1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (c) 2007-2017 Nicira, Inc. 4 */ 5 6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 7 8 #include <linux/skbuff.h> 9 #include <linux/in.h> 10 #include <linux/ip.h> 11 #include <linux/openvswitch.h> 12 #include <linux/sctp.h> 13 #include <linux/tcp.h> 14 #include <linux/udp.h> 15 #include <linux/in6.h> 16 #include <linux/if_arp.h> 17 #include <linux/if_vlan.h> 18 19 #include <net/dst.h> 20 #include <net/gso.h> 21 #include <net/ip.h> 22 #include <net/ipv6.h> 23 #include <net/ip6_fib.h> 24 #include <net/checksum.h> 25 #include <net/dsfield.h> 26 #include <net/mpls.h> 27 28 #if IS_ENABLED(CONFIG_PSAMPLE) 29 #include <net/psample.h> 30 #endif 31 32 #include <net/sctp/checksum.h> 33 34 #include "datapath.h" 35 #include "drop.h" 36 #include "flow.h" 37 #include "conntrack.h" 38 #include "vport.h" 39 #include "flow_netlink.h" 40 #include "openvswitch_trace.h" 41 42 DEFINE_PER_CPU(struct ovs_pcpu_storage, ovs_pcpu_storage) = { 43 .bh_lock = INIT_LOCAL_LOCK(bh_lock), 44 }; 45 46 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys' 47 * space. Return NULL if out of key spaces. 48 */ 49 static struct sw_flow_key *clone_key(const struct sw_flow_key *key_) 50 { 51 struct ovs_pcpu_storage *ovs_pcpu = this_cpu_ptr(&ovs_pcpu_storage); 52 struct action_flow_keys *keys = &ovs_pcpu->flow_keys; 53 int level = ovs_pcpu->exec_level; 54 struct sw_flow_key *key = NULL; 55 56 if (level <= OVS_DEFERRED_ACTION_THRESHOLD) { 57 key = &keys->key[level - 1]; 58 *key = *key_; 59 } 60 61 return key; 62 } 63 64 static void action_fifo_init(struct action_fifo *fifo) 65 { 66 fifo->head = 0; 67 fifo->tail = 0; 68 } 69 70 static bool action_fifo_is_empty(const struct action_fifo *fifo) 71 { 72 return (fifo->head == fifo->tail); 73 } 74 75 static struct deferred_action *action_fifo_get(struct action_fifo *fifo) 76 { 77 if (action_fifo_is_empty(fifo)) 78 return NULL; 79 80 return &fifo->fifo[fifo->tail++]; 81 } 82 83 static struct deferred_action *action_fifo_put(struct action_fifo *fifo) 84 { 85 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1) 86 return NULL; 87 88 return &fifo->fifo[fifo->head++]; 89 } 90 91 /* Return true if fifo is not full */ 92 static struct deferred_action *add_deferred_actions(struct sk_buff *skb, 93 const struct sw_flow_key *key, 94 const struct nlattr *actions, 95 const int actions_len) 96 { 97 struct action_fifo *fifo = this_cpu_ptr(&ovs_pcpu_storage.action_fifos); 98 struct deferred_action *da; 99 100 da = action_fifo_put(fifo); 101 if (da) { 102 da->skb = skb; 103 da->actions = actions; 104 da->actions_len = actions_len; 105 da->pkt_key = *key; 106 } 107 108 return da; 109 } 110 111 static void invalidate_flow_key(struct sw_flow_key *key) 112 { 113 key->mac_proto |= SW_FLOW_KEY_INVALID; 114 } 115 116 static bool is_flow_key_valid(const struct sw_flow_key *key) 117 { 118 return !(key->mac_proto & SW_FLOW_KEY_INVALID); 119 } 120 121 static int clone_execute(struct datapath *dp, struct sk_buff *skb, 122 struct sw_flow_key *key, 123 u32 recirc_id, 124 const struct nlattr *actions, int len, 125 bool last, bool clone_flow_key); 126 127 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb, 128 struct sw_flow_key *key, 129 const struct nlattr *attr, int len); 130 131 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key, 132 __be32 mpls_lse, __be16 mpls_ethertype, __u16 mac_len) 133 { 134 int err; 135 136 err = skb_mpls_push(skb, mpls_lse, mpls_ethertype, mac_len, !!mac_len); 137 if (err) 138 return err; 139 140 if (!mac_len) 141 key->mac_proto = MAC_PROTO_NONE; 142 143 invalidate_flow_key(key); 144 return 0; 145 } 146 147 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key, 148 const __be16 ethertype) 149 { 150 int err; 151 152 err = skb_mpls_pop(skb, ethertype, skb->mac_len, 153 ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET); 154 if (err) 155 return err; 156 157 if (ethertype == htons(ETH_P_TEB)) 158 key->mac_proto = MAC_PROTO_ETHERNET; 159 160 invalidate_flow_key(key); 161 return 0; 162 } 163 164 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key, 165 const __be32 *mpls_lse, const __be32 *mask) 166 { 167 struct mpls_shim_hdr *stack; 168 __be32 lse; 169 int err; 170 171 if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN)) 172 return -ENOMEM; 173 174 stack = mpls_hdr(skb); 175 lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask); 176 err = skb_mpls_update_lse(skb, lse); 177 if (err) 178 return err; 179 180 flow_key->mpls.lse[0] = lse; 181 return 0; 182 } 183 184 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key) 185 { 186 int err; 187 188 err = skb_vlan_pop(skb); 189 if (skb_vlan_tag_present(skb)) { 190 invalidate_flow_key(key); 191 } else { 192 key->eth.vlan.tci = 0; 193 key->eth.vlan.tpid = 0; 194 } 195 return err; 196 } 197 198 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key, 199 const struct ovs_action_push_vlan *vlan) 200 { 201 int err; 202 203 if (skb_vlan_tag_present(skb)) { 204 invalidate_flow_key(key); 205 } else { 206 key->eth.vlan.tci = vlan->vlan_tci; 207 key->eth.vlan.tpid = vlan->vlan_tpid; 208 } 209 err = skb_vlan_push(skb, vlan->vlan_tpid, 210 ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK); 211 skb_reset_mac_len(skb); 212 return err; 213 } 214 215 /* 'src' is already properly masked. */ 216 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_) 217 { 218 u16 *dst = (u16 *)dst_; 219 const u16 *src = (const u16 *)src_; 220 const u16 *mask = (const u16 *)mask_; 221 222 OVS_SET_MASKED(dst[0], src[0], mask[0]); 223 OVS_SET_MASKED(dst[1], src[1], mask[1]); 224 OVS_SET_MASKED(dst[2], src[2], mask[2]); 225 } 226 227 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key, 228 const struct ovs_key_ethernet *key, 229 const