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 void update_ethertype(struct sk_buff *skb, struct ethhdr *hdr, 141 __be16 ethertype) 142 { 143 if (skb->ip_summed == CHECKSUM_COMPLETE) { 144 __be16 diff[] = { ~(hdr->h_proto), ethertype }; 145 146 skb->csum = ~csum_partial((char *)diff, sizeof(diff), 147 ~skb->csum); 148 } 149 150 hdr->h_proto = ethertype; 151 } 152 153 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key, 154 const struct ovs_action_push_mpls *mpls) 155 { 156 __be32 *new_mpls_lse; 157 158 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */ 159 if (skb->encapsulation) 160 return -ENOTSUPP; 161 162 if (skb_cow_head(skb, MPLS_HLEN) < 0) 163 return -ENOMEM; 164 165 skb_push(skb, MPLS_HLEN); 166 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb), 167 skb->mac_len); 168 skb_reset_mac_header(skb); 169 170 new_mpls_lse = (__be32 *)skb_mpls_header(skb); 171 *new_mpls_lse = mpls->mpls_lse; 172 173 skb_postpush_rcsum(skb, new_mpls_lse, MPLS_HLEN); 174 175 update_ethertype(skb, eth_hdr(skb), mpls->mpls_ethertype); 176 if (!skb->inner_protocol) 177 skb_set_inner_protocol(skb, skb->protocol); 178 skb->protocol = mpls->mpls_ethertype; 179 180 invalidate_flow_key(key); 181 return 0; 182 } 183 184 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key, 185 const __be16 ethertype) 186 { 187 struct ethhdr *hdr; 188 int err; 189 190 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN); 191 if (unlikely(err)) 192 return err; 193 194 skb_postpull_rcsum(skb, skb_mpls_header(skb), MPLS_HLEN); 195 196 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb), 197 skb->mac_len); 198 199 __skb_pull(skb, MPLS_HLEN); 200 skb_reset_mac_header(skb); 201 202 /* skb_mpls_header() is used to locate the ethertype 203 * field correctly in the presence of VLAN tags. 204 */ 205 hdr = (struct ethhdr *)(skb_mpls_header(skb) - ETH_HLEN); 206 update_ethertype(skb, hdr, ethertype); 207 if (eth_p_mpls(skb->protocol)) 208 skb->protocol = ethertype; 209 210 invalidate_flow_key(key); 211 return 0; 212 } 213 214 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key, 215 const __be32 *mpls_lse, const __be32 *mask) 216 { 217 __be32 *stack; 218 __be32 lse; 219 int err; 220 221 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN); 222 if (unlikely(err)) 223 return err; 224 225 stack = (__be32 *)skb_mpls_header(skb); 226 lse = OVS_MASKED(*stack, *mpls_lse, *mask); 227 if (skb->ip_summed == CHECKSUM_COMPLETE) { 228 __be32 diff[] = { ~(*stack), lse }; 229 230 skb->csum = ~csum_partial((char *)diff, sizeof(diff), 231 ~skb->csum); 232 } 233 234 *stack = lse; 235 flow_key->mpls.top_lse = lse; 236 return 0; 237 } 238 239 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key) 240 { 241 int err; 242 243 err = skb_vlan_pop(skb); 244 if (skb_vlan_tag_present(skb)) 245 invalidate_flow_key(key); 246 else 247 key->eth.tci = 0; 248 return err; 249 } 250 251 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key, 252 const struct ovs_action_push_vlan *vlan) 253 { 254 if (skb_vlan_tag_present(skb)) 255 invalidate_flow_key(key); 256 else 257 key->eth.tci = vlan->vlan_tci; 258 return skb_vlan_push(skb, vlan->vlan_tpid, 259 ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT); 260 } 261 262 /* 'src' is already properly masked. */ 263 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_) 264 { 265 u16 *dst = (u16 *)dst_; 266 const u16 *src = (const u16 *)src_; 267 const u16 *mask = (const u16 *)mask_; 268 269 OVS_SET_MASKED(dst[0], src[0], mask[0]); 270 OVS_SET_MASKED(dst[1], src[1], mask[1]); 271 OVS_SET_MASKED(dst[2], src[2], mask[2]); 272 } 273 274 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key, 