RDMA/addr: Be strict with gid size

[ Upstream commit d1c803a ]

The nla_len() is less than or equal to 16.  If it's less than 16 then end
of the "gid" buffer is uninitialized.

Fixes: ae43f82 ("IB/core: Add IP to GID netlink offload")
Link: https://lore.kernel.org/r/20210405074434.264221-1-leon@kernel.org
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

drivers/infiniband/core/addr.c

@@ -76,7 +76,9 @@ static struct workqueue_struct *addr_wq;
 
 static const struct nla_policy ib_nl_addr_policy[LS_NLA_TYPE_MAX] = {
 	[LS_NLA_TYPE_DGID] = {.type = NLA_BINARY,
-		.len = sizeof(struct rdma_nla_ls_gid)},
+		.len = sizeof(struct rdma_nla_ls_gid),
+		.validation_type = NLA_VALIDATE_MIN,
+		.min = sizeof(struct rdma_nla_ls_gid)},
 };
 
 static inline bool ib_nl_is_good_ip_resp(const struct nlmsghdr *nlh)