1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (c) 2007-2017 Nicira, Inc. 4 */ 5 6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 7 8 #include <linux/skbuff.h> 9 #include <linux/in.h> 10 #include <linux/ip.h> 11 #include <linux/openvswitch.h> 12 #include <linux/sctp.h> 13 #include <linux/tcp.h> 14 #include <linux/udp.h> 15 #include <linux/in6.h> 16 #include <linux/if_arp.h> 17 #include <linux/if_vlan.h> 18 19 #include <net/dst.h> 20 #include <net/ip.h> 21 #include <net/ipv6.h> 22 #include <net/ip6_fib.h> 23 #include <net/checksum.h> 24 #include <net/dsfield.h> 25 #include <net/mpls.h> 26 #include <net/sctp/checksum.h> 27 28 #include "datapath.h" 29 #include "flow.h" 30 #include "conntrack.h" 31 #include "vport.h" 32 #include "flow_netlink.h" 33 34 struct deferred_action { 35 struct sk_buff *skb; 36 const struct nlattr *actions; 37 int actions_len; 38 39 /* Store pkt_key clone when creating deferred action. */ 40 struct sw_flow_key pkt_key; 41 }; 42 43 #define MAX_L2_LEN (VLAN_ETH_HLEN + 3 * MPLS_HLEN) 44 struct ovs_frag_data { 45 unsigned long dst; 46 struct vport *vport; 47 struct ovs_skb_cb cb; 48 __be16 inner_protocol; 49 u16 network_offset; /* valid only for MPLS */ 50 u16 vlan_tci; 51 __be16 vlan_proto; 52 unsigned int l2_len; 53 u8 mac_proto; 54 u8 l2_data[MAX_L2_LEN]; 55 }; 56 57 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage); 58 59 #define DEFERRED_ACTION_FIFO_SIZE 10 60 #define OVS_RECURSION_LIMIT 5 61 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2) 62 struct action_fifo { 63 int head; 64 int tail; 65 /* Deferred action fifo queue storage. */ 66 struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE]; 67 }; 68 69 struct action_flow_keys { 70 struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD]; 71 }; 72 73 static struct action_fifo __percpu *action_fifos; 74 static struct action_flow_keys __percpu *flow_keys; 75 static DEFINE_PER_CPU(int, exec_actions_level); 76 77 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys' 78 * space. Return NULL if out of key spaces. 79 */ 80 static struct sw_flow_key *clone_key(const struct sw_flow_key *key_) 81 { 82 struct action_flow_keys *keys = this_cpu_ptr(flow_keys); 83 int level = this_cpu_read(exec_actions_level); 84 struct sw_flow_key *key = NULL; 85 86 if (level <= OVS_DEFERRED_ACTION_THRESHOLD) { 87 key = &keys->key[level - 1]; 88 *key = *key_; 89 } 90 91 return key; 92 } 93 94 static void action_fifo_init(struct action_fifo *fifo) 95 { 96 fifo->head = 0; 97 fifo->tail = 0; 98 } 99 100 static bool action_fifo_is_empty(const struct action_fifo *fifo) 101 { 102 return (fifo->head == fifo->tail); 103 } 104 105 static struct deferred_action *action_fifo_get(struct action_fifo *fifo) 106 { 107 if (action_fifo_is_empty(fifo)) 108 return NULL; 109 110 return &fifo->fifo[fifo->tail++]; 111 } 112 113 static struct deferred_action *action_fifo_put(struct action_fifo *fifo) 114 { 115 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1) 116 return NULL; 117 118 return &fifo->fifo[fifo->head++]; 119 } 120 121 /* Return true if fifo is not full */ 122 static struct deferred_action *add_deferred_actions(struct sk_buff *skb, 123 const struct sw_flow_key *key, 124 const struct nlattr *actions, 125 const int actions_len) 126 { 127 struct action_fifo *fifo; 128 struct deferred_action *da; 129 130 fifo = this_cpu_ptr(action_fifos); 131 da = action_fifo_put(fifo); 132 if (da) { 133 da->skb = skb; 134 da->actions = actions; 135 da->actions_len = actions_len; 136 da->pkt_key = *key; 137 } 138 139 return da; 140 } 141 142 static void invalidate_flow_key(struct sw_flow_key *key) 143 { 144 key->mac_proto |= SW_FLOW_KEY_INVALID; 145 } 146 147 static bool is_flow_key_valid(const struct sw_flow_key *key) 148 { 149 return !(key->mac_proto & SW_FLOW_KEY_INVALID); 150 } 151 152 static int clone_execute(struct datapath *dp, struct sk_buff *skb, 153 struct sw_flow_key *key, 154 u32 recirc_id, 155 const struct nlattr *actions, int len, 156 bool last, bool clone_flow_key); 157 158 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb, 159 struct sw_flow_key *key, 160 const struct nlattr *attr, int len); 161 162 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key, 163 __be32 mpls_lse, __be16 mpls_ethertype, __u16 mac_len) 164 { 165 int err; 166 167 err = skb_mpls_push(skb, mpls_lse, mpls_ethertype, mac_len, !!mac_len); 168 if (err) 169 return err; 170 171 if (!mac_len) 172 key->mac_proto = MAC_PROTO_NONE; 173 174 invalidate_flow_key(key); 175 return 0; 176 } 177 178 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key, 179 const __be16 ethertype) 180 { 181 int err; 182 183 err = skb_mpls_pop(skb, ethertype, skb->mac_len, 184 ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET); 185 if (err) 186 return err; 187 188 if (ethertype == htons(ETH_P_TEB)) 189 key->mac_proto = MAC_PROTO_ETHERNET; 190 191 invalidate_flow_key(key); 192 return 0; 193 } 194 195 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key, 196 const __be32 *mpls_lse, const __be32 *mask) 197 { 198 struct mpls_shim_hdr *stack; 199 __be32 lse; 200 int err; 201 202 if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN)) 203 return -ENOMEM; 204 205 stack = mpls_hdr(skb); 206 lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask); 207 err = skb_mpls_update_lse(skb, lse); 208 if (err) 209 return err; 210 211 