On Tue, Feb 01, 2011 at 05:16:29PM +0300, Konstantin Khorenko wrote:
> Hi All,
>
> i've found a bug which makes NFSD to corrupt memory due to writing beyond the stat array.
>
> Initially this issue was found on 2.6.18-x RHEL5 kernel, but the same seems to be true for the current mainstream kernel (checked on 2.6.38-rc3).
>
> ./fs/nfsd/vfs.c:
> static inline struct raparms *
> nfsd_get_raparms(dev_t dev, ino_t ino)
> {
> ...
> // here we searched our file in a readahead cache
> // but not found it
> ...
> depth = nfsdstats.ra_size*11/10;
> ...
> // some stuff, but "depth" is not changed
> ...
> nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
> ...
> }
>
> And here we write to the 12th array element
> (nfsdstats.ra_size*11/10*10/nfsdstats.ra_size == 11),
Actually, I think that's true only if ra_size is divisible by 10; if
not, then depth will be something *less* than ra_size*11/10, so
depth*10/ra_size will be something less than 11 (namely 10).
> This means that each time we did not find a file in a readahead cache we
> corrupt memory.
... and ra_size is set to 16 * min(2, numthreads/8), so I think you'll
hit this only if min(2, numthreads/8) (rounded down) is a multiple of 5?
So for example someone using a number of threads that's a power of 2
shouldn't hit this case.
> Fortunately in a kernel with NFSDv4 compiled in it corrupts
> (inc) NFS4 ops counter, but in a kernel with NFSDv4 disabled it corrupts (inc)
> some other data, that lays in the memory beyond nfsdstats.
>
> Proposed fix is attached.
So after the fix:
> - depth = nfsdstats.ra_size*11/10;
> + depth = nfsdstats.ra_size;
depth*10/nfsdstats will always be 10 in this case. Looks good.
Thanks for finding this!
--b.
--
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
