net/openvswitch/actions.c

   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/gso.h>
  21 #include <net/ip.h>
  22 #include <net/ipv6.h>
  23 #include <net/ip6_fib.h>
  24 #include <net/checksum.h>
  25 #include <net/dsfield.h>
  26 #include <net/mpls.h>
  27 #include <net/sctp/checksum.h>
  28
  29 #include "datapath.h"
  30 #include "flow.h"
  31 #include "conntrack.h"
  32 #include "vport.h"
  33 #include "flow_netlink.h"
  34 #include "openvswitch_trace.h"
  35
  36 struct deferred_action {
  37         struct sk_buff *skb;
  38         const struct nlattr *actions;
  39         int actions_len;
  40
  41         /* Store pkt_key clone when creating deferred action. */
  42         struct sw_flow_key pkt_key;
  43 };
  44
  45 #define MAX_L2_LEN      (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
  46 struct ovs_frag_data {
  47         unsigned long dst;
  48         struct vport *vport;
  49         struct ovs_skb_cb cb;
  50         __be16 inner_protocol;
  51         u16 network_offset;     /* valid only for MPLS */
  52         u16 vlan_tci;
  53         __be16 vlan_proto;
  54         unsigned int l2_len;
  55         u8 mac_proto;
  56         u8 l2_data[MAX_L2_LEN];
  57 };
  58
  59 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
  60
  61 #define DEFERRED_ACTION_FIFO_SIZE 10
  62 #define OVS_RECURSION_LIMIT 5
  63 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
  64 struct action_fifo {
  65         int head;
  66         int tail;
  67         /* Deferred action fifo queue storage. */
  68         struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
  69 };
  70
  71 struct action_flow_keys {
  72         struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
  73 };
  74
  75 static struct action_fifo __percpu *action_fifos;
  76 static struct action_flow_keys __percpu *flow_keys;
  77 static DEFINE_PER_CPU(int, exec_actions_level);
  78
  79 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
  80  * space. Return NULL if out of key spaces.
  81  */
  82 static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
  83 {
  84         struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
  85         int level = this_cpu_read(exec_actions_level);
  86         struct sw_flow_key *key = NULL;
  87
  88         if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
  89                 key = &keys->key[level - 1];
  90                 *key = *key_;
  91         }
  92
  93         return key;
  94 }
  95
  96 static void action_fifo_init(struct action_fifo *fifo)
  97 {
  98         fifo->head = 0;
  99         fifo->tail = 0;
 100 }
 101
 102 static bool action_fifo_is_empty(const struct action_fifo *fifo)
 103 {
 104         return (fifo->head == fifo->tail);
 105 }
 106
 107 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
 108 {
 109         if (action_fifo_is_empty(fifo))
 110                 return NULL;
 111
 112         return &fifo->fifo[fifo->tail++];
 113 }
 114
 115 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
 116 {
 117         if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
 118                 return NULL;
 119
 120         return &fifo->fifo[fifo->head++];
 121 }
 122
 123 /* Return true if fifo is not full */
 124 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
 125                                     const struct sw_flow_key *key,
 126                                     const struct nlattr *actions,
 127                                     const int actions_len)
 128 {
 129         struct action_fifo *fifo;
 130         struct deferred_action *da;
 131
 132         fifo = this_cpu_ptr(action_fifos);
 133         da = action_fifo_put(fifo);
 134         if (da) {
 135                 da->skb = skb;
 136                 da->actions = actions;
 137                 da->actions_len = actions_len;
 138                 da->pkt_key = *key;
 139         }
 140
 141         return da;
 142 }
 143
 144 static void invalidate_flow_key(struct sw_flow_key *key)
 145 {
 146         key->mac_proto |= SW_FLOW_KEY_INVALID;
 147 }
 148
 149 static bool is_flow_key_valid(const struct sw_flow_key *key)
 150 {
 151         return !(key->mac_proto & SW_FLOW_KEY_INVALID);
 152 }
 153
 154 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
 155                          struct sw_flow_key *key,
