1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (c) 2007-2014 Nicira, Inc. 4 */ 5 6 #include <linux/uaccess.h> 7 #include <linux/netdevice.h> 8 #include <linux/etherdevice.h> 9 #include <linux/if_ether.h> 10 #include <linux/if_vlan.h> 11 #include <net/llc_pdu.h> 12 #include <linux/kernel.h> 13 #include <linux/jhash.h> 14 #include <linux/jiffies.h> 15 #include <linux/llc.h> 16 #include <linux/module.h> 17 #include <linux/in.h> 18 #include <linux/rcupdate.h> 19 #include <linux/cpumask.h> 20 #include <linux/if_arp.h> 21 #include <linux/ip.h> 22 #include <linux/ipv6.h> 23 #include <linux/mpls.h> 24 #include <linux/sctp.h> 25 #include <linux/smp.h> 26 #include <linux/tcp.h> 27 #include <linux/udp.h> 28 #include <linux/icmp.h> 29 #include <linux/icmpv6.h> 30 #include <linux/rculist.h> 31 #include <net/ip.h> 32 #include <net/ip_tunnels.h> 33 #include <net/ipv6.h> 34 #include <net/mpls.h> 35 #include <net/ndisc.h> 36 #include <net/nsh.h> 37 #include <net/pkt_cls.h> 38 #include <net/netfilter/nf_conntrack_zones.h> 39 40 #include "conntrack.h" 41 #include "datapath.h" 42 #include "flow.h" 43 #include "flow_netlink.h" 44 #include "vport.h" 45 46 u64 ovs_flow_used_time(unsigned long flow_jiffies) 47 { 48 struct timespec64 cur_ts; 49 u64 cur_ms, idle_ms; 50 51 ktime_get_ts64(&cur_ts); 52 idle_ms = jiffies_to_msecs(jiffies - flow_jiffies); 53 cur_ms = (u64)(u32)cur_ts.tv_sec * MSEC_PER_SEC + 54 cur_ts.tv_nsec / NSEC_PER_MSEC; 55 56 return cur_ms - idle_ms; 57 } 58 59 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF)) 60 61 void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags, 62 const struct sk_buff *skb) 63 { 64 struct sw_flow_stats *stats; 65 unsigned int cpu = smp_processor_id(); 66 int len = skb->len + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0); 67 68 stats = rcu_dereference(flow->stats[cpu]); 69 70 /* Check if already have CPU-specific stats. */ 71 if (likely(stats)) { 72 spin_lock(&stats->lock); 73 /* Mark if we write on the pre-allocated stats. */ 74 if (cpu == 0 && unlikely(flow->stats_last_writer != cpu)) 75 flow->stats_last_writer = cpu; 76 } else { 77 stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */ 78 spin_lock(&stats->lock); 79 80 /* If the current CPU is the only writer on the 81 * pre-allocated stats keep using them. 82 */ 83 if (unlikely(flow->stats_last_writer != cpu)) { 84 /* A previous locker may have already allocated the 85 * stats, so we need to check again. If CPU-specific 86 * stats were already allocated, we update the pre- 87 * allocated stats as we have already locked them. 88 */ 89 if (likely(flow->stats_last_writer != -1) && 90 likely(!rcu_access_pointer(flow->stats[cpu]))) { 91 /* Try to allocate CPU-specific stats. */ 92 struct sw_flow_stats *new_stats; 93 94 new_stats = 95 kmem_cache_alloc_node(flow_stats_cache, 96 GFP_NOWAIT | 97 __GFP_THISNODE | 98 __GFP_NOWARN | 99 __GFP_NOMEMALLOC, 100 numa_node_id()); 101 if (likely(new_stats)) { 102 new_stats->used = jiffies; 103 new_stats->packet_count = 1; 104 new_stats->byte_count = len; 105 new_stats->tcp_flags = tcp_flags; 106 spin_lock_init(&new_stats->lock); 107 108 rcu_assign_pointer(flow->stats[cpu], 109 new_stats); 110 cpumask_set_cpu(cpu, &flow->cpu_used_mask); 111 goto unlock; 112 } 113 } 114 flow->stats_last_writer = cpu; 115 } 116 } 117 118 stats->used = jiffies; 119 stats->packet_count++; 120 stats->byte_count += len; 121 