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 /* 'KEY' must not have any bits set outside of the 'MASK' */ 189 #define MASKED(OLD, KEY, MASK) ((KEY) | ((OLD) & ~(MASK))) 190 #define SET_MASKED(OLD, KEY, MASK) ((OLD) = MASKED(OLD, KEY, MASK)) 191 192 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key, 193 const __be32 *mpls_lse, const __be32 *mask) 194 { 195 __be32 *stack; 196 __be32 lse; 197 int err; 198 199 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN); 200 if (unlikely(err)) 201 return err; 202 203 stack = (__be32 *)skb_mpls_header(skb); 204 lse = MASKED(*stack, *mpls_lse, *mask); 205 if (skb->ip_summed == CHECKSUM_COMPLETE) { 206 __be32 diff[] = { ~(*stack), lse }; 207 208 skb->csum = ~csum_partial((char *)diff, sizeof(diff), 209 ~skb->csum); 210 } 211 212 *stack = lse; 213 flow_key->mpls.top_lse = lse; 214 return 0; 215 } 216 217 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key) 218 { 219 int err; 220 221 err = skb_vlan_pop(skb); 222 if (skb_vlan_tag_present(skb)) 223 invalidate_flow_key(key); 224 else 225 key->eth.tci = 0; 226 return err; 227 } 228 229 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key, 230 const struct ovs_action_push_vlan *vlan) 231 { 232 if (skb_vlan_tag_present(skb)) 233 invalidate_flow_key(key); 234 else 235 key->eth.tci = vlan->vlan_tci; 236 return skb_vlan_push(skb, vlan->vlan_tpid, 237 ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT); 238 } 239 240 /* 'src' is already properly masked. */ 241 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_) 242 { 243 u16 *dst = (u16 *)dst_; 244 const u16 *src = (const u16 *)src_; 245 const u16 *mask = (const u16 *)mask_; 246 247 SET_MASKED(dst[0], src[0], mask[0]); 248 SET_MASKED(dst[1], src[1], mask[1]); 249 SET_MASKED(dst[2], src[2], mask[2]); 250 } 251 252 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key, 253 const struct ovs_key_ethernet *key, 254 const struct ovs_key_ethernet *mask) 255 { 256 int err; 257 258 err = skb_ensure_writable(skb, ETH_HLEN); 259 if (unlikely(err)) 260 return err; 261 262 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2); 263 264 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src, 265 mask->eth_src); 266 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst, 267 mask->eth_dst); 268 269 ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2); 270 271 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source); 272 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest); 273 return 0; 274 } 275 276 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh, 277 __be32 addr, __be32 new_addr) 278 { 279 int transport_len = skb->len - skb_transport_offset(skb); 280 281 if (nh->frag_off & htons(IP_OFFSET)) 282 return; 283 284 if (nh->protocol == IPPROTO_TCP) { 285 if (likely(transport_len >= sizeof(struct tcphdr))) 286 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb, 287 addr, new_addr, 1); 288 } else if (nh->protocol == IPPROTO_UDP) { 289 if (likely(transport_len >= sizeof(struct udphdr))) { 290 struct udphdr *uh = udp_hdr(skb); 291 292 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) { 293 inet_proto_csum_replace4(&uh->check, skb, 294 addr, new_addr, 1); 295 if (!uh->check) 296 uh->check = CSUM_MANGLED_0; 297 } 298 } 299 } 300 } 301 302 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh, 303 __be32 *addr, __be32 new_addr) 304 { 305 update_ip_l4_checksum(skb, nh, *addr, new_addr); 306 csum_replace4(&nh->check, *addr, new_addr); 307 skb_clear_hash(skb); 308 *addr = new_addr; 309 } 310 311 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto, 312 __be32 addr[4], const __be32 new_addr[4]) 313 { 314 int transport_len = skb->len - skb_transport_offset(skb); 315 316 if (l4_proto == NEXTHDR_TCP) { 