 156                          u32 recirc_id,
 157                          const struct nlattr *actions, int len,
 158                          bool last, bool clone_flow_key);
 159
 160 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 161                               struct sw_flow_key *key,
 162                               const struct nlattr *attr, int len);
 163
 164 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 165                      __be32 mpls_lse, __be16 mpls_ethertype, __u16 mac_len)
 166 {
 167         int err;
 168
 169         err = skb_mpls_push(skb, mpls_lse, mpls_ethertype, mac_len, !!mac_len);
 170         if (err)
 171                 return err;
 172
 173         if (!mac_len)
 174                 key->mac_proto = MAC_PROTO_NONE;
 175
 176         invalidate_flow_key(key);
 177         return 0;
 178 }
 179
 180 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 181                     const __be16 ethertype)
 182 {
 183         int err;
 184
 185         err = skb_mpls_pop(skb, ethertype, skb->mac_len,
 186                            ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET);
 187         if (err)
 188                 return err;
 189
 190         if (ethertype == htons(ETH_P_TEB))
 191                 key->mac_proto = MAC_PROTO_ETHERNET;
 192
 193         invalidate_flow_key(key);
 194         return 0;
 195 }
 196
 197 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
 198                     const __be32 *mpls_lse, const __be32 *mask)
 199 {
 200         struct mpls_shim_hdr *stack;
 201         __be32 lse;
 202         int err;
 203
 204         if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
 205                 return -ENOMEM;
 206
 207         stack = mpls_hdr(skb);
 208         lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
 209         err = skb_mpls_update_lse(skb, lse);
 210         if (err)
 211                 return err;
 212
 213         flow_key->mpls.lse[0] = lse;
 214         return 0;
 215 }
 216
 217 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 218 {
 219         int err;
 220
 221         err = skb_vlan_pop(skb);
 222         if (skb_vlan_tag_present(skb)) {
 223                 invalidate_flow_key(key);
 224         } else {
 225                 key->eth.vlan.tci = 0;
 226                 key->eth.vlan.tpid = 0;
 227         }
 228         return err;
 229 }
 230
 231 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
 232                      const struct ovs_action_push_vlan *vlan)
 233 {
 234         if (skb_vlan_tag_present(skb)) {
 235                 invalidate_flow_key(key);
 236         } else {
 237                 key->eth.vlan.tci = vlan->vlan_tci;
 238                 key->eth.vlan.tpid = vlan->vlan_tpid;
 239         }
 240         return skb_vlan_push(skb, vlan->vlan_tpid,
 241                              ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
 242 }
 243
 244 /* 'src' is already properly masked. */
 245 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
 246 {
 247         u16 *dst = (u16 *)dst_;
 248         const u16 *src = (const u16 *)src_;
 249         const u16 *mask = (const u16 *)mask_;
 250
 251         OVS_SET_MASKED(dst[0], src[0], mask[0]);
 252         OVS_SET_MASKED(dst[1], src[1], mask[1]);
 253         OVS_SET_MASKED(dst[2], src[2], mask[2]);
 254 }
 255
 256 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
 257                         const struct ovs_key_ethernet *key,
 258                         const struct ovs_key_ethernet *mask)
 259 {
 260         int err;
 261
 262         err = skb_ensure_writable(skb, ETH_HLEN);
 263         if (unlikely(err))
 264                 return err;
 265
 266         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 267
 268         ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
 269                                mask->eth_src);
 270         ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
 271                                mask->eth_dst);
 272
 273         skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 274
 275         ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
 276         ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
 277         return 0;
 278 }
 279
 280 /* pop_eth does not support VLAN packets as this action is never called
 281  * for them.