stats->tcp_flags |= tcp_flags; 122 unlock: 123 spin_unlock(&stats->lock); 124 } 125 126 /* Must be called with rcu_read_lock or ovs_mutex. */ 127 void ovs_flow_stats_get(const struct sw_flow *flow, 128 struct ovs_flow_stats *ovs_stats, 129 unsigned long *used, __be16 *tcp_flags) 130 { 131 int cpu; 132 133 *used = 0; 134 *tcp_flags = 0; 135 memset(ovs_stats, 0, sizeof(*ovs_stats)); 136 137 /* We open code this to make sure cpu 0 is always considered */ 138 for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, &flow->cpu_used_mask)) { 139 struct sw_flow_stats *stats = rcu_dereference_ovsl(flow->stats[cpu]); 140 141 if (stats) { 142 /* Local CPU may write on non-local stats, so we must 143 * block bottom-halves here. 144 */ 145 spin_lock_bh(&stats->lock); 146 if (!*used || time_after(stats->used, *used)) 147 *used = stats->used; 148 *tcp_flags |= stats->tcp_flags; 149 ovs_stats->n_packets += stats->packet_count; 150 ovs_stats->n_bytes += stats->byte_count; 151 spin_unlock_bh(&stats->lock); 152 } 153 } 154 } 155 156 /* Called with ovs_mutex. */ 157 void ovs_flow_stats_clear(struct sw_flow *flow) 158 { 159 int cpu; 160 161 /* We open code this to make sure cpu 0 is always considered */ 162 for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, &flow->cpu_used_mask)) { 163 struct sw_flow_stats *stats = ovsl_dereference(flow->stats[cpu]); 164 165 if (stats) { 166 spin_lock_bh(&stats->lock); 167 stats->used = 0; 168 stats->packet_count = 0; 169 stats->byte_count = 0; 170 stats->tcp_flags = 0; 171 spin_unlock_bh(&stats->lock); 172 } 173 } 174 } 175 176 static int check_header(struct sk_buff *skb, int len) 177 { 178 if (unlikely(skb->len < len)) 179 return -EINVAL; 180 if (unlikely(!pskb_may_pull(skb, len))) 181 return -ENOMEM; 182 return 0; 183 } 184 185 static bool arphdr_ok(struct sk_buff *skb) 186 { 187 return pskb_may_pull(skb, skb_network_offset(skb) + 188 sizeof(struct arp_eth_header)); 189 } 190 191 static int check_iphdr(struct sk_buff *skb) 192 { 193 unsigned int nh_ofs = skb_network_offset(skb); 194 unsigned int ip_len; 195 int err; 196 197 err = check_header(skb, nh_ofs + sizeof(struct iphdr)); 198 if (unlikely(err)) 199 return err; 200 201 ip_len = ip_hdrlen(skb); 202 if (unlikely(ip_len < sizeof(struct iphdr) || 203 skb->len < nh_ofs + ip_len)) 204 return -EINVAL; 205 206 skb_set_transport_header(skb, nh_ofs + ip_len); 207 return 0; 208 } 209 210 static bool tcphdr_ok(struct sk_buff *skb) 211 { 212 int th_ofs = skb_transport_offset(skb); 213 int tcp_len; 214 215 if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr)))) 216 return false; 217 218 tcp_len = tcp_hdrlen(skb); 219 if (unlikely(tcp_len < sizeof(struct tcphdr) || 220 skb->len < th_ofs + tcp_len)) 221 return false; 222 223 return true; 224 } 225 226 static bool udphdr_ok(struct sk_buff *skb) 227 { 228 return pskb_may_pull(skb, skb_transport_offset(skb) + 229 sizeof(struct udphdr)); 230 } 231 232 static bool sctphdr_ok(struct sk_buff *skb) 233 { 234 return pskb_may_pull(skb, skb_transport_offset(skb) + 235 sizeof(struct sctphdr)); 236 } 237 238 static bool icmphdr_ok(struct sk_buff *skb) 239 { 240 return pskb_may_pull(skb, skb_transport_offset(skb) + 241 sizeof(struct icmphdr)); 242 } 243 244 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key) 245 { 246 unsigned short frag_off; 247 unsigned int payload_ofs = 0; 248 unsigned int nh_ofs = skb_network_offset(skb); 