1 /* 2 * Copyright (c) 2007-2014 Nicira, Inc. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of version 2 of the GNU General Public 6 * License as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope that it will be useful, but 9 * WITHOUT ANY WARRANTY; without even the implied warranty of 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 11 * General Public License for more details. 12 * 13 * You should have received a copy of the GNU General Public License 14 * along with this program; if not, write to the Free Software 15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 16 * 02110-1301, USA 17 */ 18 19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 20 21 #include <linux/skbuff.h> 22 #include <linux/in.h> 23 #include <linux/ip.h> 24 #include <linux/openvswitch.h> 25 #include <linux/sctp.h> 26 #include <linux/tcp.h> 27 #include <linux/udp.h> 28 #include <linux/in6.h> 29 #include <linux/if_arp.h> 30 #include <linux/if_vlan.h> 31 32 #include <net/ip.h> 33 #include <net/ipv6.h> 34 #include <net/checksum.h> 35 #include <net/dsfield.h> 36 #include <net/mpls.h> 37 #include <net/sctp/checksum.h> 38 39 #include "datapath.h" 40 #include "flow.h" 41 #include "vport.h" 42 43 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb, 44 struct sw_flow_key *key, 45 const struct nlattr *attr, int len); 46 47 struct deferred_action { 48 struct sk_buff *skb; 49 const struct nlattr *actions; 50 51 /* Store pkt_key clone when creating deferred action. */ 52 struct sw_flow_key pkt_key; 53 }; 54 55 #define DEFERRED_ACTION_FIFO_SIZE 10 56 struct action_fifo { 57 int head; 58 int tail; 59 /* Deferred action fifo queue storage. */ 60 struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE]; 61 }; 62 63 static struct action_fifo __percpu *action_fifos; 64 static DEFINE_PER_CPU(int, exec_actions_level); 65 66 static void action_fifo_init(struct action_fifo *fifo) 67 { 68 fifo->head = 0; 69 fifo->tail = 0; 70 } 71 72 static bool action_fifo_is_empty(const struct action_fifo *fifo) 73 { 74 return (fifo->head == fifo->tail); 75 } 76 77 static struct deferred_action *action_fifo_get(struct action_fifo *fifo) 78 { 79 if (action_fifo_is_empty(fifo)) 80 return NULL; 81 82 return &fifo->fifo[fifo->tail++]; 83 } 84 85 static struct deferred_action *action_fifo_put(struct action_fifo *fifo) 86 { 87 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1) 88 return NULL; 89 90 return &fifo->fifo[fifo->head++]; 91 } 92 93 /* Return true if fifo is not full */ 94 static struct deferred_action *add_deferred_actions(struct sk_buff *skb, 95 const struct sw_flow_key *key, 96 const struct nlattr *attr) 97 { 98 struct action_fifo *fifo; 99 struct deferred_action *da; 100 101 fifo = this_cpu_ptr(action_fifos); 102 da = action_fifo_put(fifo); 103 if (da) { 104 da->skb = skb; 105 da->actions = attr; 106 da->pkt_key = *key; 107 } 108 109 return da; 110 } 111 112 static void invalidate_flow_key(struct sw_flow_key *key) 113 { 114 key->eth.type = htons(0); 115 } 116 117 static bool is_flow_key_valid(const struct sw_flow_key *key) 118 { 119 return !!key->eth.type; 120 } 121 122 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key, 123 const struct ovs_action_push_mpls *mpls) 124 { 125 __be32 *new_mpls_lse; 126 struct ethhdr *hdr; 127 128 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */ 129 if (skb->encapsulation) 