struct ovs_key_ethernet *mask) 230 { 231 int err; 232 233 err = skb_ensure_writable(skb, ETH_HLEN); 234 if (unlikely(err)) 235 return err; 236 237 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2); 238 239 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src, 240 mask->eth_src); 241 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst, 242 mask->eth_dst); 243 244 skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2); 245 246 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source); 247 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest); 248 return 0; 249 } 250 251 /* pop_eth does not support VLAN packets as this action is never called 252 * for them. 253 */ 254 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key) 255 { 256 int err; 257 258 err = skb_eth_pop(skb); 259 if (err) 260 return err; 261 262 /* safe right before invalidate_flow_key */ 263 key->mac_proto = MAC_PROTO_NONE; 264 invalidate_flow_key(key); 265 return 0; 266 } 267 268 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key, 269 const struct ovs_action_push_eth *ethh) 270 { 271 int err; 272 273 err = skb_eth_push(skb, ethh->addresses.eth_dst, 274 ethh->addresses.eth_src); 275 if (err) 276 return err; 277 278 /* safe right before invalidate_flow_key */ 279 key->mac_proto = MAC_PROTO_ETHERNET; 280 invalidate_flow_key(key); 281 return 0; 282 } 283 284 static noinline_for_stack int push_nsh(struct sk_buff *skb, 285 struct sw_flow_key *key, 286 const struct nlattr *a) 287 { 288 u8 buffer[NSH_HDR_MAX_LEN]; 289 struct nshhdr *nh = (struct nshhdr *)buffer; 290 int err; 291 292 err = nsh_hdr_from_nlattr(a, nh, NSH_HDR_MAX_LEN); 293 if (err) 294 return err; 295 296 err = nsh_push(skb, nh); 297 if (err) 298 return err; 299 300 /* safe right before invalidate_flow_key */ 301 key->mac_proto = MAC_PROTO_NONE; 302 invalidate_flow_key(key); 303 return 0; 304 } 305 306 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key) 307 { 308 int err; 309 310 err = nsh_pop(skb); 311 if (err) 312 return err; 313 314 /* safe right before invalidate_flow_key */ 315 if (skb->protocol == htons(ETH_P_TEB)) 316 key->mac_proto = MAC_PROTO_ETHERNET; 317 else 318 key->mac_proto = MAC_PROTO_NONE; 319 invalidate_flow_key(key); 320 return 0; 321 } 322 323 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh, 324 __be32 addr, __be32 new_addr) 325 { 326 int transport_len = skb->len - skb_transport_offset(skb); 327 328 if (nh->frag_off & htons(IP_OFFSET)) 329 return; 330 331 if (nh->protocol == IPPROTO_TCP) { 332 if (likely(transport_len >= sizeof(struct tcphdr))) 333 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb, 334 addr, new_addr, true); 335 } else if (nh->protocol == IPPROTO_UDP) { 336 if (likely(transport_len >= sizeof(struct udphdr))) { 337 struct udphdr *uh = udp_hdr(skb); 338 339 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) { 340 inet_proto_csum_replace4(&uh->check, skb, 341 addr, new_addr, true); 342 if (!uh->check) 343 uh->check = CSUM_MANGLED_0; 344 } 345 } 346 } 347 } 348 349 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh, 350 __be32 *addr, __be32 new_addr) 351 { 352 update_ip_l4_checksum(skb, nh, *addr, new_addr); 353 csum_replace4(&nh->check, *addr, new_addr); 354 skb_clear_hash(skb); 355 ovs_ct_clear(skb, NULL); 356 *addr = new_addr; 357 } 358 359 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto, 360 __be32 addr[4], const __be32 new_addr[4]) 361 { 362 int transport_len = skb->len - skb_transport_offset(skb); 363 364 if (l4_proto == NEXTHDR_TCP) { 365 if (likely(transport_len >= sizeof(struct tcphdr))) 366 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb, 367 addr, new_addr, true); 368 } else if (l4_proto == NEXTHDR_UDP) { 369 if (likely(transport_len >= sizeof(struct udphdr))) { 370 struct udphdr *uh = udp_hdr(skb); 371 372 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) { 373 inet_proto_csum_replace16(&uh->check, skb, 374 addr, new_addr, true); 375 if (!uh->check) 376 uh->check = CSUM_MANGLED_0; 377 } 378 } 379 } else if (l4_proto == NEXTHDR_ICMP) { 380 if (likely(transport_len >= sizeof(struct icmp6hdr))) 381 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum, 382 skb, addr, new_addr, true); 383 } 384 } 385 386 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4], 387 const __be32 mask[4], __be32 masked[4]) 388 { 389 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]); 390 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]); 391 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]); 392 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]); 393 } 394 395 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto, 396 __be32 addr[4], const __be32 new_addr[4], 397 bool recalculate_csum) 398 { 399 if (recalculate_csum) 400 update_ipv6_checksum(skb, l4_proto, addr, new_addr); 401 402 skb_clear_hash(skb); 403 ovs_ct_clear(skb, NULL); 404 memcpy(addr, new_addr, sizeof(__be32[4])); 405 } 406 407 static void set_ipv6_dsfield(struct sk_buff *skb, struct ipv6hdr *nh, u8 ipv6_tclass, u8 mask) 408 { 409 u8 old_ipv6_tclass = ipv6_get_dsfield(nh); 410 411 ipv6_tclass = OVS_MASKED(old_ipv6_tclass, ipv6_tclass, mask); 412 413 if (skb->ip_summed == CHECKSUM_COMPLETE) 414 csum_replace(&skb->csum, (__force __wsum)(old_ipv6_tclass << 12), 415 (__force __wsum)(ipv6_tclass << 12)); 416 417 ipv6_change_dsfield(nh, ~mask, ipv6_tclass); 418 } 419 420 static void set_ipv6_fl(struct sk_buff *skb, struct ipv6hdr *nh, u32 fl, u32 mask) 421 { 422 u32 ofl; 423 424 ofl = nh->flow_lbl[0] << 16 | nh->flow_lbl[1] << 8 | nh->flow_lbl[2]; 425 fl = OVS_MASKED(ofl, fl, mask); 426 427 /* Bits 21-24 are always unmasked, so this retains their values. */ 428 nh->flow_lbl[0] = (u8)(fl >> 16); 429 nh->flow_lbl[1] = (u8)(fl >> 8); 430 