275 const struct ovs_key_ethernet *key, 276 const struct ovs_key_ethernet *mask) 277 { 278 int err; 279 280 err = skb_ensure_writable(skb, ETH_HLEN); 281 if (unlikely(err)) 282 return err; 283 284 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2); 285 286 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src, 287 mask->eth_src); 288 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst, 289 mask->eth_dst); 290 291 skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2); 292 293 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source); 294 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest); 295 return 0; 296 } 297 298 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh, 299 __be32 addr, __be32 new_addr) 300 { 301 int transport_len = skb->len - skb_transport_offset(skb); 302 303 if (nh->frag_off & htons(IP_OFFSET)) 304 return; 305 306 if (nh->protocol == IPPROTO_TCP) { 307 if (likely(transport_len >= sizeof(struct tcphdr))) 308 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb, 309 addr, new_addr, true); 310 } else if (nh->protocol == IPPROTO_UDP) { 311 if (likely(transport_len >= sizeof(struct udphdr))) { 312 struct udphdr *uh = udp_hdr(skb); 313 314 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) { 315 inet_proto_csum_replace4(&uh->check, skb, 316 addr, new_addr, true); 317 if (!uh->check) 318 uh->check = CSUM_MANGLED_0; 319 } 320 } 321 } 322 } 323 324 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh, 325 __be32 *addr, __be32 new_addr) 326 { 327 update_ip_l4_checksum(skb, nh, *addr, new_addr); 328 csum_replace4(&nh->check, *addr, new_addr); 329 skb_clear_hash(skb); 330 *addr = new_addr; 331 } 332 333 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto, 334 __be32 addr[4], const __be32 new_addr[4]) 335 { 336 int transport_len = skb->len - skb_transport_offset(skb); 337 338 if (l4_proto == NEXTHDR_TCP) { 339 if (likely(transport_len >= sizeof(struct tcphdr))) 340 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb, 341 addr, new_addr, true); 342 } else if (l4_proto == NEXTHDR_UDP) { 343 if (likely(transport_len >= sizeof(struct udphdr))) { 344 struct udphdr *uh = udp_hdr(skb); 345 346 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) { 347 inet_proto_csum_replace16(&uh->check, skb, 348 addr, new_addr, true); 349 if (!uh->check) 350 uh->check = CSUM_MANGLED_0; 351 } 352 } 353 } else if (l4_proto == NEXTHDR_ICMP) { 354 if (likely(transport_len >= sizeof(struct icmp6hdr))) 355 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum, 356 skb, addr, new_addr, true); 357 } 358 } 359 360 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4], 361 const __be32 mask[4], __be32 masked[4]) 362 { 363 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]); 364 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]); 365 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]); 366 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]); 367 } 368 369 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto, 370 __be32 addr[4], const __be32 new_addr[4], 371 bool recalculate_csum) 372 { 373 if (recalculate_csum) 374 update_ipv6_checksum(skb, l4_proto, addr, new_addr); 375 376 skb_clear_hash(skb); 377 memcpy(addr, new_addr, sizeof(__be32[4])); 378 } 379 380 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask) 381 { 382 /* Bits 21-24 are always unmasked, so this retains their values. */ 383 OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16)); 384 OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8)); 385 OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask); 386 } 387 388 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl, 