flow_key->mpls.lse[0] = lse; 212 return 0; 213 } 214 215 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key) 216 { 217 int err; 218 219 err = skb_vlan_pop(skb); 220 if (skb_vlan_tag_present(skb)) { 221 invalidate_flow_key(key); 222 } else { 223 key->eth.vlan.tci = 0; 224 key->eth.vlan.tpid = 0; 225 } 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.vlan.tci = vlan->vlan_tci; 236 key->eth.vlan.tpid = vlan->vlan_tpid; 237 } 238 return skb_vlan_push(skb, vlan->vlan_tpid, 239 ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK); 240 } 241 242 /* 'src' is already properly masked. */ 243 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_) 244 { 245 u16 *dst = (u16 *)dst_; 246 const u16 *src = (const u16 *)src_; 247 const u16 *mask = (const u16 *)mask_; 248 249 OVS_SET_MASKED(dst[0], src[0], mask[0]); 250 OVS_SET_MASKED(dst[1], src[1], mask[1]); 251 OVS_SET_MASKED(dst[2], src[2], mask[2]); 252 } 253 254 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key, 255 const struct ovs_key_ethernet *key, 256 const struct ovs_key_ethernet *mask) 257 { 258 int err; 259 260 err = skb_ensure_writable(skb, ETH_HLEN); 261 if (unlikely(err)) 262 return err; 263 264 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2); 265 266 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src, 267 mask->eth_src); 268 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst, 269 mask->eth_dst); 270 271 skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2); 272 273 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source); 274 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest); 275 return 0; 276 } 277 278 /* pop_eth does not support VLAN packets as this action is never called 279 * for them. 280 */ 281 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key) 282 { 283 int err; 284 285 err = skb_eth_pop(skb); 286 if (err) 287 return err; 288 289 /* safe right before invalidate_flow_key */ 290 key->mac_proto = MAC_PROTO_NONE; 291 invalidate_flow_key(key); 292 return 0; 293 } 294 295 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key, 296 const struct ovs_action_push_eth *ethh) 297 { 298 int err; 299 300 err = skb_eth_push(skb, ethh->addresses.eth_dst, 301 ethh->addresses.eth_src); 302 if (err) 303 return err; 304 305 /* safe right before invalidate_flow_key */ 306 key->mac_proto = MAC_PROTO_ETHERNET; 307 invalidate_flow_key(key); 308 return 0; 309 } 310 311 static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key, 312 const struct nshhdr *nh) 313 { 314 int err; 315 316 err = nsh_push(skb, nh); 317 if (err) 318 return err; 319 320 /* safe right before invalidate_flow_key */ 321 key->mac_proto = MAC_PROTO_NONE; 322 invalidate_flow_key(key); 323 return 0; 324 } 325 326 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key) 327 { 328 int err; 329 330 err = nsh_pop(skb); 331 if (err) 332 return err; 333 334 /* safe right before invalidate_flow_key */ 335 if (skb->protocol == htons(ETH_P_TEB)) 336 key->mac_proto = MAC_PROTO_ETHERNET; 337 else 338 key->mac_proto = MAC_PROTO_NONE; 339 invalidate_flow_key(key); 340 return 0; 341 } 342 343 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh, 344 __be32 addr, __be32 new_addr) 345 { 346 int transport_len = skb->len - skb_transport_offset(skb); 347 348 if (nh->frag_off & htons(IP_OFFSET)) 349 return; 350 351 if (nh->protocol == IPPROTO_TCP) { 352 if (likely(transport_len >= sizeof(struct tcphdr))) 353 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb, 354 addr, new_addr, true); 355 } else if (nh->protocol == IPPROTO_UDP) { 356 if (likely(transport_len >= sizeof(struct udphdr))) { 357 struct udphdr *uh = udp_hdr(skb); 358 359 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) { 360 inet_proto_csum_replace4(&uh->check, skb, 361 addr, new_addr, true); 362 if (!uh->check) 363 uh->check = CSUM_MANGLED_0; 364 } 365 } 366 } 367 } 368 369 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh, 370 __be32 *addr, __be32 new_addr) 371 { 372 update_ip_l4_checksum(skb, nh, *addr, new_addr); 373 csum_replace4(&nh->check, *addr, new_addr); 374 skb_clear_hash(skb); 375 *addr = new_addr; 376 } 377 378 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto, 379 __be32 addr[4], const __be32 new_addr[4]) 380 { 381 int transport_len = skb->len - skb_transport_offset(skb); 382 383 if (l4_proto == NEXTHDR_TCP) { 384 if (likely(transport_len >= sizeof(struct tcphdr))) 385 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb, 386 addr, new_addr, true); 387 } else if (l4_proto == NEXTHDR_UDP) { 388 if (likely(transport_len >= sizeof(struct udphdr))) { 389 struct udphdr *uh = udp_hdr(skb); 390 391 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) { 392 inet_proto_csum_replace16(&uh->check, skb, 393 addr, new_addr, true); 394 if (!uh->check) 395 uh->check = CSUM_MANGLED_0; 396 } 397 } 398 } else if (l4_proto == NEXTHDR_ICMP) { 399 if (likely(transport_len >= sizeof(struct icmp6hdr))) 400 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum, 401 skb, addr, new_addr, true); 402 } 403 } 404 405 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4], 406 const __be32 mask[4], __be32 masked[4]) 407 { 408 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]); 409 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]); 410 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]); 411 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]); 