[PATCH net v2] net/rds: ib: drop fragments shorter than the header-declared length

WARM1d REVIEWED: 1 (1M)

From: Shubham Antil <hidden>
Date: 2026-10-07 06:21:46
Also in: linux-rdma, lkml
Subsystem: networking [general], rds - reliable datagram sockets, the rest · Maintainers: "David S. Miller", Eric Dumazet, Jakub Kicinski, Paolo Abeni, Allison Henderson, Linus Torvalds

1 review trailer (1 from subsystem maintainers).

rds_ib_process_recv() accepts an incoming RDS/IB fragment once the
receive completion reports at least an RDS header
(data_len >= sizeof(struct rds_header)).  It then trusts the
header-declared total message length h_len: for the first fragment of
a message it stores be32_to_cpu(hdr->h_len) in ic->i_recv_data_rem,
and rds_ib_inc_copy_to_user() later copies up to h_len bytes from the
fragment pages to userspace on recvmsg().

The number of payload bytes actually received into the fragment page is
data_len (after subtracting the header), but it is never checked against
the amount the fragment is accounted to contribute to the message,
min(i_recv_data_rem, RDS_FRAG_SIZE).  A fragment whose header advertises
a larger h_len than the payload it delivers is still linked onto the
reassembly list.  The fragment page comes from the per-CPU receive cache
and is not zeroed, so rds_ib_inc_copy_to_user() then copies up to h_len
bytes to the PF_RDS reader, including the uninitialized tail the receive
never wrote.

Reject a fragment that carries fewer payload bytes than it is accounted
to contribute before linking it onto the reassembly list.

The issue is reproducible under KMSAN with two hosts over rdma_rxe
(Soft-RoCE); the same reproducer confirms the fix stops it.

Fixes: 1e23b3ee0e94 ("RDS/IB: Receive datagrams via IB")
Reported-by: sungbyeongchan <redacted>
Closes: https://lore.kernel.org/netdev/20261006205204.1322102-1-tjdqudcks0424@naver.com/ (local)
Reported-by: Shubham Antil <redacted>
Reported-by: Giovanni Vignone <redacted>
Reported-by: Robert van Eijk <redacted>
Reported-by: Paolo Gentry <redacted>
Assisted-by: Claude:claude-opus-4-8
Signed-off-by: Shubham Antil <redacted>
Reviewed-by: Allison Henderson <achender@kernel.org>
---
v2:
 - No functional change.
 - Add Reported-by for sungbyeongchan (first to report; see Closes link)
   and for the Octane Security reporters.
 - Add Allison's Reviewed-by.
 net/rds/ib_recv.c | 9 +++++++++
 1 file changed, 9 insertions(+)
diff --git a/net/rds/ib_recv.c b/net/rds/ib_recv.c
index bd6cb3ffa..fee77b6d7 100644
--- a/net/rds/ib_recv.c
+++ b/net/rds/ib_recv.c
@@ -949,6 +949,15 @@ static void rds_ib_process_recv(struct rds_connection *conn,
 		}
 	}
 
+	/* h_len must be backed by the payload actually received (data_len),
+	 * else the unwritten frag-page tail is copied to userspace.
+	 */
+	if (data_len < min_t(u32, ic->i_recv_data_rem, RDS_FRAG_SIZE)) {
+		rds_ib_conn_error(conn,
+			"fragment shorter than header-declared length; forcing reconnect\n");
+		goto done;
+	}
+
 	list_add_tail(&recv->r_frag->f_item, &ibinc->ii_frags);
 	recv->r_frag = NULL;
 
-- 
2.43.0
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help