 282  */
 283 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
 284 {
 285         int err;
 286
 287         err = skb_eth_pop(skb);
 288         if (err)
 289                 return err;
 290
 291         /* safe right before invalidate_flow_key */
 292         key->mac_proto = MAC_PROTO_NONE;
 293         invalidate_flow_key(key);
 294         return 0;
 295 }
 296
 297 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
 298                     const struct ovs_action_push_eth *ethh)
 299 {
 300         int err;
 301
 302         err = skb_eth_push(skb, ethh->addresses.eth_dst,
 303                            ethh->addresses.eth_src);
 304         if (err)
 305                 return err;
 306
 307         /* safe right before invalidate_flow_key */
 308         key->mac_proto = MAC_PROTO_ETHERNET;
 309         invalidate_flow_key(key);
 310         return 0;
 311 }
 312
 313 static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
 314                     const struct nshhdr *nh)
 315 {
 316         int err;
 317
 318         err = nsh_push(skb, nh);
 319         if (err)
 320                 return err;
 321
 322         /* safe right before invalidate_flow_key */
 323         key->mac_proto = MAC_PROTO_NONE;
 324         invalidate_flow_key(key);
 325         return 0;
 326 }
 327
 328 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
 329 {
 330         int err;
 331
 332         err = nsh_pop(skb);
 333         if (err)
 334                 return err;
 335
 336         /* safe right before invalidate_flow_key */
 337         if (skb->protocol == htons(ETH_P_TEB))
 338                 key->mac_proto = MAC_PROTO_ETHERNET;
 339         else
 340                 key->mac_proto = MAC_PROTO_NONE;
 341         invalidate_flow_key(key);
 342         return 0;
 343 }
 344
 345 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
 346                                   __be32 addr, __be32 new_addr)
 347 {
 348         int transport_len = skb->len - skb_transport_offset(skb);
 349
 350         if (nh->frag_off & htons(IP_OFFSET))
 351                 return;
 352
 353         if (nh->protocol == IPPROTO_TCP) {
 354                 if (likely(transport_len >= sizeof(struct tcphdr)))
 355                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 356                                                  addr, new_addr, true);
 357         } else if (nh->protocol == IPPROTO_UDP) {
 358                 if (likely(transport_len >= sizeof(struct udphdr))) {
 359                         struct udphdr *uh = udp_hdr(skb);
 360
 361                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 362                                 inet_proto_csum_replace4(&uh->check, skb,
 363                                                          addr, new_addr, true);
 364                                 if (!uh->check)
 365                                         uh->check = CSUM_MANGLED_0;
 366                         }
 367                 }
 368         }
 369 }
 370
 371 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 372                         __be32 *addr, __be32 new_addr)
 373 {
 374         update_ip_l4_checksum(skb, nh, *addr, new_addr);
 375         csum_replace4(&nh->check, *addr, new_addr);
 376         skb_clear_hash(skb);
 377         ovs_ct_clear(skb, NULL);
 378         *addr = new_addr;
 379 }
 380
 381 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 382                                  __be32 addr[4], const __be32 new_addr[4])
 383 {
 384         int transport_len = skb->len - skb_transport_offset(skb);
 385
 386         if (l4_proto == NEXTHDR_TCP) {
 387                 if (likely(transport_len >= sizeof(struct tcphdr)))
 388                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 389                                                   addr, new_addr, true);
 390         } else if (l4_proto == NEXTHDR_UDP) {
 391                 if (likely(transport_len >= sizeof(struct udphdr))) {
 392                         struct udphdr *uh = udp_hdr(skb);
 393
 394                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 395                                 inet_proto_csum_replace16(&uh->check, skb,
 396                                                           addr, new_addr, true);
 397                                 if (!uh->check)
 398                                         uh->check = CSUM_MANGLED_0;
 399                         }
 400                 }
 401         } else if (l4_proto == NEXTHDR_ICMP) {
 402                 if (likely(transport_len >= sizeof(struct icmp6hdr)))
 403                         inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 404                                                   skb, addr, new_addr, true);
 405         }
 406 }
 407
 408 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
 409                            const __be32 mask[4], __be32 masked[4])
 410 {
 411         masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
 412         masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
 413         masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