nh->flow_lbl[2] = (u8)fl; 431 432 if (skb->ip_summed == CHECKSUM_COMPLETE) 433 csum_replace(&skb->csum, (__force __wsum)htonl(ofl), (__force __wsum)htonl(fl)); 434 } 435 436 static void set_ipv6_ttl(struct sk_buff *skb, struct ipv6hdr *nh, u8 new_ttl, u8 mask) 437 { 438 new_ttl = OVS_MASKED(nh->hop_limit, new_ttl, mask); 439 440 if (skb->ip_summed == CHECKSUM_COMPLETE) 441 csum_replace(&skb->csum, (__force __wsum)(nh->hop_limit << 8), 442 (__force __wsum)(new_ttl << 8)); 443 nh->hop_limit = new_ttl; 444 } 445 446 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl, 447 u8 mask) 448 { 449 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask); 450 451 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8)); 452 nh->ttl = new_ttl; 453 } 454 455 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key, 456 const struct ovs_key_ipv4 *key, 457 const struct ovs_key_ipv4 *mask) 458 { 459 struct iphdr *nh; 460 __be32 new_addr; 461 int err; 462 463 err = skb_ensure_writable(skb, skb_network_offset(skb) + 464 sizeof(struct iphdr)); 465 if (unlikely(err)) 466 return err; 467 468 nh = ip_hdr(skb); 469 470 /* Setting an IP addresses is typically only a side effect of 471 * matching on them in the current userspace implementation, so it 472 * makes sense to check if the value actually changed. 473 */ 474 if (mask->ipv4_src) { 475 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src); 476 477 if (unlikely(new_addr != nh->saddr)) { 478 set_ip_addr(skb, nh, &nh->saddr, new_addr); 479 flow_key->ipv4.addr.src = new_addr; 480 } 481 } 482 if (mask->ipv4_dst) { 483 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst); 484 485 if (unlikely(new_addr != nh->daddr)) { 486 set_ip_addr(skb, nh, &nh->daddr, new_addr); 487 flow_key->ipv4.addr.dst = new_addr; 488 } 489 } 490 if (mask->ipv4_tos) { 491 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos); 492 flow_key->ip.tos = nh->tos; 493 } 494 if (mask->ipv4_ttl) { 495 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl); 496 flow_key->ip.ttl = nh->ttl; 497 } 498 499 return 0; 500 } 501 502 static bool is_ipv6_mask_nonzero(const __be32 addr[4]) 503 { 504 return !!(addr[0] | addr[1] | addr[2] | addr[3]); 505 } 506 507 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key, 508 const struct ovs_key_ipv6 *key, 509 const struct ovs_key_ipv6 *mask) 510 { 511 struct ipv6hdr *nh; 512 int err; 513 514 err = skb_ensure_writable(skb, skb_network_offset(skb) + 515 sizeof(struct ipv6hdr)); 516 if (unlikely(err)) 517 return err; 518 519 nh = ipv6_hdr(skb); 520 521 /* Setting an IP addresses is typically only a side effect of 522 * matching on them in the current userspace implementation, so it 523 * makes sense to check if the value actually changed. 524 */ 525 if (is_ipv6_mask_nonzero(mask->ipv6_src)) { 526 __be32 *saddr = (__be32 *)&nh->saddr; 527 __be32 masked[4]; 528 529 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked); 530 531 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) { 532 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked, 533 true); 534 memcpy(&flow_key->ipv6.addr.src, masked, 535 sizeof(flow_key->ipv6.addr.src)); 536 } 537 } 538 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) { 539 unsigned int offset = 0; 540 int flags = IP6_FH_F_SKIP_RH; 541 bool recalc_csum = true; 542 __be32 *daddr = (__be32 *)&nh->daddr; 543 __be32 masked[4]; 544 545 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked); 546 547 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) { 548 if (ipv6_ext_hdr(nh->nexthdr)) 549 recalc_csum = (ipv6_find_hdr(skb, &offset, 550 NEXTHDR_ROUTING, 551 NULL, &flags) 552 != NEXTHDR_ROUTING); 553 554 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked, 555 recalc_csum); 556 memcpy(&flow_key->ipv6.addr.dst, masked, 557 sizeof(flow_key->ipv6.addr.dst)); 558 } 559 } 560 if (mask->ipv6_tclass) { 561 set_ipv6_dsfield(skb, nh, key->ipv6_tclass, mask->ipv6_tclass); 562 flow_key->ip.tos = ipv6_get_dsfield(nh); 563 } 564 if (mask->ipv6_label) { 565 set_ipv6_fl(skb, nh, ntohl(key->ipv6_label), 566 ntohl(mask->ipv6_label)); 567 flow_key->ipv6.label = 568 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL); 569 } 570 if (mask->ipv6_hlimit) { 571 set_ipv6_ttl(skb, nh, key->ipv6_hlimit, mask->ipv6_hlimit); 572 flow_key->ip.ttl = nh->hop_limit; 573 } 574 return 0; 575 } 576 577 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key, 578 const struct nlattr *a) 579 { 580 struct nshhdr *nh; 581 size_t length; 582 int err; 583 u8 flags; 584 u8 ttl; 585 int i; 586 587 struct ovs_key_nsh key; 588 struct ovs_key_nsh mask; 589 590 err = nsh_key_from_nlattr(a, &key, &mask); 591 if (err) 592 return err; 593 594 /* Make sure the NSH base header is there */ 595 if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN)) 596 return -ENOMEM; 597 598 nh = nsh_hdr(skb); 599 length = nsh_hdr_len(nh); 600 601 /* Make sure the whole NSH header is there */ 602 err = skb_ensure_writable(skb, skb_network_offset(skb) + 603 length); 604 if (unlikely(err)) 605 return err; 606 607 nh = nsh_hdr(skb); 608 skb_postpull_rcsum(skb, nh, length); 609 flags = nsh_get_flags(nh); 610 flags = OVS_MASKED(flags, key.base.flags, mask.base.flags); 611 flow_key->nsh.base.flags = flags; 612 ttl = nsh_get_ttl(nh); 613 ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl); 614 flow_key->nsh.base.ttl = ttl; 615 nsh_set_flags_and_ttl(nh, flags, ttl); 616 nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr, 617 mask.base.path_hdr); 618 flow_key->nsh.base.path_hdr = nh->path_hdr; 619 switch (nh->mdtype) { 620 case NSH_M_TYPE1: 621 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) { 622 nh->md1.context[i] = 623 OVS_MASKED(nh->md1.context[i], key.context[i], 624 mask.context[i]); 625 } 626 memcpy(flow_key->nsh.context, nh->md1.context, 