249 unsigned int nh_len; 250 struct ipv6hdr *nh; 251 int err, nexthdr, flags = 0; 252 253 err = check_header(skb, nh_ofs + sizeof(*nh)); 254 if (unlikely(err)) 255 return err; 256 257 nh = ipv6_hdr(skb); 258 259 key->ip.proto = NEXTHDR_NONE; 260 key->ip.tos = ipv6_get_dsfield(nh); 261 key->ip.ttl = nh->hop_limit; 262 key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL); 263 key->ipv6.addr.src = nh->saddr; 264 key->ipv6.addr.dst = nh->daddr; 265 266 nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags); 267 if (flags & IP6_FH_F_FRAG) { 268 if (frag_off) { 269 key->ip.frag = OVS_FRAG_TYPE_LATER; 270 key->ip.proto = nexthdr; 271 return 0; 272 } 273 key->ip.frag = OVS_FRAG_TYPE_FIRST; 274 } else { 275 key->ip.frag = OVS_FRAG_TYPE_NONE; 276 } 277 278 /* Delayed handling of error in ipv6_find_hdr() as it 279 * always sets flags and frag_off to a valid value which may be 280 * used to set key->ip.frag above. 281 */ 282 if (unlikely(nexthdr < 0)) 283 return -EPROTO; 284 285 nh_len = payload_ofs - nh_ofs; 286 skb_set_transport_header(skb, nh_ofs + nh_len); 287 key->ip.proto = nexthdr; 288 return nh_len; 289 } 290 291 static bool icmp6hdr_ok(struct sk_buff *skb) 292 { 293 return pskb_may_pull(skb, skb_transport_offset(skb) + 294 sizeof(struct icmp6hdr)); 295 } 296 297 /** 298 * parse_vlan_tag - Parse vlan tag from vlan header. 299 * @skb: skb containing frame to parse 300 * @key_vh: pointer to parsed vlan tag 301 * @untag_vlan: should the vlan header be removed from the frame 302 * 303 * Return: ERROR on memory error. 304 * %0 if it encounters a non-vlan or incomplete packet. 305 * %1 after successfully parsing vlan tag. 306 */ 307 static int parse_vlan_tag(struct sk_buff *skb, struct vlan_head *key_vh, 308 bool untag_vlan) 309 { 310 struct vlan_head *vh = (struct vlan_head *)skb->data; 311 312 if (likely(!eth_type_vlan(vh->tpid))) 313 return 0; 314 315 if (unlikely(skb->len < sizeof(struct vlan_head) + sizeof(__be16))) 316 return 0; 317 318 if (unlikely(!pskb_may_pull(skb, sizeof(struct vlan_head) + 319 sizeof(__be16)))) 320 return -ENOMEM; 321 322 vh = (struct vlan_head *)skb->data; 323 key_vh->tci = vh->tci | htons(VLAN_CFI_MASK); 324 key_vh->tpid = vh->tpid; 325 326 if (unlikely(untag_vlan)) { 327 int offset = skb->data - skb_mac_header(skb); 328 u16 tci; 329 int err; 330 331 __skb_push(skb, offset); 332 err = __skb_vlan_pop(skb, &tci); 333 __skb_pull(skb, offset); 334 if (err) 335 return err; 336 __vlan_hwaccel_put_tag(skb, key_vh->tpid, tci); 337 } else { 338 __skb_pull(skb, sizeof(struct vlan_head)); 339 } 340 return 1; 341 } 342 343 static void clear_vlan(struct sw_flow_key *key) 344 { 345 key->eth.vlan.tci = 0; 346 key->eth.vlan.tpid = 0; 347 key->eth.cvlan.tci = 0; 348 key->eth.cvlan.tpid = 0; 349 } 350 351 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key) 352 { 353 int res; 354 355 if (skb_vlan_tag_present(skb)) { 356 key->eth.vlan.tci = htons(skb->vlan_tci) | htons(VLAN_CFI_MASK); 357 key->eth.vlan.tpid = skb->vlan_proto; 358 } else { 359 /* Parse outer vlan tag in the non-accelerated case. */ 360 res = parse_vlan_tag(skb, &key->eth.vlan, true); 361 if (res <= 0) 362 return res; 363 } 364 365 /* Parse inner vlan tag. */ 366 res = parse_vlan_tag(skb, &key->eth.cvlan, false); 367 if (res <= 0) 368 return res; 369 370 return 0; 371 } 372 373 static __be16 parse_ethertype(struct sk_buff *skb) 374 { 375 struct llc_snap_hdr { 376 u8 dsap; /* Always 0xAA */ 377 u8 ssap; /* Always 0xAA */ 378 u8 ctrl; 379 u8 oui[3]; 380 __be16 ethertype; 381 }; 382 struct llc_snap_hdr *llc; 383 __be16 proto; 384 385 proto = *(__be16 *) skb->data; 386 __skb_pull(skb, sizeof(__be16)); 387 388 if (eth_proto_is_802_3(proto)) 389 return proto; 390 391 if (skb->len < sizeof(struct llc_snap_hdr)) 392 return htons(ETH_P_802_2); 393 394 if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr)))) 395 return htons(0); 396 397 llc = (struct llc_snap_hdr *) skb->data; 398 if (llc->dsap != LLC_SAP_SNAP || 399 llc->ssap != LLC_SAP_SNAP || 400 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0) 401 return htons(ETH_P_802_2); 402 403 __skb_pull(skb, sizeof(struct llc_snap_hdr)); 404 405 if (eth_proto_is_802_3(llc->ethertype)) 406 return llc->ethertype; 407 408 return htons(ETH_P_802_2); 409 } 410 411 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key, 412 int nh_len) 413 { 414 struct icmp6hdr *icmp = icmp6_hdr(skb); 415 416 /* The ICMPv6 type and code fields use the 16-bit transport port 417 * fields, so we need to store them in 16-bit network byte order. 418 */ 419 key->tp.src = htons(icmp->icmp6_type); 420 key->tp.dst = htons(icmp->icmp6_code); 421 memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd)); 422 423 if (icmp->icmp6_code == 0 && 424 (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION || 425 icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) { 426 int icmp_len = skb->len - skb_transport_offset(skb); 427 struct nd_msg *nd; 428 int offset; 429 430 /* In order to process neighbor discovery options, we need the 431 * entire packet. 432 */ 433 if (unlikely(icmp_len < sizeof(*nd))) 434 return 0; 435 436 if (unlikely(skb_linearize(skb))) 437 return -ENOMEM; 438 439 nd = (struct nd_msg *)skb_transport_header(skb); 440 key->ipv6.nd.target = nd->target; 441 442 icmp_len -= sizeof(*nd); 443 offset = 0; 444 while (icmp_len >= 8) { 445 struct nd_opt_hdr *nd_opt = 446 (struct nd_opt_hdr *)(nd->opt + offset); 447 int opt_len = nd_opt->nd_opt_len * 8; 448 449 if (unlikely(!opt_len || opt_len > icmp_len)) 450 return 0; 451 452 /* Store the link layer address if the appropriate 453 * option is provided. It is considered an error if 454 * the same link layer option is specified twice. 455 */ 456 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR 457 && opt_len == 8) { 458 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll))) 459 goto invalid; 460 ether_addr_copy(key->ipv6.nd.sll, 461 &nd->opt[offset+sizeof(*nd_opt)]); 462 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR 463 && opt_len == 8) { 464 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll))) 465 goto invalid; 466 ether_addr_copy(key->ipv6.nd.tll, 467 &nd->opt[offset+sizeof(*nd_opt)]); 468 } 469 470 icmp_len -= opt_len; 471 offset += opt_len; 472 } 473 } 474 475 return 0; 476 477 invalid: 478 memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target)); 479 memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll)); 480 memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll)); 481 482 return 0; 483 } 484 485 static int parse_nsh(struct sk_buff *skb, struct sw_flow_key *key) 486 { 487 struct nshhdr *nh; 488 unsigned int nh_ofs = skb_network_offset(skb); 489 u8 version, length; 490 int err; 491 492 err = check_header(skb, nh_ofs + NSH_BASE_HDR_LEN); 493 if (unlikely(err)) 494 