On Mon, May 15, 2017 at 07:26:38PM +0800, Eryu Guan wrote:
> I found xfstests generic/053 generic/105 generic/307 and generic/319
> started to fail on NFSv3 mount against Linux server running v4.12-rc1
> kernel. They all try to set acl (xattr) on file/dir reside on NFSv3
> mount. And this is a regression, 4.11 Linux server worked fine. A simple
> reproducer would be:
>
> mount linux-server:/export /mnt/nfs
> touch /mnt/nfs/testfile
> setfattr -n user.test -v helloworld /mnt/nfs/testfile
>
> The last setfattr returned EACCES
>
> setfattr: /mnt/nfs/testfile: Operation not supported
>
> And I bisected this to commit 51f56777
Oh, yikes, very sorry about that, the patch should not have gone
upstream.
This was my original approach to the problem, but I abandoned it in part
because someone pointed out the problem you found. And Neil suggested
the approach used in the better fix, which is upstream as e6838a29ecb
"nfsd: check for oversized NFSv2/v3 arguments".
But somehow the old version must have been left in the tree branch I
gave Linus to pull?
Apologies, I'll take another look to make sure I'm not still confused
about somethign, and then send a revert. And I'll try to keep out an
eye for the stable backports to NAK those....
--b.
>
> commit 51f567777799c9d85a778302b9eb61cf15214a98
> Author: J. Bruce Fields <bfields@xxxxxxxxxx>
> Date: Thu Apr 6 22:36:31 2017 -0400
>
> nfsd: check for oversized NFSv2/v3 arguments
>
> And I only need to revert this hunk to "fix" the regression:
>
> diff --git a/include/linux/sunrpc/svc.h b/include/linux/sunrpc/svc.h
> index e770abe..6ef19cf 100644
> --- a/include/linux/sunrpc/svc.h
> +++ b/include/linux/sunrpc/svc.h
> @@ -336,8 +336,7 @@ xdr_argsize_check(struct svc_rqst *rqstp, __be32 *p)
> {
> char *cp = (char *)p;
> struct kvec *vec = &rqstp->rq_arg.head[0];
> - return cp >= (char*)vec->iov_base
> - && cp <= (char*)vec->iov_base + vec->iov_len;
> + return cp == (char *)vec->iov_base + vec->iov_len;
> }
>
> static inline int
>
> Debug log printing out cp, iov_base and iov_base + iov_len confirmed
> that cp can be greater than iov_base but smaller than iov_base + iov_len
>
> [10868.819939] xdr_argsize_check: cp == ffff88020ea6f094, iov_base == ffff88020ea6f068, iov_base + iov_len == ffff88020ea6f104
>
> Please help take a look. If you need more info please let me know.
>
> Thanks,
> Eryu
--
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