627 sizeof(nh->md1.context)); 628 break; 629 case NSH_M_TYPE2: 630 memset(flow_key->nsh.context, 0, 631 sizeof(flow_key->nsh.context)); 632 break; 633 default: 634 return -EINVAL; 635 } 636 skb_postpush_rcsum(skb, nh, length); 637 return 0; 638 } 639 640 /* Must follow skb_ensure_writable() since that can move the skb data. */ 641 static void set_tp_port(struct sk_buff *skb, __be16 *port, 642 __be16 new_port, __sum16 *check) 643 { 644 ovs_ct_clear(skb, NULL); 645 inet_proto_csum_replace2(check, skb, *port, new_port, false); 646 *port = new_port; 647 } 648 649 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key, 650 const struct ovs_key_udp *key, 651 const struct ovs_key_udp *mask) 652 { 653 struct udphdr *uh; 654 __be16 src, dst; 655 int err; 656 657 err = skb_ensure_writable(skb, skb_transport_offset(skb) + 658 sizeof(struct udphdr)); 659 if (unlikely(err)) 660 return err; 661 662 uh = udp_hdr(skb); 663 /* Either of the masks is non-zero, so do not bother checking them. */ 664 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src); 665 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst); 666 667 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) { 668 if (likely(src != uh->source)) { 669 set_tp_port(skb, &uh->source, src, &uh->check); 670 flow_key->tp.src = src; 671 } 672 if (likely(dst != uh->dest)) { 673 set_tp_port(skb, &uh->dest, dst, &uh->check); 674 flow_key->tp.dst = dst; 675 } 676 677 if (unlikely(!uh->check)) 678 uh->check = CSUM_MANGLED_0; 679 } else { 680 uh->source = src; 681 uh->dest = dst; 682 flow_key->tp.src = src; 683 flow_key->tp.dst = dst; 684 ovs_ct_clear(skb, NULL); 685 } 686 687 skb_clear_hash(skb); 688 689 return 0; 690 } 691 692 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key, 693 const struct ovs_key_tcp *key, 694 const struct ovs_key_tcp *mask) 695 { 696 struct tcphdr *th; 697 __be16 src, dst; 698 int err; 699 700 err = skb_ensure_writable(skb, skb_transport_offset(skb) + 701 sizeof(struct tcphdr)); 702 if (unlikely(err)) 703 return err; 704 705 th = tcp_hdr(skb); 706 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src); 707 if (likely(src != th->source)) { 708 set_tp_port(skb, &th->source, src, &th->check); 709 flow_key->tp.src = src; 710 } 711 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst); 712 if (likely(dst != th->dest)) { 713 set_tp_port(skb, &th->dest, dst, &th->check); 714 flow_key->tp.dst = dst; 715 } 716 skb_clear_hash(skb); 717 718 return 0; 719 } 720 721 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key, 722 const struct ovs_key_sctp *key, 723 const struct ovs_key_sctp *mask) 724 { 725 unsigned int sctphoff = skb_transport_offset(skb); 726 struct sctphdr *sh; 727 __le32 old_correct_csum, new_csum, old_csum; 728 int err; 729 730 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr)); 731 if (unlikely(err)) 732 return err; 733 734 sh = sctp_hdr(skb); 735 old_csum = sh->checksum; 736 old_correct_csum = sctp_compute_cksum(skb, sctphoff); 737 738 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src); 739 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst); 740 741 new_csum = sctp_compute_cksum(skb, sctphoff); 742 743 /* Carry any checksum errors through. */ 744 sh->checksum = old_csum ^ old_correct_csum ^ new_csum; 745 746 skb_clear_hash(skb); 747 ovs_ct_clear(skb, NULL); 748 749 flow_key->tp.src = sh->source; 750 flow_key->tp.dst = sh->dest; 751 752 return 0; 753 } 754 755 static int ovs_vport_output(struct net *net, struct sock *sk, 756 struct sk_buff *skb) 757 { 758 struct ovs_frag_data *data = this_cpu_ptr(&ovs_pcpu_storage.frag_data); 759 struct vport *vport = data->vport; 760 761 if (skb_cow_head(skb, data->l2_len) < 0) { 762 kfree_skb_reason(skb, SKB_DROP_REASON_NOMEM); 763 return -ENOMEM; 764 } 765 766 __skb_dst_copy(skb, data->dst); 767 *OVS_CB(skb) = data->cb; 768 skb->inner_protocol = data->inner_protocol; 769 if (data->vlan_tci & VLAN_CFI_MASK) 770 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK); 771 else 772 __vlan_hwaccel_clear_tag(skb); 773 774 /* Reconstruct the MAC header. */ 775 skb_push(skb, data->l2_len); 776 memcpy(skb->data, &data->l2_data, data->l2_len); 777 skb_postpush_rcsum(skb, skb->data, data->l2_len); 778 skb_reset_mac_header(skb); 779 780 if (eth_p_mpls(skb->protocol)) { 781 skb->inner_network_header = skb->network_header; 782 skb_set_network_header(skb, data->network_offset); 783 skb_reset_mac_len(skb); 784 } 785 786 ovs_vport_send(vport, skb, data->mac_proto); 787 return 0; 788 } 789 790 static unsigned int 791 ovs_dst_get_mtu(const struct dst_entry *dst) 792 { 793 return dst->dev->mtu; 794 } 795 796 static struct dst_ops ovs_dst_ops = { 797 .family = AF_UNSPEC, 798 .mtu = ovs_dst_get_mtu, 799 }; 800 801 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is 802 * ovs_vport_output(), which is called once per fragmented packet. 803 */ 804 static void prepare_frag(struct vport *vport, struct sk_buff *skb, 805 u16 orig_network_offset, u8 mac_proto) 806 { 807 unsigned int hlen = skb_network_offset(skb); 808 struct ovs_frag_data *data; 809 810 data = this_cpu_ptr(&ovs_pcpu_storage.frag_data); 811 data->dst = skb->_skb_refdst; 812 data->vport = vport; 813 data->cb = *OVS_CB(skb); 814 data->inner_protocol = skb->inner_protocol; 815 data->network_offset = orig_network_offset; 816 if (skb_vlan_tag_present(skb)) 817 data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK; 818 else 819 data->vlan_tci = 0; 820 data->vlan_proto = skb->vlan_proto; 821 data->mac_proto = mac_proto; 822 data->l2_len = hlen; 823 memcpy(&data->l2_data, skb->data, hlen); 824 825 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); 826 skb_pull(skb, hlen); 827 } 828 829 static void ovs_fragment(struct net *net, struct vport *vport, 830 struct sk_buff *skb, u16 mru, 831 struct sw_flow_key *key) 832 { 833 enum ovs_drop_reason reason; 834 u16 orig_network_offset = 0; 835 836 if (eth_p_mpls(skb->protocol)) { 837 orig_network_offset = skb_network_offset(skb); 838 skb->network_header = skb->inner_network_header; 839 } 840 841 if (skb_network_offset(skb) > MAX_L2_LEN) { 842 OVS_NLERR(1, "L2 header too long to fragment"); 843 reason = OVS_DROP_FRAG_L2_TOO_LONG; 844 goto err; 845 } 846 847 if (key->eth.type == htons(ETH_P_IP)) { 848 struct rtable ovs_rt = { 0 }; 849 unsigned long orig_dst; 850 851 prepare_frag(vport, skb, orig_network_offset, 852 ovs_key_mac_proto(key)); 853 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 854 DST_OBSOLETE_NONE, DST_NOCOUNT); 855 ovs_rt.dst.dev = vport->dev; 856 857 orig_dst = skb->_skb_refdst; 858 skb_dst_set_noref(skb, &ovs_rt.dst); 859 IPCB(skb)->frag_max_size = mru; 860 861 ip_do_fragment(net, skb->sk, skb, ovs_vport_output); 862 refdst_drop(orig_dst); 863 } else if (key->eth.type == htons(ETH_P_IPV6)) { 864 unsigned long orig_dst; 865 struct rt6_info ovs_rt; 866 867 prepare_frag(vport, skb, orig_network_offset, 868 ovs_key_mac_proto(key)); 869 memset(&ovs_rt, 0, sizeof(ovs_rt)); 870 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 871 DST_OBSOLETE_NONE, DST_NOCOUNT); 872 ovs_rt.dst.dev = vport->dev; 873 874 orig_dst = skb->_skb_refdst; 875 skb_dst_set_noref(skb, &ovs_rt.dst); 876 IP6CB(skb)->frag_max_size = mru; 877 878 ipv6_stub->ipv6_fragment(net, skb->sk, skb, ovs_vport_output); 879 refdst_drop(orig_dst); 880 } else { 881 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.", 882 ovs_vport_name(vport), ntohs(key->eth.type), mru, 883 vport->dev->mtu); 884 reason = OVS_DROP_FRAG_INVALID_PROTO; 885 goto err; 886 } 887 888 return; 889 err: 890 ovs_kfree_skb_reason(skb, reason); 891 } 892 893 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port, 894 struct sw_flow_key *key) 895 { 896 struct vport *vport = ovs_vport_rcu(dp, out_port); 897 898 if (likely(vport && 899 netif_running(vport->dev) && 900 netif_carrier_ok(vport->dev))) { 901 u16 mru = OVS_CB(skb)->mru; 902 u32 cutlen = OVS_CB(skb)->cutlen; 903 904 if (unlikely(cutlen > 0)) { 905 if (skb->len - cutlen > ovs_mac_header_len(key)) 906 pskb_trim(skb, skb->len - cutlen); 907 else 908 pskb_trim(skb, ovs_mac_header_len(key)); 909 } 910 911 if (likely(!mru || 912 (skb->len <= mru + vport->dev->hard_header_len))) { 913 ovs_vport_send(vport, skb, ovs_key_mac_proto(key)); 914 } else if (mru <= vport->dev->mtu) { 915 struct net *net = read_pnet(&dp->net); 916 917 ovs_fragment(net, vport, skb, mru, key); 918 } else { 919 kfree_skb_reason(skb, SKB_DROP_REASON_PKT_TOO_BIG); 920 } 921 } else { 922 kfree_skb_reason(skb, SKB_DROP_REASON_DEV_READY); 923 } 924 } 925 926 static int output_userspace(struct datapath *dp, struct sk_buff *skb, 927 struct sw_flow_key *key, const struct nlattr *attr, 928 const struct nlattr *actions, int actions_len, 929 uint32_t cutlen) 930 { 931 struct dp_upcall_info upcall; 932 const struct nlattr *a; 933 int rem; 934 935 memset(&upcall, 0, sizeof(upcall)); 936 upcall.cmd = OVS_PACKET_CMD_ACTION; 937 upcall.mru = OVS_CB(skb)->mru; 938 939 nla_for_each_nested(a, attr, rem) { 940 switch (nla_type(a)) { 941 case OVS_USERSPACE_ATTR_USERDATA: 942 upcall.userdata = a; 943 break; 944 945 case OVS_USERSPACE_ATTR_PID: 946 if (dp->user_features & 947 OVS_DP_F_DISPATCH_UPCALL_PER_CPU) 948 upcall.portid = 949 ovs_dp_get_upcall_portid(dp, 950 smp_processor_id()); 951 else 952 upcall.portid = nla_get_u32(a); 953 break; 954 955 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: { 956 /* Get out tunnel info. */ 957 struct vport *vport; 958 959 vport = ovs_vport_rcu(dp, nla_get_u32(a)); 960 if (vport) { 961 int err; 962 963 err = dev_fill_metadata_dst(vport->dev, skb); 964 if (!err) 965 upcall.egress_tun_info = skb_tunnel_info(skb); 966 } 967 968 break; 969 } 970 971 case OVS_USERSPACE_ATTR_ACTIONS: { 972 /* Include actions. */ 973 upcall.actions = actions; 974 upcall.actions_len = actions_len; 975 break; 976 } 977 978 } /* End of switch. */ 979 } 980 981 return ovs_dp_upcall(dp, skb, key, &upcall, cutlen); 982 } 983 984 static int dec_ttl_exception_handler(struct datapath *dp, struct sk_buff *skb, 985 struct sw_flow_key *key, 986 const struct nlattr *attr) 987 { 988 /* The first attribute is always 'OVS_DEC_TTL_ATTR_ACTION'. */ 989 struct nlattr *actions = nla_data(attr); 990 991 if (nla_len(actions)) 992 return clone_execute(dp, skb, key, 0, nla_data(actions), 993 nla_len(actions), true, false); 994 995 ovs_kfree_skb_reason(skb, OVS_DROP_IP_TTL); 996 return 0; 997 } 998 999 /* When 'last' is true, sample() should always consume the 'skb'. 1000 * Otherwise, sample() should keep 'skb' intact regardless what 1001 * actions are executed within sample(). 1002 */ 1003 static int sample(struct datapath *dp, struct sk_buff *skb, 1004 struct sw_flow_key *key, const struct nlattr *attr, 1005 bool last) 1006 { 1007 struct nlattr *actions; 1008 struct nlattr *sample_arg; 1009 int rem = nla_len(attr); 1010 const struct sample_arg *arg; 1011 u32 init_probability; 1012 bool clone_flow_key; 1013 int err; 1014 1015 /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */ 1016 sample_arg = nla_data(attr); 1017 arg = nla_data(sample_arg); 1018 actions = nla_next(sample_arg, &rem); 1019 init_probability = OVS_CB(skb)->probability; 1020 1021 if ((arg->probability != U32_MAX) && 1022 (!arg->probability || get_random_u32() > arg->probability)) { 1023 if (last) 1024 ovs_kfree_skb_reason(skb, OVS_DROP_LAST_ACTION); 1025 return 0; 1026 } 1027 1028 OVS_CB(skb)->probability = arg->probability; 1029 1030 clone_flow_key = !arg->exec; 1031 err = clone_execute(dp, skb, key, 0, actions, rem, last, 1032 clone_flow_key); 1033 1034 if (!last) 1035 OVS_CB(skb)->probability = init_probability; 1036 1037 return err; 1038 } 1039 1040 /* When 'last' is true, clone() should always consume the 'skb'. 