
> > 4-byte memory access works only when pointer is 4-byte
> aligned. Use memcpy
> > function, since 4-byte pointer alignment is not guaranteed.
> The same applies
> > to 2-byte memory access.
>
>
> What hardware do you see this issue? At least the Intel and
> ARM devices
> I have used work just fine as is.
>
>
>
> I see the problem on Marvell Kirkwood 88F6282 ARM SOC. The issue is
> reproducible.
> The pointer 'c' points to a 2-byte boundary. As a result the domain
> question Class
> field, that is before TTL field is set to zero and thus the cache
> contents is invalid.
Ok, good to know.
>
>
> I would prefer the use << operator instead of memcpy here. See
> usage
> examples in dnsproxy.c for details.
>
>
> Should I change it like this?
> /* now the 4 byte TTL field */
> i = htonl(new_ttl);
> c[0] = (unsigned char)((i) & 0xFF);
>
> c[1] = (unsigned char)((i>>8) & 0xFF);
> c[2] = (unsigned char)((i>>16) & 0xFF);
> c[3] = (unsigned char)((i>>24) & 0xFF);
What about just doing
c[0] = new_ttl >> 24 & 0xff;
c[1] = new_ttl >> 16 & 0xff;
c[2] = new_ttl >> 8 & 0xff;
c[3] = new_ttl & 0xff;
That should give the same result and avoids htonl() call, also () are
not needed as >> is done before AND.
> c += 4;
> len -= 4;
> if (len < 0)
> break;
>
>
> /* now the 2 byte rdlen field */
> w = c[0] | c[1] << 8;
>
> c += ntohs(w) + 2;
> len -= ntohs(w) + 2;
We can avoid ntohs() by this
w = c[0] << 8 | c[1];
c += w + 2;
len -= w + 2;
and so on.
>
>
> And
>
>
> /* now the query, which is a name and 2 16 bit words */
> l = dns_name_length(c) + 1;
> c += l;
> w = c[0] | c[1] << 8;
>
> type = ntohs(w);
Could be written as
type = c[0] << 8 | c[1];
Cheers,
Jukka
_______________________________________________
connman mailing list
[email protected]
https://lists.connman.net/mailman/listinfo/connman
