1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Handle bridge arp/nd proxy/suppress 4 * 5 * Copyright (C) 2017 Cumulus Networks 6 * Copyright (c) 2017 Roopa Prabhu <roopa@cumulusnetworks.com> 7 * 8 * Authors: 9 * Roopa Prabhu <roopa@cumulusnetworks.com> 10 */ 11 12 #include <linux/kernel.h> 13 #include <linux/netdevice.h> 14 #include <linux/etherdevice.h> 15 #include <linux/neighbour.h> 16 #include <net/arp.h> 17 #include <linux/if_vlan.h> 18 #include <linux/inetdevice.h> 19 #include <net/addrconf.h> 20 #if IS_ENABLED(CONFIG_IPV6) 21 #include <net/ip6_checksum.h> 22 #include <net/ndisc.h> 23 #endif 24 25 #include "br_private.h" 26 27 void br_recalculate_neigh_suppress_enabled(struct net_bridge *br) 28 { 29 struct net_bridge_port *p; 30 bool neigh_suppress = false; 31 32 list_for_each_entry(p, &br->port_list, list) { 33 if (READ_ONCE(p->flags) & (BR_NEIGH_SUPPRESS | BR_NEIGH_VLAN_SUPPRESS)) { 34 neigh_suppress = true; 35 break; 36 } 37 } 38 39 br_opt_toggle(br, BROPT_NEIGH_SUPPRESS_ENABLED, neigh_suppress); 40 } 41 42 #if IS_ENABLED(CONFIG_INET) 43 static void br_arp_send(struct net_bridge *br, struct net_bridge_port *p, 44 struct net_device *dev, __be32 dest_ip, __be32 src_ip, 45 const unsigned char *dest_hw, 46 const unsigned char *src_hw, 47 const unsigned char *target_hw, 48 __be16 vlan_proto, u16 vlan_tci) 49 { 50 struct net_bridge_vlan_group *vg; 51 struct sk_buff *skb; 52 u16 pvid; 53 54 netdev_dbg(dev, "arp send dev %s dst %pI4 dst_hw %pM src %pI4 src_hw %pM\n", 55 dev->name, &dest_ip, dest_hw, &src_ip, src_hw); 56 57 if (!vlan_tci) { 58 arp_send(ARPOP_REPLY, ETH_P_ARP, dest_ip, dev, src_ip, 59 dest_hw, src_hw, target_hw); 60 return; 61 } 62 63 skb = arp_create(ARPOP_REPLY, ETH_P_ARP, dest_ip, dev, src_ip, 64 dest_hw, src_hw, target_hw); 65 if (!skb) 66 return; 67 68 if (p) 69 vg = nbp_vlan_group_rcu(p); 70 else 71 vg = br_vlan_group_rcu(br); 72 pvid = br_get_pvid(vg); 73 if (pvid == (vlan_tci & VLAN_VID_MASK)) 74 vlan_tci = 0; 75 76 if (vlan_tci) 77 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci); 78 79 if (p) { 80 arp_xmit(skb); 81 } else { 82 skb_reset_mac_header(skb); 83 __skb_pull(skb, skb_network_offset(skb)); 84 skb->ip_summed = CHECKSUM_UNNECESSARY; 85 skb->pkt_type = PACKET_HOST; 86 87 netif_rx(skb); 88 } 89 } 90 91 static int br_chk_addr_ip(struct net_device *dev, 92 struct netdev_nested_priv *priv) 93 { 94 __be32 ip = *(__be32 *)priv->data; 95 struct in_device *in_dev; 96 __be32 addr = 0; 97 98 in_dev = __in_dev_get_rcu(dev); 99 if (in_dev) 100 addr = inet_confirm_addr(dev_net(dev), in_dev, 0, ip, 101 RT_SCOPE_HOST); 102 103 if (addr == ip) 104 return 1; 105 106 return 0; 107 } 108 109 static bool br_is_local_ip(struct net_device *dev, __be32 ip) 110 { 111 struct netdev_nested_priv priv = { 112 .data = (void *)&ip, 113 }; 114 115 if (br_chk_addr_ip(dev, &priv)) 116 return true; 117 118 /* check if ip is