412 } 413 414 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto, 415 __be32 addr[4], const __be32 new_addr[4], 416 bool recalculate_csum) 417 { 418 if (recalculate_csum) 419 update_ipv6_checksum(skb, l4_proto, addr, new_addr); 420 421 skb_clear_hash(skb); 422 memcpy(addr, new_addr, sizeof(__be32[4])); 423 } 424 425 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask) 426 { 427 /* Bits 21-24 are always unmasked, so this retains their values. */ 428 OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16)); 429 OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8)); 430 OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask); 431 } 432 433 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl, 434 u8 mask) 435 { 436 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask); 437 438 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8)); 439 nh->ttl = new_ttl; 440 } 441 442 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key, 443 const struct ovs_key_ipv4 *key, 444 const struct ovs_key_ipv4 *mask) 445 { 446 struct iphdr *nh; 447 __be32 new_addr; 448 int err; 449 450 err = skb_ensure_writable(skb, skb_network_offset(skb) + 451 sizeof(struct iphdr)); 452 if (unlikely(err)) 453 return err; 454 455 nh = ip_hdr(skb); 456 457 /* Setting an IP addresses is typically only a side effect of 458 * matching on them in the current userspace implementation, so it 459 * makes sense to check if the value actually changed. 460 */ 461 if (mask->ipv4_src) { 462 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src); 463 464 if (unlikely(new_addr != nh->saddr)) { 465 set_ip_addr(skb, nh, &nh->saddr, new_addr); 466 flow_key->ipv4.addr.src = new_addr; 467 } 468 } 469 if (mask->ipv4_dst) { 470 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst); 471 472 if (unlikely(new_addr != nh->daddr)) { 473 set_ip_addr(skb, nh, &nh->daddr, new_addr); 474 flow_key->ipv4.addr.dst = new_addr; 475 } 476 } 477 if (mask->ipv4_tos) { 478 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos); 479 flow_key->ip.tos = nh->tos; 480 } 481 if (mask->ipv4_ttl) { 482 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl); 483 flow_key->ip.ttl = nh->ttl; 484 } 485 486 return 0; 487 } 488 489 static bool is_ipv6_mask_nonzero(const __be32 addr[4]) 490 { 491 return !!(addr[0] | addr[1] | addr[2] | addr[3]); 492 } 493 494 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key, 495 const struct ovs_key_ipv6 *key, 496 const struct ovs_key_ipv6 *mask) 497 { 498 struct ipv6hdr *nh; 499 int err; 500 501 err = skb_ensure_writable(skb, skb_network_offset(skb) + 502 sizeof(struct ipv6hdr)); 503 if (unlikely(err)) 504 return err; 505 506 nh = ipv6_hdr(skb); 507 508 /* Setting an IP addresses is typically only a side effect of 509 * matching on them in the current userspace implementation, so it 510 * makes sense to check if the value actually changed. 511 */ 512 if (is_ipv6_mask_nonzero(mask->ipv6_src)) { 513 __be32 *saddr = (__be32 *)&nh->saddr; 514 __be32 masked[4]; 515 516 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked); 517 518 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) { 519 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked, 520 true); 521 memcpy(&flow_key->ipv6.addr.src, masked, 522 sizeof(flow_key->ipv6.addr.src)); 523 } 524 } 525 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) { 526 unsigned int offset = 0; 527 int flags = IP6_FH_F_SKIP_RH; 528 bool recalc_csum = true; 529 __be32 *daddr = (__be32 *)&nh->daddr; 530 __be32 masked[4]; 531 532 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked); 533 534 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) { 535 if (ipv6_ext_hdr(nh->nexthdr)) 536 recalc_csum = (ipv6_find_hdr(skb, &offset, 537 NEXTHDR_ROUTING, 538 NULL, &flags) 539 != NEXTHDR_ROUTING); 540 541 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked, 542 recalc_csum); 543 memcpy(&flow_key->ipv6.addr.dst, masked, 544 sizeof(flow_key->ipv6.addr.dst)); 545 } 546 } 547 if (mask->ipv6_tclass) { 548 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass); 549 flow_key->ip.tos = ipv6_get_dsfield(nh); 550 } 551 if (mask->ipv6_label) { 552 set_ipv6_fl(nh, ntohl(key->ipv6_label), 553 ntohl(mask->ipv6_label)); 554 flow_key->ipv6.label = 555 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL); 556 } 557 if (mask->ipv6_hlimit) { 558 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit, 559 mask->ipv6_hlimit); 560 flow_key->ip.ttl = nh->hop_limit; 561 } 562 return 0; 563 } 564 565 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key, 566 const struct nlattr *a) 567 { 568 struct nshhdr *nh; 569 size_t length; 570 int err; 571 u8 flags; 572 u8 ttl; 573 int i; 574 575 struct ovs_key_nsh key; 576 struct ovs_key_nsh mask; 577 578 err = nsh_key_from_nlattr(a, &key, &mask); 579 if (err) 580 return err; 581 582 /* Make sure the NSH base header is there */ 583 if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN)) 584 return -ENOMEM; 585 586 nh = nsh_hdr(skb); 587 length = nsh_hdr_len(nh); 588 589 /* Make sure the whole NSH header is there */ 590 err = skb_ensure_writable(skb, skb_network_offset(skb) + 591 length); 592 if (unlikely(err)) 593 return err; 594 595 nh = nsh_hdr(skb); 596 skb_postpull_rcsum(skb, nh, length); 597 flags = nsh_get_flags(nh); 598 flags = OVS_MASKED(flags, key.base.flags, mask.base.flags); 599 