317 if (likely(transport_len >= sizeof(struct tcphdr))) 318 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb, 319 addr, new_addr, 1); 320 } else if (l4_proto == NEXTHDR_UDP) { 321 if (likely(transport_len >= sizeof(struct udphdr))) { 322 struct udphdr *uh = udp_hdr(skb); 323 324 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) { 325 inet_proto_csum_replace16(&uh->check, skb, 326 addr, new_addr, 1); 327 if (!uh->check) 328 uh->check = CSUM_MANGLED_0; 329 } 330 } 331 } else if (l4_proto == NEXTHDR_ICMP) { 332 if (likely(transport_len >= sizeof(struct icmp6hdr))) 333 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum, 334 skb, addr, new_addr, 1); 335 } 336 } 337 338 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4], 339 const __be32 mask[4], __be32 masked[4]) 340 { 341 masked[0] = MASKED(old[0], addr[0], mask[0]); 342 masked[1] = MASKED(old[1], addr[1], mask[1]); 343 masked[2] = MASKED(old[2], addr[2], mask[2]); 344 masked[3] = MASKED(old[3], addr[3], mask[3]); 345 } 346 347 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto, 348 __be32 addr[4], const __be32 new_addr[4], 349 bool recalculate_csum) 350 { 351 if (recalculate_csum) 352 update_ipv6_checksum(skb, l4_proto, addr, new_addr); 353 354 skb_clear_hash(skb); 355 memcpy(addr, new_addr, sizeof(__be32[4])); 356 } 357 358 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask) 359 { 360 /* Bits 21-24 are always unmasked, so this retains their values. */ 361 SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16)); 362 SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8)); 363 SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask); 364 } 365 366 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl, 367 u8 mask) 368 { 369 new_ttl = MASKED(nh->ttl, new_ttl, mask); 370 371 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8)); 372 nh->ttl = new_ttl; 373 } 374 375 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key, 376 const struct ovs_key_ipv4 *key, 377 const struct ovs_key_ipv4 *mask) 378 { 379 struct iphdr *nh; 380 __be32 new_addr; 381 int err; 382 383 err = skb_ensure_writable(skb, skb_network_offset(skb) + 384 sizeof(struct iphdr)); 385 if (unlikely(err)) 386 return err; 387 388 nh = ip_hdr(skb); 389 390 /* Setting an IP addresses is typically only a side effect of 391 * matching on them in the current userspace implementation, so it 392 * makes sense to check if the value actually changed. 393 */ 394 if (mask->ipv4_src) { 395 new_addr = MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src); 396 397 if (unlikely(new_addr != nh->saddr)) { 398 set_ip_addr(skb, nh, &nh->saddr, new_addr); 399 flow_key->ipv4.addr.src = new_addr; 400 } 401 } 402 if (mask->ipv4_dst) { 403 new_addr = MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst); 404 405 if (unlikely(new_addr != nh->daddr)) { 406 set_ip_addr(skb, nh, &nh->daddr, new_addr); 407 flow_key->ipv4.addr.dst = new_addr; 408 } 409 } 410 if (mask->ipv4_tos) { 411 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos); 412 flow_key->ip.tos = nh->tos; 413 } 414 if (mask->ipv4_ttl) { 415 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl); 416 flow_key->ip.ttl = nh->ttl; 417 } 418 419 return 0; 420 } 421 422 static bool is_ipv6_mask_nonzero(const __be32 addr[4]) 423 { 424 return !!(addr[0] | addr[1] | addr[2] | addr[3]); 425 } 426 427 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key, 428 const struct ovs_key_ipv6 *key, 429 const struct ovs_key_ipv6 *mask) 430 { 431 struct ipv6hdr *nh; 432 int err; 433 434 err = skb_ensure_writable(skb, skb_network_offset(skb) + 435 sizeof(struct ipv6hdr)); 436 if (unlikely(err)) 437 return err; 438 439 nh = ipv6_hdr(skb); 440 441 /* Setting an IP addresses is typically only a side effect of 442 * matching on them in the current userspace implementation, so it 443 * makes sense to check if the value actually changed. 