NFSD: da_addr_body field missing in some GETDEVICEINFO replies
authorChuck Lever <chuck.lever@oracle.com>
Wed, 16 Aug 2023 14:20:52 +0000 (10:20 -0400)
committerChuck Lever <chuck.lever@oracle.com>
Tue, 29 Aug 2023 21:45:22 +0000 (17:45 -0400)
commit6372e2ee629894433fe6107d7048536a3280a284
tree047b6b66caa3b8483fea8e9a50ddfdd82863d96f
parent2a4557452aacf9e7168cb83bc102467094ff9391
NFSD: da_addr_body field missing in some GETDEVICEINFO replies

The XDR specification in RFC 8881 looks like this:

struct device_addr4 {
layouttype4 da_layout_type;
opaque da_addr_body<>;
};

struct GETDEVICEINFO4resok {
device_addr4 gdir_device_addr;
bitmap4 gdir_notification;
};

union GETDEVICEINFO4res switch (nfsstat4 gdir_status) {
case NFS4_OK:
GETDEVICEINFO4resok gdir_resok4;
case NFS4ERR_TOOSMALL:
count4 gdir_mincount;
default:
void;
};

Looking at nfsd4_encode_getdeviceinfo() ....

When the client provides a zero gd_maxcount, then the Linux NFS
server implementation encodes the da_layout_type field and then
skips the da_addr_body field completely, proceeding directly to
encode gdir_notification field.

There does not appear to be an option in the specification to skip
encoding da_addr_body. Moreover, Section 18.40.3 says:

> If the client wants to just update or turn off notifications, it
> MAY send a GETDEVICEINFO operation with gdia_maxcount set to zero.
> In that event, if the device ID is valid, the reply's da_addr_body
> field of the gdir_device_addr field will be of zero length.

Since the layout drivers are responsible for encoding the
da_addr_body field, put this fix inside the ->encode_getdeviceinfo
methods.

Fixes: 9cf514ccfacb ("nfsd: implement pNFS operations")
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: Tom Haynes <loghyr@gmail.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
fs/nfsd/blocklayoutxdr.c
fs/nfsd/flexfilelayoutxdr.c
fs/nfsd/nfs4xdr.c