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/netfilter_ipv6.h> 13 #include <linux/sctp.h> 14 #include <linux/tcp.h> 15 #include <linux/udp.h> 16 #include <linux/in6.h> 17 #include <linux/if_arp.h> 18 #include <linux/if_vlan.h> 19 20 #include <net/dst.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 #include <net/sctp/checksum.h> 28 29 #include "datapath.h" 30 #include "flow.h" 31 #include "conntrack.h" 32 #include "vport.h" 33 #include "flow_netlink.h" 34 35 struct deferred_action { 36 struct sk_buff *skb; 37 const struct nlattr *actions; 38 int actions_len; 39 40 /* Store pkt_key clone when creating deferred action. */ 41 struct sw_flow_key pkt_key; 42 }; 43 44 #define MAX_L2_LEN (VLAN_ETH_HLEN + 3 * MPLS_HLEN) 45 struct ovs_frag_data { 46 unsigned long dst; 47 struct vport *vport; 48 struct ovs_skb_cb cb; 49 __be16 inner_protocol; 50 u16 network_offset; /* valid only for MPLS */ 51 u16 vlan_tci; 52 __be16 vlan_proto; 53 unsigned int l2_len; 54 u8 mac_proto; 55 u8 l2_data[MAX_L2_LEN]; 56 }; 57 58 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage); 59 60 #define DEFERRED_ACTION_FIFO_SIZE 10 61 #define OVS_RECURSION_LIMIT 5 62 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2) 63 struct action_fifo { 64 int head; 65 int tail; 66 /* Deferred action fifo queue storage. */ 67 struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE]; 68 }; 69 70 struct action_flow_keys { 71 struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD]; 72 }; 73 74 static struct action_fifo __percpu *action_fifos; 75 static struct action_flow_keys __percpu *flow_keys; 76 static DEFINE_PER_CPU(int, exec_actions_level); 77 78 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys' 79 * space. Return NULL if out of key spaces. 80 */ 81 static struct sw_flow_key *clone_key(const struct sw_flow_key *key_) 82 { 83 struct action_flow_keys *keys = this_cpu_ptr(flow_keys); 84 int level = this_cpu_read(exec_actions_level); 85 struct sw_flow_key *key = NULL; 86 87 if (level <= OVS_DEFERRED_ACTION_THRESHOLD) { 88 key = &keys->key[level - 1]; 89 *key = *key_; 90 } 91 92 return key; 93 } 94 95 static void action_fifo_init(struct action_fifo *fifo) 96 { 97 fifo->head = 0; 98 fifo->tail = 0; 99 } 100 101 static bool action_fifo_is_empty(const struct action_fifo *fifo) 102 { 103 return (fifo->head == fifo->tail); 104 } 105 106 static struct deferred_action *action_fifo_get(struct action_fifo *fifo) 107 { 108 if (action_fifo_is_empty(fifo)) 109 return NULL; 110 111 return &fifo->fifo[fifo->tail++]; 112 } 113 114 static struct deferred_action *action_fifo_put(struct action_fifo *fifo) 115 { 116 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1) 117 return NULL; 118 119 return &fifo->fifo[fifo->head++]; 120 } 121 122 /* Return true if fifo is not full */ 123 static struct deferred_action *add_deferred_actions(struct sk_buff *skb, 124 const struct sw_flow_key *key, 125 const struct nlattr *actions, 126 const int actions_len) 127 { 128 struct action_fifo *fifo; 129 struct deferred_action *da; 130 131 fifo = this_cpu_ptr(action_fifos); 132 da = action_fifo_put(fifo); 133 if (da) { 134 da->skb = skb; 135 da->actions = actions; 136 da->actions_len = actions_len; 137 da->pkt_key = *key; 138 } 139 140 return da; 141 } 142 143 static void invalidate_flow_key(struct sw_flow_key *key) 144 { 145 key->mac_proto |= SW_FLOW_KEY_INVALID; 146 } 147 148 static bool is_flow_key_valid(const struct sw_flow_key *key) 149 { 150 return !(key->mac_proto & SW_FLOW_KEY_INVALID); 151 } 152 153 static int clone_execute(struct datapath *dp, struct sk_buff *skb, 154 struct sw_flow_key *key, 155 u32 recirc_id, 156 const struct nlattr *actions, int len, 157 bool last, bool clone_flow_key); 158 159 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb, 160 struct sw_flow_key *key, 161 const struct nlattr *attr, int len); 162 163 static void update_ethertype(struct sk_buff *skb, struct ethhdr *hdr, 164 __be16 ethertype) 165 { 166 if (skb->ip_summed == CHECKSUM_COMPLETE) { 167 __be16 diff[] = { ~(hdr->h_proto), ethertype }; 168 169 skb->csum = ~csum_partial((char *)diff, sizeof(diff), 170 ~skb->csum); 171 } 172 173 hdr->h_proto = ethertype; 174 } 175 176 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key, 177 const struct ovs_action_push_mpls *mpls) 178 { 179 struct mpls_shim_hdr *new_mpls_lse; 180 181 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */ 182 if (skb->encapsulation) 183 return -ENOTSUPP; 184 185 if (skb_cow_head(skb, MPLS_HLEN) < 0) 186 return -ENOMEM; 187 188 if (!skb->inner_protocol) { 189 skb_set_inner_network_header(skb, skb->mac_len); 190 skb_set_inner_protocol(skb, skb->protocol); 191 } 192 193 skb_push(skb, MPLS_HLEN); 194 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb), 195 skb->mac_len); 196 skb_reset_mac_header(skb); 197 skb_set_network_header(skb, skb->mac_len); 198 199 new_mpls_lse = mpls_hdr(skb); 200 new_mpls_lse->label_stack_entry = mpls->mpls_lse; 201 202 skb_postpush_rcsum(skb, new_mpls_lse, MPLS_HLEN); 203 204 if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET) 205 update_ethertype(skb, eth_hdr(skb), mpls->mpls_ethertype); 206 skb->protocol = mpls->mpls_ethertype; 207 208 invalidate_flow_key(key); 209 return 0; 210 } 211 212 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key, 213 const __be16 ethertype) 