1041 * Otherwise, clone() should keep 'skb' intact regardless what 1042 * actions are executed within clone(). 1043 */ 1044 static int clone(struct datapath *dp, struct sk_buff *skb, 1045 struct sw_flow_key *key, const struct nlattr *attr, 1046 bool last) 1047 { 1048 struct nlattr *actions; 1049 struct nlattr *clone_arg; 1050 int rem = nla_len(attr); 1051 bool dont_clone_flow_key; 1052 1053 /* The first action is always 'OVS_CLONE_ATTR_EXEC'. */ 1054 clone_arg = nla_data(attr); 1055 dont_clone_flow_key = nla_get_u32(clone_arg); 1056 actions = nla_next(clone_arg, &rem); 1057 1058 return clone_execute(dp, skb, key, 0, actions, rem, last, 1059 !dont_clone_flow_key); 1060 } 1061 1062 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key, 1063 const struct nlattr *attr) 1064 { 1065 struct ovs_action_hash *hash_act = nla_data(attr); 1066 u32 hash = 0; 1067 1068 if (hash_act->hash_alg == OVS_HASH_ALG_L4) { 1069 /* OVS_HASH_ALG_L4 hasing type. */ 1070 hash = skb_get_hash(skb); 1071 } else if (hash_act->hash_alg == OVS_HASH_ALG_SYM_L4) { 1072 /* OVS_HASH_ALG_SYM_L4 hashing type. NOTE: this doesn't 1073 * extend past an encapsulated header. 1074 */ 1075 hash = __skb_get_hash_symmetric(skb); 1076 } 1077 1078 hash = jhash_1word(hash, hash_act->hash_basis); 1079 if (!hash) 1080 hash = 0x1; 1081 1082 key->ovs_flow_hash = hash; 1083 } 1084 1085 static int execute_set_action(struct sk_buff *skb, 1086 struct sw_flow_key *flow_key, 1087 const struct nlattr *a) 1088 { 1089 /* Only tunnel set execution is supported without a mask. */ 1090 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) { 1091 struct ovs_tunnel_info *tun = nla_data(a); 1092 1093 skb_dst_drop(skb); 1094 dst_hold((struct dst_entry *)tun->tun_dst); 1095 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst); 1096 return 0; 1097 } 1098 1099 return -EINVAL; 1100 } 1101 1102 /* Mask is at the midpoint of the data. */ 1103 #define get_mask(a, type) ((const type)nla_data(a) + 1) 1104 1105 static int execute_masked_set_action(struct sk_buff *skb, 1106 struct sw_flow_key *flow_key, 1107 const struct nlattr *a) 1108 { 1109 int err = 0; 1110 1111 switch (nla_type(a)) { 1112 case OVS_KEY_ATTR_PRIORITY: 1113 OVS_SET_MASKED(skb->priority, nla_get_u32(a), 1114 *get_mask(a, u32 *)); 1115 flow_key->phy.priority = skb->priority; 1116 break; 1117 1118 case OVS_KEY_ATTR_SKB_MARK: 1119 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *)); 1120 flow_key->phy.skb_mark = skb->mark; 1121 break; 1122 1123 case OVS_KEY_ATTR_TUNNEL_INFO: 1124 /* Masked data not supported for tunnel. */ 1125 err = -EINVAL; 1126 break; 1127 1128 case OVS_KEY_ATTR_ETHERNET: 1129 err = set_eth_addr(skb, flow_key, nla_data(a), 1130 get_mask(a, struct ovs_key_ethernet *)); 1131 break; 1132 1133 case OVS_KEY_ATTR_NSH: 1134 err = set_nsh(skb, flow_key, a); 1135 break; 1136 1137 case OVS_KEY_ATTR_IPV4: 1138 err = set_ipv4(skb, flow_key, nla_data(a), 1139 get_mask(a, struct ovs_key_ipv4 *)); 1140 break; 1141 1142 case OVS_KEY_ATTR_IPV6: 1143 err = set_ipv6(skb, flow_key, nla_data(a), 1144 get_mask(a, struct ovs_key_ipv6 *)); 1145 break; 1146 1147 case OVS_KEY_ATTR_TCP: 1148 err = set_tcp(skb, flow_key, nla_data(a), 1149 get_mask(a, struct ovs_key_tcp *)); 1150 break; 1151 1152 case OVS_KEY_ATTR_UDP: 1153 err = set_udp(skb, flow_key, nla_data(a), 1154 get_mask(a, struct ovs_key_udp *)); 1155 break; 1156 1157 case OVS_KEY_ATTR_SCTP: 1158 err = set_sctp(skb, flow_key, nla_data(a), 1159 get_mask(a, struct ovs_key_sctp *)); 1160 break; 1161 1162 case OVS_KEY_ATTR_MPLS: 1163 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a, 1164 __be32 *)); 1165 break; 1166 1167 case OVS_KEY_ATTR_CT_STATE: 1168 case OVS_KEY_ATTR_CT_ZONE: 1169 case OVS_KEY_ATTR_CT_MARK: 1170 case OVS_KEY_ATTR_CT_LABELS: 1171 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4: 1172 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6: 1173 err = -EINVAL; 1174 break; 1175 } 1176 1177 return err; 1178 } 1179 1180 static int execute_recirc(struct datapath *dp, struct sk_buff *skb, 1181 struct sw_flow_key *key, 1182 const struct nlattr *a, bool last) 1183 { 1184 u32 recirc_id; 1185 1186 if (!is_flow_key_valid(key)) { 1187 int err; 1188 1189 err = ovs_flow_key_update(skb, key); 1190 if (err) 1191 return err; 1192 } 1193 BUG_ON(!is_flow_key_valid(key)); 1194 1195 recirc_id = nla_get_u32(a); 1196 return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true); 1197 } 1198 1199 static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb, 1200 struct sw_flow_key *key, 1201 const struct nlattr *attr, bool last) 1202 { 1203 struct ovs_skb_cb *ovs_cb = OVS_CB(skb); 1204 const struct nlattr *actions, *cpl_arg; 1205 int len, max_len, rem = nla_len(attr); 1206 const struct check_pkt_len_arg *arg; 1207 bool clone_flow_key; 1208 1209 /* The first netlink attribute in 'attr' is always 1210 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'. 1211 */ 1212 cpl_arg = nla_data(attr); 1213 arg = nla_data(cpl_arg); 1214 1215 len = ovs_cb->mru ? ovs_cb->mru + skb->mac_len : skb->len; 1216 max_len = arg->pkt_len; 1217 1218 if ((skb_is_gso(skb) && skb_gso_validate_mac_len(skb, max_len)) || 1219 len <= max_len) { 1220 /* Second netlink attribute in 'attr' is always 1221 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'. 1222 */ 1223 actions = nla_next(cpl_arg, &rem); 1224 clone_flow_key = !arg->exec_for_lesser_equal; 1225 } else { 1226 /* Third netlink attribute in 'attr' is always 1227 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'. 