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/netfilter_ipv6.h> 26 #include <linux/sctp.h> 27 #include <linux/tcp.h> 28 #include <linux/udp.h> 29 #include <linux/in6.h> 30 #include <linux/if_arp.h> 31 #include <linux/if_vlan.h> 32 33 #include <net/dst.h> 34 #include <net/ip.h> 35 #include <net/ipv6.h> 36 #include <net/ip6_fib.h> 37 #include <net/checksum.h> 38 #include <net/dsfield.h> 39 #include <net/mpls.h> 40 #include <net/sctp/checksum.h> 41 42 #include "datapath.h" 43 #include "flow.h" 44 #include "conntrack.h" 45 #include "vport.h" 46 47 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb, 48 struct sw_flow_key *key, 49 const struct nlattr *attr, int len); 50 51 struct deferred_action { 52 struct sk_buff *skb; 53 const struct nlattr *actions; 54 55 /* Store pkt_key clone when creating deferred action. */ 56 struct sw_flow_key pkt_key; 57 }; 58 59 #define MAX_L2_LEN (VLAN_ETH_HLEN + 3 * MPLS_HLEN) 60 struct ovs_frag_data { 61 unsigned long dst; 62 struct vport *vport; 63 struct ovs_skb_cb cb; 64 __be16 inner_protocol; 65 __u16 vlan_tci; 66 __be16 vlan_proto; 67 unsigned int l2_len; 68 u8 l2_data[MAX_L2_LEN]; 69 }; 70 71 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage); 72 73 #define DEFERRED_ACTION_FIFO_SIZE 10 74 struct action_fifo { 75 int head; 76 int tail; 77 /* Deferred action fifo queue storage. */ 78 struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE]; 79 }; 80 81 static struct action_fifo __percpu *action_fifos; 82 static DEFINE_PER_CPU(int, exec_actions_level); 83 84 static void action_fifo_init(struct action_fifo *fifo) 85 { 86 fifo->head = 0; 87 fifo->tail = 0; 88 } 89 90 static bool action_fifo_is_empty(const struct action_fifo *fifo) 91 { 92 return (fifo->head == fifo->tail); 93 } 94 95 static struct deferred_action *action_fifo_get(struct action_fifo *fifo) 96 { 97 if (action_fifo_is_empty(fifo)) 98 return NULL; 99 100 return &fifo->fifo[fifo->tail++]; 101 } 102 103 static struct deferred_action *action_fifo_put(struct action_fifo *fifo) 104 { 105 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1) 106 return NULL; 107 108 return &fifo->fifo[fifo->head++]; 109 } 110 111 /* Return true if fifo is not full */ 112 static struct deferred_action *add_deferred_actions(struct sk_buff *skb, 113 const struct sw_flow_key *key, 114 const struct nlattr *attr) 115 { 116 struct action_fifo *fifo; 117 struct deferred_action *da; 118 119 fifo = this_cpu_ptr(action_fifos); 120 da = action_fifo_put(fifo); 121 if (da) { 122 da->skb = skb; 123 da->actions = attr; 124 da->pkt_key = *key; 125 } 126 127 return da; 128 } 129 130 static void invalidate_flow_key(struct sw_flow_key *key) 131 { 132 key->eth.type = htons(0); 133 } 134 135 static bool is_flow_key_valid(const struct sw_flow_key *key) 136 { 137 return !!key->eth.type; 138 } 139 140 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key, 141 const struct ovs_action_push_mpls *mpls) 142 { 143 __be32 *new_mpls_lse; 144 struct ethhdr *hdr; 145 146 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */ 147 if (skb->encapsulation) 148 return -ENOTSUPP; 149 150 if (skb_cow_head(skb, MPLS_HLEN) < 0) 151 return -ENOMEM; 152 153 skb_push(skb, MPLS_HLEN); 154 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb), 155 skb->mac_len); 156 skb_reset_mac_header(skb); 157 158 new_mpls_lse = (__be32 *)skb_mpls_header(skb); 159 *new_mpls_lse = mpls->mpls_lse; 160 161 if (skb->ip_summed == CHECKSUM_COMPLETE) 162 skb->csum = csum_add(skb->csum, csum_partial(new_mpls_lse, 163 MPLS_HLEN, 0)); 164 165 hdr = eth_hdr(skb); 166 hdr->h_proto = mpls->mpls_ethertype; 167 168 if (!skb->inner_protocol) 169 skb_set_inner_protocol(skb, skb->protocol); 170 skb->protocol = mpls->mpls_ethertype; 171 172 invalidate_flow_key(key); 173 return 0; 174 } 175 176 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key, 177 const __be16 ethertype) 178 { 179 struct ethhdr *hdr; 180 int err; 181 