389 u8 mask) 390 { 391 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask); 392 393 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8)); 394 nh->ttl = new_ttl; 395 } 396 397 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key, 398 const struct ovs_key_ipv4 *key, 399 const struct ovs_key_ipv4 *mask) 400 { 401 struct iphdr *nh; 402 __be32 new_addr; 403 int err; 404 405 err = skb_ensure_writable(skb, skb_network_offset(skb) + 406 sizeof(struct iphdr)); 407 if (unlikely(err)) 408 return err; 409 410 nh = ip_hdr(skb); 411 412 /* Setting an IP addresses is typically only a side effect of 413 * matching on them in the current userspace implementation, so it 414 * makes sense to check if the value actually changed. 415 */ 416 if (mask->ipv4_src) { 417 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src); 418 419 if (unlikely(new_addr != nh->saddr)) { 420 set_ip_addr(skb, nh, &nh->saddr, new_addr); 421 flow_key->ipv4.addr.src = new_addr; 422 } 423 } 424 if (mask->ipv4_dst) { 425 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst); 426 427 if (unlikely(new_addr != nh->daddr)) { 428 set_ip_addr(skb, nh, &nh->daddr, new_addr); 429 flow_key->ipv4.addr.dst = new_addr; 430 } 431 } 432 if (mask->ipv4_tos) { 433 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos); 434 flow_key->ip.tos = nh->tos; 435 } 436 if (mask->ipv4_ttl) { 437 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl); 438 flow_key->ip.ttl = nh->ttl; 439 } 440 441 return 0; 442 } 443 444 static bool is_ipv6_mask_nonzero(const __be32 addr[4]) 445 { 446 return !!(addr[0] | addr[1] | addr[2] | addr[3]); 447 } 448 449 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key, 450 const struct ovs_key_ipv6 *key, 451 const struct ovs_key_ipv6 *mask) 452 { 453 struct ipv6hdr *nh; 454 int err; 455 456 err = skb_ensure_writable(skb, skb_network_offset(skb) + 457 sizeof(struct ipv6hdr)); 458 if (unlikely(err)) 459 return err; 460 461 nh = ipv6_hdr(skb); 462 463 /* Setting an IP addresses is typically only a side effect of 464 * matching on them in the current userspace implementation, so it 465 * makes sense to check if the value actually changed. 466 */ 467 if (is_ipv6_mask_nonzero(mask->ipv6_src)) { 468 __be32 *saddr = (__be32 *)&nh->saddr; 469 __be32 masked[4]; 470 471 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked); 472 473 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) { 474 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked, 475 true); 476 memcpy(&flow_key->ipv6.addr.src, masked, 477 sizeof(flow_key->ipv6.addr.src)); 478 } 479 } 480 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) { 481 unsigned int offset = 0; 482 int flags = IP6_FH_F_SKIP_RH; 483 bool recalc_csum = true; 484 __be32 *daddr = (__be32 *)&nh->daddr; 485 __be32 masked[4]; 486 487 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked); 488 489 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) { 490 if (ipv6_ext_hdr(nh->nexthdr)) 491 recalc_csum = (ipv6_find_hdr(skb, &offset, 492 NEXTHDR_ROUTING, 493 NULL, &flags) 494 != NEXTHDR_ROUTING); 495 496 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked, 497 recalc_csum); 498 memcpy(&flow_key->ipv6.addr.dst, masked, 499 sizeof(flow_key->ipv6.addr.dst)); 500 } 501 } 502 if (mask->ipv6_tclass) { 503 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass); 504 flow_key->ip.tos = ipv6_get_dsfield(nh); 505 } 506 if (mask->ipv6_label) { 507 set_ipv6_fl(nh, ntohl(key->ipv6_label), 508 ntohl(mask->ipv6_label)); 509 flow_key->ipv6.label = 510 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL); 511 } 512 if (mask->ipv6_hlimit) { 513 