Skip to contentSkip to conversation

xprtrdma: Fix Write chunk payload corruption on a short reply header

PATCH New Chuck Lever 1 participant 1 message Aug 26

Subsystems (3)
lore
Share & tools
lore

Chuck Lever

author maintainer
#0 lore raw

PATCH v2 1/1 xprtrdma: Fix Write chunk payload corruption on a short reply header

An NFSv3 server may return READ3resok with attributes_follow set to FALSE. The client has already reserved enough head iovec space for the largest possible reply header, so such a reply falls 84 bytes short of that estimate.

rpcrdma_inline_fixup() points the head iovec at the received message but leaves its length at the reserved size, so it runs past the header into stale receive buffer content. xdr_realign_pages() then shifts those bytes into the page list, and the READ returns transport data where READ payload belongs.

nfs3_read_done() records the shorter header size, so on an ordinary NFSv3 mount, only the first READ is corrupted. However, the FlexFiles layout driver installs its own completion callback, so every READ from a data server is corrupt.

To address this, reduce the head iovec to the header bytes actually received when a Write chunk carries the READ payload. An auth flavor that transforms the payload cannot use a Write chunk, so scoping it to Write chunks leaves krb5[ip] alone.

---
Changes in v2:
- Reduce the head iovec to the inline header bytes received.
- Exclude a Responder's implicit XDR roundup from that length.
- Link to v1: https://lore.kernel.org/r/[email protected]

---
 net/sunrpc/xprtrdma/rpc_rdma.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)
net/sunrpc/xprtrdma/rpc_rdma.c +12 -2
index 1285f04cdac1..ba8bdca8a49f 100644
@@ -1243,9 +1243,19 @@ rpcrdma_decode_msg(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep,
 
 	/* Build the RPC reply's Payload stream in rqst->rq_rcv_buf */
 	base = (char *)xdr_inline_decode(xdr, 0);
-	rpclen = xdr_stream_remaining(xdr);
+	rpclen = rep->rr_hdrbuf.len - xdr_stream_pos(xdr);
+	if (writelist)
+		rpclen &= ~3;
+	if (unlikely(!rpclen))
+		return -EIO;
+	if (writelist && rpclen < rqst->rq_rcv_buf.head[0].iov_len) {
+		rqst->rq_rcv_buf.head[0].iov_len = rpclen;
+		rqst->rq_private_buf.head[0].iov_len = rpclen;
+	}
+
 	r_xprt->rx_stats.fixup_copy_count +=
-		rpcrdma_inline_fixup(rqst, base, rpclen, writelist & 3);
+		rpcrdma_inline_fixup(rqst, base, rpclen,
+				     xdr_pad_size(writelist));
 
 	r_xprt->rx_stats.total_rdma_reply += writelist;
 	return rpclen + xdr_align_size(writelist);
Signature · 1 line
2.55.0