flow_key->nsh.base.flags = flags; 600 ttl = nsh_get_ttl(nh); 601 ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl); 602 flow_key->nsh.base.ttl = ttl; 603 nsh_set_flags_and_ttl(nh, flags, ttl); 604 nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr, 605 mask.base.path_hdr); 606 flow_key->nsh.base.path_hdr = nh->path_hdr; 607 switch (nh->mdtype) { 608 case NSH_M_TYPE1: 609 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) { 610 nh->md1.context[i] = 611 OVS_MASKED(nh->md1.context[i], key.context[i], 612 mask.context[i]); 613 } 614 memcpy(flow_key->nsh.context, nh->md1.context, 615 sizeof(nh->md1.context)); 616 break; 617 case NSH_M_TYPE2: 618 memset(flow_key->nsh.context, 0, 619 sizeof(flow_key->nsh.context)); 620 break; 621 default: 622 return -EINVAL; 623 } 624 skb_postpush_rcsum(skb, nh, length); 625 return 0; 626 } 627 628 /* Must follow skb_ensure_writable() since that can move the skb data. */ 629 static void set_tp_port(struct sk_buff *skb, __be16 *port, 630 __be16 new_port, __sum16 *check) 631 { 632 inet_proto_csum_replace2(check, skb, *port, new_port, false); 633 *port = new_port; 634 } 635 636 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key, 637 const struct ovs_key_udp *key, 638 const struct ovs_key_udp *mask) 639 { 640 struct udphdr *uh; 641 __be16 src, dst; 642 int err; 643 644 err = skb_ensure_writable(skb, skb_transport_offset(skb) + 645 sizeof(struct udphdr)); 646 if (unlikely(err)) 647 return err; 648 649 uh = udp_hdr(skb); 650 /* Either of the masks is non-zero, so do not bother checking them. */ 651 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src); 652 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst); 653 654 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) { 655 if (likely(src != uh->source)) { 656 set_tp_port(skb, &uh->source, src, &uh->check); 657 flow_key->tp.src = src; 658 } 659 if (likely(dst != uh->dest)) { 660 set_tp_port(skb, &uh->dest, dst, &uh->check); 661 flow_key->tp.dst = dst; 662 } 663 664 if (unlikely(!uh->check)) 665 uh->check = CSUM_MANGLED_0; 666 } else { 667 uh->source = src; 668 uh->dest = dst; 669 flow_key->tp.src = src; 670 flow_key->tp.dst = dst; 671 } 672 673 skb_clear_hash(skb); 674 675 return 0; 676 } 677 678 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key, 679 const struct ovs_key_tcp *key, 680 const struct ovs_key_tcp *mask) 681 { 682 struct tcphdr *th; 683 __be16 src, dst; 684 int err; 685 686 err = skb_ensure_writable(skb, skb_transport_offset(skb) + 687 sizeof(struct tcphdr)); 688 if (unlikely(err)) 689 return err; 690 691 th = tcp_hdr(skb); 692 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src); 693 if (likely(src != th->source)) { 694 set_tp_port(skb, &th->source, src, &th->check); 695 flow_key->tp.src = src; 696 } 697 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst); 698 if (likely(dst != th->dest)) { 699 set_tp_port(skb, &th->dest, dst, &th->check); 700 flow_key->tp.dst = dst; 701 } 702 skb_clear_hash(skb); 703 704 return 0; 705 } 706 707 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key, 708 const struct ovs_key_sctp *key, 709 const struct ovs_key_sctp *mask) 710 { 711 unsigned int sctphoff = skb_transport_offset(skb); 712 struct sctphdr *sh; 713 __le32 old_correct_csum, new_csum, old_csum; 714 int err; 715 716 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr)); 717 if (unlikely(err)) 718 return err; 719 720 sh = sctp_hdr(skb); 721 old_csum = sh->checksum; 722 old_correct_csum = sctp_compute_cksum(skb, sctphoff); 723 724 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src); 725 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst); 726 727 new_csum = sctp_compute_cksum(skb, sctphoff); 728 729 /* Carry any checksum errors through. */ 730 sh->checksum = old_csum ^ old_correct_csum ^ new_csum; 731 732 skb_clear_hash(skb); 733 flow_key->tp.src = sh->source; 734 flow_key->tp.dst = sh->dest; 735 736 return 0; 737 } 738 739 static int ovs_vport_output(struct net *net, struct sock *sk, 740 struct sk_buff *skb) 741 { 742 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage); 743 struct vport *vport = data->vport; 744 745 if (skb_cow_head(skb, data->l2_len) < 0) { 746 kfree_skb(skb); 747 return -ENOMEM; 748 } 749 750 __skb_dst_copy(skb, data->dst); 751 *OVS_CB(skb) = data->cb; 752 skb->inner_protocol = data->inner_protocol; 753 if (data->vlan_tci & VLAN_CFI_MASK) 754 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK); 755 else 756 __vlan_hwaccel_clear_tag(skb); 757 758 /* Reconstruct the MAC header. */ 759 skb_push(skb, data->l2_len); 760 memcpy(skb->data, &data->l2_data, data->l2_len); 761 skb_postpush_rcsum(skb, skb->data, data->l2_len); 762 skb_reset_mac_header(skb); 763 764 if (eth_p_mpls(skb->protocol)) { 765 skb->inner_network_header = skb->network_header; 766 skb_set_network_header(skb, data->network_offset); 767 skb_reset_mac_len(skb); 768 } 769 770 ovs_vport_send(vport, skb, data->mac_proto); 771 return 0; 772 } 773 774 static unsigned int 775 ovs_dst_get_mtu(const struct dst_entry *dst) 776 { 777 return dst->dev->mtu; 778 } 779 780 static struct dst_ops ovs_dst_ops = { 781 .family = AF_UNSPEC, 782 .mtu = ovs_dst_get_mtu, 783 }; 784 785 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is 786 * ovs_vport_output(), which is called once per fragmented packet. 