214 { 215 int err; 216 217 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN); 218 if (unlikely(err)) 219 return err; 220 221 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN); 222 223 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb), 224 skb->mac_len); 225 226 __skb_pull(skb, MPLS_HLEN); 227 skb_reset_mac_header(skb); 228 skb_set_network_header(skb, skb->mac_len); 229 230 if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET) { 231 struct ethhdr *hdr; 232 233 /* mpls_hdr() is used to locate the ethertype field correctly in the 234 * presence of VLAN tags. 235 */ 236 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN); 237 update_ethertype(skb, hdr, ethertype); 238 } 239 if (eth_p_mpls(skb->protocol)) 240 skb->protocol = ethertype; 241 242 invalidate_flow_key(key); 243 return 0; 244 } 245 246 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key, 247 const __be32 *mpls_lse, const __be32 *mask) 248 { 249 struct mpls_shim_hdr *stack; 250 __be32 lse; 251 int err; 252 253 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN); 254 if (unlikely(err)) 255 return err; 256 257 stack = mpls_hdr(skb); 258 lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask); 259 if (skb->ip_summed == CHECKSUM_COMPLETE) { 260 __be32 diff[] = { ~(stack->label_stack_entry), lse }; 261 262 skb->csum = ~csum_partial((char *)diff, sizeof(diff), 263 ~skb->csum); 264 } 265 266 stack->label_stack_entry = lse; 267 flow_key->mpls.top_lse = lse; 268 return 0; 269 } 270 271 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key) 272 { 273 int err; 274 275 err = skb_vlan_pop(skb); 276 if (skb_vlan_tag_present(skb)) { 277 invalidate_flow_key(key); 278 } else { 279 key->eth.vlan.tci = 0; 280 key->eth.vlan.tpid = 0; 281 } 282 return err; 283 } 284 285 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key, 286 const struct ovs_action_push_vlan *vlan) 287 { 288 if (skb_vlan_tag_present(skb)) { 289 invalidate_flow_key(key); 290 } else { 291 key->eth.vlan.tci = vlan->vlan_tci; 292 key->eth.vlan.tpid = vlan->vlan_tpid; 293 } 294 return skb_vlan_push(skb, vlan->vlan_tpid, 295 ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK); 296 } 297 298 /* 'src' is already properly masked. */ 299 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_) 300 { 301 u16 *dst = (u16 *)dst_; 302 const u16 *src = (const u16 *)src_; 303 const u16 *mask = (const u16 *)mask_; 304 305 OVS_SET_MASKED(dst[0], src[0], mask[0]); 306 OVS_SET_MASKED(dst[1], src[1], mask[1]); 307 OVS_SET_MASKED(dst[2], src[2], mask[2]); 308 } 309 310 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key, 311 const struct ovs_key_ethernet *key, 312 const struct ovs_key_ethernet *mask) 313 { 314 int err; 315 316 err = skb_ensure_writable(skb, ETH_HLEN); 317 if (unlikely(err)) 318 return err; 319 320 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2); 321 322 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src, 323 mask->eth_src); 324 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst, 325 mask->eth_dst); 326 327 skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2); 328 329 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source); 330 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest); 331 return 0; 332 } 333 334 /* pop_eth does not support VLAN packets as this action is never called 335 * for them. 336 */ 337 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key) 338 { 339 skb_pull_rcsum(skb, ETH_HLEN); 340 skb_reset_mac_header(skb); 341 skb_reset_mac_len(skb); 342 343 /* safe right before invalidate_flow_key */ 344 key->mac_proto = MAC_PROTO_NONE; 345 invalidate_flow_key(key); 346 return 0; 347 } 348 349 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key, 350 const struct ovs_action_push_eth *ethh) 351 { 352 struct ethhdr *hdr; 353 354 /* Add the new Ethernet header */ 355 if (skb_cow_head(skb, ETH_HLEN) < 0) 356 return -ENOMEM; 357 358 skb_push(skb, ETH_HLEN); 359 skb_reset_mac_header(skb); 360 skb_reset_mac_len(skb); 361 362 hdr = eth_hdr(skb); 363 ether_addr_copy(hdr->h_source, ethh->addresses.eth_src); 364 ether_addr_copy(hdr->h_dest, ethh->addresses.eth_dst); 365 hdr->h_proto = skb->protocol; 366 367 skb_postpush_rcsum(skb, hdr, ETH_HLEN); 368 369 /* safe right before invalidate_flow_key */ 370 key->mac_proto = MAC_PROTO_ETHERNET; 371 invalidate_flow_key(key); 372 return 0; 373 } 374 375 static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key, 376 const struct nshhdr *nh) 377 { 378 int err; 379 380 err = nsh_push(skb, nh); 381 if (err) 382 return err; 383 384 /* safe right before invalidate_flow_key */ 385 key->mac_proto = MAC_PROTO_NONE; 386 invalidate_flow_key(key); 387 return 0; 388 } 389 390 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key) 391 { 392 int err; 393 394 err = nsh_pop(skb); 395 if (err) 396 return err; 397 398 /* safe right before invalidate_flow_key */ 399 if (skb->protocol == htons(ETH_P_TEB)) 400 key->mac_proto = MAC_PROTO_ETHERNET; 401 else 402 key->mac_proto = MAC_PROTO_NONE; 403 invalidate_flow_key(key); 404 return 0; 405 } 406 407 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh, 408 __be32 addr, __be32 new_addr) 409 { 410 int transport_len = skb->len - skb_transport_offset(skb); 411 412 if (nh->frag_off & htons(IP_OFFSET)) 413 return; 414 415 if (nh->protocol == IPPROTO_TCP) { 416 if (likely(transport_len >= sizeof(struct tcphdr))) 417 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb, 418 addr, new_addr, true); 419 } else if (nh->protocol == IPPROTO_UDP) { 420 if (likely(transport_len >= sizeof(struct udphdr))) { 421 struct udphdr *uh = udp_hdr(skb); 422 423 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) { 424 inet_proto_csum_replace4(&uh->check, skb, 425 addr, new_addr, true); 426 if (!uh->check) 427 uh->check = CSUM_MANGLED_0; 428 } 429 } 430 } 431 } 432 433 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh, 434 __be32 *addr, __be32 new_addr) 435 { 436 update_ip_l4_checksum(skb, nh, *addr, new_addr); 437 csum_replace4(&nh->check, *addr, new_addr); 438 skb_clear_hash(skb); 439 *addr = new_addr; 440 } 441 442 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto, 443 __be32 addr[4], const __be32 new_addr[4]) 444 { 445 int transport_len = skb->len - skb_transport_offset(skb); 446 447 if (l4_proto == NEXTHDR_TCP) { 448 if (likely(transport_len >= sizeof(struct tcphdr))) 449 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb, 450 addr, new_addr, true); 451 } else if (l4_proto == NEXTHDR_UDP) { 452 if (likely(transport_len >= sizeof(struct udphdr))) { 453 struct udphdr *uh = udp_hdr(skb); 454 455 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) { 456 inet_proto_csum_replace16(&uh->check, skb, 457 addr, new_addr, true); 458 if (!uh->check) 459 uh->check = CSUM_MANGLED_0; 460 } 461 } 462 } else if (l4_proto == NEXTHDR_ICMP) { 463 if (likely(transport_len >= sizeof(struct icmp6hdr))) 464 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum, 465 skb, addr, new_addr, true); 466 } 467 } 468 469 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4], 470 const __be32 mask[4], __be32 masked[4]) 471 { 472 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]); 473 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]); 474 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]); 475 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]); 476 } 477 478 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto, 479 __be32 addr[4], const __be32 new_addr[4], 480 bool recalculate_csum) 481 { 482 if (recalculate_csum) 483 update_ipv6_checksum(skb, l4_proto, addr, new_addr); 484 485 skb_clear_hash(skb); 486 memcpy(addr, new_addr, sizeof(__be32[4])); 487 } 488 489 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask) 490 { 491 /* Bits 21-24 are always unmasked, so this retains their values. */ 492 OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16)); 493 OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8)); 494 OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask); 495 } 496 497 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl, 498 u8 mask) 499 { 500 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask); 501 502 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8)); 503 nh->ttl = new_ttl; 504 } 505 506 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key, 507 const struct ovs_key_ipv4 *key, 508 const struct ovs_key_ipv4 *mask) 509 { 510 struct iphdr *nh; 511 __be32 new_addr; 512 int err; 513 514 err = skb_ensure_writable(skb, skb_network_offset(skb) + 515 sizeof(struct iphdr)); 516 if (unlikely(err)) 517 return err; 518 519 nh = ip_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 (mask->ipv4_src) { 526 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src); 527 528 if (unlikely(new_addr != nh->saddr)) { 529 set_ip_addr(skb, nh, &nh->saddr, new_addr); 530 flow_key->ipv4.addr.src = new_addr; 531 } 532 } 533 if (mask->ipv4_dst) { 534 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst); 535 536 if (unlikely(new_addr != nh->daddr)) { 537 set_ip_addr(skb, nh, &nh->daddr, new_addr); 538 flow_key->ipv4.addr.dst = new_addr; 539 } 540 } 541 if (mask->ipv4_tos) { 542 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos); 543 flow_key->ip.tos = nh->tos; 544 } 545 if (mask->ipv4_ttl) { 546 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl); 547 flow_key->ip.ttl = nh->ttl; 548 } 549 550 return 0; 551 } 552 553 static bool is_ipv6_mask_nonzero(const __be32 addr[4]) 554 { 555 return !!(addr[0] | addr[1] | addr[2] | addr[3]); 556 } 557 558 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key, 559 const struct ovs_key_ipv6 *key, 560 const struct ovs_key_ipv6 *mask) 561 { 562 struct ipv6hdr *nh; 563 int err; 564 565 err = skb_ensure_writable(skb, skb_network_offset(skb) + 566 sizeof(struct ipv6hdr)); 567 if (unlikely(err)) 568 return err; 569 570 nh = ipv6_hdr(skb); 571 572 /* Setting an IP addresses is typically only a side effect of 573 * matching on them in the current userspace implementation, so it 574 * makes sense to check if the value actually changed. 