182 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN); 183 if (unlikely(err)) 184 return err; 185 186 skb_postpull_rcsum(skb, skb_mpls_header(skb), MPLS_HLEN); 187 188 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb), 189 skb->mac_len); 190 191 __skb_pull(skb, MPLS_HLEN); 192 skb_reset_mac_header(skb); 193 194 /* skb_mpls_header() is used to locate the ethertype 195 * field correctly in the presence of VLAN tags. 196 */ 197 hdr = (struct ethhdr *)(skb_mpls_header(skb) - ETH_HLEN); 198 hdr->h_proto = ethertype; 199 if (eth_p_mpls(skb->protocol)) 200 skb->protocol = ethertype; 201 202 invalidate_flow_key(key); 203 return 0; 204 } 205 206 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key, 207 const __be32 *mpls_lse, const __be32 *mask) 208 { 209 __be32 *stack; 210 __be32 lse; 211 int err; 212 213 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN); 214 if (unlikely(err)) 215 return err; 216 217 stack = (__be32 *)skb_mpls_header(skb); 218 lse = OVS_MASKED(*stack, *mpls_lse, *mask); 219 if (skb->ip_summed == CHECKSUM_COMPLETE) { 220 __be32 diff[] = { ~(*stack), lse }; 221 222 skb->csum = ~csum_partial((char *)diff, sizeof(diff), 223 ~skb->csum); 224 } 225 226 *stack = lse; 227 flow_key->mpls.top_lse = lse; 228 return 0; 229 } 230 231 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key) 232 { 233 int err; 234 235 err = skb_vlan_pop(skb); 236 if (skb_vlan_tag_present(skb)) 237 invalidate_flow_key(key); 238 else 239 key->eth.tci = 0; 240 return err; 241 } 242 243 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key, 244 const struct ovs_action_push_vlan *vlan) 245 { 246 if (skb_vlan_tag_present(skb)) 247 invalidate_flow_key(key); 248 else 249 key->eth.tci = vlan->vlan_tci; 250 return skb_vlan_push(skb, vlan->vlan_tpid, 251 ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT); 252 } 253 254 /* 'src' is already properly masked. */ 255 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_) 256 { 257 u16 *dst = (u16 *)dst_; 258 const u16 *src = (const u16 *)src_; 259 const u16 *mask = (const u16 *)mask_; 260 261 OVS_SET_MASKED(dst[0], src[0], mask[0]); 262 OVS_SET_MASKED(dst[1], src[1], mask[1]); 263 OVS_SET_MASKED(dst[2], src[2], mask[2]); 264 } 265 266 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key, 267 const struct ovs_key_ethernet *key, 268 const struct ovs_key_ethernet *mask) 269 { 270 int err; 271 272 err = skb_ensure_writable(skb, ETH_HLEN); 273 if (unlikely(err)) 274 return err; 275 276 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2); 277 278 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src, 279 mask->eth_src); 280 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst, 281 mask->eth_dst); 282 283 ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2); 284 285 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source); 286 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest); 287 return 0; 288 } 289 290 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh, 291 __be32 addr, __be32 new_addr) 292 { 293 int transport_len = skb->len - skb_transport_offset(skb); 294 295 if (nh->frag_off & htons(IP_OFFSET)) 296 return; 297 298 if (nh->protocol == IPPROTO_TCP) { 299 if (likely(transport_len >= sizeof(struct tcphdr))) 300 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb, 301 addr, new_addr, true); 302 } else if (nh->protocol == IPPROTO_UDP) { 303 if (likely(transport_len >= sizeof(struct udphdr))) { 304 struct udphdr *uh = udp_hdr(skb); 305 306 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) { 307 inet_proto_csum_replace4(&uh->check, skb, 308 addr, new_addr, true); 309 if (!uh->check) 310 uh->check = CSUM_MANGLED_0; 311 } 312 } 313 } 314 } 315 316 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh, 317 __be32 *addr, __be32 new_addr) 318 { 319 update_ip_l4_checksum(skb, nh, *addr, new_addr); 320 csum_replace4(&nh->check, *addr, new_addr); 321 skb_clear_hash(skb); 