CVE-2026-52995

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> net/rds: zero per-item info buffer before handing it to visitors<br /> <br /> rds_for_each_conn_info() and rds_walk_conn_path_info() both hand a<br /> caller-allocated on-stack u64 buffer to a per-connection visitor and<br /> then copy the full item_len bytes back to user space via<br /> rds_info_copy() regardless of how much of the buffer the visitor<br /> actually wrote.<br /> <br /> rds_ib_conn_info_visitor() and rds6_ib_conn_info_visitor() only<br /> write a subset of their output struct when the underlying<br /> rds_connection is not in state RDS_CONN_UP (src/dst addr, tos, sl<br /> and the two GIDs via explicit memsets). Several u32 fields<br /> (max_send_wr, max_recv_wr, max_send_sge, rdma_mr_max, rdma_mr_size,<br /> cache_allocs) and the 2-byte alignment hole between sl and<br /> cache_allocs remain as whatever stack contents preceded the visitor<br /> call and are then memcpy_to_user()&amp;#39;d out to user space.<br /> <br /> struct rds_info_rdma_connection and struct rds6_info_rdma_connection<br /> are the only rds_info_* structs in include/uapi/linux/rds.h that are<br /> not marked __attribute__((packed)), so they have a real alignment<br /> hole. The other info visitors (rds_conn_info_visitor,<br /> rds6_conn_info_visitor, rds_tcp_tc_info, ...) write all fields of<br /> their packed output struct today and are not known to be vulnerable,<br /> but a future visitor that adds a conditional write-path would have<br /> the same bug.<br /> <br /> Reproduction on a kernel built without CONFIG_INIT_STACK_ALL_ZERO=y:<br /> a local unprivileged user opens AF_RDS, sets SO_RDS_TRANSPORT=IB,<br /> binds to a local address on an RDMA-capable netdev (rxe soft-RoCE on<br /> any netdev is sufficient), sendto()&amp;#39;s any peer on the same subnet<br /> (fails cleanly but installs an rds_connection in the global hash in<br /> RDS_CONN_CONNECTING), then calls getsockopt(SOL_RDS,<br /> RDS_INFO_IB_CONNECTIONS). The returned 68-byte item contains 26<br /> bytes of stack garbage including kernel text/data pointers:<br /> <br /> 0..7 0a 63 00 01 0a 63 00 02 src=10.99.0.1 dst=10.99.0.2<br /> 8..39 00 ... gids (memset-zeroed)<br /> 40..47 e0 92 a3 81 ff ff ff ff kernel pointer (max_send_wr)<br /> 48..55 7f 37 b5 81 ff ff ff ff kernel pointer (rdma_mr_max)<br /> 56..59 01 00 08 00 rdma_mr_size (garbage)<br /> 60..61 00 00 tos, sl<br /> 62..63 00 00 alignment padding<br /> 64..67 18 00 00 00 cache_allocs (garbage)<br /> <br /> Fix by zeroing the per-item buffer in both rds_for_each_conn_info()<br /> and rds_walk_conn_path_info() before invoking the visitor. This<br /> covers the IPv4/IPv6 IB visitors and hardens all current and future<br /> visitors against the same class of bug.<br /> <br /> No functional change for visitors that fully populate their output.<br /> <br /> Changes in v2:<br /> - retarget at the net tree (subject prefix "[PATCH net v2]",<br /> net/rds: prefix in the title)<br /> - pick up Reviewed-by tags from Sharath Srinivasan and<br /> Allison Henderson