130 return -ENOTSUPP; 131 132 if (skb_cow_head(skb, MPLS_HLEN) < 0) 133 return -ENOMEM; 134 135 skb_push(skb, MPLS_HLEN); 136 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb), 137 skb->mac_len); 138 skb_reset_mac_header(skb); 139 140 new_mpls_lse = (__be32 *)skb_mpls_header(skb); 141 *new_mpls_lse = mpls->mpls_lse; 142 143 if (skb->ip_summed == CHECKSUM_COMPLETE) 144 skb->csum = csum_add(skb->csum, csum_partial(new_mpls_lse, 145 MPLS_HLEN, 0)); 146 147 hdr = eth_hdr(skb); 148 hdr->h_proto = mpls->mpls_ethertype; 149 150 if (!skb->inner_protocol) 151 skb_set_inner_protocol(skb, skb->protocol); 152 skb->protocol = mpls->mpls_ethertype; 153 154 invalidate_flow_key(key); 155 return 0; 156 } 157 158 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key, 159 const __be16 ethertype) 160 { 161 struct ethhdr *hdr; 162 int err; 163 164 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN); 165 if (unlikely(err)) 166 return err; 167 168 skb_postpull_rcsum(skb, skb_mpls_header(skb), MPLS_HLEN); 169 170 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb), 171 skb->mac_len); 172 173 __skb_pull(skb, MPLS_HLEN); 174 skb_reset_mac_header(skb); 175 176 /* skb_mpls_header() is used to locate the ethertype 177 * field correctly in the presence of VLAN tags. 178 */ 179 hdr = (struct ethhdr *)(skb_mpls_header(skb) - ETH_HLEN); 180 hdr->h_proto = ethertype; 181 if (eth_p_mpls(skb->protocol)) 182 skb->protocol = ethertype; 183 184 invalidate_flow_key(key); 185 return 0; 186 } 187 188 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *key, 189 const __be32 *mpls_lse) 190 { 191 __be32 *stack; 192 int err; 193 194 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN); 195 if (unlikely(err)) 196 return err; 197 198 stack = (__be32 *)skb_mpls_header(skb); 199 if (skb->ip_summed == CHECKSUM_COMPLETE) { 200 __be32 diff[] = { ~(*stack), *mpls_lse }; 201 skb->csum = ~csum_partial((char *)diff, sizeof(diff), 202 ~skb->csum); 203 } 204 205 *stack = *mpls_lse; 206 key->mpls.top_lse = *mpls_lse; 207 return 0; 208 } 209 210 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key) 211 { 212 int err; 213 214 err = skb_vlan_pop(skb); 215 if (vlan_tx_tag_present(skb)) 216 invalidate_flow_key(key); 217 else 218 key->eth.tci = 0; 219 return err; 220 } 221 222 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key, 223 const struct ovs_action_push_vlan *vlan) 224 { 225 if (vlan_tx_tag_present(skb)) 226 invalidate_flow_key(key); 227 else 228 key->eth.tci = vlan->vlan_tci; 229 return skb_vlan_push(skb, vlan->vlan_tpid, 230 ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT); 231 } 232 233 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *key, 234 const struct ovs_key_ethernet *eth_key) 235 { 236 int err; 237 err = skb_ensure_writable(skb, ETH_HLEN); 238 if (unlikely(err)) 239 return err; 240 241 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2); 242 243 ether_addr_copy(eth_hdr(skb)->h_source, eth_key->eth_src); 244 ether_addr_copy(eth_hdr(skb)->h_dest, eth_key->eth_dst); 245 246 ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2); 247 248 ether_addr_copy(key->eth.src, eth_key->eth_src); 249 ether_addr_copy(key->eth.dst, eth_key->eth_dst); 250 return 0; 251 } 252 253 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh, 254 __be32 *addr, __be32 new_addr) 255 { 256 int transport_len = skb->len - skb_transport_offset(skb); 257 258 if (nh->protocol == IPPROTO_TCP) { 259 if (likely(transport_len >= sizeof(struct tcphdr))) 260 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb, 261 *addr, new_addr, 1); 262 } else if (nh->protocol == IPPROTO_UDP) { 263 if (likely(transport_len >= sizeof(struct udphdr))) { 264 struct udphdr *uh = udp_hdr(skb); 265 266 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) { 267 inet_proto_csum_replace4(&uh->check, skb, 268 *addr, new_addr, 1); 269 if (!uh->check) 270 uh->check = CSUM_MANGLED_0; 271 } 272 } 273 } 274 275 csum_replace4(&nh->check, *addr, new_addr); 276 skb_clear_hash(skb); 277 *addr = new_addr; 278 } 279 280 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto, 281 __be32 addr[4], const __be32 new_addr[4]) 282 { 283 int transport_len = skb->len - skb_transport_offset(skb); 284 285 if (l4_proto == NEXTHDR_TCP) { 286 if (likely(transport_len >= sizeof(struct tcphdr))) 287 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb, 288 addr, new_addr, 1); 289 } else if (l4_proto == NEXTHDR_UDP) { 290 if (likely(transport_len >= sizeof(struct udphdr))) { 291 struct udphdr *uh = udp_hdr(skb); 292 293 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) { 294 inet_proto_csum_replace16(&uh->check, skb, 295 addr, new_addr, 1); 296 if (!uh->check) 297 uh->check = CSUM_MANGLED_0; 298 } 299 } 300 } else if (l4_proto == NEXTHDR_ICMP) { 301 if (likely(transport_len >= sizeof(struct icmp6hdr))) 302 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum, 303 skb, addr, new_addr, 1); 304 } 305 } 306 307 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto, 308 __be32 addr[4], const __be32 new_addr[4], 309 bool recalculate_csum) 310 { 311 if (recalculate_csum) 312 update_ipv6_checksum(skb, l4_proto, addr, new_addr); 313 314 skb_clear_hash(skb); 315 memcpy(addr, new_addr, sizeof(__be32[4])); 316 } 317 318 static void set_ipv6_tc(struct ipv6hdr *nh, u8 tc) 319 { 320 nh->priority = tc >> 4; 321 nh->flow_lbl[0] = (nh->flow_lbl[0] & 0x0F) | ((tc & 0x0F) << 4); 322 } 323 324 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl) 325 { 326 nh->flow_lbl[0] = (nh->flow_lbl[0] & 0xF0) | (fl & 0x000F0000) >> 16; 327 nh->flow_lbl[1] = (fl & 0x0000FF00) >> 8; 328 nh->flow_lbl[2] = fl & 0x000000FF; 329 } 330 331 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl) 332 { 333 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8)); 334 nh->ttl = new_ttl; 335 } 336 337 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *key, 338 const struct ovs_key_ipv4 *ipv4_key) 339 { 340 struct iphdr *nh; 341 int err; 342 343 err = skb_ensure_writable(skb, skb_network_offset(skb) + 344 sizeof(struct iphdr)); 345 if (unlikely(err)) 346 return err; 347 348 nh = ip_hdr(skb); 349 350 if (ipv4_key->ipv4_src != nh->saddr) { 351 set_ip_addr(skb, nh, &nh->saddr, ipv4_key->ipv4_src); 352 key->ipv4.addr.src = ipv4_key->ipv4_src; 353 } 354 355 if (ipv4_key->ipv4_dst != nh->daddr) { 356 set_ip_addr(skb, nh, &nh->daddr, ipv4_key->ipv4_dst); 357 key->ipv4.addr.dst = ipv4_key->ipv4_dst; 358 } 359 360 if (ipv4_key->ipv4_tos != nh->tos) { 361 ipv4_change_dsfield(nh, 0, ipv4_key->ipv4_tos); 362 key->ip.tos = nh->tos; 363 } 364 365 if (ipv4_key->ipv4_ttl != nh->ttl) { 366 set_ip_ttl(skb, nh, ipv4_key->ipv4_ttl); 367 key->ip.ttl = ipv4_key->ipv4_ttl; 368 } 369 370 return 0; 371 } 372 373 