 414         masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
 415 }
 416
 417 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 418                           __be32 addr[4], const __be32 new_addr[4],
 419                           bool recalculate_csum)
 420 {
 421         if (recalculate_csum)
 422                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 423
 424         skb_clear_hash(skb);
 425         ovs_ct_clear(skb, NULL);
 426         memcpy(addr, new_addr, sizeof(__be32[4]));
 427 }
 428
 429 static void set_ipv6_dsfield(struct sk_buff *skb, struct ipv6hdr *nh, u8 ipv6_tclass, u8 mask)
 430 {
 431         u8 old_ipv6_tclass = ipv6_get_dsfield(nh);
 432
 433         ipv6_tclass = OVS_MASKED(old_ipv6_tclass, ipv6_tclass, mask);
 434
 435         if (skb->ip_summed == CHECKSUM_COMPLETE)
 436                 csum_replace(&skb->csum, (__force __wsum)(old_ipv6_tclass << 12),
 437                              (__force __wsum)(ipv6_tclass << 12));
 438
 439         ipv6_change_dsfield(nh, ~mask, ipv6_tclass);
 440 }
 441
 442 static void set_ipv6_fl(struct sk_buff *skb, struct ipv6hdr *nh, u32 fl, u32 mask)
 443 {
 444         u32 ofl;
 445
 446         ofl = nh->flow_lbl[0] << 16 |  nh->flow_lbl[1] << 8 |  nh->flow_lbl[2];
 447         fl = OVS_MASKED(ofl, fl, mask);
 448
 449         /* Bits 21-24 are always unmasked, so this retains their values. */
 450         nh->flow_lbl[0] = (u8)(fl >> 16);
 451         nh->flow_lbl[1] = (u8)(fl >> 8);
 452         nh->flow_lbl[2] = (u8)fl;
 453
 454         if (skb->ip_summed == CHECKSUM_COMPLETE)
 455                 csum_replace(&skb->csum, (__force __wsum)htonl(ofl), (__force __wsum)htonl(fl));
 456 }
 457
 458 static void set_ipv6_ttl(struct sk_buff *skb, struct ipv6hdr *nh, u8 new_ttl, u8 mask)
 459 {
 460         new_ttl = OVS_MASKED(nh->hop_limit, new_ttl, mask);
 461
 462         if (skb->ip_summed == CHECKSUM_COMPLETE)
 463                 csum_replace(&skb->csum, (__force __wsum)(nh->hop_limit << 8),
 464                              (__force __wsum)(new_ttl << 8));
 465         nh->hop_limit = new_ttl;
 466 }
 467
 468 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
 469                        u8 mask)
 470 {
 471         new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
 472
 473         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 474         nh->ttl = new_ttl;
 475 }
 476
 477 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
 478                     const struct ovs_key_ipv4 *key,
 479                     const struct ovs_key_ipv4 *mask)
 480 {
 481         struct iphdr *nh;
 482         __be32 new_addr;
 483         int err;
 484
 485         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 486                                   sizeof(struct iphdr));
 487         if (unlikely(err))
 488                 return err;
 489
 490         nh = ip_hdr(skb);
 491
 492         /* Setting an IP addresses is typically only a side effect of
 493          * matching on them in the current userspace implementation, so it
 494          * makes sense to check if the value actually changed.
 495          */
 496         if (mask->ipv4_src) {
 497                 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
 498
 499                 if (unlikely(new_addr != nh->saddr)) {
 500                         set_ip_addr(skb, nh, &nh->saddr, new_addr);
 501                         flow_key->ipv4.addr.src = new_addr;
 502                 }
 503         }
 504         if (mask->ipv4_dst) {
 505                 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
 506
 507                 if (unlikely(new_addr != nh->daddr)) {
 508                         set_ip_addr(skb, nh, &nh->daddr, new_addr);
 509                         flow_key->ipv4.addr.dst = new_addr;
 510                 }
 511         }
 512         if (mask->ipv4_tos) {
 513                 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
 514                 flow_key->ip.tos = nh->tos;
 515         }
 516         if (mask->ipv4_ttl) {
 517                 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
 518                 flow_key->ip.ttl = nh->ttl;
 519         }
 520
 521         return 0;
 522 }
 523
 524 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
 525 {
 526         return !!(addr[0] | addr[1] | addr[2] | addr[3]);
 527 }
 528
 529 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
 530                     const struct ovs_key_ipv6 *key,
 531                     const struct ovs_key_ipv6 *mask)
 532 {
 533         struct ipv6hdr *nh;
 534         int err;
 535
 536         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 537                                   sizeof(struct ipv6hdr));
 538         if (unlikely(err))
 539                 return err;
 540
 541         nh = ipv6_hdr(skb);
 542
 543         /* Setting an IP addresses is typically only a side effect of
 544          * matching on them in the current userspace implementation, so it
 545          * makes sense to check if the value actually changed.