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit, 514 mask->ipv6_hlimit); 515 flow_key->ip.ttl = nh->hop_limit; 516 } 517 return 0; 518 } 519 520 /* Must follow skb_ensure_writable() since that can move the skb data. */ 521 static void set_tp_port(struct sk_buff *skb, __be16 *port, 522 __be16 new_port, __sum16 *check) 523 { 524 inet_proto_csum_replace2(check, skb, *port, new_port, false); 525 *port = new_port; 526 } 527 528 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key, 529 const struct ovs_key_udp *key, 530 const struct ovs_key_udp *mask) 531 { 532 struct udphdr *uh; 533 __be16 src, dst; 534 int err; 535 536 err = skb_ensure_writable(skb, skb_transport_offset(skb) + 537 sizeof(struct udphdr)); 538 if (unlikely(err)) 539 return err; 540 541 uh = udp_hdr(skb); 542 /* Either of the masks is non-zero, so do not bother checking them. */ 543 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src); 544 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst); 545 546 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) { 547 if (likely(src != uh->source)) { 548 set_tp_port(skb, &uh->source, src, &uh->check); 549 flow_key->tp.src = src; 550 } 551 if (likely(dst != uh->dest)) { 552 set_tp_port(skb, &uh->dest, dst, &uh->check); 553 flow_key->tp.dst = dst; 554 } 555 556 if (unlikely(!uh->check)) 557 uh->check = CSUM_MANGLED_0; 558 } else { 559 uh->source = src; 560 uh->dest = dst; 561 flow_key->tp.src = src; 562 flow_key->tp.dst = dst; 563 } 564 565 skb_clear_hash(skb); 566 567 return 0; 568 } 569 570 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key, 571 const struct ovs_key_tcp *key, 572 const struct ovs_key_tcp *mask) 573 { 574 struct tcphdr *th; 575 __be16 src, dst; 576 int err; 577 578 err = skb_ensure_writable(skb, skb_transport_offset(skb) + 579 sizeof(struct tcphdr)); 580 if (unlikely(err)) 581 return err; 582 583 th = tcp_hdr(skb); 584 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src); 585 if (likely(src != th->source)) { 586 set_tp_port(skb, &th->source, src, &th->check); 587 flow_key->tp.src = src; 588 } 589 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst); 590 if (likely(dst != th->dest)) { 591 set_tp_port(skb, &th->dest, dst, &th->check); 592 flow_key->tp.dst = dst; 593 } 594 skb_clear_hash(skb); 595 596 return 0; 597 } 598 599 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key, 600 const struct ovs_key_sctp *key, 601 const struct ovs_key_sctp *mask) 602 { 603 unsigned int sctphoff = skb_transport_offset(skb); 604 struct sctphdr *sh; 605 __le32 old_correct_csum, new_csum, old_csum; 606 int err; 607 608 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr)); 609 if (unlikely(err)) 610 return err; 611 612 sh = sctp_hdr(skb); 613 old_csum = sh->checksum; 614 old_correct_csum = sctp_compute_cksum(skb, sctphoff); 615 616 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src); 617 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst); 618 619 new_csum = sctp_compute_cksum(skb, sctphoff); 620 621 /* Carry any checksum errors through. */ 622 sh->checksum = old_csum ^ old_correct_csum ^ new_csum; 623 624 skb_clear_hash(skb); 625 flow_key->tp.src = sh->source; 626 flow_key->tp.dst = sh->dest; 627 628 return 0; 629 } 630 631 static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb) 632 { 633 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage); 634 struct vport *vport = data->vport; 635 636 if (skb_cow_head(skb, data->l2_len) < 0) { 637 kfree_skb(skb); 638 return -ENOMEM; 639 } 640 641 __skb_dst_copy(skb, data->dst); 642 *OVS_CB(skb) = data->cb; 643 skb->inner_protocol = data->inner_protocol; 644 skb->vlan_tci = data->vlan_tci; 645 