configured on upper dev */ 119 if (netdev_walk_all_upper_dev_rcu(dev, br_chk_addr_ip, &priv)) 120 return true; 121 122 return false; 123 } 124 125 void br_do_proxy_suppress_arp(struct sk_buff *skb, struct net_bridge *br, 126 u16 vid, struct net_bridge_port *p) 127 { 128 struct net_device *dev = br->dev; 129 struct net_device *vlandev = dev; 130 struct neighbour *n; 131 struct arphdr *parp; 132 u8 *arpptr, *sha; 133 __be32 sip, tip; 134 135 BR_INPUT_SKB_CB(skb)->proxyarp_replied = 0; 136 BR_INPUT_SKB_CB(skb)->grat_arp = 0; 137 138 if ((dev->flags & IFF_NOARP) || 139 !pskb_may_pull(skb, arp_hdr_len(dev))) 140 return; 141 142 parp = arp_hdr(skb); 143 144 if (parp->ar_pro != htons(ETH_P_IP) || 145 parp->ar_hln != dev->addr_len || 146 parp->ar_pln != 4) 147 return; 148 149 arpptr = (u8 *)parp + sizeof(struct arphdr); 150 sha = arpptr; 151 arpptr += dev->addr_len; /* sha */ 152 memcpy(&sip, arpptr, sizeof(sip)); 153 arpptr += sizeof(sip); 154 arpptr += dev->addr_len; /* tha */ 155 memcpy(&tip, arpptr, sizeof(tip)); 156 157 if (ipv4_is_loopback(tip) || 158 ipv4_is_multicast(tip)) 159 return; 160 161 if (br_opt_get(br, BROPT_NEIGH_SUPPRESS_ENABLED)) { 162 if (br_is_neigh_suppress_enabled(p, vid)) 163 return; 164 if (is_unicast_ether_addr(eth_hdr(skb)->h_dest) && 165 parp->ar_op == htons(ARPOP_REQUEST)) 166 return; 167 if (parp->ar_op != htons(ARPOP_RREQUEST) && 168 parp->ar_op != htons(ARPOP_RREPLY) && 169 sip == tip) { 170 /* prevent flooding to neigh suppress ports */ 171 BR_INPUT_SKB_CB(skb)->proxyarp_replied = 1; 172 BR_INPUT_SKB_CB(skb)->grat_arp = 1; 173 return; 174 } 175 } 176 177 if (parp->ar_op != htons(ARPOP_REQUEST)) 178 return; 179 180 if (vid != 0) { 181 vlandev = __vlan_find_dev_deep_rcu(br->dev, skb->vlan_proto, 182 vid); 183 if (!vlandev) 184 return; 185 } 186 187 if (br_opt_get(br, BROPT_NEIGH_SUPPRESS_ENABLED) && 188 br_is_local_ip(vlandev, tip)) { 189 /* its our local ip, so don't proxy reply 190 * and don't forward to neigh suppress ports 191 */ 192 BR_INPUT_SKB_CB(skb)->proxyarp_replied = 1; 193 return; 194 } 195 196 n = neigh_lookup(&arp_tbl, &tip, vlandev); 197 if (n) { 198 struct net_bridge_fdb_entry *f; 199 200 if (!(READ_ONCE(n->nud_state) & NUD_VALID)) { 201 neigh_release(n); 202 return; 203 } 204 205 f = br_fdb_find_rcu(br, n->ha, vid); 206 if (f) { 207 const struct net_bridge_port *dst = READ_ONCE(f->dst); 208 bool replied = false; 209 210 if ((p && test_bit(BR_PROXYARP_BIT, &p->flags)) || 211 (dst && test_bit(BR_PROXYARP_WIFI_BIT, &dst->flags)) || 212 br_is_neigh_suppress_enabled(dst, vid)) { 213 if (!vid) 214 br_arp_send(br, p, skb->dev, sip, tip, 215 sha, n->ha, sha, 0, 0); 216 else 217 br_arp_send(br, p, skb->dev, sip, tip, 218 sha, n->ha, sha, 219 skb->vlan_proto, 220 skb_vlan_tag_get(skb)); 221 replied = true; 222 } 223 224 /* If we have replied or as long as we know the 225 * mac, indicate to arp replied 226 */ 227 if (replied || 228 br_opt_get(br, BROPT_NEIGH_SUPPRESS_ENABLED)) 229 BR_INPUT_SKB_CB(skb)->proxyarp_replied = 1; 230 } 231 232 neigh_release(n); 233 } 234 } 235 #endif 236 237 #if IS_ENABLED(CONFIG_IPV6) 238 /* Validate skb as an NS/NA and linearize it for br_nd_send()'s ND 239 * option parsing; returns the nd_msg, or NULL on failure. 