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *key, 374 const struct ovs_key_ipv6 *ipv6_key) 375 { 376 struct ipv6hdr *nh; 377 int err; 378 __be32 *saddr; 379 __be32 *daddr; 380 381 err = skb_ensure_writable(skb, skb_network_offset(skb) + 382 sizeof(struct ipv6hdr)); 383 if (unlikely(err)) 384 return err; 385 386 nh = ipv6_hdr(skb); 387 saddr = (__be32 *)&nh->saddr; 388 daddr = (__be32 *)&nh->daddr; 389 390 if (memcmp(ipv6_key->ipv6_src, saddr, sizeof(ipv6_key->ipv6_src))) { 391 set_ipv6_addr(skb, ipv6_key->ipv6_proto, saddr, 392 ipv6_key->ipv6_src, true); 393 memcpy(&key->ipv6.addr.src, ipv6_key->ipv6_src, 394 sizeof(ipv6_key->ipv6_src)); 395 } 396 397 if (memcmp(ipv6_key->ipv6_dst, daddr, sizeof(ipv6_key->ipv6_dst))) { 398 unsigned int offset = 0; 399 int flags = IP6_FH_F_SKIP_RH; 400 bool recalc_csum = true; 401 402 if (ipv6_ext_hdr(nh->nexthdr)) 403 recalc_csum = ipv6_find_hdr(skb, &offset, 404 NEXTHDR_ROUTING, NULL, 405 &flags) != NEXTHDR_ROUTING; 406 407 set_ipv6_addr(skb, ipv6_key->ipv6_proto, daddr, 408 ipv6_key->ipv6_dst, recalc_csum); 409 memcpy(&key->ipv6.addr.dst, ipv6_key->ipv6_dst, 410 sizeof(ipv6_key->ipv6_dst)); 411 } 412 413 set_ipv6_tc(nh, ipv6_key->ipv6_tclass); 414 key->ip.tos = ipv6_get_dsfield(nh); 415 416 set_ipv6_fl(nh, ntohl(ipv6_key->ipv6_label)); 417 key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL); 418 419 nh->hop_limit = ipv6_key->ipv6_hlimit; 420 key->ip.ttl = ipv6_key->ipv6_hlimit; 421 return 0; 422 } 423 424 /* Must follow skb_ensure_writable() since that can move the skb data. */ 425 static void set_tp_port(struct sk_buff *skb, __be16 *port, 426 __be16 new_port, __sum16 *check) 427 { 428 inet_proto_csum_replace2(check, skb, *port, new_port, 0); 429 *port = new_port; 430 skb_clear_hash(skb); 431 } 432 433 static void set_udp_port(struct sk_buff *skb, __be16 *port, __be16 new_port) 434 { 435 struct udphdr *uh = udp_hdr(skb); 436 437 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) { 438 set_tp_port(skb, port, new_port, &uh->check); 439 440 if (!uh->check) 441 uh->check = CSUM_MANGLED_0; 442 } else { 443 *port = new_port; 444 skb_clear_hash(skb); 445 } 446 } 447 448 static int set_udp(struct sk_buff *skb, struct sw_flow_key *key, 449 const struct ovs_key_udp *udp_port_key) 450 { 451 struct udphdr *uh; 452 int err; 453 454 err = skb_ensure_writable(skb, skb_transport_offset(skb) + 455 sizeof(struct udphdr)); 456 if (unlikely(err)) 457 return err; 458 459 uh = udp_hdr(skb); 460 if (udp_port_key->udp_src != uh->source) { 461 set_udp_port(skb, &uh->source, udp_port_key->udp_src); 462 key->tp.src = udp_port_key->udp_src; 463 } 464 465 if (udp_port_key->udp_dst != uh->dest) { 466 set_udp_port(skb, &uh->dest, udp_port_key->udp_dst); 467 key->tp.dst = udp_port_key->udp_dst; 468 } 469 470 return 0; 471 } 472 473 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *key, 474 const struct ovs_key_tcp *tcp_port_key) 475 { 476 struct tcphdr *th; 477 int err; 478 479 err = skb_ensure_writable(skb, skb_transport_offset(skb) + 480 sizeof(struct tcphdr)); 481 if (unlikely(err)) 482 return err; 483 484 th = tcp_hdr(skb); 485 if (tcp_port_key->tcp_src != th->source) { 486 set_tp_port(skb, &th->source, tcp_port_key->tcp_src, &th->check); 487 key->tp.src = tcp_port_key->tcp_src; 488 } 489 490 if (tcp_port_key->tcp_dst != th->dest) { 491 