575 */ 576 if (is_ipv6_mask_nonzero(mask->ipv6_src)) { 577 __be32 *saddr = (__be32 *)&nh->saddr; 578 __be32 masked[4]; 579 580 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked); 581 582 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) { 583 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked, 584 true); 585 memcpy(&flow_key->ipv6.addr.src, masked, 586 sizeof(flow_key->ipv6.addr.src)); 587 } 588 } 589 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) { 590 unsigned int offset = 0; 591 int flags = IP6_FH_F_SKIP_RH; 592 bool recalc_csum = true; 593 __be32 *daddr = (__be32 *)&nh->daddr; 594 __be32 masked[4]; 595 596 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked); 597 598 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) { 599 if (ipv6_ext_hdr(nh->nexthdr)) 600 recalc_csum = (ipv6_find_hdr(skb, &offset, 601 NEXTHDR_ROUTING, 602 NULL, &flags) 603 != NEXTHDR_ROUTING); 604 605 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked, 606 recalc_csum); 607 memcpy(&flow_key->ipv6.addr.dst, masked, 608 sizeof(flow_key->ipv6.addr.dst)); 609 } 610 } 611 if (mask->ipv6_tclass) { 612 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass); 613 flow_key->ip.tos = ipv6_get_dsfield(nh); 614 } 615 if (mask->ipv6_label) { 616 set_ipv6_fl(nh, ntohl(key->ipv6_label), 617 ntohl(mask->ipv6_label)); 618 flow_key->ipv6.label = 619 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL); 620 } 621 if (mask->ipv6_hlimit) { 622 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit, 623 mask->ipv6_hlimit); 624 flow_key->ip.ttl = nh->hop_limit; 625 } 626 return 0; 627 } 628 629 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key, 630 const struct nlattr *a) 631 { 632 struct nshhdr *nh; 633 size_t length; 634 int err; 635 u8 flags; 636 u8 ttl; 637 int i; 638 639 struct ovs_key_nsh key; 640 struct ovs_key_nsh mask; 641 642 err = nsh_key_from_nlattr(a, &key, &mask); 643 if (err) 644 return err; 645 646 /* Make sure the NSH base header is there */ 647 if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN)) 648 return -ENOMEM; 649 650 nh = nsh_hdr(skb); 651 length = nsh_hdr_len(nh); 652 653 /* Make sure the whole NSH header is there */ 654 err = skb_ensure_writable(skb, skb_network_offset(skb) + 655 length); 656 if (unlikely(err)) 657 return err; 658 659 nh = nsh_hdr(skb); 660 skb_postpull_rcsum(skb, nh, length); 661 flags = nsh_get_flags(nh); 662 flags = OVS_MASKED(flags, key.base.flags, mask.base.flags); 663 flow_key->nsh.base.flags = flags; 664 ttl = nsh_get_ttl(nh); 665 ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl); 666 flow_key->nsh.base.ttl = ttl; 667 nsh_set_flags_and_ttl(nh, flags, ttl); 668 nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr, 669 mask.base.path_hdr); 670 flow_key->nsh.base.path_hdr = nh->path_hdr; 671 switch (nh->mdtype) { 672 case NSH_M_TYPE1: 673 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) { 674 nh->md1.context[i] = 675 OVS_MASKED(nh->md1.context[i], key.context[i], 676 mask.context[i]); 677 } 678 memcpy(flow_key->nsh.context, nh->md1.context, 679 sizeof(nh->md1.context)); 680 break; 681 case NSH_M_TYPE2: 682 memset(flow_key->nsh.context, 0, 683 sizeof(flow_key->nsh.context)); 684 break; 685 default: 686 return -EINVAL; 687 } 688 skb_postpush_rcsum(skb, nh, length); 689 return 0; 690 } 691 692 /* Must follow skb_ensure_writable() since that can move the skb data. */ 693 static void set_tp_port(struct sk_buff *skb, __be16 *port, 694 __be16 new_port, __sum16 *check) 695 { 696 inet_proto_csum_replace2(check, skb, *port, new_port, false); 697 *port = new_port; 698 } 699 700 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key, 701 const struct ovs_key_udp *key, 702 const struct ovs_key_udp *mask) 703 { 704 struct udphdr *uh; 705 __be16 src, dst; 706 int err; 707 708 err = skb_ensure_writable(skb, skb_transport_offset(skb) + 709 sizeof(struct udphdr)); 710 if (unlikely(err)) 711 return err; 712 713 uh = udp_hdr(skb); 714 /* Either of the masks is non-zero, so do not bother checking them. */ 715 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src); 716 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst); 717 718 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) { 719 if (likely(src != uh->source)) { 720 set_tp_port(skb, &uh->source, src, &uh->check); 721 flow_key->tp.src = src; 722 } 723 if (likely(dst != uh->dest)) { 724 set_tp_port(skb, &uh->dest, dst, &uh->check); 725 flow_key->tp.dst = dst; 726 } 727 728 if (unlikely(!uh->check)) 729 uh->check = CSUM_MANGLED_0; 730 } else { 731 uh->source = src; 732 uh->dest = dst; 733 flow_key->tp.src = src; 734 flow_key->tp.dst = dst; 735 } 736 737 skb_clear_hash(skb); 738 739 return 0; 740 } 741 742 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key, 743 const struct ovs_key_tcp *key, 744 const struct ovs_key_tcp *mask) 745 { 746 struct tcphdr *th; 747 __be16 src, dst; 748 int err; 749 750 err = skb_ensure_writable(skb, skb_transport_offset(skb) + 751 sizeof(struct tcphdr)); 752 if (unlikely(err)) 753 return err; 754 755 th = tcp_hdr(skb); 756 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src); 757 if (likely(src != th->source)) { 758 set_tp_port(skb, &th->source, src, &th->check); 759 flow_key->tp.src = src; 760 } 761 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst); 762 if (likely(dst != th->dest)) { 763 set_tp_port(skb, &th->dest, dst, &th->check); 764 flow_key->tp.dst = dst; 765 } 766 skb_clear_hash(skb); 767 768 return 0; 769 } 770 771 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key, 772 const struct ovs_key_sctp *key, 773 const struct ovs_key_sctp *mask) 774 { 775 unsigned int sctphoff = skb_transport_offset(skb); 776 struct sctphdr *sh; 777 __le32 old_correct_csum, new_csum, old_csum; 778 int err; 779 780 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr)); 781 if (unlikely(err)) 782 return err; 783 784 sh = sctp_hdr(skb); 785 old_csum = sh->checksum; 