787 */ 788 static void prepare_frag(struct vport *vport, struct sk_buff *skb, 789 u16 orig_network_offset, u8 mac_proto) 790 { 791 unsigned int hlen = skb_network_offset(skb); 792 struct ovs_frag_data *data; 793 794 data = this_cpu_ptr(&ovs_frag_data_storage); 795 data->dst = skb->_skb_refdst; 796 data->vport = vport; 797 data->cb = *OVS_CB(skb); 798 data->inner_protocol = skb->inner_protocol; 799 data->network_offset = orig_network_offset; 800 if (skb_vlan_tag_present(skb)) 801 data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK; 802 else 803 data->vlan_tci = 0; 804 data->vlan_proto = skb->vlan_proto; 805 data->mac_proto = mac_proto; 806 data->l2_len = hlen; 807 memcpy(&data->l2_data, skb->data, hlen); 808 809 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); 810 skb_pull(skb, hlen); 811 } 812 813 static void ovs_fragment(struct net *net, struct vport *vport, 814 struct sk_buff *skb, u16 mru, 815 struct sw_flow_key *key) 816 { 817 u16 orig_network_offset = 0; 818 819 if (eth_p_mpls(skb->protocol)) { 820 orig_network_offset = skb_network_offset(skb); 821 skb->network_header = skb->inner_network_header; 822 } 823 824 if (skb_network_offset(skb) > MAX_L2_LEN) { 825 OVS_NLERR(1, "L2 header too long to fragment"); 826 goto err; 827 } 828 829 if (key->eth.type == htons(ETH_P_IP)) { 830 struct dst_entry ovs_dst; 831 unsigned long orig_dst; 832 833 prepare_frag(vport, skb, orig_network_offset, 834 ovs_key_mac_proto(key)); 835 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1, 836 DST_OBSOLETE_NONE, DST_NOCOUNT); 837 ovs_dst.dev = vport->dev; 838 839 orig_dst = skb->_skb_refdst; 840 skb_dst_set_noref(skb, &ovs_dst); 841 IPCB(skb)->frag_max_size = mru; 842 843 ip_do_fragment(net, skb->sk, skb, ovs_vport_output); 844 refdst_drop(orig_dst); 845 } else if (key->eth.type == htons(ETH_P_IPV6)) { 846 unsigned long orig_dst; 847 struct rt6_info ovs_rt; 848 849 prepare_frag(vport, skb, orig_network_offset, 850 ovs_key_mac_proto(key)); 851 memset(&ovs_rt, 0, sizeof(ovs_rt)); 852 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1, 853 DST_OBSOLETE_NONE, DST_NOCOUNT); 854 ovs_rt.dst.dev = vport->dev; 855 856 orig_dst = skb->_skb_refdst; 857 skb_dst_set_noref(skb, &ovs_rt.dst); 858 IP6CB(skb)->frag_max_size = mru; 859 860 ipv6_stub->ipv6_fragment(net, skb->sk, skb, ovs_vport_output); 861 refdst_drop(orig_dst); 862 } else { 863 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.", 864 ovs_vport_name(vport), ntohs(key->eth.type), mru, 865 vport->dev->mtu); 866 goto err; 867 } 868 869 return; 870 err: 871 kfree_skb(skb); 872 } 873 874 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port, 875 struct sw_flow_key *key) 876 { 877 struct vport *vport = ovs_vport_rcu(dp, out_port); 878 879 if (likely(vport)) { 880 u16 mru = OVS_CB(skb)->mru; 881 u32 cutlen = OVS_CB(skb)->cutlen; 882 883 if (unlikely(cutlen > 0)) { 884 if (skb->len - cutlen > ovs_mac_header_len(key)) 885 pskb_trim(skb, skb->len - cutlen); 886 else 887 pskb_trim(skb, ovs_mac_header_len(key)); 888 } 889 890 if (likely(!mru || 891 (skb->len <= mru + vport->dev->hard_header_len))) { 892 ovs_vport_send(vport, skb, ovs_key_mac_proto(key)); 893 } else if (mru <= vport->dev->mtu) { 894 struct net *net = read_pnet(&dp->net); 895 896 ovs_fragment(net, vport, skb, mru, key); 897 } else { 898 kfree_skb(skb); 899 } 900 } else { 901 kfree_skb(skb); 902 } 903 } 904 905 static int output_userspace(struct datapath *dp, struct sk_buff *skb, 906 struct sw_flow_key *key, const struct nlattr *attr, 907 const struct nlattr *actions, int actions_len, 908 uint32_t cutlen) 909 { 910 struct dp_upcall_info upcall; 911 const struct nlattr *a; 912 int rem; 913 914 memset(&upcall, 0, sizeof(upcall)); 915 upcall.cmd = OVS_PACKET_CMD_ACTION; 916 upcall.mru = OVS_CB(skb)->mru; 917 918 for (a = nla_data(attr), rem = nla_len(attr); rem > 0; 919 a = nla_next(a, &rem)) { 920 switch (nla_type(a)) { 921 case OVS_USERSPACE_ATTR_USERDATA: 922 upcall.userdata = a; 923 break; 924 925 case OVS_USERSPACE_ATTR_PID: 926 upcall.portid = nla_get_u32(a); 927 break; 928 929 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: { 930 /* Get out tunnel info. */ 931 struct vport *vport; 932 933 vport = ovs_vport_rcu(dp, nla_get_u32(a)); 934 if (vport) { 935 int err; 936 937 err = dev_fill_metadata_dst(vport->dev, skb); 938 if (!err) 939 upcall.egress_tun_info = skb_tunnel_info(skb); 940 } 941 942 break; 943 } 944 945 case OVS_USERSPACE_ATTR_ACTIONS: { 946 /* Include actions. */ 947 upcall.actions = actions; 948 upcall.actions_len = actions_len; 949 break; 950 } 951 952 } /* End of switch. */ 953 } 954 955 return ovs_dp_upcall(dp, skb, key, &upcall, cutlen); 956 } 957 958 static int dec_ttl_exception_handler(struct datapath *dp, struct sk_buff *skb, 959 struct sw_flow_key *key, 960 const struct nlattr *attr, bool last) 961 { 962 /* The first attribute is always 'OVS_DEC_TTL_ATTR_ACTION'. */ 963 struct nlattr *actions = nla_data(attr); 964 965 if (nla_len(actions)) 966 return clone_execute(dp, skb, key, 0, nla_data(actions), 967 nla_len(actions), last, false); 968 969 consume_skb(skb); 970 return 0; 971 } 972 973 /* When 'last' is true, sample() should always consume the 'skb'. 974 * Otherwise, sample() should keep 'skb' intact regardless what 975 * actions are executed within sample(). 