322 *addr = new_addr; 323 } 324 325 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto, 326 __be32 addr[4], const __be32 new_addr[4]) 327 { 328 int transport_len = skb->len - skb_transport_offset(skb); 329 330 if (l4_proto == NEXTHDR_TCP) { 331 if (likely(transport_len >= sizeof(struct tcphdr))) 332 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb, 333 addr, new_addr, true); 334 } else if (l4_proto == NEXTHDR_UDP) { 335 if (likely(transport_len >= sizeof(struct udphdr))) { 336 struct udphdr *uh = udp_hdr(skb); 337 338 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) { 339 inet_proto_csum_replace16(&uh->check, skb, 340 addr, new_addr, true); 341 if (!uh->check) 342 uh->check = CSUM_MANGLED_0; 343 } 344 } 345 } else if (l4_proto == NEXTHDR_ICMP) { 346 if (likely(transport_len >= sizeof(struct icmp6hdr))) 347 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum, 348 skb, addr, new_addr, true); 349 } 350 } 351 352 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4], 353 const __be32 mask[4], __be32 masked[4]) 354 { 355 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]); 356 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]); 357 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]); 358 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]); 359 } 360 361 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto, 362 __be32 addr[4], const __be32 new_addr[4], 363 bool recalculate_csum) 364 { 365 if (recalculate_csum) 366 update_ipv6_checksum(skb, l4_proto, addr, new_addr); 367 368 skb_clear_hash(skb); 369 memcpy(addr, new_addr, sizeof(__be32[4])); 370 } 371 372 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask) 373 { 374 /* Bits 21-24 are always unmasked, so this retains their values. */ 375 OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16)); 376 OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8)); 377 OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask); 378 } 379 380 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl, 381 u8 mask) 382 { 383 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask); 384 385 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8)); 386 nh->ttl = new_ttl; 387 } 388 389 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key, 390 const struct ovs_key_ipv4 *key, 391 const struct ovs_key_ipv4 *mask) 392 { 393 struct iphdr *nh; 394 __be32 new_addr; 395 int err; 396 397 err = skb_ensure_writable(skb, skb_network_offset(skb) + 398 sizeof(struct iphdr)); 399 if (unlikely(err)) 400 return err; 401 402 nh = ip_hdr(skb); 403 404 /* Setting an IP addresses is typically only a side effect of 405 * matching on them in the current userspace implementation, so it 406 * makes sense to check if the value actually changed. 407 */ 408 if (mask->ipv4_src) { 409 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src); 410 411 if (unlikely(new_addr != nh->saddr)) { 412 set_ip_addr(skb, nh, &nh->saddr, new_addr); 413 flow_key->ipv4.addr.src = new_addr; 414 } 415 } 416 if (mask->ipv4_dst) { 417 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst); 418 419 if (unlikely(new_addr != nh->daddr)) { 420 set_ip_addr(skb, nh, &nh->daddr, new_addr); 421 flow_key->ipv4.addr.dst = new_addr; 422 } 423 } 424 if (mask->ipv4_tos) { 425 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos); 426 flow_key->ip.tos = nh->tos; 427 } 428 if (mask->ipv4_ttl) { 429 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl); 430 flow_key->ip.ttl = nh->ttl; 431 } 432 433 return 0; 434 } 435 436 static bool is_ipv6_mask_nonzero(const __be32 addr[4]) 437 { 438 return !!(addr[0] | addr[1] | addr[2] | addr[3]); 439 } 440 441 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key, 442 const struct ovs_key_ipv6 *key, 443 const struct ovs_key_ipv6 *mask) 444 { 445 struct ipv6hdr *nh; 446 int err; 447 448 err = skb_ensure_writable(skb, skb_network_offset(skb) + 449 sizeof(struct ipv6hdr)); 450 if (unlikely(err)) 451 return err; 452 453 nh = ipv6_hdr(skb); 454 455 /* Setting an IP addresses is typically only a side effect of 456 * matching on them in the current userspace implementation, so it 457 * makes sense to check if the value actually changed. 