 546          */
 547         if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
 548                 __be32 *saddr = (__be32 *)&nh->saddr;
 549                 __be32 masked[4];
 550
 551                 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
 552
 553                 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
 554                         set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
 555                                       true);
 556                         memcpy(&flow_key->ipv6.addr.src, masked,
 557                                sizeof(flow_key->ipv6.addr.src));
 558                 }
 559         }
 560         if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
 561                 unsigned int offset = 0;
 562                 int flags = IP6_FH_F_SKIP_RH;
 563                 bool recalc_csum = true;
 564                 __be32 *daddr = (__be32 *)&nh->daddr;
 565                 __be32 masked[4];
 566
 567                 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
 568
 569                 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
 570                         if (ipv6_ext_hdr(nh->nexthdr))
 571                                 recalc_csum = (ipv6_find_hdr(skb, &offset,
 572                                                              NEXTHDR_ROUTING,
 573                                                              NULL, &flags)
 574                                                != NEXTHDR_ROUTING);
 575
 576                         set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
 577                                       recalc_csum);
 578                         memcpy(&flow_key->ipv6.addr.dst, masked,
 579                                sizeof(flow_key->ipv6.addr.dst));
 580                 }
 581         }
 582         if (mask->ipv6_tclass) {
 583                 set_ipv6_dsfield(skb, nh, key->ipv6_tclass, mask->ipv6_tclass);
 584                 flow_key->ip.tos = ipv6_get_dsfield(nh);
 585         }
 586         if (mask->ipv6_label) {
 587                 set_ipv6_fl(skb, nh, ntohl(key->ipv6_label),
 588                             ntohl(mask->ipv6_label));
 589                 flow_key->ipv6.label =
 590                     *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 591         }
 592         if (mask->ipv6_hlimit) {
 593                 set_ipv6_ttl(skb, nh, key->ipv6_hlimit, mask->ipv6_hlimit);
 594                 flow_key->ip.ttl = nh->hop_limit;
 595         }
 596         return 0;
 597 }
 598
 599 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
 600                    const struct nlattr *a)
 601 {
 602         struct nshhdr *nh;
 603         size_t length;
 604         int err;
 605         u8 flags;
 606         u8 ttl;
 607         int i;
 608
 609         struct ovs_key_nsh key;
 610         struct ovs_key_nsh mask;
 611
 612         err = nsh_key_from_nlattr(a, &key, &mask);
 613         if (err)
 614                 return err;
 615
 616         /* Make sure the NSH base header is there */
 617         if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
 618                 return -ENOMEM;
 619
 620         nh = nsh_hdr(skb);
 621         length = nsh_hdr_len(nh);
 622
 623         /* Make sure the whole NSH header is there */
 624         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 625                                        length);
 626         if (unlikely(err))
 627                 return err;
 628
 629         nh = nsh_hdr(skb);
 630         skb_postpull_rcsum(skb, nh, length);
 631         flags = nsh_get_flags(nh);
 632         flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
 633         flow_key->nsh.base.flags = flags;
 634         ttl = nsh_get_ttl(nh);
 635         ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
 636         flow_key->nsh.base.ttl = ttl;
 637         nsh_set_flags_and_ttl(nh, flags, ttl);
 638         nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
 639                                   mask.base.path_hdr);
 640         flow_key->nsh.base.path_hdr = nh->path_hdr;
 641         switch (nh->mdtype) {
 642         case NSH_M_TYPE1:
 643                 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
 644                         nh->md1.context[i] =
 645                             OVS_MASKED(nh->md1.context[i], key.context[i],
 646                                        mask.context[i]);
 647                 }
 648                 memcpy(flow_key->nsh.context, nh->md1.context,
 649                        sizeof(nh->md1.context));
 650                 break;
 651         case NSH_M_TYPE2:
 652                 memset(flow_key->nsh.context, 0,
 653                        sizeof(flow_key->nsh.context));
 654                 break;
 655         default:
 656                 return -EINVAL;
 657         }
 658         skb_postpush_rcsum(skb, nh, length);
 659         return 0;
 660 }
 661
 662 /* Must follow skb_ensure_writable() since that can move the skb data. */
 663 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 664                         __be16 new_port, __sum16 *check)
 665 {
 666         ovs_ct_clear(skb, NULL);
 667         inet_proto_csum_replace2(check, skb, *port, new_port, false);
 668         *port = new_port;
 669 }
 670
 671 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 672                    const struct ovs_key_udp *key,