444 */ 445 if (is_ipv6_mask_nonzero(mask->ipv6_src)) { 446 __be32 *saddr = (__be32 *)&nh->saddr; 447 __be32 masked[4]; 448 449 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked); 450 451 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) { 452 set_ipv6_addr(skb, key->ipv6_proto, saddr, masked, 453 true); 454 memcpy(&flow_key->ipv6.addr.src, masked, 455 sizeof(flow_key->ipv6.addr.src)); 456 } 457 } 458 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) { 459 unsigned int offset = 0; 460 int flags = IP6_FH_F_SKIP_RH; 461 bool recalc_csum = true; 462 __be32 *daddr = (__be32 *)&nh->daddr; 463 __be32 masked[4]; 464 465 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked); 466 467 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) { 468 if (ipv6_ext_hdr(nh->nexthdr)) 469 recalc_csum = (ipv6_find_hdr(skb, &offset, 470 NEXTHDR_ROUTING, 471 NULL, &flags) 472 != NEXTHDR_ROUTING); 473 474 set_ipv6_addr(skb, key->ipv6_proto, daddr, masked, 475 recalc_csum); 476 memcpy(&flow_key->ipv6.addr.dst, masked, 477 sizeof(flow_key->ipv6.addr.dst)); 478 } 479 } 480 if (mask->ipv6_tclass) { 481 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass); 482 flow_key->ip.tos = ipv6_get_dsfield(nh); 483 } 484 if (mask->ipv6_label) { 485 set_ipv6_fl(nh, ntohl(key->ipv6_label), 486 ntohl(mask->ipv6_label)); 487 flow_key->ipv6.label = 488 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL); 489 } 490 if (mask->ipv6_hlimit) { 491 SET_MASKED(nh->hop_limit, key->ipv6_hlimit, mask->ipv6_hlimit); 492 flow_key->ip.ttl = nh->hop_limit; 493 } 494 return 0; 495 } 496 497 /* Must follow skb_ensure_writable() since that can move the skb data. */ 498 static void set_tp_port(struct sk_buff *skb, __be16 *port, 499 __be16 new_port, __sum16 *check) 500 { 501 inet_proto_csum_replace2(check, skb, *port, new_port, 0); 502 *port = new_port; 503 } 504 505 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key, 506 const struct ovs_key_udp *key, 507 const struct ovs_key_udp *mask) 508 { 509 struct udphdr *uh; 510 __be16 src, dst; 511 int err; 512 513 err = skb_ensure_writable(skb, skb_transport_offset(skb) + 514 sizeof(struct udphdr)); 515 if (unlikely(err)) 516 return err; 517 518 uh = udp_hdr(skb); 519 /* Either of the masks is non-zero, so do not bother checking them. */ 520 src = MASKED(uh->source, key->udp_src, mask->udp_src); 521 dst = MASKED(uh->dest, key->udp_dst, mask->udp_dst); 522 523 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) { 524 if (likely(src != uh->source)) { 525 set_tp_port(skb, &uh->source, src, &uh->check); 526 flow_key->tp.src = src; 527 } 528 if (likely(dst != uh->dest)) { 529 set_tp_port(skb, &uh->dest, dst, &uh->check); 530 flow_key->tp.dst = dst; 531 } 532 533 if (unlikely(!uh->check)) 534 uh->check = CSUM_MANGLED_0; 535 } else { 536 uh->source = src; 537 uh->dest = dst; 538 flow_key->tp.src = src; 539 flow_key->tp.dst = dst; 540 } 541 542 skb_clear_hash(skb); 543 544 return 0; 545 } 546 547 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key, 548 const struct ovs_key_tcp *key, 549 const struct ovs_key_tcp *mask) 550 { 551 struct tcphdr *th; 552 __be16 src, dst; 553 int err; 554 555 err = skb_ensure_writable(skb, skb_transport_offset(skb) + 556 sizeof(struct tcphdr)); 557 if (unlikely(err)) 558 return err; 559 560 th = tcp_hdr(skb); 561 src = MASKED(th->source, key->tcp_src, mask->tcp_src); 562 if (likely(src != th->source)) { 563 set_tp_port(skb, &th->source, src, &th->check); 564 flow_key->tp.src = src; 565 } 566 dst = MASKED(th->dest, key->tcp_dst, mask->tcp_dst); 567 if (likely(dst != th->dest)) { 568 