458 */ 459 if (is_ipv6_mask_nonzero(mask->ipv6_src)) { 460 __be32 *saddr = (__be32 *)&nh->saddr; 461 __be32 masked[4]; 462 463 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked); 464 465 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) { 466 set_ipv6_addr(skb, key->ipv6_proto, saddr, masked, 467 true); 468 memcpy(&flow_key->ipv6.addr.src, masked, 469 sizeof(flow_key->ipv6.addr.src)); 470 } 471 } 472 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) { 473 unsigned int offset = 0; 474 int flags = IP6_FH_F_SKIP_RH; 475 bool recalc_csum = true; 476 __be32 *daddr = (__be32 *)&nh->daddr; 477 __be32 masked[4]; 478 479 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked); 480 481 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) { 482 if (ipv6_ext_hdr(nh->nexthdr)) 483 recalc_csum = (ipv6_find_hdr(skb, &offset, 484 NEXTHDR_ROUTING, 485 NULL, &flags) 486 != NEXTHDR_ROUTING); 487 488 set_ipv6_addr(skb, key->ipv6_proto, daddr, masked, 489 recalc_csum); 490 memcpy(&flow_key->ipv6.addr.dst, masked, 491 sizeof(flow_key->ipv6.addr.dst)); 492 } 493 } 494 if (mask->ipv6_tclass) { 495 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass); 496 flow_key->ip.tos = ipv6_get_dsfield(nh); 497 } 498 if (mask->ipv6_label) { 499 set_ipv6_fl(nh, ntohl(key->ipv6_label), 500 ntohl(mask->ipv6_label)); 501 flow_key->ipv6.label = 502 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL); 503 } 504 if (mask->ipv6_hlimit) { 505 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit, 506 mask->ipv6_hlimit); 507 flow_key->ip.ttl = nh->hop_limit; 508 } 509 return 0; 510 } 511 512 /* Must follow skb_ensure_writable() since that can move the skb data. */ 513 static void set_tp_port(struct sk_buff *skb, __be16 *port, 514 __be16 new_port, __sum16 *check) 515 { 516 inet_proto_csum_replace2(check, skb, *port, new_port, false); 517 *port = new_port; 518 } 519 520 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key, 521 const struct ovs_key_udp *key, 522 const struct ovs_key_udp *mask) 523 { 524 struct udphdr *uh; 525 __be16 src, dst; 526 int err; 527 528 err = skb_ensure_writable(skb, skb_transport_offset(skb) + 529 sizeof(struct udphdr)); 530 if (unlikely(err)) 531 return err; 532 533 uh = udp_hdr(skb); 534 /* Either of the masks is non-zero, so do not bother checking them. */ 535 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src); 536 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst); 537 538 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) { 539 if (likely(src != uh->source)) { 540 set_tp_port(skb, &uh->source, src, &uh->check); 541 flow_key->tp.src = src; 542 } 543 if (likely(dst != uh->dest)) { 544 set_tp_port(skb, &uh->dest, dst, &uh->check); 545 flow_key->tp.dst = dst; 546 } 547 548 if (unlikely(!uh->check)) 549 uh->check = CSUM_MANGLED_0; 550 } else { 551 uh->source = src; 552 uh->dest = dst; 553 flow_key->tp.src = src; 554 flow_key->tp.dst = dst; 555 } 556 557 skb_clear_hash(skb); 558 559 return 0; 560 } 561 562 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key, 563 const struct ovs_key_tcp *key, 564 const struct ovs_key_tcp *mask) 565 { 566 struct tcphdr *th; 567 __be16 src, dst; 568 int err; 569 570 err = skb_ensure_writable(skb, skb_transport_offset(skb) + 571 sizeof(struct tcphdr)); 572 if (unlikely(err)) 573 return err; 574 575 th = tcp_hdr(skb); 576 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src); 577 if (likely(src != th->source)) { 578 set_tp_port(skb, &th->source, src, &th->check); 579 flow_key->tp.src = src; 580 } 581 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst); 582 if (likely(dst != th->dest)) { 583 set_tp_port(skb, &th->dest, dst, &th->check); 