240 */ 241 struct nd_msg *br_is_nd_neigh_msg(struct sk_buff *skb) 242 { 243 if (ndisc_check_ns_na(skb)) 244 return NULL; 245 246 if (skb_linearize(skb)) 247 return NULL; 248 249 return (struct nd_msg *)skb_transport_header(skb); 250 } 251 252 static void br_nd_send(struct net_bridge *br, struct net_bridge_port *p, 253 struct sk_buff *request, struct neighbour *n, 254 __be16 vlan_proto, u16 vlan_tci) 255 { 256 struct net_device *dev = request->dev; 257 struct net_bridge_vlan_group *vg; 258 struct ndisc_options ndopts; 259 struct nd_msg *na, *ns; 260 struct sk_buff *reply; 261 struct ipv6hdr *pip6; 262 int na_olen = 8; /* opt hdr + ETH_ALEN for target */ 263 int ns_olen; 264 u8 *daddr; 265 bool dad; 266 u16 pvid; 267 int len; 268 269 if (!dev) 270 return; 271 272 len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) + 273 sizeof(*na) + na_olen + dev->needed_tailroom; 274 275 reply = alloc_skb(len, GFP_ATOMIC); 276 if (!reply) 277 return; 278 279 reply->protocol = htons(ETH_P_IPV6); 280 reply->dev = dev; 281 skb_reserve(reply, LL_RESERVED_SPACE(dev)); 282 skb_push(reply, sizeof(struct ethhdr)); 283 skb_set_mac_header(reply, 0); 284 285 daddr = eth_hdr(request)->h_source; 286 ns = (struct nd_msg *)skb_transport_header(request); 287 288 /* Derive the option length from the IPv6 payload length so that any 289 * trailing L2 padding in the skb is not parsed as ND options. 290 */ 291 ns_olen = ntohs(ipv6_hdr(request)->payload_len) - sizeof(*ns); 292 if (!ndisc_parse_options(dev, ns->opt, ns_olen, &ndopts)) { 293 kfree_skb(reply); 294 return; 295 } 296 297 if (ndopts.nd_opts_src_lladdr) { 298 u8 *lladdr; 299 300 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_src_lladdr, dev); 301 if (lladdr) 302 daddr = lladdr; 303 } 304 305 dad = ipv6_addr_any(&ipv6_hdr(request)->saddr); 306 307 /* Ethernet header */ 308 if (dad) 309 ipv6_eth_mc_map(&in6addr_linklocal_allnodes, eth_hdr(reply)->h_dest); 310 else 311 ether_addr_copy(eth_hdr(reply)->h_dest, daddr); 312 ether_addr_copy(eth_hdr(reply)->h_source, n->ha); 313 eth_hdr(reply)->h_proto = htons(ETH_P_IPV6); 314 reply->protocol = htons(ETH_P_IPV6); 315 316 skb_pull(reply, sizeof(struct ethhdr)); 317 skb_set_network_header(reply, 0); 318 skb_put(reply, sizeof(struct ipv6hdr)); 319 320 /* IPv6 header */ 321 pip6 = ipv6_hdr(reply); 322 memset(pip6, 0, sizeof(struct ipv6hdr)); 323 pip6->version = 6; 324 pip6->priority = ipv6_hdr(request)->priority; 325 pip6->nexthdr = IPPROTO_ICMPV6; 326 pip6->hop_limit = 255; 327 pip6->daddr = dad ? in6addr_linklocal_allnodes : ipv6_hdr(request)->saddr; 328 pip6->saddr = *(struct in6_addr *)n->primary_key; 329 330 skb_pull(reply, sizeof(struct ipv6hdr)); 331 skb_set_transport_header(reply, 0); 332 333 na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen); 334 335 /* Neighbor Advertisement */ 336 memset(na, 0, sizeof(*na) + na_olen); 337 na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT; 338 na->icmph.icmp6_router = (n->flags & NTF_ROUTER) ? 