786 old_correct_csum = sctp_compute_cksum(skb, sctphoff); 787 788 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src); 789 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst); 790 791 new_csum = sctp_compute_cksum(skb, sctphoff); 792 793 /* Carry any checksum errors through. */ 794 sh->checksum = old_csum ^ old_correct_csum ^ new_csum; 795 796 skb_clear_hash(skb); 797 flow_key->tp.src = sh->source; 798 flow_key->tp.dst = sh->dest; 799 800 return 0; 801 } 802 803 static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb) 804 { 805 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage); 806 struct vport *vport = data->vport; 807 808 if (skb_cow_head(skb, data->l2_len) < 0) { 809 kfree_skb(skb); 810 return -ENOMEM; 811 } 812 813 __skb_dst_copy(skb, data->dst); 814 *OVS_CB(skb) = data->cb; 815 skb->inner_protocol = data->inner_protocol; 816 if (data->vlan_tci & VLAN_CFI_MASK) 817 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK); 818 else 819 __vlan_hwaccel_clear_tag(skb); 820 821 /* Reconstruct the MAC header. */ 822 skb_push(skb, data->l2_len); 823 memcpy(skb->data, &data->l2_data, data->l2_len); 824 skb_postpush_rcsum(skb, skb->data, data->l2_len); 825 skb_reset_mac_header(skb); 826 827 if (eth_p_mpls(skb->protocol)) { 828 skb->inner_network_header = skb->network_header; 829 skb_set_network_header(skb, data->network_offset); 830 skb_reset_mac_len(skb); 831 } 832 833 ovs_vport_send(vport, skb, data->mac_proto); 834 return 0; 835 } 836 837 static unsigned int 838 ovs_dst_get_mtu(const struct dst_entry *dst) 839 { 840 return dst->dev->mtu; 841 } 842 843 static struct dst_ops ovs_dst_ops = { 844 .family = AF_UNSPEC, 845 .mtu = ovs_dst_get_mtu, 846 }; 847 848 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is 849 * ovs_vport_output(), which is called once per fragmented packet. 850 */ 851 static void prepare_frag(struct vport *vport, struct sk_buff *skb, 852 u16 orig_network_offset, u8 mac_proto) 853 { 854 unsigned int hlen = skb_network_offset(skb); 855 struct ovs_frag_data *data; 856 857 data = this_cpu_ptr(&ovs_frag_data_storage); 858 data->dst = skb->_skb_refdst; 859 data->vport = vport; 860 data->cb = *OVS_CB(skb); 861 data->inner_protocol = skb->inner_protocol; 862 data->network_offset = orig_network_offset; 863 if (skb_vlan_tag_present(skb)) 864 data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK; 865 else 866 data->vlan_tci = 0; 867 data->vlan_proto = skb->vlan_proto; 868 data->mac_proto = mac_proto; 869 data->l2_len = hlen; 870 memcpy(&data->l2_data, skb->data, hlen); 871 872 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); 873 skb_pull(skb, hlen); 874 } 875 876 static void ovs_fragment(struct net *net, struct vport *vport, 877 struct sk_buff *skb, u16 mru, 878 struct sw_flow_key *key) 879 { 880 u16 orig_network_offset = 0; 881 882 if (eth_p_mpls(skb->protocol)) { 883 orig_network_offset = skb_network_offset(skb); 884 skb->network_header = skb->inner_network_header; 885 } 886 887 if (skb_network_offset(skb) > MAX_L2_LEN) { 888 OVS_NLERR(1, "L2 header too long to fragment"); 889 goto err; 890 } 891 892 if (key->eth.type == htons(ETH_P_IP)) { 893 struct dst_entry ovs_dst; 894 unsigned long orig_dst; 895 896 prepare_frag(vport, skb, orig_network_offset, 897 ovs_key_mac_proto(key)); 898 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1, 899 DST_OBSOLETE_NONE, DST_NOCOUNT); 900 ovs_dst.dev = vport->dev; 901 902 orig_dst = skb->_skb_refdst; 903 skb_dst_set_noref(skb, &ovs_dst); 904 IPCB(skb)->frag_max_size = mru; 905 906 ip_do_fragment(net, skb->sk, skb, ovs_vport_output); 907 refdst_drop(orig_dst); 908 } else if (key->eth.type == htons(ETH_P_IPV6)) { 909 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops(); 910 unsigned long orig_dst; 911 struct rt6_info ovs_rt; 912 913 if (!v6ops) 914 goto err; 915 916 prepare_frag(vport, skb, orig_network_offset, 917 ovs_key_mac_proto(key)); 918 memset(&ovs_rt, 0, sizeof(ovs_rt)); 919 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1, 920 DST_OBSOLETE_NONE, DST_NOCOUNT); 921 ovs_rt.dst.dev = vport->dev; 922 923 orig_dst = skb->_skb_refdst; 924 skb_dst_set_noref(skb, &ovs_rt.dst); 925 IP6CB(skb)->frag_max_size = mru; 926 927 v6ops->fragment(net, skb->sk, skb, ovs_vport_output); 928 refdst_drop(orig_dst); 929 } else { 930 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.", 931 ovs_vport_name(vport), ntohs(key->eth.type), mru, 932 vport->dev->mtu); 933 goto err; 934 } 935 936 return; 937 err: 938 kfree_skb(skb); 939 } 940 941 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port, 942 struct sw_flow_key *key) 943 { 944 struct vport *vport = ovs_vport_rcu(dp, out_port); 945 946 if (likely(vport)) { 947 u16 mru = OVS_CB(skb)->mru; 948 u32 cutlen = OVS_CB(skb)->cutlen; 949 950 if (unlikely(cutlen > 0)) { 951 if (skb->len - cutlen > ovs_mac_header_len(key)) 952 pskb_trim(skb, skb->len - cutlen); 953 else 954 pskb_trim(skb, ovs_mac_header_len(key)); 955 } 956 957 if (likely(!mru || 958 (skb->len <= mru + vport->dev->hard_header_len))) { 959 ovs_vport_send(vport, skb, ovs_key_mac_proto(key)); 960 } else if (mru <= vport->dev->mtu) { 961 struct net *net = read_pnet(&dp->net); 962 963 ovs_fragment(net, vport, skb, mru, key); 964 } else { 965 kfree_skb(skb); 966 } 967 } else { 968 kfree_skb(skb); 969 } 970 } 971 972 static int