return err; 495 496 nh = nsh_hdr(skb); 497 version = nsh_get_ver(nh); 498 length = nsh_hdr_len(nh); 499 500 if (version != 0) 501 return -EINVAL; 502 503 err = check_header(skb, nh_ofs + length); 504 if (unlikely(err)) 505 return err; 506 507 nh = nsh_hdr(skb); 508 key->nsh.base.flags = nsh_get_flags(nh); 509 key->nsh.base.ttl = nsh_get_ttl(nh); 510 key->nsh.base.mdtype = nh->mdtype; 511 key->nsh.base.np = nh->np; 512 key->nsh.base.path_hdr = nh->path_hdr; 513 switch (key->nsh.base.mdtype) { 514 case NSH_M_TYPE1: 515 if (length != NSH_M_TYPE1_LEN) 516 return -EINVAL; 517 memcpy(key->nsh.context, nh->md1.context, 518 sizeof(nh->md1)); 519 break; 520 case NSH_M_TYPE2: 521 memset(key->nsh.context, 0, 522 sizeof(nh->md1)); 523 break; 524 default: 525 return -EINVAL; 526 } 527 528 return 0; 529 } 530 531 /** 532 * key_extract_l3l4 - extracts L3/L4 header information. 533 * @skb: sk_buff that contains the frame, with skb->data pointing to the 534 * L3 header 535 * @key: output flow key 536 * 537 * Return: %0 if successful, otherwise a negative errno value. 538 */ 539 static int key_extract_l3l4(struct sk_buff *skb, struct sw_flow_key *key) 540 { 541 int error; 542 543 /* Network layer. */ 544 if (key->eth.type == htons(ETH_P_IP)) { 545 struct iphdr *nh; 546 __be16 offset; 547 548 error = check_iphdr(skb); 549 if (unlikely(error)) { 550 memset(&key->ip, 0, sizeof(key->ip)); 551 memset(&key->ipv4, 0, sizeof(key->ipv4)); 552 if (error == -EINVAL) { 553 skb->transport_header = skb->network_header; 554 error = 0; 555 } 556 return error; 557 } 558 559 nh = ip_hdr(skb); 560 key->ipv4.addr.src = nh->saddr; 561 key->ipv4.addr.dst = nh->daddr; 562 563 key->ip.proto = nh->protocol; 564 key->ip.tos = nh->tos; 565 key->ip.ttl = nh->ttl; 566 567 offset = nh->frag_off & htons(IP_OFFSET); 568 if (offset) { 569 key->ip.frag = OVS_FRAG_TYPE_LATER; 570 memset(&key->tp, 0, sizeof(key->tp)); 571 return 0; 572 } 573 if (nh->frag_off & htons(IP_MF) || 574 skb_shinfo(skb)->gso_type & SKB_GSO_UDP) 575 key->ip.frag = OVS_FRAG_TYPE_FIRST; 576 else 577 key->ip.frag = OVS_FRAG_TYPE_NONE; 578 579 /* Transport layer. */ 580 if (key->ip.proto == IPPROTO_TCP) { 581 if (tcphdr_ok(skb)) { 582 struct tcphdr *tcp = tcp_hdr(skb); 583 key->tp.src = tcp->source; 584 key->tp.dst = tcp->dest; 585 key->tp.flags = TCP_FLAGS_BE16(tcp); 586 } else { 587 memset(&key->tp, 0, sizeof(key->tp)); 588 } 589 590 } else if (key->ip.proto == IPPROTO_UDP) { 591 if (udphdr_ok(skb)) { 592 struct udphdr *udp = udp_hdr(skb); 593 key->tp.src = udp->source; 594 key->tp.dst = udp->dest; 595 } else { 596 memset(&key->tp, 0, sizeof(key->tp)); 597 } 598 } else if (key->ip.proto == IPPROTO_SCTP) { 599 if (sctphdr_ok(skb)) { 600 struct sctphdr *sctp = sctp_hdr(skb); 601 key->tp.src = sctp->source; 602 key->tp.dst = sctp->dest; 603 } else { 604 memset(&key->tp, 0, sizeof(key->tp)); 605 } 606 } else if (key->ip.proto == IPPROTO_ICMP) { 607 if (icmphdr_ok(skb)) { 608 struct icmphdr *icmp = icmp_hdr(skb); 609 /* The ICMP type and code fields use the 16-bit 610 * transport port fields, so we need to store 611 * them in 16-bit network byte order. */ 612 key->tp.src = htons(icmp->type); 613 key->tp.dst = htons(icmp->code); 614 } else { 615 memset(&key->tp, 0, sizeof(key->tp)); 616 } 617 } 618 619 } else if (key->eth.type == htons(ETH_P_ARP) || 620 key->eth.type == htons(ETH_P_RARP)) { 621 struct arp_eth_header *arp; 