set_tp_port(skb, &th->dest, dst, &th->check); 569 flow_key->tp.dst = dst; 570 } 571 skb_clear_hash(skb); 572 573 return 0; 574 } 575 576 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key, 577 const struct ovs_key_sctp *key, 578 const struct ovs_key_sctp *mask) 579 { 580 unsigned int sctphoff = skb_transport_offset(skb); 581 struct sctphdr *sh; 582 __le32 old_correct_csum, new_csum, old_csum; 583 int err; 584 585 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr)); 586 if (unlikely(err)) 587 return err; 588 589 sh = sctp_hdr(skb); 590 old_csum = sh->checksum; 591 old_correct_csum = sctp_compute_cksum(skb, sctphoff); 592 593 sh->source = MASKED(sh->source, key->sctp_src, mask->sctp_src); 594 sh->dest = MASKED(sh->dest, key->sctp_dst, mask->sctp_dst); 595 596 new_csum = sctp_compute_cksum(skb, sctphoff); 597 598 /* Carry any checksum errors through. */ 599 sh->checksum = old_csum ^ old_correct_csum ^ new_csum; 600 601 skb_clear_hash(skb); 602 flow_key->tp.src = sh->source; 603 flow_key->tp.dst = sh->dest; 604 605 return 0; 606 } 607 608 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port) 609 { 610 struct vport *vport = ovs_vport_rcu(dp, out_port); 611 612 if (likely(vport)) 613 ovs_vport_send(vport, skb); 614 else 615 kfree_skb(skb); 616 } 617 618 static int output_userspace(struct datapath *dp, struct sk_buff *skb, 619 struct sw_flow_key *key, const struct nlattr *attr, 620 const struct nlattr *actions, int actions_len) 621 { 622 struct ovs_tunnel_info info; 623 struct dp_upcall_info upcall; 624 const struct nlattr *a; 625 int rem; 626 627 memset(&upcall, 0, sizeof(upcall)); 628 upcall.cmd = OVS_PACKET_CMD_ACTION; 629 630 for (a = nla_data(attr), rem = nla_len(attr); rem > 0; 631 a = nla_next(a, &rem)) { 632 switch (nla_type(a)) { 633 case OVS_USERSPACE_ATTR_USERDATA: 634 upcall.userdata = a; 635 break; 636 637 case OVS_USERSPACE_ATTR_PID: 638 upcall.portid = nla_get_u32(a); 639 break; 640 641 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: { 642 /* Get out tunnel info. */ 643 struct vport *vport; 644 645 vport = ovs_vport_rcu(dp, nla_get_u32(a)); 646 if (vport) { 647 int err; 648 649 err = ovs_vport_get_egress_tun_info(vport, skb, 650 &info); 651 if (!err) 652 upcall.egress_tun_info = &info; 653 } 654 break; 655 } 656 657 case OVS_USERSPACE_ATTR_ACTIONS: { 658 /* Include actions. */ 659 upcall.actions = actions; 660 upcall.actions_len = actions_len; 661 break; 662 } 663 664 } /* End of switch. */ 665 } 666 667 return ovs_dp_upcall(dp, skb, key, &upcall); 668 } 669 670 static int sample(struct datapath *dp, struct sk_buff *skb, 671 struct sw_flow_key *key, const struct nlattr *attr, 672 const struct nlattr *actions, int actions_len) 673 { 674 const struct nlattr *acts_list = NULL; 675 const struct nlattr *a; 676 int rem; 677 678 for (a = nla_data(attr), rem = nla_len(attr); rem > 0; 679 a = nla_next(a, &rem)) { 680 switch (nla_type(a)) { 681 case OVS_SAMPLE_ATTR_PROBABILITY: 682 if (prandom_u32() >= nla_get_u32(a)) 683 return 0; 684 break; 685 686 case OVS_SAMPLE_ATTR_ACTIONS: 687 acts_list = a; 688 break; 689 } 690 } 691 692 rem = nla_len(acts_list); 693 a = nla_data(acts_list); 694 695 /* Actions list is empty, do nothing */ 696 if (unlikely(!rem)) 697 return 0; 698 699 /* The only known usage of sample action is having a single user-space 700 * action. Treat this usage as a special case. 701 * The output_userspace() should clone the skb to be sent to the 702 * user space. This skb will be consumed by its caller. 