skb->vlan_proto = data->vlan_proto; 646 647 /* Reconstruct the MAC header. */ 648 skb_push(skb, data->l2_len); 649 memcpy(skb->data, &data->l2_data, data->l2_len); 650 skb_postpush_rcsum(skb, skb->data, data->l2_len); 651 skb_reset_mac_header(skb); 652 653 ovs_vport_send(vport, skb); 654 return 0; 655 } 656 657 static unsigned int 658 ovs_dst_get_mtu(const struct dst_entry *dst) 659 { 660 return dst->dev->mtu; 661 } 662 663 static struct dst_ops ovs_dst_ops = { 664 .family = AF_UNSPEC, 665 .mtu = ovs_dst_get_mtu, 666 }; 667 668 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is 669 * ovs_vport_output(), which is called once per fragmented packet. 670 */ 671 static void prepare_frag(struct vport *vport, struct sk_buff *skb) 672 { 673 unsigned int hlen = skb_network_offset(skb); 674 struct ovs_frag_data *data; 675 676 data = this_cpu_ptr(&ovs_frag_data_storage); 677 data->dst = skb->_skb_refdst; 678 data->vport = vport; 679 data->cb = *OVS_CB(skb); 680 data->inner_protocol = skb->inner_protocol; 681 data->vlan_tci = skb->vlan_tci; 682 data->vlan_proto = skb->vlan_proto; 683 data->l2_len = hlen; 684 memcpy(&data->l2_data, skb->data, hlen); 685 686 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); 687 skb_pull(skb, hlen); 688 } 689 690 static void ovs_fragment(struct net *net, struct vport *vport, 691 struct sk_buff *skb, u16 mru, __be16 ethertype) 692 { 693 if (skb_network_offset(skb) > MAX_L2_LEN) { 694 OVS_NLERR(1, "L2 header too long to fragment"); 695 goto err; 696 } 697 698 if (ethertype == htons(ETH_P_IP)) { 699 struct dst_entry ovs_dst; 700 unsigned long orig_dst; 701 702 prepare_frag(vport, skb); 703 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1, 704 DST_OBSOLETE_NONE, DST_NOCOUNT); 705 ovs_dst.dev = vport->dev; 706 707 orig_dst = skb->_skb_refdst; 708 skb_dst_set_noref(skb, &ovs_dst); 709 IPCB(skb)->frag_max_size = mru; 710 711 ip_do_fragment(net, skb->sk, skb, ovs_vport_output); 712 refdst_drop(orig_dst); 713 } else if (ethertype == htons(ETH_P_IPV6)) { 714 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops(); 715 unsigned long orig_dst; 716 struct rt6_info ovs_rt; 717 718 if (!v6ops) { 719 goto err; 720 } 721 722 prepare_frag(vport, skb); 723 memset(&ovs_rt, 0, sizeof(ovs_rt)); 724 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1, 725 DST_OBSOLETE_NONE, DST_NOCOUNT); 726 ovs_rt.dst.dev = vport->dev; 727 728 orig_dst = skb->_skb_refdst; 729 skb_dst_set_noref(skb, &ovs_rt.dst); 730 IP6CB(skb)->frag_max_size = mru; 731 732 v6ops->fragment(net, skb->sk, skb, ovs_vport_output); 733 refdst_drop(orig_dst); 734 } else { 735 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.", 736 ovs_vport_name(vport), ntohs(ethertype), mru, 737 vport->dev->mtu); 738 goto err; 739 } 740 741 return; 742 err: 743 kfree_skb(skb); 744 } 745 746 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port, 747 struct sw_flow_key *key) 748 { 749 struct vport *vport = ovs_vport_rcu(dp, out_port); 750 751 if (likely(vport)) { 752 u16 mru = OVS_CB(skb)->mru; 753 754 if (likely(!mru || (skb->len <= mru + ETH_HLEN))) { 755 ovs_vport_send(vport, skb); 756 } else if (mru <= vport->dev->mtu) { 757 struct net *net = read_pnet(&dp->net); 758 __be16 ethertype = key->eth.type; 759 760 if (!is_flow_key_valid(key)) { 761 if (eth_p_mpls(skb->protocol)) 762 ethertype = skb->inner_protocol; 763 else 764 ethertype = vlan_get_protocol(skb); 765 } 766 767 ovs_fragment(net, vport, skb, mru, ethertype); 768 } else { 769 kfree_skb(skb); 770 } 771 } else { 772 kfree_skb(skb); 773 } 774 } 775 776 static int output_userspace(struct datapath *dp, struct sk_buff *skb, 777 struct sw_flow_key *key, const struct