976 */ 977 static int sample(struct datapath *dp, struct sk_buff *skb, 978 struct sw_flow_key *key, const struct nlattr *attr, 979 bool last) 980 { 981 struct nlattr *actions; 982 struct nlattr *sample_arg; 983 int rem = nla_len(attr); 984 const struct sample_arg *arg; 985 bool clone_flow_key; 986 987 /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */ 988 sample_arg = nla_data(attr); 989 arg = nla_data(sample_arg); 990 actions = nla_next(sample_arg, &rem); 991 992 if ((arg->probability != U32_MAX) && 993 (!arg->probability || prandom_u32() > arg->probability)) { 994 if (last) 995 consume_skb(skb); 996 return 0; 997 } 998 999 clone_flow_key = !arg->exec; 1000 return clone_execute(dp, skb, key, 0, actions, rem, last, 1001 clone_flow_key); 1002 } 1003 1004 /* When 'last' is true, clone() should always consume the 'skb'. 1005 * Otherwise, clone() should keep 'skb' intact regardless what 1006 * actions are executed within clone(). 1007 */ 1008 static int clone(struct datapath *dp, struct sk_buff *skb, 1009 struct sw_flow_key *key, const struct nlattr *attr, 1010 bool last) 1011 { 1012 struct nlattr *actions; 1013 struct nlattr *clone_arg; 1014 int rem = nla_len(attr); 1015 bool dont_clone_flow_key; 1016 1017 /* The first action is always 'OVS_CLONE_ATTR_ARG'. */ 1018 clone_arg = nla_data(attr); 1019 dont_clone_flow_key = nla_get_u32(clone_arg); 1020 actions = nla_next(clone_arg, &rem); 1021 1022 return clone_execute(dp, skb, key, 0, actions, rem, last, 1023 !dont_clone_flow_key); 1024 } 1025 1026 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key, 1027 const struct nlattr *attr) 1028 { 1029 struct ovs_action_hash *hash_act = nla_data(attr); 1030 u32 hash = 0; 1031 1032 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */ 1033 hash = skb_get_hash(skb); 1034 hash = jhash_1word(hash, hash_act->hash_basis); 1035 if (!hash) 1036 hash = 0x1; 1037 1038 key->ovs_flow_hash = hash; 1039 } 1040 1041 static int execute_set_action(struct sk_buff *skb, 1042 struct sw_flow_key *flow_key, 1043 const struct nlattr *a) 1044 { 1045 /* Only tunnel set execution is supported without a mask. */ 1046 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) { 1047 struct ovs_tunnel_info *tun = nla_data(a); 1048 1049 skb_dst_drop(skb); 1050 dst_hold((struct dst_entry *)tun->tun_dst); 1051 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst); 1052 return 0; 1053 } 1054 1055 return -EINVAL; 1056 } 1057 1058 /* Mask is at the midpoint of the data. */ 1059 #define get_mask(a, type) ((const type)nla_data(a) + 1) 1060 1061 static int execute_masked_set_action(struct sk_buff *skb, 1062 struct sw_flow_key *flow_key, 1063 const struct nlattr *a) 1064 { 1065 int err = 0; 1066 1067 switch (nla_type(a)) { 1068 case OVS_KEY_ATTR_PRIORITY: 1069 OVS_SET_MASKED(skb->priority, nla_get_u32(a), 1070 *get_mask(a, u32 *)); 1071 flow_key->phy.priority = skb->priority; 1072 break; 1073 1074 case OVS_KEY_ATTR_SKB_MARK: 1075 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *)); 1076 flow_key->phy.skb_mark = skb->mark; 1077 break; 1078 1079 case OVS_KEY_ATTR_TUNNEL_INFO: 1080 /* Masked data not supported for tunnel. */ 1081 err = -EINVAL; 1082 break; 1083 1084 case OVS_KEY_ATTR_ETHERNET: 1085 err = set_eth_addr(skb, flow_key, nla_data(a), 1086 get_mask(a, struct ovs_key_ethernet *)); 1087 break; 1088 1089 case OVS_KEY_ATTR_NSH: 1090 err = set_nsh(skb, flow_key, a); 1091 break; 1092 1093 case OVS_KEY_ATTR_IPV4: 1094 err = set_ipv4(skb, flow_key, nla_data(a), 1095 get_mask(a, struct ovs_key_ipv4 *)); 1096 break; 1097 1098 case OVS_KEY_ATTR_IPV6: 1099 err = set_ipv6(skb, flow_key, nla_data(a), 1100 get_mask(a, struct ovs_key_ipv6 *)); 1101 break; 1102 1103 case OVS_KEY_ATTR_TCP: 1104 err = set_tcp(skb, flow_key, nla_data(a), 1105 get_mask(a, struct ovs_key_tcp *)); 1106 break; 1107 1108 case OVS_KEY_ATTR_UDP: 1109 err = set_udp(skb, flow_key, nla_data(a), 1110 get_mask(a, struct ovs_key_udp *)); 1111 break; 1112 1113 case OVS_KEY_ATTR_SCTP: 1114 err = set_sctp(skb, flow_key, nla_data(a), 1115 get_mask(a, struct ovs_key_sctp *)); 1116 break; 1117 1118 case OVS_KEY_ATTR_MPLS: 1119 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a, 1120 __be32 *)); 1121 break; 1122 1123 case OVS_KEY_ATTR_CT_STATE: 1124 case OVS_KEY_ATTR_CT_ZONE: 1125 case OVS_KEY_ATTR_CT_MARK: 1126 case OVS_KEY_ATTR_CT_LABELS: 1127 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4: 1128 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6: 1129 err = -EINVAL; 1130 break; 1131 } 1132 1133 return err; 1134 } 1135 1136 static int execute_recirc(struct datapath *dp, struct sk_buff *skb, 1137 struct sw_flow_key *key, 1138 const struct nlattr *a, bool last) 1139 { 1140 u32 recirc_id; 1141 1142 if (!is_flow_key_valid(key)) { 1143 int err; 1144 1145 err = ovs_flow_key_update(skb, key); 1146 if (err) 1147 return err; 1148 } 1149 BUG_ON(!is_flow_key_valid(key)); 1150 1151 recirc_id = nla_get_u32(a); 1152 return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true); 1153 } 1154 1155 static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb, 1156 struct sw_flow_key *key, 1157 const struct nlattr *attr, bool last) 1158 { 1159 struct ovs_skb_cb *ovs_cb = OVS_CB(skb); 1160 const struct nlattr *actions, *cpl_arg; 1161 int len, max_len, rem = nla_len(attr); 1162 const struct check_pkt_len_arg *arg; 1163 bool clone_flow_key; 1164 1165 /* The first netlink attribute in 'attr' is always 1166 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'. 