1 : 0; 339 na->icmph.icmp6_override = 1; 340 na->icmph.icmp6_solicited = dad ? 0 : 1; 341 na->target = ns->target; 342 ether_addr_copy(&na->opt[2], n->ha); 343 na->opt[0] = ND_OPT_TARGET_LL_ADDR; 344 na->opt[1] = na_olen >> 3; 345 346 na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr, 347 &pip6->daddr, 348 sizeof(*na) + na_olen, 349 IPPROTO_ICMPV6, 350 csum_partial(na, sizeof(*na) + na_olen, 0)); 351 352 pip6->payload_len = htons(sizeof(*na) + na_olen); 353 354 skb_push(reply, sizeof(struct ipv6hdr)); 355 skb_push(reply, sizeof(struct ethhdr)); 356 357 reply->ip_summed = CHECKSUM_UNNECESSARY; 358 359 if (p) 360 vg = nbp_vlan_group_rcu(p); 361 else 362 vg = br_vlan_group_rcu(br); 363 pvid = br_get_pvid(vg); 364 if (pvid == (vlan_tci & VLAN_VID_MASK)) 365 vlan_tci = 0; 366 367 if (vlan_tci) 368 __vlan_hwaccel_put_tag(reply, vlan_proto, vlan_tci); 369 370 netdev_dbg(dev, "nd send dev %s dst %pI6 dst_hw %pM src %pI6 src_hw %pM\n", 371 dev->name, &pip6->daddr, daddr, &pip6->saddr, n->ha); 372 373 if (p) { 374 dev_queue_xmit(reply); 375 } else { 376 skb_reset_mac_header(reply); 377 __skb_pull(reply, skb_network_offset(reply)); 378 reply->ip_summed = CHECKSUM_UNNECESSARY; 379 reply->pkt_type = PACKET_HOST; 380 381 netif_rx(reply); 382 } 383 } 384 385 static int br_chk_addr_ip6(struct net_device *dev, 386 struct netdev_nested_priv *priv) 387 { 388 struct in6_addr *addr = (struct in6_addr *)priv->data; 389 390 if (ipv6_chk_addr(dev_net(dev), addr, dev, 0)) 391 return 1; 392 393 return 0; 394 } 395 396 static bool br_is_local_ip6(struct net_device *dev, struct in6_addr *addr) 397 398 { 399 struct netdev_nested_priv priv = { 400 .data = (void *)addr, 401 }; 402 403 if (br_chk_addr_ip6(dev, &priv)) 404 return true; 405 406 /* check if ip is configured on upper dev */ 407 if (netdev_walk_all_upper_dev_rcu(dev, br_chk_addr_ip6, &priv)) 408 return true; 409 410 return false; 411 } 412 413 void br_do_suppress_nd(struct sk_buff *skb, struct net_bridge *br, 414 u16 vid, struct net_bridge_port *p, struct nd_msg *msg) 415 { 416 struct net_device *dev = br->dev; 417 struct net_device *vlandev = NULL; 418 struct in6_addr *saddr, *daddr; 419 struct ipv6hdr *iphdr; 420 struct neighbour *n; 421 422 BR_INPUT_SKB_CB(skb)->proxyarp_replied = 0; 423 BR_INPUT_SKB_CB(skb)->grat_arp = 0; 424 425 if (br_is_neigh_suppress_enabled(p, vid)) 426 return; 427 428 if (is_unicast_ether_addr(eth_hdr(skb)->h_dest) && 429 msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) 