nlattr *attr, 778 const struct nlattr *actions, int actions_len) 779 { 780 struct dp_upcall_info upcall; 781 const struct nlattr *a; 782 int rem; 783 784 memset(&upcall, 0, sizeof(upcall)); 785 upcall.cmd = OVS_PACKET_CMD_ACTION; 786 upcall.mru = OVS_CB(skb)->mru; 787 788 for (a = nla_data(attr), rem = nla_len(attr); rem > 0; 789 a = nla_next(a, &rem)) { 790 switch (nla_type(a)) { 791 case OVS_USERSPACE_ATTR_USERDATA: 792 upcall.userdata = a; 793 break; 794 795 case OVS_USERSPACE_ATTR_PID: 796 upcall.portid = nla_get_u32(a); 797 break; 798 799 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: { 800 /* Get out tunnel info. */ 801 struct vport *vport; 802 803 vport = ovs_vport_rcu(dp, nla_get_u32(a)); 804 if (vport) { 805 int err; 806 807 err = dev_fill_metadata_dst(vport->dev, skb); 808 if (!err) 809 upcall.egress_tun_info = skb_tunnel_info(skb); 810 } 811 812 break; 813 } 814 815 case OVS_USERSPACE_ATTR_ACTIONS: { 816 /* Include actions. */ 817 upcall.actions = actions; 818 upcall.actions_len = actions_len; 819 break; 820 } 821 822 } /* End of switch. */ 823 } 824 825 return ovs_dp_upcall(dp, skb, key, &upcall); 826 } 827 828 static int sample(struct datapath *dp, struct sk_buff *skb, 829 struct sw_flow_key *key, const struct nlattr *attr, 830 const struct nlattr *actions, int actions_len) 831 { 832 const struct nlattr *acts_list = NULL; 833 const struct nlattr *a; 834 int rem; 835 836 for (a = nla_data(attr), rem = nla_len(attr); rem > 0; 837 a = nla_next(a, &rem)) { 838 u32 probability; 839 840 switch (nla_type(a)) { 841 case OVS_SAMPLE_ATTR_PROBABILITY: 842 probability = nla_get_u32(a); 843 if (!probability || prandom_u32() > probability) 844 return 0; 845 break; 846 847 case OVS_SAMPLE_ATTR_ACTIONS: 848 acts_list = a; 849 break; 850 } 851 } 852 853 rem = nla_len(acts_list); 854 a = nla_data(acts_list); 855 856 /* Actions list is empty, do nothing */ 857 if (unlikely(!rem)) 858 return 0; 859 860 /* The only known usage of sample action is having a single user-space 861 * action. Treat this usage as a special case. 862 * The output_userspace() should clone the skb to be sent to the 863 * user space. This skb will be consumed by its caller. 864 */ 865 if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE && 866 nla_is_last(a, rem))) 867 return output_userspace(dp, skb, key, a, actions, actions_len); 868 869 skb = skb_clone(skb, GFP_ATOMIC); 870 if (!skb) 871 /* Skip the sample action when out of memory. */ 872 return 0; 873 874 if (!add_deferred_actions(skb, key, a)) { 875 if (net_ratelimit()) 876 pr_warn("%s: deferred actions limit reached, dropping sample action\n", 877 ovs_dp_name(dp)); 878 879 kfree_skb(skb); 880 } 881 return 0; 882 } 883 884 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key, 885 const struct nlattr *attr) 886 { 887 struct ovs_action_hash *hash_act = nla_data(attr); 888 u32 hash = 0; 889 890 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */ 891 hash = skb_get_hash(skb); 892 hash = jhash_1word(hash, hash_act->hash_basis); 893 if (!hash) 894 hash = 0x1; 895 896 key->ovs_flow_hash = hash; 897 } 898 899 static int execute_set_action(struct sk_buff *skb, 900 struct sw_flow_key *flow_key, 901 const struct nlattr *a) 902 { 903 /* Only tunnel set execution is supported without a mask. */ 904 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) { 905 struct ovs_tunnel_info *tun = nla_data(a); 906 907 skb_dst_drop(skb); 908 dst_hold((struct dst_entry *)tun->tun_dst); 909 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst); 910 return 0; 911 } 912 913 return -EINVAL; 914 } 915 916 /* Mask is at the midpoint of the data. */ 917 #define