Hi Bruce/Chuck,
There is a problem with nfsd doing a WRITE of size 4093,4094,4095. To
reproduce, mount with RDMA and do a simple dd "dd if=/dev/zero
of=/mnt/testfile bs=4093 count=1". What happens is that nfsd fails to
parse GETATTR after the WRITE in the compound. It fails the operation
with ERR_OP_ILLEGAL.
The problem happens for the values where XDR round up ends up rounding
up to the page size. I don't know if my fix is the appropriate way to
fix this but with it I don't get the error:
diff --git a/fs/nfsd/nfs4xdr.c b/fs/nfsd/nfs4xdr.c
index 2c61c6b..a8489c3 100644
--- a/fs/nfsd/nfs4xdr.c
+++ b/fs/nfsd/nfs4xdr.c
@@ -1289,11 +1289,12 @@ static __be32 nfsd4_decode_opaque(struct nfsd4_compounda
len = XDR_QUADLEN(write->wr_buflen) << 2;
if (len >= avail) {
- int pages;
+ int pages = 0;
len -= avail;
- pages = len >> PAGE_SHIFT;
+ if (write->wr_buflen >= PAGE_SIZE)
+ pages = len >> PAGE_SHIFT;
argp->pagelist += pages;
argp->pagelen -= pages * PAGE_SIZE;
len -= pages * PAGE_SIZE;
So the problem is the using "len" instead of "write->wr_buflen" leads
for the values 4093,4094,4095 that are rounded up to 4096, it makes
pages=1 and the argp->pagelen ends up being a negative value leading
to problems of parsing the GETATTR.
If this looks OK, I can send a patch.
--
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
