On Feb 2, 2009, at Feb 2, 2009, 4:41 PM, Ian Dall wrote:
On Mon, 2009-02-02 at 12:00 -0500, Chuck Lever wrote:
On Jan 31, 2009, at 10:45 PM, Ian Dall wrote:
The following patch is relative to 2.6.29-rc3. The crux of the
matter is
that sock_addr structures can't be reliably compared using memcmp()
because there are padding bytes in the structure which can't be
guaranteed to be the same even when the sock_addr structures refer
to
the same thing.
There have been significant changes to the way address comparison is
done in client.c since my original patch (Kernel Bug Tracker #11061,
comment 19). In particular ipv4 addresses are now mapped to ipv6
addresses for comparison (but only if the kernel is configured for
IPV6). The conditional compilation blocks were quite big and
contained
duplication. Adding my changes in their original style would have
only
made matters worse. So I have taken to opportunity to clean up. By
inverting the logic, I was able to reduce the depth of nested ifs,
reduce the number of trivial helper functions and reduce the size of
the
conditional compilation blocks. This is a mater of opinion of
course,
but I think the result is cleaner and easier to read and maintain.
This has been tested with kernels configured with and without IPV6.
As far as I can tell, this is only an issue in nfs_match_client(), so
there isn't any need to rewrite the other pieces.
True. and I can do it that way easily enough if that is more
acceptable.
However I felt it better to keep all the address comparison code with
the switches on sa_family and compilation conditional on CONFIG_IPV6
etc
together. That way it is less likely that something gets fixed in one
place and not another and easier to add more network address types,
should that be necessary at some time down the track. Also, there
would
be a lack of symmetry in that nfs_match_client would then need to
"know"
about the details of a sock_addr, whereas nfs_find_client and
nfs_find_client_next can treat the sockaddr as an opaque object. Small
points perhaps.
Simply define a helper near nfs_sockaddr_match_ipaddr(), included
inside the "#if defined(CONFIG_IPV6)," that handles port comparison,
and call that from nfs_match_client().
Eventually we will probably extract all of this address family-
specific kruft into something that can be shared amongst all of the
NFS-related components, so it will likely end up split out this way in
any event.
--
Chuck Lever
chuck[dot]lever[at]oracle[dot]com
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