584 flow_key->tp.dst = dst; 585 } 586 skb_clear_hash(skb); 587 588 return 0; 589 } 590 591 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key, 592 const struct ovs_key_sctp *key, 593 const struct ovs_key_sctp *mask) 594 { 595 unsigned int sctphoff = skb_transport_offset(skb); 596 struct sctphdr *sh; 597 __le32 old_correct_csum, new_csum, old_csum; 598 int err; 599 600 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr)); 601 if (unlikely(err)) 602 return err; 603 604 sh = sctp_hdr(skb); 605 old_csum = sh->checksum; 606 old_correct_csum = sctp_compute_cksum(skb, sctphoff); 607 608 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src); 609 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst); 610 611 new_csum = sctp_compute_cksum(skb, sctphoff); 612 613 /* Carry any checksum errors through. */ 614 sh->checksum = old_csum ^ old_correct_csum ^ new_csum; 615 616 skb_clear_hash(skb); 617 flow_key->tp.src = sh->source; 618 flow_key->tp.dst = sh->dest; 619 620 return 0; 621 } 622 623 static int ovs_vport_output(struct sock *sock, struct sk_buff *skb) 624 { 625 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage); 626 struct vport *vport = data->vport; 627 628 if (skb_cow_head(skb, data->l2_len) < 0) { 629 kfree_skb(skb); 630 return -ENOMEM; 631 } 632 633 __skb_dst_copy(skb, data->dst); 634 *OVS_CB(skb) = data->cb; 635 skb->inner_protocol = data->inner_protocol; 636 skb->vlan_tci = data->vlan_tci; 637 skb->vlan_proto = data->vlan_proto; 638 639 /* Reconstruct the MAC header. */ 640 skb_push(skb, data->l2_len); 641 memcpy(skb->data, &data->l2_data, data->l2_len); 642 ovs_skb_postpush_rcsum(skb, skb->data, data->l2_len); 643 skb_reset_mac_header(skb); 644 645 ovs_vport_send(vport, skb); 646 return 0; 647 } 648 649 static unsigned int 650 ovs_dst_get_mtu(const struct dst_entry *dst) 651 { 652 return dst->dev->mtu; 653 } 654 655 static struct dst_ops ovs_dst_ops = { 656 .family = AF_UNSPEC, 657 .mtu = ovs_dst_get_mtu, 658 }; 659 660 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is 661 * ovs_vport_output(), which is called once per fragmented packet. 662 */ 663 static void prepare_frag(struct vport *vport, struct sk_buff *skb) 664 { 665 unsigned int hlen = skb_network_offset(skb); 666 struct ovs_frag_data *data; 667 668 data = this_cpu_ptr(&ovs_frag_data_storage); 669 data->dst = skb->_skb_refdst; 670 data->vport = vport; 671 data->cb = *OVS_CB(skb); 672 data->inner_protocol = skb->inner_protocol; 673 data->vlan_tci = skb->vlan_tci; 674 data->vlan_proto = skb->vlan_proto; 675 data->l2_len = hlen; 676 memcpy(&data->l2_data, skb->data, hlen); 677 678 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); 679 skb_pull(skb, hlen); 680 } 681 682 static void ovs_fragment(struct vport *vport, struct sk_buff *skb, u16 mru, 683 __be16 ethertype) 684 { 685 if (skb_network_offset(skb) > MAX_L2_LEN) { 686 OVS_NLERR(1, "L2 header too long to fragment"); 687 return; 688 } 689 690 if (ethertype == htons(ETH_P_IP)) { 691 struct dst_entry ovs_dst; 692 unsigned long orig_dst; 693 694 prepare_frag(vport, skb); 695 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1, 696 DST_OBSOLETE_NONE, DST_NOCOUNT); 697 ovs_dst.dev = vport->dev; 698 699 orig_dst = skb->_skb_refdst; 700 skb_dst_set_noref(skb, &ovs_dst); 701 IPCB(skb)->frag_max_size = mru; 702 703 ip_do_fragment(skb->sk, skb, ovs_vport_output); 704 refdst_drop(orig_dst); 705 } else if (ethertype == htons(ETH_P_IPV6)) { 706 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops(); 707 unsigned long orig_dst; 708 struct rt6_info ovs_rt; 709 710 if (!v6ops) { 711 kfree_skb(skb); 712 return; 713 } 714 715 prepare_frag(vport, skb); 716 memset(&ovs_rt, 0, sizeof(ovs_rt)); 717 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1, 718 DST_OBSOLETE_NONE, DST_NOCOUNT); 719 ovs_rt.dst.dev = vport->dev; 720 721 orig_dst = skb->_skb_refdst; 722 skb_dst_set_noref(skb, &ovs_rt.dst); 