get_mask(a, type) ((const type)nla_data(a) + 1) 918 919 static int execute_masked_set_action(struct sk_buff *skb, 920 struct sw_flow_key *flow_key, 921 const struct nlattr *a) 922 { 923 int err = 0; 924 925 switch (nla_type(a)) { 926 case OVS_KEY_ATTR_PRIORITY: 927 OVS_SET_MASKED(skb->priority, nla_get_u32(a), 928 *get_mask(a, u32 *)); 929 flow_key->phy.priority = skb->priority; 930 break; 931 932 case OVS_KEY_ATTR_SKB_MARK: 933 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *)); 934 flow_key->phy.skb_mark = skb->mark; 935 break; 936 937 case OVS_KEY_ATTR_TUNNEL_INFO: 938 /* Masked data not supported for tunnel. */ 939 err = -EINVAL; 940 break; 941 942 case OVS_KEY_ATTR_ETHERNET: 943 err = set_eth_addr(skb, flow_key, nla_data(a), 944 get_mask(a, struct ovs_key_ethernet *)); 945 break; 946 947 case OVS_KEY_ATTR_IPV4: 948 err = set_ipv4(skb, flow_key, nla_data(a), 949 get_mask(a, struct ovs_key_ipv4 *)); 950 break; 951 952 case OVS_KEY_ATTR_IPV6: 953 err = set_ipv6(skb, flow_key, nla_data(a), 954 get_mask(a, struct ovs_key_ipv6 *)); 955 break; 956 957 case OVS_KEY_ATTR_TCP: 958 err = set_tcp(skb, flow_key, nla_data(a), 959 get_mask(a, struct ovs_key_tcp *)); 960 break; 961 962 case OVS_KEY_ATTR_UDP: 963 err = set_udp(skb, flow_key, nla_data(a), 964 get_mask(a, struct ovs_key_udp *)); 965 break; 966 967 case OVS_KEY_ATTR_SCTP: 968 err = set_sctp(skb, flow_key, nla_data(a), 969 get_mask(a, struct ovs_key_sctp *)); 970 break; 971 972 case OVS_KEY_ATTR_MPLS: 973 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a, 974 __be32 *)); 975 break; 976 977 case OVS_KEY_ATTR_CT_STATE: 978 case OVS_KEY_ATTR_CT_ZONE: 979 case OVS_KEY_ATTR_CT_MARK: 980 case OVS_KEY_ATTR_CT_LABELS: 981 err = -EINVAL; 982 break; 983 } 984 985 return err; 986 } 987 988 static int execute_recirc(struct datapath *dp, struct sk_buff *skb, 989 struct sw_flow_key *key, 990 const struct nlattr *a, int rem) 991 { 992 struct deferred_action *da; 993 994 if (!is_flow_key_valid(key)) { 995 int err; 996 997 err = ovs_flow_key_update(skb, key); 998 if (err) 999 return err; 1000 } 1001 BUG_ON(!is_flow_key_valid(key)); 1002 1003 if (!nla_is_last(a, rem)) { 1004 /* Recirc action is the not the last action 1005 * of the action list, need to clone the skb. 1006 */ 1007 skb = skb_clone(skb, GFP_ATOMIC); 1008 1009 /* Skip the recirc action when out of memory, but 1010 * continue on with the rest of the action list. 1011 */ 1012 if (!skb) 1013 return 0; 1014 } 1015 1016 da = add_deferred_actions(skb, key, NULL); 1017 if (da) { 1018 da->pkt_key.recirc_id = nla_get_u32(a); 1019 } else { 1020 kfree_skb(skb); 1021 1022 if (net_ratelimit()) 1023 pr_warn("%s: deferred action limit reached, drop recirc action\n", 1024 ovs_dp_name(dp)); 1025 } 1026 1027 return 0; 1028 } 1029 1030 /* Execute a list of actions against 'skb'. */ 1031 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb, 1032 struct sw_flow_key *key, 1033 const struct nlattr *attr, int len) 1034 { 1035 /* Every output action needs a separate clone of 'skb', but the common 1036 * case is just a single output action, so that doing a clone and 1037 * then freeing the original skbuff is wasteful. So the following code 1038 * is slightly obscure just to avoid that. 1039 */ 1040 int prev_port = -1; 1041 const struct nlattr *a; 1042 int rem; 1043 1044 for (a = attr, rem = len; rem > 0; 1045 a = nla_next(a, &rem)) { 1046 int err = 0; 1047 1048 if (unlikely(prev_port != -1)) { 1049 struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC); 1050 1051 if (out_skb) 1052 do_output(dp, out_skb, prev_port, key); 1053 1054 prev_port = -1; 1055 } 1056 1057 switch (nla_type(a)) { 1058 case