output_userspace(struct datapath *dp, struct sk_buff *skb, 973 struct sw_flow_key *key, const struct nlattr *attr, 974 const struct nlattr *actions, int actions_len, 975 uint32_t cutlen) 976 { 977 struct dp_upcall_info upcall; 978 const struct nlattr *a; 979 int rem; 980 981 memset(&upcall, 0, sizeof(upcall)); 982 upcall.cmd = OVS_PACKET_CMD_ACTION; 983 upcall.mru = OVS_CB(skb)->mru; 984 985 for (a = nla_data(attr), rem = nla_len(attr); rem > 0; 986 a = nla_next(a, &rem)) { 987 switch (nla_type(a)) { 988 case OVS_USERSPACE_ATTR_USERDATA: 989 upcall.userdata = a; 990 break; 991 992 case OVS_USERSPACE_ATTR_PID: 993 upcall.portid = nla_get_u32(a); 994 break; 995 996 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: { 997 /* Get out tunnel info. */ 998 struct vport *vport; 999 1000 vport = ovs_vport_rcu(dp, nla_get_u32(a)); 1001 if (vport) { 1002 int err; 1003 1004 err = dev_fill_metadata_dst(vport->dev, skb); 1005 if (!err) 1006 upcall.egress_tun_info = skb_tunnel_info(skb); 1007 } 1008 1009 break; 1010 } 1011 1012 case OVS_USERSPACE_ATTR_ACTIONS: { 1013 /* Include actions. */ 1014 upcall.actions = actions; 1015 upcall.actions_len = actions_len; 1016 break; 1017 } 1018 1019 } /* End of switch. */ 1020 } 1021 1022 return ovs_dp_upcall(dp, skb, key, &upcall, cutlen); 1023 } 1024 1025 /* When 'last' is true, sample() should always consume the 'skb'. 1026 * Otherwise, sample() should keep 'skb' intact regardless what 1027 * actions are executed within sample(). 1028 */ 1029 static int sample(struct datapath *dp, struct sk_buff *skb, 1030 struct sw_flow_key *key, const struct nlattr *attr, 1031 bool last) 1032 { 1033 struct nlattr *actions; 1034 struct nlattr *sample_arg; 1035 int rem = nla_len(attr); 1036 const struct sample_arg *arg; 1037 bool clone_flow_key; 1038 1039 /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */ 1040 sample_arg = nla_data(attr); 1041 arg = nla_data(sample_arg); 1042 actions = nla_next(sample_arg, &rem); 1043 1044 if ((arg->probability != U32_MAX) && 1045 (!arg->probability || prandom_u32() > arg->probability)) { 1046 if (last) 1047 consume_skb(skb); 1048 return 0; 1049 } 1050 1051 clone_flow_key = !arg->exec; 1052 return clone_execute(dp, skb, key, 0, actions, rem, last, 1053 clone_flow_key); 1054 } 1055 1056 /* When 'last' is true, clone() should always consume the 'skb'. 1057 * Otherwise, clone() should keep 'skb' intact regardless what 1058 * actions are executed within clone(). 1059 */ 1060 static int clone(struct datapath *dp, struct sk_buff *skb, 1061 struct sw_flow_key *key, const struct nlattr *attr, 1062 bool last) 1063 { 1064 struct nlattr *actions; 1065 struct nlattr *clone_arg; 1066 int rem = nla_len(attr); 1067 bool dont_clone_flow_key; 1068 1069 /* The first action is always 'OVS_CLONE_ATTR_ARG'. */ 1070 clone_arg = nla_data(attr); 1071 dont_clone_flow_key = nla_get_u32(clone_arg); 1072 actions = nla_next(clone_arg, &rem); 1073 1074 return clone_execute(dp, skb, key, 0, actions, rem, last, 1075 !dont_clone_flow_key); 1076 } 1077 1078 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key, 1079 const struct nlattr *attr) 1080 { 1081 struct ovs_action_hash *hash_act = nla_data(attr); 1082 u32 hash = 0; 1083 1084 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */ 1085 hash = skb_get_hash(skb); 1086 hash = jhash_1word(hash, hash_act->hash_basis); 1087 if (!hash) 1088 hash = 0x1; 1089 1090 key->ovs_flow_hash = hash; 1091 } 1092 1093 static int execute_set_action(struct sk_buff *skb, 1094 struct sw_flow_key *flow_key, 1095 const struct nlattr *a) 1096 { 1097 /* Only tunnel set execution is supported without a mask. */ 1098 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) { 1099 struct ovs_tunnel_info *tun = nla_data(a); 1100 1101 skb_dst_drop(skb); 1102 dst_hold((struct dst_entry *)tun->tun_dst); 1103 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst); 1104 return 0; 1105 } 1106 1107 return -EINVAL; 1108 } 1109 1110 /* Mask is at the midpoint of the data. */ 1111 #define get_mask(a, type) ((const type)nla_data(a) + 1) 1112 1113 static int execute_masked_set_action(struct sk_buff *skb, 1114 struct sw_flow_key *flow_key, 1115 const struct nlattr *a) 1116 { 1117 int err = 0; 1118 1119 switch (nla_type(a)) { 1120 case OVS_KEY_ATTR_PRIORITY: 1121 OVS_SET_MASKED(skb->priority, nla_get_u32(a), 1122 *get_mask(a, u32 *)); 1123 flow_key->phy.priority = skb->priority; 1124 break; 1125 1126 case OVS_KEY_ATTR_SKB_MARK: 1127 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *)); 1128 flow_key->phy.skb_mark = skb->mark; 1129 break; 1130 1131 case OVS_KEY_ATTR_TUNNEL_INFO: 1132 /* Masked data not supported for tunnel. */ 1133 err = -EINVAL; 1134 break; 1135 1136 case OVS_KEY_ATTR_ETHERNET: 1137 err = set_eth_addr(skb, flow_key, nla_data(a), 1138 get_mask(a, struct ovs_key_ethernet *)); 1139 break; 1140 1141 case OVS_KEY_ATTR_NSH: 1142 err = set_nsh(skb, flow_key, a); 1143 break; 1144 1145 case OVS_KEY_ATTR_IPV4: 1146 err = set_ipv4(skb, flow_key, nla_data(a), 1147 get_mask(a, struct ovs_key_ipv4 *)); 1148 break; 1149 1150 case OVS_KEY_ATTR_IPV6: 1151 err = set_ipv6(skb, flow_key, nla_data(a), 1152 get_mask(a, struct ovs_key_ipv6 *)); 1153 break; 1154 1155 case OVS_KEY_ATTR_TCP: 1156 err = set_tcp(skb, flow_key, nla_data(a), 1157 get_mask(a, struct ovs_key_tcp *)); 1158 break; 1159 1160 