set_tp_port(skb, &th->dest, tcp_port_key->tcp_dst, &th->check); 492 key->tp.dst = tcp_port_key->tcp_dst; 493 } 494 495 return 0; 496 } 497 498 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *key, 499 const struct ovs_key_sctp *sctp_port_key) 500 { 501 struct sctphdr *sh; 502 int err; 503 unsigned int sctphoff = skb_transport_offset(skb); 504 505 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr)); 506 if (unlikely(err)) 507 return err; 508 509 sh = sctp_hdr(skb); 510 if (sctp_port_key->sctp_src != sh->source || 511 sctp_port_key->sctp_dst != sh->dest) { 512 __le32 old_correct_csum, new_csum, old_csum; 513 514 old_csum = sh->checksum; 515 old_correct_csum = sctp_compute_cksum(skb, sctphoff); 516 517 sh->source = sctp_port_key->sctp_src; 518 sh->dest = sctp_port_key->sctp_dst; 519 520 new_csum = sctp_compute_cksum(skb, sctphoff); 521 522 /* Carry any checksum errors through. */ 523 sh->checksum = old_csum ^ old_correct_csum ^ new_csum; 524 525 skb_clear_hash(skb); 526 key->tp.src = sctp_port_key->sctp_src; 527 key->tp.dst = sctp_port_key->sctp_dst; 528 } 529 530 return 0; 531 } 532 533 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port) 534 { 535 struct vport *vport = ovs_vport_rcu(dp, out_port); 536 537 if (likely(vport)) 538 ovs_vport_send(vport, skb); 539 else 540 kfree_skb(skb); 541 } 542 543 static int output_userspace(struct datapath *dp, struct sk_buff *skb, 544 struct sw_flow_key *key, const struct nlattr *attr) 545 { 546 struct ovs_tunnel_info info; 547 struct dp_upcall_info upcall; 548 const struct nlattr *a; 549 int rem; 550 551 upcall.cmd = OVS_PACKET_CMD_ACTION; 552 upcall.userdata = NULL; 553 upcall.portid = 0; 554 upcall.egress_tun_info = NULL; 555 556 for (a = nla_data(attr), rem = nla_len(attr); rem > 0; 557 a = nla_next(a, &rem)) { 558 switch (nla_type(a)) { 559 case OVS_USERSPACE_ATTR_USERDATA: 560 upcall.userdata = a; 561 break; 562 563 case OVS_USERSPACE_ATTR_PID: 564 upcall.portid = nla_get_u32(a); 565 break; 566 567 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: { 568 /* Get out tunnel info. */ 569 struct vport *vport; 570 571 vport = ovs_vport_rcu(dp, nla_get_u32(a)); 572 if (vport) { 573 int err; 574 575 err = ovs_vport_get_egress_tun_info(vport, skb, 576 &info); 577 if (!err) 578 upcall.egress_tun_info = &info; 579 } 580 break; 581 } 582 583 } /* End of switch. */ 584 } 585 586 return ovs_dp_upcall(dp, skb, key, &upcall); 587 } 588 589 static int sample(struct datapath *dp, struct sk_buff *skb, 590 struct sw_flow_key *key, const struct nlattr *attr) 591 { 592 const struct nlattr *acts_list = NULL; 593 const struct nlattr *a; 594 int rem; 595 596 for (a = nla_data(attr), rem = nla_len(attr); rem > 0; 597 a = nla_next(a, &rem)) { 598 switch (nla_type(a)) { 599 case OVS_SAMPLE_ATTR_PROBABILITY: 600 if (prandom_u32() >= nla_get_u32(a)) 601 return 0; 602 break; 603 604 case OVS_SAMPLE_ATTR_ACTIONS: 605 acts_list = a; 606 break; 607 } 608 } 609 610 rem = nla_len(acts_list); 611 a = nla_data(acts_list); 612 613 /* Actions list is empty, do nothing */ 614 if (unlikely(!rem)) 615 return 0; 616 617 /* The only known usage of sample action is having a single user-space 618 * action. Treat this usage as a special case. 619 * The output_userspace() should clone the skb to be sent to the 620 * user space. This skb will be consumed by its caller. 