OVS_ACTION_ATTR_OUTPUT: 1059 prev_port = nla_get_u32(a); 1060 break; 1061 1062 case OVS_ACTION_ATTR_USERSPACE: 1063 output_userspace(dp, skb, key, a, attr, len); 1064 break; 1065 1066 case OVS_ACTION_ATTR_HASH: 1067 execute_hash(skb, key, a); 1068 break; 1069 1070 case OVS_ACTION_ATTR_PUSH_MPLS: 1071 err = push_mpls(skb, key, nla_data(a)); 1072 break; 1073 1074 case OVS_ACTION_ATTR_POP_MPLS: 1075 err = pop_mpls(skb, key, nla_get_be16(a)); 1076 break; 1077 1078 case OVS_ACTION_ATTR_PUSH_VLAN: 1079 err = push_vlan(skb, key, nla_data(a)); 1080 break; 1081 1082 case OVS_ACTION_ATTR_POP_VLAN: 1083 err = pop_vlan(skb, key); 1084 break; 1085 1086 case OVS_ACTION_ATTR_RECIRC: 1087 err = execute_recirc(dp, skb, key, a, rem); 1088 if (nla_is_last(a, rem)) { 1089 /* If this is the last action, the skb has 1090 * been consumed or freed. 1091 * Return immediately. 1092 */ 1093 return err; 1094 } 1095 break; 1096 1097 case OVS_ACTION_ATTR_SET: 1098 err = execute_set_action(skb, key, nla_data(a)); 1099 break; 1100 1101 case OVS_ACTION_ATTR_SET_MASKED: 1102 case OVS_ACTION_ATTR_SET_TO_MASKED: 1103 err = execute_masked_set_action(skb, key, nla_data(a)); 1104 break; 1105 1106 case OVS_ACTION_ATTR_SAMPLE: 1107 err = sample(dp, skb, key, a, attr, len); 1108 break; 1109 1110 case OVS_ACTION_ATTR_CT: 1111 if (!is_flow_key_valid(key)) { 1112 err = ovs_flow_key_update(skb, key); 1113 if (err) 1114 return err; 1115 } 1116 1117 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key, 1118 nla_data(a)); 1119 1120 /* Hide stolen IP fragments from user space. */ 1121 if (err) 1122 return err == -EINPROGRESS ? 0 : err; 1123 break; 1124 } 1125 1126 if (unlikely(err)) { 1127 kfree_skb(skb); 1128 return err; 1129 } 1130 } 1131 1132 if (prev_port != -1) 1133 do_output(dp, skb, prev_port, key); 1134 else 1135 consume_skb(skb); 1136 1137 return 0; 1138 } 1139 1140 static void process_deferred_actions(struct datapath *dp) 1141 { 1142 struct action_fifo *fifo = this_cpu_ptr(action_fifos); 1143 1144 /* Do not touch the FIFO in case there is no deferred actions. */ 1145 if (action_fifo_is_empty(fifo)) 1146 return; 1147 1148 /* Finishing executing all deferred actions. */ 1149 do { 1150 struct deferred_action *da = action_fifo_get(fifo); 1151 struct sk_buff *skb = da->skb; 1152 struct sw_flow_key *key = &da->pkt_key; 1153 const struct nlattr *actions = da->actions; 1154 1155 if (actions) 1156 do_execute_actions(dp, skb, key, actions, 1157 nla_len(actions)); 1158 else 1159 ovs_dp_process_packet(skb, key); 1160 } while (!action_fifo_is_empty(fifo)); 1161 1162 /* Reset FIFO for the next packet. */ 1163 action_fifo_init(fifo); 1164 } 1165 1166 /* Execute a list of actions against 'skb'. */ 1167 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb, 1168 const struct sw_flow_actions *acts, 1169 struct sw_flow_key *key) 1170 { 1171 static const int ovs_recursion_limit = 5; 1172 int err, level; 1173 1174 level = __this_cpu_inc_return(exec_actions_level); 1175 if (unlikely(level > ovs_recursion_limit)) { 1176 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n", 1177 ovs_dp_name(dp)); 1178 kfree_skb(skb); 1179 err = -ENETDOWN; 1180 goto out; 1181 } 1182 1183 err = do_execute_actions(dp, skb, key, 1184 acts->actions, acts->actions_len); 1185 1186 if (level == 1) 1187 process_deferred_actions(dp); 1188 1189 out: 1190 __this_cpu_dec(exec_actions_level); 1191 return err; 1192 } 1193 1194 int action_fifos_init(void) 1195 { 1196 action_fifos = alloc_percpu(struct action_fifo); 1197 if (!action_fifos) 1198 return -ENOMEM; 1199 1200 return 0; 1201 } 1202 1203 void action_fifos_exit(void) 1204 { 1205 free_percpu(action_fifos); 1206 } 1207