case OVS_KEY_ATTR_UDP: 1161 err = set_udp(skb, flow_key, nla_data(a), 1162 get_mask(a, struct ovs_key_udp *)); 1163 break; 1164 1165 case OVS_KEY_ATTR_SCTP: 1166 err = set_sctp(skb, flow_key, nla_data(a), 1167 get_mask(a, struct ovs_key_sctp *)); 1168 break; 1169 1170 case OVS_KEY_ATTR_MPLS: 1171 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a, 1172 __be32 *)); 1173 break; 1174 1175 case OVS_KEY_ATTR_CT_STATE: 1176 case OVS_KEY_ATTR_CT_ZONE: 1177 case OVS_KEY_ATTR_CT_MARK: 1178 case OVS_KEY_ATTR_CT_LABELS: 1179 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4: 1180 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6: 1181 err = -EINVAL; 1182 break; 1183 } 1184 1185 return err; 1186 } 1187 1188 static int execute_recirc(struct datapath *dp, struct sk_buff *skb, 1189 struct sw_flow_key *key, 1190 const struct nlattr *a, bool last) 1191 { 1192 u32 recirc_id; 1193 1194 if (!is_flow_key_valid(key)) { 1195 int err; 1196 1197 err = ovs_flow_key_update(skb, key); 1198 if (err) 1199 return err; 1200 } 1201 BUG_ON(!is_flow_key_valid(key)); 1202 1203 recirc_id = nla_get_u32(a); 1204 return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true); 1205 } 1206 1207 static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb, 1208 struct sw_flow_key *key, 1209 const struct nlattr *attr, bool last) 1210 { 1211 const struct nlattr *actions, *cpl_arg; 1212 const struct check_pkt_len_arg *arg; 1213 int rem = nla_len(attr); 1214 bool clone_flow_key; 1215 1216 /* The first netlink attribute in 'attr' is always 1217 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'. 1218 */ 1219 cpl_arg = nla_data(attr); 1220 arg = nla_data(cpl_arg); 1221 1222 if (skb->len <= arg->pkt_len) { 1223 /* Second netlink attribute in 'attr' is always 1224 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'. 1225 */ 1226 actions = nla_next(cpl_arg, &rem); 1227 clone_flow_key = !arg->exec_for_lesser_equal; 1228 } else { 1229 /* Third netlink attribute in 'attr' is always 1230 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'. 1231 */ 1232 actions = nla_next(cpl_arg, &rem); 1233 actions = nla_next(actions, &rem); 1234 clone_flow_key = !arg->exec_for_greater; 1235 } 1236 1237 return clone_execute(dp, skb, key, 0, nla_data(actions), 1238 nla_len(actions), last, clone_flow_key); 1239 } 1240 1241 /* Execute a list of actions against 'skb'. */ 1242 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb, 1243 struct sw_flow_key *key, 1244 const struct nlattr *attr, int len) 1245 { 1246 const struct nlattr *a; 1247 int rem; 1248 1249 for (a = attr, rem = len; rem > 0; 1250 a = nla_next(a, &rem)) { 1251 int err = 0; 1252 1253 switch (nla_type(a)) { 1254 case OVS_ACTION_ATTR_OUTPUT: { 1255 int port = nla_get_u32(a); 1256 struct sk_buff *clone; 1257 1258 /* Every output action needs a separate clone 1259 * of 'skb', In case the output action is the 1260 * last action, cloning can be avoided. 1261 */ 1262 if (nla_is_last(a, rem)) { 1263 do_output(dp, skb, port, key); 1264 /* 'skb' has been used for output. 1265 */ 1266 return 0; 1267 } 1268 1269 clone = skb_clone(skb, GFP_ATOMIC); 1270 if (clone) 1271 do_output(dp, clone, port, key); 1272 OVS_CB(skb)->cutlen = 0; 1273 break; 1274 } 1275 1276 case OVS_ACTION_ATTR_TRUNC: { 1277 struct ovs_action_trunc *trunc = nla_data(a); 1278 1279 if (skb->len > trunc->max_len) 1280 OVS_CB(skb)->cutlen = skb->len - trunc->max_len; 1281 break; 1282 } 1283 1284 case OVS_ACTION_ATTR_USERSPACE: 1285 output_userspace(dp, skb, key, a, attr, 1286 len, OVS_CB(skb)->cutlen); 1287 OVS_CB(skb)->cutlen = 0; 1288 break; 1289 1290 case OVS_ACTION_ATTR_HASH: 1291 execute_hash(skb, key, a); 1292 break; 1293 1294 case OVS_ACTION_ATTR_PUSH_MPLS: 1295 err = push_mpls(skb, key, nla_data(a)); 1296 break; 1297 1298 case OVS_ACTION_ATTR_POP_MPLS: 1299 err = pop_mpls(skb, key, nla_get_be16(a)); 1300 break; 1301 1302 case OVS_ACTION_ATTR_PUSH_VLAN: 1303 err = push_vlan(skb, key, nla_data(a)); 1304 break; 1305 1306 case OVS_ACTION_ATTR_POP_VLAN: 1307 err = pop_vlan(skb, key); 1308 break; 1309 1310 case OVS_ACTION_ATTR_RECIRC: { 1311 bool last = nla_is_last(a, rem); 1312 1313 err = execute_recirc(dp, skb, key, a, last); 1314 if (last) { 1315 /* If this is the last action, the skb has 1316 * been consumed or freed. 1317 * Return immediately. 1318 */ 1319 return err; 1320 } 1321 break; 1322 } 1323 1324 case OVS_ACTION_ATTR_SET: 1325 err = execute_set_action(skb, key, nla_data(a)); 1326 break; 1327 1328 case OVS_ACTION_ATTR_SET_MASKED: 1329 case OVS_ACTION_ATTR_SET_TO_MASKED: 1330 err = execute_masked_set_action(skb, key, nla_data(a)); 1331 break; 1332 1333 case OVS_ACTION_ATTR_SAMPLE: { 1334 bool last = nla_is_last(a, rem); 1335 1336 err = sample(dp, skb, key, a, last); 1337 if (last) 1338 return err; 1339 1340 break; 1341 } 1342 1343 case OVS_ACTION_ATTR_CT: 1344 if (!is_flow_key_valid(key)) { 1345 err = ovs_flow_key_update(skb, key); 1346 if (err) 1347 return err; 1348 } 1349 1350 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key, 1351 nla_data(a)); 1352 1353 /* Hide stolen IP fragments from user space. */ 1354 if (err) 1355 return err == -EINPROGRESS ? 