723 IP6CB(skb)->frag_max_size = mru; 724 725 v6ops->fragment(skb->sk, skb, ovs_vport_output); 726 refdst_drop(orig_dst); 727 } else { 728 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.", 729 ovs_vport_name(vport), ntohs(ethertype), mru, 730 vport->dev->mtu); 731 kfree_skb(skb); 732 } 733 } 734 735 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port, 736 struct sw_flow_key *key) 737 { 738 struct vport *vport = ovs_vport_rcu(dp, out_port); 739 740 if (likely(vport)) { 741 u16 mru = OVS_CB(skb)->mru; 742 743 if (likely(!mru || (skb->len <= mru + ETH_HLEN))) { 744 ovs_vport_send(vport, skb); 745 } else if (mru <= vport->dev->mtu) { 746 __be16 ethertype = key->eth.type; 747 748 if (!is_flow_key_valid(key)) { 749 if (eth_p_mpls(skb->protocol)) 750 ethertype = skb->inner_protocol; 751 else 752 ethertype = vlan_get_protocol(skb); 753 } 754 755 ovs_fragment(vport, skb, mru, ethertype); 756 } else { 757 kfree_skb(skb); 758 } 759 } else { 760 kfree_skb(skb); 761 } 762 } 763 764 static int output_userspace(struct datapath *dp, struct sk_buff *skb, 765 struct sw_flow_key *key, const struct nlattr *attr, 766 const struct nlattr *actions, int actions_len) 767 { 768 struct ip_tunnel_info info; 769 struct dp_upcall_info upcall; 770 const struct nlattr *a; 771 int rem; 772 773 memset(&upcall, 0, sizeof(upcall)); 774 upcall.cmd = OVS_PACKET_CMD_ACTION; 775 upcall.mru = OVS_CB(skb)->mru; 776 777 for (a = nla_data(attr), rem = nla_len(attr); rem > 0; 778 a = nla_next(a, &rem)) { 779 switch (nla_type(a)) { 780 case OVS_USERSPACE_ATTR_USERDATA: 781 upcall.userdata = a; 782 break; 783 784 case OVS_USERSPACE_ATTR_PID: 785 upcall.portid = nla_get_u32(a); 786 break; 787 788 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: { 789 /* Get out tunnel info. */ 790 struct vport *vport; 791 792 vport = ovs_vport_rcu(dp, nla_get_u32(a)); 793 if (vport) { 794 int err; 795 796 upcall.egress_tun_info = &info; 797 err = ovs_vport_get_egress_tun_info(vport, skb, 798 &upcall); 799 if (err) 800 upcall.egress_tun_info = NULL; 801 } 802 803 break; 804 } 805 806 case OVS_USERSPACE_ATTR_ACTIONS: { 807 /* Include actions. */ 808 upcall.actions = actions; 809 upcall.actions_len = actions_len; 810 break; 811 } 812 813 } /* End of switch. */ 814 } 815 816 return ovs_dp_upcall(dp, skb, key, &upcall); 817 } 818 819 static int sample(struct datapath *dp, struct sk_buff *skb, 820 struct sw_flow_key *key, const struct nlattr *attr, 821 const struct nlattr *actions, int actions_len) 822 { 823 const struct nlattr *acts_list = NULL; 824 const struct nlattr *a; 825 int rem; 826 827 for (a = nla_data(attr), rem = nla_len(attr); rem > 0; 828 a = nla_next(a, &rem)) { 829 u32 probability; 830 831 switch (nla_type(a)) { 832 case OVS_SAMPLE_ATTR_PROBABILITY: 833 probability = nla_get_u32(a); 834 if (!probability || prandom_u32() > probability) 835 return 0; 836 break; 837 838 case OVS_SAMPLE_ATTR_ACTIONS: 839 acts_list = a; 840 break; 841 } 842 } 843 844 rem = nla_len(acts_list); 845 a = nla_data(acts_list); 846 847 /* Actions list is empty, do nothing */ 848 if (unlikely(!rem)) 849 return 0; 850 851 /* The only known usage of sample action is having a single user-space 852 * action. Treat this usage as a special case. 853 * The output_userspace() should clone the skb to be sent to the 854 * user space. This skb will be consumed by its caller. 