622 bool arp_available = arphdr_ok(skb); 623 624 arp = (struct arp_eth_header *)skb_network_header(skb); 625 626 if (arp_available && 627 arp->ar_hrd == htons(ARPHRD_ETHER) && 628 arp->ar_pro == htons(ETH_P_IP) && 629 arp->ar_hln == ETH_ALEN && 630 arp->ar_pln == 4) { 631 632 /* We only match on the lower 8 bits of the opcode. */ 633 if (ntohs(arp->ar_op) <= 0xff) 634 key->ip.proto = ntohs(arp->ar_op); 635 else 636 key->ip.proto = 0; 637 638 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src)); 639 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst)); 640 ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha); 641 ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha); 642 } else { 643 memset(&key->ip, 0, sizeof(key->ip)); 644 memset(&key->ipv4, 0, sizeof(key->ipv4)); 645 } 646 } else if (eth_p_mpls(key->eth.type)) { 647 u8 label_count = 1; 648 649 memset(&key->mpls, 0, sizeof(key->mpls)); 650 skb_set_inner_network_header(skb, skb->mac_len); 651 while (1) { 652 __be32 lse; 653 654 error = check_header(skb, skb->mac_len + 655 label_count * MPLS_HLEN); 656 if (unlikely(error)) 657 return 0; 658 659 memcpy(&lse, skb_inner_network_header(skb), MPLS_HLEN); 660 661 if (label_count <= MPLS_LABEL_DEPTH) 662 memcpy(&key->mpls.lse[label_count - 1], &lse, 663 MPLS_HLEN); 664 665 skb_set_inner_network_header(skb, skb->mac_len + 666 label_count * MPLS_HLEN); 667 if (lse & htonl(MPLS_LS_S_MASK)) 668 break; 669 670 label_count++; 671 } 672 if (label_count > MPLS_LABEL_DEPTH) 673 label_count = MPLS_LABEL_DEPTH; 674 675 key->mpls.num_labels_mask = GENMASK(label_count - 1, 0); 676 } else if (key->eth.type == htons(ETH_P_IPV6)) { 677 int nh_len; /* IPv6 Header + Extensions */ 678 679 nh_len = parse_ipv6hdr(skb, key); 680 if (unlikely(nh_len < 0)) { 681 switch (nh_len) { 682 case -EINVAL: 683 memset(&key->ip, 0, sizeof(key->ip)); 684 memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr)); 685 fallthrough; 686 case -EPROTO: 687 skb->transport_header = skb->network_header; 688 error = 0; 689 break; 690 default: 691 error = nh_len; 692 } 693 return error; 694 } 695 696 if (key->ip.frag == OVS_FRAG_TYPE_LATER) { 697 memset(&key->tp, 0, sizeof(key->tp)); 698 return 0; 699 } 700 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP) 701 key->ip.frag = OVS_FRAG_TYPE_FIRST; 702 703 /* Transport layer. */ 704 if (key->ip.proto == NEXTHDR_TCP) { 705 if (tcphdr_ok(skb)) { 706 struct tcphdr *tcp = tcp_hdr(skb); 707 key->tp.src = tcp->source; 708 key->tp.dst = tcp->dest; 709 key->tp.flags = TCP_FLAGS_BE16(tcp); 710 } else { 711 memset(&key->tp, 0, sizeof(key->tp)); 712 } 713 } else if (key->ip.proto == NEXTHDR_UDP) { 714 if (udphdr_ok(skb)) { 715 struct udphdr *udp = udp_hdr(skb); 716 key->tp.src = udp->source; 717 key->tp.dst = udp->dest; 718 } else { 719 memset(&key->tp, 0, sizeof(key->tp)); 720 } 721 } else if (key->ip.proto == NEXTHDR_SCTP) { 722 if (sctphdr_ok(skb)) { 723 struct sctphdr *sctp = sctp_hdr(skb); 724 key->tp.src = sctp->source; 725 key->tp.dst = sctp->dest; 726 } else { 727 memset(&key->tp, 0, sizeof(key->tp)); 728 } 729 } else if (key->ip.proto == NEXTHDR_ICMP) { 730 if (icmp6hdr_ok(skb)) { 731 error = parse_icmpv6(skb, key, nh_len); 732 if (error) 733 return error; 734 } else { 735 memset(&key->tp, 0, sizeof(key->tp)); 736 } 737 } 738 } else if (key->eth.type == htons(ETH_P_NSH)) { 739 error = parse_nsh(skb, key); 740 if (error) 741 return error; 742 } 743 return 0; 744 } 745 746 /** 747 * key_extract - extracts a flow key from an Ethernet frame. 