1228 */ 1229 actions = nla_next(cpl_arg, &rem); 1230 actions = nla_next(actions, &rem); 1231 clone_flow_key = !arg->exec_for_greater; 1232 } 1233 1234 return clone_execute(dp, skb, key, 0, nla_data(actions), 1235 nla_len(actions), last, clone_flow_key); 1236 } 1237 1238 static int execute_dec_ttl(struct sk_buff *skb, struct sw_flow_key *key) 1239 { 1240 int err; 1241 1242 if (skb->protocol == htons(ETH_P_IPV6)) { 1243 struct ipv6hdr *nh; 1244 1245 err = skb_ensure_writable(skb, skb_network_offset(skb) + 1246 sizeof(*nh)); 1247 if (unlikely(err)) 1248 return err; 1249 1250 nh = ipv6_hdr(skb); 1251 1252 if (nh->hop_limit <= 1) 1253 return -EHOSTUNREACH; 1254 1255 key->ip.ttl = --nh->hop_limit; 1256 } else if (skb->protocol == htons(ETH_P_IP)) { 1257 struct iphdr *nh; 1258 u8 old_ttl; 1259 1260 err = skb_ensure_writable(skb, skb_network_offset(skb) + 1261 sizeof(*nh)); 1262 if (unlikely(err)) 1263 return err; 1264 1265 nh = ip_hdr(skb); 1266 if (nh->ttl <= 1) 1267 return -EHOSTUNREACH; 1268 1269 old_ttl = nh->ttl--; 1270 csum_replace2(&nh->check, htons(old_ttl << 8), 1271 htons(nh->ttl << 8)); 1272 key->ip.ttl = nh->ttl; 1273 } 1274 return 0; 1275 } 1276 1277 #if IS_ENABLED(CONFIG_PSAMPLE) 1278 static void execute_psample(struct datapath *dp, struct sk_buff *skb, 1279 const struct nlattr *attr) 1280 { 1281 struct psample_group psample_group = {}; 1282 struct psample_metadata md = {}; 1283 const struct nlattr *a; 1284 u32 rate; 1285 int rem; 1286 1287 nla_for_each_attr(a, nla_data(attr), nla_len(attr), rem) { 1288 switch (nla_type(a)) { 1289 case OVS_PSAMPLE_ATTR_GROUP: 1290 psample_group.group_num = nla_get_u32(a); 1291 break; 1292 1293 case OVS_PSAMPLE_ATTR_COOKIE: 1294 md.user_cookie = nla_data(a); 1295 md.user_cookie_len = nla_len(a); 1296 break; 1297 } 1298 } 1299 1300 psample_group.net = ovs_dp_get_net(dp); 1301 md.in_ifindex = OVS_CB(skb)->input_vport->dev->ifindex; 1302 md.trunc_size = skb->len - OVS_CB(skb)->cutlen; 1303 md.rate_as_probability = 1; 1304 1305 rate = OVS_CB(skb)->probability ? OVS_CB(skb)->probability : U32_MAX; 1306 1307 psample_sample_packet(&psample_group, skb, rate, &md); 1308 } 1309 #else 1310 static void execute_psample(struct datapath *dp, struct sk_buff *skb, 1311 const struct nlattr *attr) 1312 {} 1313 #endif 1314 1315 /* Execute a list of actions against 'skb'. */ 1316 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb, 1317 struct sw_flow_key *key, 1318 const struct nlattr *attr, int len) 1319 { 1320 const struct nlattr *a; 1321 int rem; 1322 1323 for (a = attr, rem = len; rem > 0; 1324 a = nla_next(a, &rem)) { 1325 int err = 0; 1326 1327 if (trace_ovs_do_execute_action_enabled()) 1328 trace_ovs_do_execute_action(dp, skb, key, a, rem); 1329 1330 /* Actions that rightfully have to consume the skb should do it 1331 * and return directly. 1332 */ 1333 switch (nla_type(a)) { 1334 case OVS_ACTION_ATTR_OUTPUT: { 1335 int port = nla_get_u32(a); 1336 struct sk_buff *clone; 1337 1338 /* Every output action needs a separate clone 1339 * of 'skb', In case the output action is the 1340 * last action, cloning can be avoided. 1341 */ 1342 if (nla_is_last(a, rem)) { 1343 do_output(dp, skb, port, key); 1344 /* 'skb' has been used for output. 1345 */ 1346 return 0; 1347 } 1348 1349 clone = skb_clone(skb, GFP_ATOMIC); 1350 if (clone) 1351 do_output(dp, clone, port, key); 1352 OVS_CB(skb)->cutlen = 0; 1353 break; 1354 } 1355 1356 case OVS_ACTION_ATTR_TRUNC: { 1357 struct ovs_action_trunc *trunc = nla_data(a); 1358 1359 if (skb->len > trunc->max_len) 1360 OVS_CB(skb)->cutlen = skb->len - trunc->max_len; 1361 break; 1362 } 1363 1364 case OVS_ACTION_ATTR_USERSPACE: 1365 output_userspace(dp, skb, key, a, attr, 1366 len, OVS_CB(skb)->cutlen); 1367 OVS_CB(skb)->cutlen = 0; 1368 if (nla_is_last(a, rem)) { 1369 consume_skb(skb); 1370 return 0; 1371 } 1372 break; 1373 1374 case OVS_ACTION_ATTR_HASH: 1375 execute_hash(skb, key, a); 1376 break; 1377 1378 case OVS_ACTION_ATTR_PUSH_MPLS: { 1379 struct ovs_action_push_mpls *mpls = nla_data(a); 1380 1381 err = push_mpls(skb, key, mpls->mpls_lse, 1382 mpls->mpls_ethertype, skb->mac_len); 1383 break; 1384 } 1385 case OVS_ACTION_ATTR_ADD_MPLS: { 1386 struct ovs_action_add_mpls *mpls = nla_data(a); 1387 __u16 mac_len = 0; 1388 1389 if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK) 1390 mac_len = skb->mac_len; 1391 1392 err = push_mpls(skb, key, mpls->mpls_lse, 1393 mpls->mpls_ethertype, mac_len); 1394 break; 1395 } 1396 case OVS_ACTION_ATTR_POP_MPLS: 1397 err = pop_mpls(skb, key, nla_get_be16(a)); 1398 break; 1399 1400 case OVS_ACTION_ATTR_PUSH_VLAN: 1401 err = push_vlan(skb, key, nla_data(a)); 1402 break; 1403 1404 case OVS_ACTION_ATTR_POP_VLAN: 1405 err = pop_vlan(skb, key); 1406 break; 1407 1408 case OVS_ACTION_ATTR_RECIRC: { 1409 bool last = nla_is_last(a, rem); 1410 1411 err = execute_recirc(dp, skb, key, a, last); 1412 if (last) { 1413 /* If this is the last action, the skb has 1414 * been consumed or freed. 1415 * Return immediately. 1416 */ 1417 return err; 1418 } 1419 break; 1420 } 1421 1422 case OVS_ACTION_ATTR_SET: 1423 err = execute_set_action(skb, key, nla_data(a)); 1424 break; 1425 1426 case OVS_ACTION_ATTR_SET_MASKED: 1427 case OVS_ACTION_ATTR_SET_TO_MASKED: 1428 err = execute_masked_set_action(skb, key, nla_data(a)); 1429 break; 1430 1431 case OVS_ACTION_ATTR_SAMPLE: { 1432 bool last = nla_is_last(a, rem); 1433 1434 err = sample(dp, skb, key, a, last); 1435 if (last) 1436 return err; 1437 1438 break; 1439 } 1440 1441 case OVS_ACTION_ATTR_CT: 1442 if (!is_flow_key_valid(key)) { 1443 err = ovs_flow_key_update(skb, key); 1444 if (err) 1445 return err; 1446 } 1447 1448 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key, 1449 nla_data(a)); 1450 1451 /* Hide stolen IP fragments from user space. */ 1452 if (err) 1453 return err == -EINPROGRESS ? 