 673                    const struct ovs_key_udp *mask)
 674 {
 675         struct udphdr *uh;
 676         __be16 src, dst;
 677         int err;
 678
 679         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 680                                   sizeof(struct udphdr));
 681         if (unlikely(err))
 682                 return err;
 683
 684         uh = udp_hdr(skb);
 685         /* Either of the masks is non-zero, so do not bother checking them. */
 686         src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
 687         dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
 688
 689         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 690                 if (likely(src != uh->source)) {
 691                         set_tp_port(skb, &uh->source, src, &uh->check);
 692                         flow_key->tp.src = src;
 693                 }
 694                 if (likely(dst != uh->dest)) {
 695                         set_tp_port(skb, &uh->dest, dst, &uh->check);
 696                         flow_key->tp.dst = dst;
 697                 }
 698
 699                 if (unlikely(!uh->check))
 700                         uh->check = CSUM_MANGLED_0;
 701         } else {
 702                 uh->source = src;
 703                 uh->dest = dst;
 704                 flow_key->tp.src = src;
 705                 flow_key->tp.dst = dst;
 706                 ovs_ct_clear(skb, NULL);
 707         }
 708
 709         skb_clear_hash(skb);
 710
 711         return 0;
 712 }
 713
 714 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 715                    const struct ovs_key_tcp *key,
 716                    const struct ovs_key_tcp *mask)
 717 {
 718         struct tcphdr *th;
 719         __be16 src, dst;
 720         int err;
 721
 722         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 723                                   sizeof(struct tcphdr));
 724         if (unlikely(err))
 725                 return err;
 726
 727         th = tcp_hdr(skb);
 728         src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
 729         if (likely(src != th->source)) {
 730                 set_tp_port(skb, &th->source, src, &th->check);
 731                 flow_key->tp.src = src;
 732         }
 733         dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
 734         if (likely(dst != th->dest)) {
 735                 set_tp_port(skb, &th->dest, dst, &th->check);
 736                 flow_key->tp.dst = dst;
 737         }
 738         skb_clear_hash(skb);
 739
 740         return 0;
 741 }
 742
 743 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 744                     const struct ovs_key_sctp *key,
 745                     const struct ovs_key_sctp *mask)
 746 {
 747         unsigned int sctphoff = skb_transport_offset(skb);
 748         struct sctphdr *sh;
 749         __le32 old_correct_csum, new_csum, old_csum;
 750         int err;
 751
 752         err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
 753         if (unlikely(err))
 754                 return err;
 755
 756         sh = sctp_hdr(skb);
 757         old_csum = sh->checksum;
 758         old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 759
 760         sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
 761         sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
 762
 763         new_csum = sctp_compute_cksum(skb, sctphoff);
 764
 765         /* Carry any checksum errors through. */
 766         sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 767
 768         skb_clear_hash(skb);
 769         ovs_ct_clear(skb, NULL);
 770
 771         flow_key->tp.src = sh->source;
 772         flow_key->tp.dst = sh->dest;
 773
 774         return 0;
 775 }
 776
 777 static int ovs_vport_output(struct net *net, struct sock *sk,
 778                             struct sk_buff *skb)
 779 {
 780         struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
 781         struct vport *vport = data->vport;
 782
 783         if (skb_cow_head(skb, data->l2_len) < 0) {
 784                 kfree_skb(skb);
 785                 return -ENOMEM;
 786         }
 787
 788         __skb_dst_copy(skb, data->dst);
 789         *OVS_CB(skb) = data->cb;
 790         skb->inner_protocol = data->inner_protocol;
 791         if (data->vlan_tci & VLAN_CFI_MASK)
 792                 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
 793         else
 794                 __vlan_hwaccel_clear_tag(skb);
 795
 796         /* Reconstruct the MAC header.  */
 797         skb_push(skb, data->l2_len);
 798         memcpy(skb->data, &data->l2_data, data->l2_len);
 799         skb_postpush_rcsum(skb, skb->data, data->l2_len);
 800         skb_reset_mac_header(skb);
 801
 802         if (eth_p_mpls(skb->protocol)) {
 803                 skb->inner_network_header = skb->network_header;
 804                 skb_set_network_header(skb, data->network_offset);
 805                 skb_reset_mac_len(skb);
 806         }
 807
 808         ovs_vport_send(vport, skb, data->mac_proto);
 809         return 0;
 810 }
 811
 812 static unsigned int
 813 ovs_dst_get_mtu(const struct dst_entry *dst)
 814 {
 815         return dst->dev->mtu;
 816 }
 817
 818 static struct dst_ops ovs_dst_ops = {
 819         .family = AF_UNSPEC,
 820         .mtu = ovs_dst_get_mtu,
 821 };
 822
 823 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
 824  * ovs_vport_output(), which is called once per fragmented packet.