748 * @skb: sk_buff that contains the frame, with skb->data pointing to the 749 * Ethernet header 750 * @key: output flow key 751 * 752 * The caller must ensure that skb->len >= ETH_HLEN. 753 * 754 * Initializes @skb header fields as follows: 755 * 756 * - skb->mac_header: the L2 header. 757 * 758 * - skb->network_header: just past the L2 header, or just past the 759 * VLAN header, to the first byte of the L2 payload. 760 * 761 * - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6 762 * on output, then just past the IP header, if one is present and 763 * of a correct length, otherwise the same as skb->network_header. 764 * For other key->eth.type values it is left untouched. 765 * 766 * - skb->protocol: the type of the data starting at skb->network_header. 767 * Equals to key->eth.type. 768 * 769 * Return: %0 if successful, otherwise a negative errno value. 770 */ 771 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key) 772 { 773 struct ethhdr *eth; 774 775 /* Flags are always used as part of stats */ 776 key->tp.flags = 0; 777 778 skb_reset_mac_header(skb); 779 780 /* Link layer. */ 781 clear_vlan(key); 782 if (ovs_key_mac_proto(key) == MAC_PROTO_NONE) { 783 if (unlikely(eth_type_vlan(skb->protocol))) 784 return -EINVAL; 785 786 skb_reset_network_header(skb); 787 key->eth.type = skb->protocol; 788 } else { 789 eth = eth_hdr(skb); 790 ether_addr_copy(key->eth.src, eth->h_source); 791 ether_addr_copy(key->eth.dst, eth->h_dest); 792 793 __skb_pull(skb, 2 * ETH_ALEN); 794 /* We are going to push all headers that we pull, so no need to 795 * update skb->csum here. 796 */ 797 798 if (unlikely(parse_vlan(skb, key))) 799 return -ENOMEM; 800 801 key->eth.type = parse_ethertype(skb); 802 if (unlikely(key->eth.type == htons(0))) 803 return -ENOMEM; 804 805 /* Multiple tagged packets need to retain TPID to satisfy 806 * skb_vlan_pop(), which will later shift the ethertype into 807 * skb->protocol. 808 */ 809 if (key->eth.cvlan.tci & htons(VLAN_CFI_MASK)) 810 skb->protocol = key->eth.cvlan.tpid; 811 else 812 skb->protocol = key->eth.type; 813 814 skb_reset_network_header(skb); 815 __skb_push(skb, skb->data - skb_mac_header(skb)); 816 } 817 818 skb_reset_mac_len(skb); 819 820 /* Fill out L3/L4 key info, if any */ 821 return key_extract_l3l4(skb, key); 822 } 823 824 /* In the case of conntrack fragment handling it expects L3 headers, 825 * add a helper. 826 */ 827 int ovs_flow_key_update_l3l4(struct sk_buff *skb, struct sw_flow_key *key) 828 { 829 return key_extract_l3l4(skb, key); 830 } 831 832 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key) 833 { 834 int res; 835 836 res = key_extract(skb, key); 837 if (!res) 838 key->mac_proto &= ~SW_FLOW_KEY_INVALID; 839 840 return res; 841 } 842 843 static int key_extract_mac_proto(struct sk_buff *skb) 844 { 845 switch (skb->dev->type) { 846 case ARPHRD_ETHER: 847 return MAC_PROTO_ETHERNET; 848 case ARPHRD_NONE: 849 if (skb->protocol == htons(ETH_P_TEB)) 850 return MAC_PROTO_ETHERNET; 851 return MAC_PROTO_NONE; 852 } 853 WARN_ON_ONCE(1); 854 return -EINVAL; 855 } 856 857 int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info, 858 struct sk_buff *skb, struct sw_flow_key *key) 859 { 860 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 861 struct tc_skb_ext *tc_ext; 