430 return; 431 432 if (msg->icmph.icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT && 433 !msg->icmph.icmp6_solicited) { 434 /* prevent flooding to neigh suppress ports */ 435 BR_INPUT_SKB_CB(skb)->proxyarp_replied = 1; 436 BR_INPUT_SKB_CB(skb)->grat_arp = 1; 437 return; 438 } 439 440 if (msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION) 441 return; 442 443 iphdr = ipv6_hdr(skb); 444 saddr = &iphdr->saddr; 445 daddr = &iphdr->daddr; 446 447 if (!ipv6_addr_cmp(saddr, daddr)) { 448 /* prevent flooding to neigh suppress ports */ 449 BR_INPUT_SKB_CB(skb)->proxyarp_replied = 1; 450 return; 451 } 452 453 if (vid != 0) { 454 /* build neigh table lookup on the vlan device */ 455 vlandev = __vlan_find_dev_deep_rcu(br->dev, skb->vlan_proto, 456 vid); 457 if (!vlandev) 458 return; 459 } else { 460 vlandev = dev; 461 } 462 463 if (br_is_local_ip6(vlandev, &msg->target)) { 464 /* its our own ip, so don't proxy reply 465 * and don't forward to arp suppress ports 466 */ 467 BR_INPUT_SKB_CB(skb)->proxyarp_replied = 1; 468 return; 469 } 470 471 n = neigh_lookup(&nd_tbl, &msg->target, vlandev); 472 if (n) { 473 struct net_bridge_fdb_entry *f; 474 475 if (!(READ_ONCE(n->nud_state) & NUD_VALID)) { 476 neigh_release(n); 477 return; 478 } 479 480 f = br_fdb_find_rcu(br, n->ha, vid); 481 if (f) { 482 const struct net_bridge_port *dst = READ_ONCE(f->dst); 483 bool replied = false; 484 485 if (br_is_neigh_suppress_enabled(dst, vid)) { 486 if (vid != 0) 487 br_nd_send(br, p, skb, n, 488 skb->vlan_proto, 489 skb_vlan_tag_get(skb)); 490 else 491 br_nd_send(br, p, skb, n, 0, 0); 492 replied = true; 493 } 494 495 /* If we have replied or as long as we know the 496 * mac, indicate to NEIGH_SUPPRESS ports that we 497 * have replied 498 */ 499 if (replied || 500 br_opt_get(br, BROPT_NEIGH_SUPPRESS_ENABLED)) 501 BR_INPUT_SKB_CB(skb)->proxyarp_replied = 1; 502 } 503 neigh_release(n); 504 } 505 } 506 #endif 507 508 bool br_is_neigh_suppress_enabled(const struct net_bridge_port *p, u16 vid) 509 { 510 if (!p) 511 return false; 512 513 if (vid && test_bit(BR_NEIGH_VLAN_SUPPRESS_BIT, &p->flags)) { 514 struct net_bridge_vlan_group *vg = nbp_vlan_group_rcu(p); 515 struct net_bridge_vlan *v; 516 517 v = br_vlan_find(vg, vid); 518 if (!v) 519 return false; 520 return !!(v->priv_flags & BR_VLFLAG_NEIGH_SUPPRESS_ENABLED); 521 } 522 return test_bit(BR_NEIGH_SUPPRESS_BIT, &p->flags); 523 } 524 525 bool br_is_neigh_forward_grat_enabled(const struct net_bridge_port *p, u16 vid) 526 { 527 if (vid && test_bit(BR_NEIGH_VLAN_SUPPRESS_BIT, &p->flags)) { 528 struct net_bridge_vlan_group *vg = nbp_vlan_group_rcu(p); 529 struct net_bridge_vlan *v; 530 531 v = br_vlan_find(vg, vid); 532 if (!v) 533 return false; 534 return !!(v->priv_flags & BR_VLFLAG_NEIGH_FORWARD_GRAT_ENABLED); 535 } 536 return test_bit(BR_NEIGH_FORWARD_GRAT_BIT, &p->flags); 537 } 538