703 */ 704 if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE && 705 nla_is_last(a, rem))) 706 return output_userspace(dp, skb, key, a, actions, actions_len); 707 708 skb = skb_clone(skb, GFP_ATOMIC); 709 if (!skb) 710 /* Skip the sample action when out of memory. */ 711 return 0; 712 713 if (!add_deferred_actions(skb, key, a)) { 714 if (net_ratelimit()) 715 pr_warn("%s: deferred actions limit reached, dropping sample action\n", 716 ovs_dp_name(dp)); 717 718 kfree_skb(skb); 719 } 720 return 0; 721 } 722 723 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key, 724 const struct nlattr *attr) 725 { 726 struct ovs_action_hash *hash_act = nla_data(attr); 727 u32 hash = 0; 728 729 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */ 730 hash = skb_get_hash(skb); 731 hash = jhash_1word(hash, hash_act->hash_basis); 732 if (!hash) 733 hash = 0x1; 734 735 key->ovs_flow_hash = hash; 736 } 737 738 static int execute_set_action(struct sk_buff *skb, 739 struct sw_flow_key *flow_key, 740 const struct nlattr *a) 741 { 742 /* Only tunnel set execution is supported without a mask. */ 743 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) { 744 OVS_CB(skb)->egress_tun_info = nla_data(a); 745 return 0; 746 } 747 748 return -EINVAL; 749 } 750 751 /* Mask is at the midpoint of the data. */ 752 #define get_mask(a, type) ((const type)nla_data(a) + 1) 753 754 static int execute_masked_set_action(struct sk_buff *skb, 755 struct sw_flow_key *flow_key, 756 const struct nlattr *a) 757 { 758 int err = 0; 759 760 switch (nla_type(a)) { 761 case OVS_KEY_ATTR_PRIORITY: 762 SET_MASKED(skb->priority, nla_get_u32(a), *get_mask(a, u32 *)); 763 flow_key->phy.priority = skb->priority; 764 break; 765 766 case OVS_KEY_ATTR_SKB_MARK: 767 SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *)); 768 flow_key->phy.skb_mark = skb->mark; 769 break; 770 771 case OVS_KEY_ATTR_TUNNEL_INFO: 772 /* Masked data not supported for tunnel. */ 773 err = -EINVAL; 774 break; 775 776 case OVS_KEY_ATTR_ETHERNET: 777 err = set_eth_addr(skb, flow_key, nla_data(a), 778 get_mask(a, struct ovs_key_ethernet *)); 779 break; 780 781 case OVS_KEY_ATTR_IPV4: 782 err = set_ipv4(skb, flow_key, nla_data(a), 783 get_mask(a, struct ovs_key_ipv4 *)); 784 break; 785 786 case OVS_KEY_ATTR_IPV6: 787 err = set_ipv6(skb, flow_key, nla_data(a), 788 get_mask(a, struct ovs_key_ipv6 *)); 789 break; 790 791 case OVS_KEY_ATTR_TCP: 792 err = set_tcp(skb, flow_key, nla_data(a), 793 get_mask(a, struct ovs_key_tcp *)); 794 break; 795 796 case OVS_KEY_ATTR_UDP: 797 err = set_udp(skb, flow_key, nla_data(a), 798 get_mask(a, struct ovs_key_udp *)); 799 break; 800 801 case OVS_KEY_ATTR_SCTP: 802 err = set_sctp(skb, flow_key, nla_data(a), 803 get_mask(a, struct ovs_key_sctp *)); 804 break; 805 806 case OVS_KEY_ATTR_MPLS: 807 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a, 808 __be32 *)); 809 break; 810 } 811 812 return err; 813 } 814 815 static int execute_recirc(struct datapath *dp, struct sk_buff *skb, 816 struct sw_flow_key *key, 817 const struct nlattr *a, int rem) 818 { 819 struct deferred_action *da; 820 821 if (!is_flow_key_valid(key)) { 822 int err; 823 824 err = ovs_flow_key_update(skb, key); 825 if (err) 826 return err; 827 } 828 BUG_ON(!is_flow_key_valid(key)); 829 830 if (!nla_is_last(a, rem)) { 831 /* Recirc action is the not the last action 832 * of the action list, need to clone the skb. 833 */ 834 skb = skb_clone(skb, GFP_ATOMIC); 835 836 /* Skip the recirc action when out of memory, but 837 * continue on with the rest of the action list. 838 */ 839 if (!skb) 840 return 0; 841 } 842 843 da = add_deferred_actions(skb, key, NULL); 844 if (da) { 845 da->pkt_key.recirc_id = nla_get_u32(a); 846 } else { 847 kfree_skb(skb); 848 849 if (net_ratelimit()) 850 pr_warn("%s: deferred action limit reached, drop recirc action\n", 851 ovs_dp_name(dp)); 852 } 853 854 return 0; 855 } 856 857 /* Execute a list of actions against 'skb'. */ 858 