855 */ 856 if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE && 857 nla_is_last(a, rem))) 858 return output_userspace(dp, skb, key, a, actions, actions_len); 859 860 skb = skb_clone(skb, GFP_ATOMIC); 861 if (!skb) 862 /* Skip the sample action when out of memory. */ 863 return 0; 864 865 if (!add_deferred_actions(skb, key, a)) { 866 if (net_ratelimit()) 867 pr_warn("%s: deferred actions limit reached, dropping sample action\n", 868 ovs_dp_name(dp)); 869 870 kfree_skb(skb); 871 } 872 return 0; 873 } 874 875 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key, 876 const struct nlattr *attr) 877 { 878 struct ovs_action_hash *hash_act = nla_data(attr); 879 u32 hash = 0; 880 881 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */ 882 hash = skb_get_hash(skb); 883 hash = jhash_1word(hash, hash_act->hash_basis); 884 if (!hash) 885 hash = 0x1; 886 887 key->ovs_flow_hash = hash; 888 } 889 890 static int execute_set_action(struct sk_buff *skb, 891 struct sw_flow_key *flow_key, 892 const struct nlattr *a) 893 { 894 /* Only tunnel set execution is supported without a mask. */ 895 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) { 896 struct ovs_tunnel_info *tun = nla_data(a); 897 898 skb_dst_drop(skb); 899 dst_hold((struct dst_entry *)tun->tun_dst); 900 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst); 901 return 0; 902 } 903 904 return -EINVAL; 905 } 906 907 /* Mask is at the midpoint of the data. */ 908 #define get_mask(a, type) ((const type)nla_data(a) + 1) 909 910 static int execute_masked_set_action(struct sk_buff *skb, 911 struct sw_flow_key *flow_key, 912 const struct nlattr *a) 913 { 914 int err = 0; 915 916 switch (nla_type(a)) { 917 case OVS_KEY_ATTR_PRIORITY: 918 OVS_SET_MASKED(skb->priority, nla_get_u32(a), 919 *get_mask(a, u32 *)); 920 flow_key->phy.priority = skb->priority; 921 break; 922 923 case OVS_KEY_ATTR_SKB_MARK: 924 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *)); 925 flow_key->phy.skb_mark = skb->mark; 926 break; 927 928 case OVS_KEY_ATTR_TUNNEL_INFO: 929 /* Masked data not supported for tunnel. */ 930 err = -EINVAL; 931 break; 932 933 case OVS_KEY_ATTR_ETHERNET: 934 err = set_eth_addr(skb, flow_key, nla_data(a), 935 get_mask(a, struct ovs_key_ethernet *)); 936 break; 937 938 case OVS_KEY_ATTR_IPV4: 939 err = set_ipv4(skb, flow_key, nla_data(a), 940 get_mask(a, struct ovs_key_ipv4 *)); 941 break; 942 943 case OVS_KEY_ATTR_IPV6: 944 err = set_ipv6(skb, flow_key, nla_data(a), 945 get_mask(a, struct ovs_key_ipv6 *)); 946 break; 947 948 case OVS_KEY_ATTR_TCP: 949 err = set_tcp(skb, flow_key, nla_data(a), 950 get_mask(a, struct ovs_key_tcp *)); 951 break; 952 953 case OVS_KEY_ATTR_UDP: 954 err = set_udp(skb, flow_key, nla_data(a), 955 get_mask(a, struct ovs_key_udp *)); 956 break; 957 958 case OVS_KEY_ATTR_SCTP: 959 err = set_sctp(skb, flow_key, nla_data(a), 960 get_mask(a, struct ovs_key_sctp *)); 961 break; 962 963 case OVS_KEY_ATTR_MPLS: 964 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a, 965 __be32 *)); 966 break; 967 968 case OVS_KEY_ATTR_CT_STATE: 969 case OVS_KEY_ATTR_CT_ZONE: 970 case OVS_KEY_ATTR_CT_MARK: 971 case OVS_KEY_ATTR_CT_LABEL: 972 err = -EINVAL; 973 break; 974 } 975 976 return err; 977 } 978 979 static int execute_recirc(struct datapath *dp, struct sk_buff *skb, 980 struct sw_flow_key *key, 981 const struct nlattr *a, int rem) 982 { 983 struct deferred_action *da; 984 985 if (!is_flow_key_valid(key)) { 986 int err; 987 988 err = ovs_flow_key_update(skb, key); 989 if (err) 990 return err; 991 } 992 BUG_ON(!is_flow_key_valid(key)); 993 994 if (!nla_is_last(a, rem)) { 995 /* Recirc action is the not the last action 996 * of the action list, need to clone the skb. 997 */ 998 skb = skb_clone(skb, GFP_ATOMIC); 999 1000 /* Skip the recirc action when out of memory, but 1001 * continue on with the rest of the action list. 1002 */ 1003 if (!skb) 1004 return 0; 1005 } 1006 1007 da = add_deferred_actions(skb, key, NULL); 1008 if (da) { 1009 da->pkt_key.recirc_id = nla_get_u32(a); 1010 } else { 1011 kfree_skb(skb); 1012 1013 if (net_ratelimit()) 1014 pr_warn("%s: deferred action limit reached, drop recirc action\n", 1015 ovs_dp_name(dp)); 1016 } 1017 1018 return 0; 1019 } 1020 1021 /* Execute a list of actions against 'skb'. */ 1022 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb, 1023 struct sw_flow_key *key, 1024 const struct nlattr *attr, int len) 1025 { 1026 /* Every output action needs a separate clone of 