0 : err; 1454 break; 1455 1456 case OVS_ACTION_ATTR_CT_CLEAR: 1457 err = ovs_ct_clear(skb, key); 1458 break; 1459 1460 case OVS_ACTION_ATTR_PUSH_ETH: 1461 err = push_eth(skb, key, nla_data(a)); 1462 break; 1463 1464 case OVS_ACTION_ATTR_POP_ETH: 1465 err = pop_eth(skb, key); 1466 break; 1467 1468 case OVS_ACTION_ATTR_PUSH_NSH: 1469 err = push_nsh(skb, key, nla_data(a)); 1470 break; 1471 1472 case OVS_ACTION_ATTR_POP_NSH: 1473 err = pop_nsh(skb, key); 1474 break; 1475 1476 case OVS_ACTION_ATTR_METER: 1477 if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) { 1478 ovs_kfree_skb_reason(skb, OVS_DROP_METER); 1479 return 0; 1480 } 1481 break; 1482 1483 case OVS_ACTION_ATTR_CLONE: { 1484 bool last = nla_is_last(a, rem); 1485 1486 err = clone(dp, skb, key, a, last); 1487 if (last) 1488 return err; 1489 1490 break; 1491 } 1492 1493 case OVS_ACTION_ATTR_CHECK_PKT_LEN: { 1494 bool last = nla_is_last(a, rem); 1495 1496 err = execute_check_pkt_len(dp, skb, key, a, last); 1497 if (last) 1498 return err; 1499 1500 break; 1501 } 1502 1503 case OVS_ACTION_ATTR_DEC_TTL: 1504 err = execute_dec_ttl(skb, key); 1505 if (err == -EHOSTUNREACH) 1506 return dec_ttl_exception_handler(dp, skb, 1507 key, a); 1508 break; 1509 1510 case OVS_ACTION_ATTR_DROP: { 1511 enum ovs_drop_reason reason = nla_get_u32(a) 1512 ? OVS_DROP_EXPLICIT_WITH_ERROR 1513 : OVS_DROP_EXPLICIT; 1514 1515 ovs_kfree_skb_reason(skb, reason); 1516 return 0; 1517 } 1518 1519 case OVS_ACTION_ATTR_PSAMPLE: 1520 execute_psample(dp, skb, a); 1521 OVS_CB(skb)->cutlen = 0; 1522 if (nla_is_last(a, rem)) { 1523 consume_skb(skb); 1524 return 0; 1525 } 1526 break; 1527 } 1528 1529 if (unlikely(err)) { 1530 ovs_kfree_skb_reason(skb, OVS_DROP_ACTION_ERROR); 1531 return err; 1532 } 1533 } 1534 1535 ovs_kfree_skb_reason(skb, OVS_DROP_LAST_ACTION); 1536 return 0; 1537 } 1538 1539 /* Execute the actions on the clone of the packet. The effect of the 1540 * execution does not affect the original 'skb' nor the original 'key'. 1541 * 1542 * The execution may be deferred in case the actions can not be executed 1543 * immediately. 1544 */ 1545 static int clone_execute(struct datapath *dp, struct sk_buff *skb, 1546 struct sw_flow_key *key, u32 recirc_id, 1547 const struct nlattr *actions, int len, 1548 bool last, bool clone_flow_key) 1549 { 1550 struct deferred_action *da; 1551 struct sw_flow_key *clone; 1552 1553 skb = last ? skb : skb_clone(skb, GFP_ATOMIC); 1554 if (!skb) { 1555 /* Out of memory, skip this action. 1556 */ 1557 return 0; 1558 } 1559 1560 /* When clone_flow_key is false, the 'key' will not be change 1561 * by the actions, then the 'key' can be used directly. 1562 * Otherwise, try to clone key from the next recursion level of 1563 * 'flow_keys'. If clone is successful, execute the actions 1564 * without deferring. 1565 */ 1566 clone = clone_flow_key ? clone_key(key) : key; 1567 if (clone) { 1568 int err = 0; 1569 1570 if (actions) { /* Sample action */ 1571 if (clone_flow_key) 1572 __this_cpu_inc(ovs_pcpu_storage.exec_level); 1573 1574 err = do_execute_actions(dp, skb, clone, 1575 actions, len); 1576 1577 if (clone_flow_key) 1578 __this_cpu_dec(ovs_pcpu_storage.exec_level); 1579 } else { /* Recirc action */ 1580 clone->recirc_id = recirc_id; 1581 ovs_dp_process_packet(skb, clone); 1582 } 1583 return err; 1584 } 1585 1586 /* Out of 'flow_keys' space. Defer actions */ 1587 da = add_deferred_actions(skb, key, actions, len); 1588 if (da) { 1589 if (!actions) { /* Recirc action */ 1590 key = &da->pkt_key; 1591 key->recirc_id = recirc_id; 1592 } 1593 } else { 1594 /* Out of per CPU action FIFO space. Drop the 'skb' and 1595 * log an error. 1596 */ 1597 ovs_kfree_skb_reason(skb, OVS_DROP_DEFERRED_LIMIT); 1598 1599 if (net_ratelimit()) { 1600 if (actions) { /* Sample action */ 1601 pr_warn("%s: deferred action limit reached, drop sample action\n", 1602 ovs_dp_name(dp)); 1603 } else { /* Recirc action */ 1604 pr_warn("%s: deferred action limit reached, drop recirc action (recirc_id=%#x)\n", 1605 ovs_dp_name(dp), recirc_id); 1606 } 1607 } 1608 } 1609 return 0; 1610 } 1611 1612 static void process_deferred_actions(struct datapath *dp) 1613 { 1614 struct action_fifo *fifo = this_cpu_ptr(&ovs_pcpu_storage.action_fifos); 1615 1616 /* Do not touch the FIFO in case there is no deferred actions. */ 1617 if (action_fifo_is_empty(fifo)) 1618 return; 1619 1620 /* Finishing executing all deferred actions. */ 1621 do { 1622 struct deferred_action *da = action_fifo_get(fifo); 1623 struct sk_buff *skb = da->skb; 1624 struct sw_flow_key *key = &da->pkt_key; 1625 const struct nlattr *actions = da->actions; 1626 int actions_len = da->actions_len; 1627 1628 if (actions) 1629 do_execute_actions(dp, skb, key, actions, actions_len); 1630 else 1631 ovs_dp_process_packet(skb, key); 1632 } while (!action_fifo_is_empty(fifo)); 1633 1634 /* Reset FIFO for the next packet. */ 1635 action_fifo_init(fifo); 1636 } 1637 1638 /* Execute a list of actions against 'skb'. */ 1639 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb, 1640 const struct sw_flow_actions *acts, 1641 struct sw_flow_key *key) 1642 { 1643 int err, level; 1644 1645 level = __this_cpu_inc_return(ovs_pcpu_storage.exec_level); 1646 if (unlikely(level > OVS_RECURSION_LIMIT)) { 1647 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n", 1648 ovs_dp_name(dp)); 1649 ovs_kfree_skb_reason(skb, OVS_DROP_RECURSION_LIMIT); 1650 err = -ENETDOWN; 1651 goto out; 1652 } 1653 1654 OVS_CB(skb)->acts_origlen = acts->orig_len; 1655 err = do_execute_actions(dp, skb, key, 1656 acts->actions, acts->actions_len); 1657 1658 if (level == 1) 1659 process_deferred_actions(dp); 1660 1661 out: 1662 __this_cpu_dec(ovs_pcpu_storage.exec_level); 1663 return err; 1664 } 1665