621 */ 622 if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE && 623 nla_is_last(a, rem))) 624 return output_userspace(dp, skb, key, a); 625 626 skb = skb_clone(skb, GFP_ATOMIC); 627 if (!skb) 628 /* Skip the sample action when out of memory. */ 629 return 0; 630 631 if (!add_deferred_actions(skb, key, a)) { 632 if (net_ratelimit()) 633 pr_warn("%s: deferred actions limit reached, dropping sample action\n", 634 ovs_dp_name(dp)); 635 636 kfree_skb(skb); 637 } 638 return 0; 639 } 640 641 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key, 642 const struct nlattr *attr) 643 { 644 struct ovs_action_hash *hash_act = nla_data(attr); 645 u32 hash = 0; 646 647 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */ 648 hash = skb_get_hash(skb); 649 hash = jhash_1word(hash, hash_act->hash_basis); 650 if (!hash) 651 hash = 0x1; 652 653 key->ovs_flow_hash = hash; 654 } 655 656 static int execute_set_action(struct sk_buff *skb, struct sw_flow_key *key, 657 const struct nlattr *nested_attr) 658 { 659 int err = 0; 660 661 switch (nla_type(nested_attr)) { 662 case OVS_KEY_ATTR_PRIORITY: 663 skb->priority = nla_get_u32(nested_attr); 664 key->phy.priority = skb->priority; 665 break; 666 667 case OVS_KEY_ATTR_SKB_MARK: 668 skb->mark = nla_get_u32(nested_attr); 669 key->phy.skb_mark = skb->mark; 670 break; 671 672 case OVS_KEY_ATTR_TUNNEL_INFO: 673 OVS_CB(skb)->egress_tun_info = nla_data(nested_attr); 674 break; 675 676 case OVS_KEY_ATTR_ETHERNET: 677 err = set_eth_addr(skb, key, nla_data(nested_attr)); 678 break; 679 680 case OVS_KEY_ATTR_IPV4: 681 err = set_ipv4(skb, key, nla_data(nested_attr)); 682 break; 683 684 case OVS_KEY_ATTR_IPV6: 685 err = set_ipv6(skb, key, nla_data(nested_attr)); 686 break; 687 688 case OVS_KEY_ATTR_TCP: 689 err = set_tcp(skb, key, nla_data(nested_attr)); 690 break; 691 692 case OVS_KEY_ATTR_UDP: 693 err = set_udp(skb, key, nla_data(nested_attr)); 694 break; 695 696 case OVS_KEY_ATTR_SCTP: 697 err = set_sctp(skb, key, nla_data(nested_attr)); 698 break; 699 700 case OVS_KEY_ATTR_MPLS: 701 err = set_mpls(skb, key, nla_data(nested_attr)); 702 break; 703 } 704 705 return err; 706 } 707 708 static int execute_recirc(struct datapath *dp, struct sk_buff *skb, 709 struct sw_flow_key *key, 710 const struct nlattr *a, int rem) 711 { 712 struct deferred_action *da; 713 714 if (!is_flow_key_valid(key)) { 715 int err; 716 717 err = ovs_flow_key_update(skb, key); 718 if (err) 719 return err; 720 } 721 BUG_ON(!is_flow_key_valid(key)); 722 723 if (!nla_is_last(a, rem)) { 724 /* Recirc action is the not the last action 725 * of the action list, need to clone the skb. 726 */ 727 skb = skb_clone(skb, GFP_ATOMIC); 728 729 /* Skip the recirc action when out of memory, but 730 * continue on with the rest of the action list. 731 */ 732 if (!skb) 733 return 0; 734 } 735 736 da = add_deferred_actions(skb, key, NULL); 737 if (da) { 738 da->pkt_key.recirc_id = nla_get_u32(a); 739 } else { 740 kfree_skb(skb); 741 742 if (net_ratelimit()) 743 pr_warn("%s: deferred action limit reached, drop recirc action\n", 744 ovs_dp_name(dp)); 745 } 746 747 return 0; 748 } 749 750 /* Execute a list of actions against 'skb'. */ 751 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb, 752 struct sw_flow_key *key, 753 const struct nlattr *attr, int len) 754 { 755 /* Every output action needs a separate clone of 'skb', but the common 756 * case is just