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb, 859 struct sw_flow_key *key, 860 const struct nlattr *attr, int len) 861 { 862 /* Every output action needs a separate clone of 'skb', but the common 863 * case is just a single output action, so that doing a clone and 864 * then freeing the original skbuff is wasteful. So the following code 865 * is slightly obscure just to avoid that. 866 */ 867 int prev_port = -1; 868 const struct nlattr *a; 869 int rem; 870 871 for (a = attr, rem = len; rem > 0; 872 a = nla_next(a, &rem)) { 873 int err = 0; 874 875 if (unlikely(prev_port != -1)) { 876 struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC); 877 878 if (out_skb) 879 do_output(dp, out_skb, prev_port); 880 881 prev_port = -1; 882 } 883 884 switch (nla_type(a)) { 885 case OVS_ACTION_ATTR_OUTPUT: 886 prev_port = nla_get_u32(a); 887 break; 888 889 case OVS_ACTION_ATTR_USERSPACE: 890 output_userspace(dp, skb, key, a, attr, len); 891 break; 892 893 case OVS_ACTION_ATTR_HASH: 894 execute_hash(skb, key, a); 895 break; 896 897 case OVS_ACTION_ATTR_PUSH_MPLS: 898 err = push_mpls(skb, key, nla_data(a)); 899 break; 900 901 case OVS_ACTION_ATTR_POP_MPLS: 902 err = pop_mpls(skb, key, nla_get_be16(a)); 903 break; 904 905 case OVS_ACTION_ATTR_PUSH_VLAN: 906 err = push_vlan(skb, key, nla_data(a)); 907 break; 908 909 case OVS_ACTION_ATTR_POP_VLAN: 910 err = pop_vlan(skb, key); 911 break; 912 913 case OVS_ACTION_ATTR_RECIRC: 914 err = execute_recirc(dp, skb, key, a, rem); 915 if (nla_is_last(a, rem)) { 916 /* If this is the last action, the skb has 917 * been consumed or freed. 918 * Return immediately. 919 */ 920 return err; 921 } 922 break; 923 924 case OVS_ACTION_ATTR_SET: 925 err = execute_set_action(skb, key, nla_data(a)); 926 break; 927 928 case OVS_ACTION_ATTR_SET_MASKED: 929 case OVS_ACTION_ATTR_SET_TO_MASKED: 930 err = execute_masked_set_action(skb, key, nla_data(a)); 931 break; 932 933 case OVS_ACTION_ATTR_SAMPLE: 934 err = sample(dp, skb, key, a, attr, len); 935 break; 936 } 937 938 if (unlikely(err)) { 939 kfree_skb(skb); 940 return err; 941 } 942 } 943 944 if (prev_port != -1) 945 do_output(dp, skb, prev_port); 946 else 947 consume_skb(skb); 948 949 return 0; 950 } 951 952 static void process_deferred_actions(struct datapath *dp) 953 { 954 struct action_fifo *fifo = this_cpu_ptr(action_fifos); 955 956 /* Do not touch the FIFO in case there is no deferred actions. */ 957 if (action_fifo_is_empty(fifo)) 958 return; 959 960 /* Finishing executing all deferred actions. */ 961 do { 962 struct deferred_action *da = action_fifo_get(fifo); 963 struct sk_buff *skb = da->skb; 964 struct sw_flow_key *key = &da->pkt_key; 965 const struct nlattr *actions = da->actions; 966 967 if (actions) 968 do_execute_actions(dp, skb, key, actions, 969 nla_len(actions)); 970 else 971 ovs_dp_process_packet(skb, key); 972 } while (!action_fifo_is_empty(fifo)); 973 974 /* Reset FIFO for the next packet. */ 975 action_fifo_init(fifo); 976 } 977 978 /* Execute a list of actions against 'skb'. */ 979 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb, 980 const struct sw_flow_actions *acts, 981 struct sw_flow_key *key) 982 { 983 int level = this_cpu_read(exec_actions_level); 984 int err; 985 986 this_cpu_inc(exec_actions_level); 987 OVS_CB(skb)->egress_tun_info = NULL; 988 err = do_execute_actions(dp, skb, key, 989 acts->actions, acts->actions_len); 990 991 if (!level) 992 process_deferred_actions(dp); 993 994 this_cpu_dec(exec_actions_level); 995 return err; 996 } 997 998 int action_fifos_init(void) 999 { 1000 action_fifos = alloc_percpu(struct action_fifo); 1001 if (!action_fifos) 1002 return -ENOMEM; 1003 1004 return 0; 1005 } 1006 1007 void action_fifos_exit(void) 1008 { 1009 free_percpu(action_fifos); 1010 } 1011