nfs: fix nfs4d readlink truncated packet
authorAvi Kivity <avi@cloudius-systems.com>
Thu, 26 Jun 2014 14:23:12 +0000 (17:23 +0300)
committerJ. Bruce Fields <bfields@redhat.com>
Wed, 2 Jul 2014 21:37:13 +0000 (17:37 -0400)
XDR requires 4-byte alignment; nfs4d READLINK reply writes out the padding,
but truncates the packet to the padding-less size.

Fix by taking the padding into consideration when truncating the packet.

Symptoms:

# ll /mnt/
ls: cannot read symbolic link /mnt/test: Input/output error
total 4
-rw-r--r--. 1 root root  0 Jun 14 01:21 123456
lrwxrwxrwx. 1 root root  6 Jul  2 03:33 test
drwxr-xr-x. 1 root root  0 Jul  2 23:50 tmp
drwxr-xr-x. 1 root root 60 Jul  2 23:44 tree

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>
Fixes: 476a7b1f4b2c (nfsd4: don't treat readlink like a zero-copy operation)
Reviewed-by: Kinglong Mee <kinglongmee@gmail.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
fs/nfsd/nfs4xdr.c

index 5b4fef55676a9feece8c2b6bf6aa418396115093..2fc7abebeb9b97e4192f04311f06752c4de92d79 100644 (file)
@@ -3278,7 +3278,7 @@ nfsd4_encode_readlink(struct nfsd4_compoundres *resp, __be32 nfserr, struct nfsd
 
        wire_count = htonl(maxcount);
        write_bytes_to_xdr_buf(xdr->buf, length_offset, &wire_count, 4);
-       xdr_truncate_encode(xdr, length_offset + 4 + maxcount);
+       xdr_truncate_encode(xdr, length_offset + 4 + ALIGN(maxcount, 4));
        if (maxcount & 3)
                write_bytes_to_xdr_buf(xdr->buf, length_offset + 4 + maxcount,
                                                &zero, 4 - (maxcount&3));