a single output action, so that doing a clone and 757 * then freeing the original skbuff is wasteful. So the following code 758 * is slightly obscure just to avoid that. 759 */ 760 int prev_port = -1; 761 const struct nlattr *a; 762 int rem; 763 764 for (a = attr, rem = len; rem > 0; 765 a = nla_next(a, &rem)) { 766 int err = 0; 767 768 if (unlikely(prev_port != -1)) { 769 struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC); 770 771 if (out_skb) 772 do_output(dp, out_skb, prev_port); 773 774 prev_port = -1; 775 } 776 777 switch (nla_type(a)) { 778 case OVS_ACTION_ATTR_OUTPUT: 779 prev_port = nla_get_u32(a); 780 break; 781 782 case OVS_ACTION_ATTR_USERSPACE: 783 output_userspace(dp, skb, key, a); 784 break; 785 786 case OVS_ACTION_ATTR_HASH: 787 execute_hash(skb, key, a); 788 break; 789 790 case OVS_ACTION_ATTR_PUSH_MPLS: 791 err = push_mpls(skb, key, nla_data(a)); 792 break; 793 794 case OVS_ACTION_ATTR_POP_MPLS: 795 err = pop_mpls(skb, key, nla_get_be16(a)); 796 break; 797 798 case OVS_ACTION_ATTR_PUSH_VLAN: 799 err = push_vlan(skb, key, nla_data(a)); 800 break; 801 802 case OVS_ACTION_ATTR_POP_VLAN: 803 err = pop_vlan(skb, key); 804 break; 805 806 case OVS_ACTION_ATTR_RECIRC: 807 err = execute_recirc(dp, skb, key, a, rem); 808 if (nla_is_last(a, rem)) { 809 /* If this is the last action, the skb has 810 * been consumed or freed. 811 * Return immediately. 812 */ 813 return err; 814 } 815 break; 816 817 case OVS_ACTION_ATTR_SET: 818 err = execute_set_action(skb, key, nla_data(a)); 819 break; 820 821 case OVS_ACTION_ATTR_SAMPLE: 822 err = sample(dp, skb, key, a); 823 break; 824 } 825 826 if (unlikely(err)) { 827 kfree_skb(skb); 828 return err; 829 } 830 } 831 832 if (prev_port != -1) 833 do_output(dp, skb, prev_port); 834 else 835 consume_skb(skb); 836 837 return 0; 838 } 839 840 static void process_deferred_actions(struct datapath *dp) 841 { 842 struct action_fifo *fifo = this_cpu_ptr(action_fifos); 843 844 /* Do not touch the FIFO in case there is no deferred actions. */ 845 if (action_fifo_is_empty(fifo)) 846 return; 847 848 /* Finishing executing all deferred actions. */ 849 do { 850 struct deferred_action *da = action_fifo_get(fifo); 851 struct sk_buff *skb = da->skb; 852 struct sw_flow_key *key = &da->pkt_key; 853 const struct nlattr *actions = da->actions; 854 855 if (actions) 856 do_execute_actions(dp, skb, key, actions, 857 nla_len(actions)); 858 else 859 ovs_dp_process_packet(skb, key); 860 } while (!action_fifo_is_empty(fifo)); 861 862 /* Reset FIFO for the next packet. */ 863 action_fifo_init(fifo); 864 } 865 866 /* Execute a list of actions against 'skb'. */ 867 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb, 868 const struct sw_flow_actions *acts, 869 struct sw_flow_key *key) 870 { 871 int level = this_cpu_read(exec_actions_level); 872 int err; 873 874 this_cpu_inc(exec_actions_level); 875 OVS_CB(skb)->egress_tun_info = NULL; 876 err = do_execute_actions(dp, skb, key, 877 acts->actions, acts->actions_len); 878 879 if (!level) 880 process_deferred_actions(dp); 881 882 this_cpu_dec(exec_actions_level); 883 return err; 884 } 885 886 int action_fifos_init(void) 887 { 888 action_fifos = alloc_percpu(struct action_fifo); 889 if (!action_fifos) 890 return -ENOMEM; 891 892 return 0; 893 } 894 895 void action_fifos_exit(void) 896 { 897 free_percpu(action_fifos); 898 } 899