1167 */ 1168 cpl_arg = nla_data(attr); 1169 arg = nla_data(cpl_arg); 1170 1171 len = ovs_cb->mru ? ovs_cb->mru + skb->mac_len : skb->len; 1172 max_len = arg->pkt_len; 1173 1174 if ((skb_is_gso(skb) && skb_gso_validate_mac_len(skb, max_len)) || 1175 len <= max_len) { 1176 /* Second netlink attribute in 'attr' is always 1177 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'. 1178 */ 1179 actions = nla_next(cpl_arg, &rem); 1180 clone_flow_key = !arg->exec_for_lesser_equal; 1181 } else { 1182 /* Third netlink attribute in 'attr' is always 1183 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'. 1184 */ 1185 actions = nla_next(cpl_arg, &rem); 1186 actions = nla_next(actions, &rem); 1187 clone_flow_key = !arg->exec_for_greater; 1188 } 1189 1190 return clone_execute(dp, skb, key, 0, nla_data(actions), 1191 nla_len(actions), last, clone_flow_key); 1192 } 1193 1194 static int execute_dec_ttl(struct sk_buff *skb, struct sw_flow_key *key) 1195 { 1196 int err; 1197 1198 if (skb->protocol == htons(ETH_P_IPV6)) { 1199 struct ipv6hdr *nh; 1200 1201 err = skb_ensure_writable(skb, skb_network_offset(skb) + 1202 sizeof(*nh)); 1203 if (unlikely(err)) 1204 return err; 1205 1206 nh = ipv6_hdr(skb); 1207 1208 if (nh->hop_limit <= 1) 1209 return -EHOSTUNREACH; 1210 1211 key->ip.ttl = --nh->hop_limit; 1212 } else if (skb->protocol == htons(ETH_P_IP)) { 1213 struct iphdr *nh; 1214 u8 old_ttl; 1215 1216 err = skb_ensure_writable(skb, skb_network_offset(skb) + 1217 sizeof(*nh)); 1218 if (unlikely(err)) 1219 return err; 1220 1221 nh = ip_hdr(skb); 1222 if (nh->ttl <= 1) 1223 return -EHOSTUNREACH; 1224 1225 old_ttl = nh->ttl--; 1226 csum_replace2(&nh->check, htons(old_ttl << 8), 1227 htons(nh->ttl << 8)); 1228 key->ip.ttl = nh->ttl; 1229 } 1230 return 0; 1231 } 1232 1233 /* Execute a list of actions against 'skb'. */ 1234 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb, 1235 struct sw_flow_key *key, 1236 const struct nlattr *attr, int len) 1237 { 1238 const struct nlattr *a; 1239 int rem; 1240 1241 for (a = attr, rem = len; rem > 0; 1242 a = nla_next(a, &rem)) { 1243 int err = 0; 1244 1245 switch (nla_type(a)) { 1246 case OVS_ACTION_ATTR_OUTPUT: { 1247 int port = nla_get_u32(a); 1248 struct sk_buff *clone; 1249 1250 /* Every output action needs a separate clone 1251 * of 'skb', In case the output action is the 1252 * last action, cloning can be avoided. 1253 */ 1254 if (nla_is_last(a, rem)) { 1255 do_output(dp, skb, port, key); 1256 /* 'skb' has been used for output. 1257 */ 1258 return 0; 1259 } 1260 1261 clone = skb_clone(skb, GFP_ATOMIC); 1262 if (clone) 1263 do_output(dp, clone, port, key); 1264 OVS_CB(skb)->cutlen = 0; 1265 break; 1266 } 1267 1268 case OVS_ACTION_ATTR_TRUNC: { 1269 struct ovs_action_trunc *trunc = nla_data(a); 1270 1271 if (skb->len > trunc->max_len) 1272 OVS_CB(skb)->cutlen = skb->len - trunc->max_len; 1273 break; 1274 } 1275 1276 case OVS_ACTION_ATTR_USERSPACE: 1277 output_userspace(dp, skb, key, a, attr, 1278 len, OVS_CB(skb)->cutlen); 1279 OVS_CB(skb)->cutlen = 0; 1280 break; 1281 1282 case OVS_ACTION_ATTR_HASH: 1283 execute_hash(skb, key, a); 1284 break; 1285 1286 case OVS_ACTION_ATTR_PUSH_MPLS: { 1287 struct ovs_action_push_mpls *mpls = nla_data(a); 1288 1289 err = push_mpls(skb, key, mpls->mpls_lse, 1290 mpls->mpls_ethertype, skb->mac_len); 1291 break; 1292 } 1293 case OVS_ACTION_ATTR_ADD_MPLS: { 1294 struct ovs_action_add_mpls *mpls = nla_data(a); 1295 __u16 mac_len = 0; 1296 1297 if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK) 1298 mac_len = skb->mac_len; 1299 1300 err = push_mpls(skb, key, mpls->mpls_lse, 1301 mpls->mpls_ethertype, mac_len); 1302 break; 1303 } 1304 case OVS_ACTION_ATTR_POP_MPLS: 1305 err = pop_mpls(skb, key, nla_get_be16(a)); 1306 break; 1307 1308 case OVS_ACTION_ATTR_PUSH_VLAN: 1309 err = push_vlan(skb, key, nla_data(a)); 1310 break; 1311 1312 case OVS_ACTION_ATTR_POP_VLAN: 1313 err = pop_vlan(skb, key); 1314 break; 1315 1316 case OVS_ACTION_ATTR_RECIRC: { 1317 bool last = nla_is_last(a, rem); 1318 1319 err = execute_recirc(dp, skb, key, a, last); 1320 if (last) { 1321 /* If this is the last action, the skb has 1322 * been consumed or freed. 1323 * Return immediately. 1324 */ 1325 return err; 1326 } 1327 break; 1328 } 1329 1330 case OVS_ACTION_ATTR_SET: 1331 err = execute_set_action(skb, key, nla_data(a)); 1332 break; 1333 1334 case OVS_ACTION_ATTR_SET_MASKED: 1335 case OVS_ACTION_ATTR_SET_TO_MASKED: 1336 err = execute_masked_set_action(skb, key, nla_data(a)); 1337 break; 1338 1339 case OVS_ACTION_ATTR_SAMPLE: { 1340 bool last = nla_is_last(a, rem); 1341 1342 err = sample(dp, skb, key, a, last); 1343 if (last) 1344 return err; 1345 1346 break; 1347 } 1348 1349 case OVS_ACTION_ATTR_CT: 1350 if (!is_flow_key_valid(key)) { 1351 err = ovs_flow_key_update(skb, key); 1352 if (err) 1353 return err; 1354 } 1355 1356 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key, 1357 nla_data(a)); 1358 1359 /* Hide stolen IP fragments from user space. */ 1360 if (err) 1361 return err == -EINPROGRESS ? 