'skb', but the common 1027 * case is just a single output action, so that doing a clone and 1028 * then freeing the original skbuff is wasteful. So the following code 1029 * is slightly obscure just to avoid that. 1030 */ 1031 int prev_port = -1; 1032 const struct nlattr *a; 1033 int rem; 1034 1035 for (a = attr, rem = len; rem > 0; 1036 a = nla_next(a, &rem)) { 1037 int err = 0; 1038 1039 if (unlikely(prev_port != -1)) { 1040 struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC); 1041 1042 if (out_skb) 1043 do_output(dp, out_skb, prev_port, key); 1044 1045 prev_port = -1; 1046 } 1047 1048 switch (nla_type(a)) { 1049 case OVS_ACTION_ATTR_OUTPUT: 1050 prev_port = nla_get_u32(a); 1051 break; 1052 1053 case OVS_ACTION_ATTR_USERSPACE: 1054 output_userspace(dp, skb, key, a, attr, len); 1055 break; 1056 1057 case OVS_ACTION_ATTR_HASH: 1058 execute_hash(skb, key, a); 1059 break; 1060 1061 case OVS_ACTION_ATTR_PUSH_MPLS: 1062 err = push_mpls(skb, key, nla_data(a)); 1063 break; 1064 1065 case OVS_ACTION_ATTR_POP_MPLS: 1066 err = pop_mpls(skb, key, nla_get_be16(a)); 1067 break; 1068 1069 case OVS_ACTION_ATTR_PUSH_VLAN: 1070 err = push_vlan(skb, key, nla_data(a)); 1071 break; 1072 1073 case OVS_ACTION_ATTR_POP_VLAN: 1074 err = pop_vlan(skb, key); 1075 break; 1076 1077 case OVS_ACTION_ATTR_RECIRC: 1078 err = execute_recirc(dp, skb, key, a, rem); 1079 if (nla_is_last(a, rem)) { 1080 /* If this is the last action, the skb has 1081 * been consumed or freed. 1082 * Return immediately. 1083 */ 1084 return err; 1085 } 1086 break; 1087 1088 case OVS_ACTION_ATTR_SET: 1089 err = execute_set_action(skb, key, nla_data(a)); 1090 break; 1091 1092 case OVS_ACTION_ATTR_SET_MASKED: 1093 case OVS_ACTION_ATTR_SET_TO_MASKED: 1094 err = execute_masked_set_action(skb, key, nla_data(a)); 1095 break; 1096 1097 case OVS_ACTION_ATTR_SAMPLE: 1098 err = sample(dp, skb, key, a, attr, len); 1099 break; 1100 1101 case OVS_ACTION_ATTR_CT: 1102 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key, 1103 nla_data(a)); 1104 1105 /* Hide stolen IP fragments from user space. */ 1106 if (err == -EINPROGRESS) 1107 return 0; 1108 break; 1109 } 1110 1111 if (unlikely(err)) { 1112 kfree_skb(skb); 1113 return err; 1114 } 1115 } 1116 1117 if (prev_port != -1) 1118 do_output(dp, skb, prev_port, key); 1119 else 1120 consume_skb(skb); 1121 1122 return 0; 1123 } 1124 1125 static void process_deferred_actions(struct datapath *dp) 1126 { 1127 struct action_fifo *fifo = this_cpu_ptr(action_fifos); 1128 1129 /* Do not touch the FIFO in case there is no deferred actions. */ 1130 if (action_fifo_is_empty(fifo)) 1131 return; 1132 1133 /* Finishing executing all deferred actions. */ 1134 do { 1135 struct deferred_action *da = action_fifo_get(fifo); 1136 struct sk_buff *skb = da->skb; 1137 struct sw_flow_key *key = &da->pkt_key; 1138 const struct nlattr *actions = da->actions; 1139 1140 if (actions) 1141 do_execute_actions(dp, skb, key, actions, 1142 nla_len(actions)); 1143 else 1144 ovs_dp_process_packet(skb, key); 1145 } while (!action_fifo_is_empty(fifo)); 1146 1147 /* Reset FIFO for the next packet. */ 1148 action_fifo_init(fifo); 1149 } 1150 1151 /* Execute a list of actions against 'skb'. */ 1152 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb, 1153 const struct sw_flow_actions *acts, 1154 struct sw_flow_key *key) 1155 { 1156 int level = this_cpu_read(exec_actions_level); 1157 int err; 1158 1159 this_cpu_inc(exec_actions_level); 1160 err = do_execute_actions(dp, skb, key, 1161 acts->actions, acts->actions_len); 1162 1163 if (!level) 1164 process_deferred_actions(dp); 1165 1166 this_cpu_dec(exec_actions_level); 1167 return err; 1168 } 1169 1170 int action_fifos_init(void) 1171 { 1172 action_fifos = alloc_percpu(struct action_fifo); 1173 if (!action_fifos) 1174 return -ENOMEM; 1175 1176 return 0; 1177 } 1178 1179 void action_fifos_exit(void) 1180 { 1181 free_percpu(action_fifos); 1182 } 1183