862 #endif 863 bool post_ct = false, post_ct_snat = false, post_ct_dnat = false; 864 int res, err; 865 u16 zone = 0; 866 867 /* Extract metadata from packet. */ 868 if (tun_info) { 869 key->tun_proto = ip_tunnel_info_af(tun_info); 870 memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key)); 871 872 if (tun_info->options_len) { 873 BUILD_BUG_ON((1 << (sizeof(tun_info->options_len) * 874 8)) - 1 875 > sizeof(key->tun_opts)); 876 877 ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key, tun_info->options_len), 878 tun_info); 879 key->tun_opts_len = tun_info->options_len; 880 } else { 881 key->tun_opts_len = 0; 882 } 883 } else { 884 key->tun_proto = 0; 885 key->tun_opts_len = 0; 886 memset(&key->tun_key, 0, sizeof(key->tun_key)); 887 } 888 889 key->phy.priority = skb->priority; 890 key->phy.in_port = OVS_CB(skb)->input_vport->port_no; 891 key->phy.skb_mark = skb->mark; 892 key->ovs_flow_hash = 0; 893 res = key_extract_mac_proto(skb); 894 if (res < 0) 895 return res; 896 key->mac_proto = res; 897 898 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 899 if (tc_skb_ext_tc_enabled()) { 900 tc_ext = skb_ext_find(skb, TC_SKB_EXT); 901 key->recirc_id = tc_ext ? tc_ext->chain : 0; 902 OVS_CB(skb)->mru = tc_ext ? tc_ext->mru : 0; 903 post_ct = tc_ext ? tc_ext->post_ct : false; 904 post_ct_snat = post_ct ? tc_ext->post_ct_snat : false; 905 post_ct_dnat = post_ct ? tc_ext->post_ct_dnat : false; 906 zone = post_ct ? tc_ext->zone : 0; 907 } else { 908 key->recirc_id = 0; 909 } 910 #else 911 key->recirc_id = 0; 912 #endif 913 914 err = key_extract(skb, key); 915 if (!err) { 916 ovs_ct_fill_key(skb, key, post_ct); /* Must be after key_extract(). */ 917 if (post_ct) { 918 if (!skb_get_nfct(skb)) { 919 key->ct_zone = zone; 920 } else { 921 if (!post_ct_dnat) 922 key->ct_state &= ~OVS_CS_F_DST_NAT; 923 if (!post_ct_snat) 924 key->ct_state &= ~OVS_CS_F_SRC_NAT; 925 } 926 } 927 } 928 return err; 929 } 930 931 int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr, 932 struct sk_buff *skb, 933 struct sw_flow_key *key, bool log) 934 { 935 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1]; 936 u64 attrs = 0; 937 int err; 938 939 err = parse_flow_nlattrs(attr, a, &attrs, log); 940 if (err) 941 return -EINVAL; 942 943 /* Extract metadata from netlink attributes. */ 944 err = ovs_nla_get_flow_metadata(net, a, attrs, key, log); 945 if (err) 946 return err; 947 948 /* key_extract assumes that skb->protocol is set-up for 949 * layer 3 packets which is the case for other callers, 950 * in particular packets received from the network stack. 951 * Here the correct value can be set from the metadata 952 * extracted above. 953 * For L2 packet key eth type would be zero. skb protocol 954 * would be set to correct value later during key-extact. 955 */ 956 957 skb->protocol = key->eth.type; 958 err = key_extract(skb, key); 959 if (err) 960 return err; 961 962 /* Check that we have conntrack original direction tuple metadata only 963 * for packets for which it makes sense. Otherwise the key may be 964 * corrupted due to overlapping key fields. 965 */ 966 if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4) && 967 key->eth.type != htons(ETH_P_IP)) 968 return -EINVAL; 969 if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6) && 970 (key->eth.type != htons(ETH_P_IPV6) || 971 sw_flow_key_is_nd(key))) 972 return -EINVAL; 973 974 return 0; 975 } 976