0 : err; 1356 break; 1357 1358 case OVS_ACTION_ATTR_CT_CLEAR: 1359 err = ovs_ct_clear(skb, key); 1360 break; 1361 1362 case OVS_ACTION_ATTR_PUSH_ETH: 1363 err = push_eth(skb, key, nla_data(a)); 1364 break; 1365 1366 case OVS_ACTION_ATTR_POP_ETH: 1367 err = pop_eth(skb, key); 1368 break; 1369 1370 case OVS_ACTION_ATTR_PUSH_NSH: { 1371 u8 buffer[NSH_HDR_MAX_LEN]; 1372 struct nshhdr *nh = (struct nshhdr *)buffer; 1373 1374 err = nsh_hdr_from_nlattr(nla_data(a), nh, 1375 NSH_HDR_MAX_LEN); 1376 if (unlikely(err)) 1377 break; 1378 err = push_nsh(skb, key, nh); 1379 break; 1380 } 1381 1382 case OVS_ACTION_ATTR_POP_NSH: 1383 err = pop_nsh(skb, key); 1384 break; 1385 1386 case OVS_ACTION_ATTR_METER: 1387 if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) { 1388 consume_skb(skb); 1389 return 0; 1390 } 1391 break; 1392 1393 case OVS_ACTION_ATTR_CLONE: { 1394 bool last = nla_is_last(a, rem); 1395 1396 err = clone(dp, skb, key, a, last); 1397 if (last) 1398 return err; 1399 1400 break; 1401 } 1402 1403 case OVS_ACTION_ATTR_CHECK_PKT_LEN: { 1404 bool last = nla_is_last(a, rem); 1405 1406 err = execute_check_pkt_len(dp, skb, key, a, last); 1407 if (last) 1408 return err; 1409 1410 break; 1411 } 1412 } 1413 1414 if (unlikely(err)) { 1415 kfree_skb(skb); 1416 return err; 1417 } 1418 } 1419 1420 consume_skb(skb); 1421 return 0; 1422 } 1423 1424 /* Execute the actions on the clone of the packet. The effect of the 1425 * execution does not affect the original 'skb' nor the original 'key'. 1426 * 1427 * The execution may be deferred in case the actions can not be executed 1428 * immediately. 1429 */ 1430 static int clone_execute(struct datapath *dp, struct sk_buff *skb, 1431 struct sw_flow_key *key, u32 recirc_id, 1432 const struct nlattr *actions, int len, 1433 bool last, bool clone_flow_key) 1434 { 1435 struct deferred_action *da; 1436 struct sw_flow_key *clone; 1437 1438 skb = last ? skb : skb_clone(skb, GFP_ATOMIC); 1439 if (!skb) { 1440 /* Out of memory, skip this action. 1441 */ 1442 return 0; 1443 } 1444 1445 /* When clone_flow_key is false, the 'key' will not be change 1446 * by the actions, then the 'key' can be used directly. 1447 * Otherwise, try to clone key from the next recursion level of 1448 * 'flow_keys'. If clone is successful, execute the actions 1449 * without deferring. 1450 */ 1451 clone = clone_flow_key ? clone_key(key) : key; 1452 if (clone) { 1453 int err = 0; 1454 1455 if (actions) { /* Sample action */ 1456 if (clone_flow_key) 1457 __this_cpu_inc(exec_actions_level); 1458 1459 err = do_execute_actions(dp, skb, clone, 1460 actions, len); 1461 1462 if (clone_flow_key) 1463 __this_cpu_dec(exec_actions_level); 1464 } else { /* Recirc action */ 1465 clone->recirc_id = recirc_id; 1466 ovs_dp_process_packet(skb, clone); 1467 } 1468 return err; 1469 } 1470 1471 /* Out of 'flow_keys' space. Defer actions */ 1472 da = add_deferred_actions(skb, key, actions, len); 1473 if (da) { 1474 if (!actions) { /* Recirc action */ 1475 key = &da->pkt_key; 1476 key->recirc_id = recirc_id; 1477 } 1478 } else { 1479 /* Out of per CPU action FIFO space. Drop the 'skb' and 1480 * log an error. 1481 */ 1482 kfree_skb(skb); 1483 1484 if (net_ratelimit()) { 1485 if (actions) { /* Sample action */ 1486 pr_warn("%s: deferred action limit reached, drop sample action\n", 1487 ovs_dp_name(dp)); 1488 } else { /* Recirc action */ 1489 pr_warn("%s: deferred action limit reached, drop recirc action\n", 1490 ovs_dp_name(dp)); 1491 } 1492 } 1493 } 1494 return 0; 1495 } 1496 1497 static void process_deferred_actions(struct datapath *dp) 1498 { 1499 struct action_fifo *fifo = this_cpu_ptr(action_fifos); 1500 1501 /* Do not touch the FIFO in case there is no deferred actions. */ 1502 if (action_fifo_is_empty(fifo)) 1503 return; 1504 1505 /* Finishing executing all deferred actions. */ 1506 do { 1507 struct deferred_action *da = action_fifo_get(fifo); 1508 struct sk_buff *skb = da->skb; 1509 struct sw_flow_key *key = &da->pkt_key; 1510 const struct nlattr *actions = da->actions; 1511 int actions_len = da->actions_len; 1512 1513 if (actions) 1514 do_execute_actions(dp, skb, key, actions, actions_len); 1515 else 1516 ovs_dp_process_packet(skb, key); 1517 } while (!action_fifo_is_empty(fifo)); 1518 1519 /* Reset FIFO for the next packet. */ 1520 action_fifo_init(fifo); 1521 } 1522 1523 /* Execute a list of actions against 'skb'. */ 1524 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb, 1525 const struct sw_flow_actions *acts, 1526 struct sw_flow_key *key) 1527 { 1528 int err, level; 1529 1530 level = __this_cpu_inc_return(exec_actions_level); 1531 if (unlikely(level > OVS_RECURSION_LIMIT)) { 1532 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n", 1533 ovs_dp_name(dp)); 1534 kfree_skb(skb); 1535 err = -ENETDOWN; 1536 goto out; 1537 } 1538 1539 OVS_CB(skb)->acts_origlen = acts->orig_len; 1540 err = do_execute_actions(dp, skb, key, 1541 acts->actions, acts->actions_len); 1542 1543 if (level == 1) 1544 process_deferred_actions(dp); 1545 1546 out: 1547 __this_cpu_dec(exec_actions_level); 1548 return err; 1549 } 1550 1551 int action_fifos_init(void) 1552 { 1553 action_fifos = alloc_percpu(struct action_fifo); 1554 if (!action_fifos) 1555 return -ENOMEM; 1556 1557 flow_keys = alloc_percpu(struct action_flow_keys); 1558 if (!flow_keys) { 1559 free_percpu(action_fifos); 1560 return -ENOMEM; 1561 } 1562 1563 return 0; 1564 } 1565 1566 void action_fifos_exit(void) 1567 { 1568 free_percpu(action_fifos); 1569 free_percpu(flow_keys); 1570 } 1571