0 : err; 1362 break; 1363 1364 case OVS_ACTION_ATTR_CT_CLEAR: 1365 err = ovs_ct_clear(skb, key); 1366 break; 1367 1368 case OVS_ACTION_ATTR_PUSH_ETH: 1369 err = push_eth(skb, key, nla_data(a)); 1370 break; 1371 1372 case OVS_ACTION_ATTR_POP_ETH: 1373 err = pop_eth(skb, key); 1374 break; 1375 1376 case OVS_ACTION_ATTR_PUSH_NSH: { 1377 u8 buffer[NSH_HDR_MAX_LEN]; 1378 struct nshhdr *nh = (struct nshhdr *)buffer; 1379 1380 err = nsh_hdr_from_nlattr(nla_data(a), nh, 1381 NSH_HDR_MAX_LEN); 1382 if (unlikely(err)) 1383 break; 1384 err = push_nsh(skb, key, nh); 1385 break; 1386 } 1387 1388 case OVS_ACTION_ATTR_POP_NSH: 1389 err = pop_nsh(skb, key); 1390 break; 1391 1392 case OVS_ACTION_ATTR_METER: 1393 if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) { 1394 consume_skb(skb); 1395 return 0; 1396 } 1397 break; 1398 1399 case OVS_ACTION_ATTR_CLONE: { 1400 bool last = nla_is_last(a, rem); 1401 1402 err = clone(dp, skb, key, a, last); 1403 if (last) 1404 return err; 1405 1406 break; 1407 } 1408 1409 case OVS_ACTION_ATTR_CHECK_PKT_LEN: { 1410 bool last = nla_is_last(a, rem); 1411 1412 err = execute_check_pkt_len(dp, skb, key, a, last); 1413 if (last) 1414 return err; 1415 1416 break; 1417 } 1418 1419 case OVS_ACTION_ATTR_DEC_TTL: 1420 err = execute_dec_ttl(skb, key); 1421 if (err == -EHOSTUNREACH) { 1422 err = dec_ttl_exception_handler(dp, skb, key, 1423 a, true); 1424 return err; 1425 } 1426 break; 1427 } 1428 1429 if (unlikely(err)) { 1430 kfree_skb(skb); 1431 return err; 1432 } 1433 } 1434 1435 consume_skb(skb); 1436 return 0; 1437 } 1438 1439 /* Execute the actions on the clone of the packet. The effect of the 1440 * execution does not affect the original 'skb' nor the original 'key'. 1441 * 1442 * The execution may be deferred in case the actions can not be executed 1443 * immediately. 1444 */ 1445 static int clone_execute(struct datapath *dp, struct sk_buff *skb, 1446 struct sw_flow_key *key, u32 recirc_id, 1447 const struct nlattr *actions, int len, 1448 bool last, bool clone_flow_key) 1449 { 1450 struct deferred_action *da; 1451 struct sw_flow_key *clone; 1452 1453 skb = last ? skb : skb_clone(skb, GFP_ATOMIC); 1454 if (!skb) { 1455 /* Out of memory, skip this action. 1456 */ 1457 return 0; 1458 } 1459 1460 /* When clone_flow_key is false, the 'key' will not be change 1461 * by the actions, then the 'key' can be used directly. 1462 * Otherwise, try to clone key from the next recursion level of 1463 * 'flow_keys'. If clone is successful, execute the actions 1464 * without deferring. 1465 */ 1466 clone = clone_flow_key ? clone_key(key) : key; 1467 if (clone) { 1468 int err = 0; 1469 1470 if (actions) { /* Sample action */ 1471 if (clone_flow_key) 1472 __this_cpu_inc(exec_actions_level); 1473 1474 err = do_execute_actions(dp, skb, clone, 1475 actions, len); 1476 1477 if (clone_flow_key) 1478 __this_cpu_dec(exec_actions_level); 1479 } else { /* Recirc action */ 1480 clone->recirc_id = recirc_id; 1481 ovs_dp_process_packet(skb, clone); 1482 } 1483 return err; 1484 } 1485 1486 /* Out of 'flow_keys' space. Defer actions */ 1487 da = add_deferred_actions(skb, key, actions, len); 1488 if (da) { 1489 if (!actions) { /* Recirc action */ 1490 key = &da->pkt_key; 1491 key->recirc_id = recirc_id; 1492 } 1493 } else { 1494 /* Out of per CPU action FIFO space. Drop the 'skb' and 1495 * log an error. 1496 */ 1497 kfree_skb(skb); 1498 1499 if (net_ratelimit()) { 1500 if (actions) { /* Sample action */ 1501 pr_warn("%s: deferred action limit reached, drop sample action\n", 1502 ovs_dp_name(dp)); 1503 } else { /* Recirc action */ 1504 pr_warn("%s: deferred action limit reached, drop recirc action\n", 1505 ovs_dp_name(dp)); 1506 } 1507 } 1508 } 1509 return 0; 1510 } 1511 1512 static void process_deferred_actions(struct datapath *dp) 1513 { 1514 struct action_fifo *fifo = this_cpu_ptr(action_fifos); 1515 1516 /* Do not touch the FIFO in case there is no deferred actions. */ 1517 if (action_fifo_is_empty(fifo)) 1518 return; 1519 1520 /* Finishing executing all deferred actions. */ 1521 do { 1522 struct deferred_action *da = action_fifo_get(fifo); 1523 struct sk_buff *skb = da->skb; 1524 struct sw_flow_key *key = &da->pkt_key; 1525 const struct nlattr *actions = da->actions; 1526 int actions_len = da->actions_len; 1527 1528 if (actions) 1529 do_execute_actions(dp, skb, key, actions, actions_len); 1530 else 1531 ovs_dp_process_packet(skb, key); 1532 } while (!action_fifo_is_empty(fifo)); 1533 1534 /* Reset FIFO for the next packet. */ 1535 action_fifo_init(fifo); 1536 } 1537 1538 /* Execute a list of actions against 'skb'. */ 1539 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb, 1540 const struct sw_flow_actions *acts, 1541 struct sw_flow_key *key) 1542 { 1543 int err, level; 1544 1545 level = __this_cpu_inc_return(exec_actions_level); 1546 if (unlikely(level > OVS_RECURSION_LIMIT)) { 1547 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n", 1548 ovs_dp_name(dp)); 1549 kfree_skb(skb); 1550 err = -ENETDOWN; 1551 goto out; 1552 } 1553 1554 OVS_CB(skb)->acts_origlen = acts->orig_len; 1555 err = do_execute_actions(dp, skb, key, 1556 acts->actions, acts->actions_len); 1557 1558 if (level == 1) 1559 process_deferred_actions(dp); 1560 1561 out: 1562 __this_cpu_dec(exec_actions_level); 1563 return err; 1564 } 1565 1566 int action_fifos_init(void) 1567 { 1568 action_fifos = alloc_percpu(struct action_fifo); 1569 if (!action_fifos) 1570 return -ENOMEM; 1571 1572 flow_keys = alloc_percpu(struct action_flow_keys); 1573 if (!flow_keys) { 1574 free_percpu(action_fifos); 1575 return -ENOMEM; 1576 } 1577 1578 return 0; 1579 } 1580 1581 void action_fifos_exit(void) 1582 { 1583 free_percpu(action_fifos); 1584 free_percpu(flow_keys); 1585 } 1586