 825  */
 826 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
 827                          u16 orig_network_offset, u8 mac_proto)
 828 {
 829         unsigned int hlen = skb_network_offset(skb);
 830         struct ovs_frag_data *data;
 831
 832         data = this_cpu_ptr(&ovs_frag_data_storage);
 833         data->dst = skb->_skb_refdst;
 834         data->vport = vport;
 835         data->cb = *OVS_CB(skb);
 836         data->inner_protocol = skb->inner_protocol;
 837         data->network_offset = orig_network_offset;
 838         if (skb_vlan_tag_present(skb))
 839                 data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
 840         else
 841                 data->vlan_tci = 0;
 842         data->vlan_proto = skb->vlan_proto;
 843         data->mac_proto = mac_proto;
 844         data->l2_len = hlen;
 845         memcpy(&data->l2_data, skb->data, hlen);
 846
 847         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 848         skb_pull(skb, hlen);
 849 }
 850
 851 static void ovs_fragment(struct net *net, struct vport *vport,
 852                          struct sk_buff *skb, u16 mru,
 853                          struct sw_flow_key *key)
 854 {
 855         u16 orig_network_offset = 0;
 856
 857         if (eth_p_mpls(skb->protocol)) {
 858                 orig_network_offset = skb_network_offset(skb);
 859                 skb->network_header = skb->inner_network_header;
 860         }
 861
 862         if (skb_network_offset(skb) > MAX_L2_LEN) {
 863                 OVS_NLERR(1, "L2 header too long to fragment");
 864                 goto err;
 865         }
 866
 867         if (key->eth.type == htons(ETH_P_IP)) {
 868                 struct rtable ovs_rt = { 0 };
 869                 unsigned long orig_dst;
 870
 871                 prepare_frag(vport, skb, orig_network_offset,
 872                              ovs_key_mac_proto(key));
 873                 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
 874                          DST_OBSOLETE_NONE, DST_NOCOUNT);
 875                 ovs_rt.dst.dev = vport->dev;
 876
 877                 orig_dst = skb->_skb_refdst;
 878                 skb_dst_set_noref(skb, &ovs_rt.dst);
 879                 IPCB(skb)->frag_max_size = mru;
 880
 881                 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
 882                 refdst_drop(orig_dst);
 883         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 884                 unsigned long orig_dst;
 885                 struct rt6_info ovs_rt;
 886
 887                 prepare_frag(vport, skb, orig_network_offset,
 888                              ovs_key_mac_proto(key));
 889                 memset(&ovs_rt, 0, sizeof(ovs_rt));
 890                 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
 891                          DST_OBSOLETE_NONE, DST_NOCOUNT);
 892                 ovs_rt.dst.dev = vport->dev;
 893
 894                 orig_dst = skb->_skb_refdst;
 895                 skb_dst_set_noref(skb, &ovs_rt.dst);
 896                 IP6CB(skb)->frag_max_size = mru;
 897
 898                 ipv6_stub->ipv6_fragment(net, skb->sk, skb, ovs_vport_output);
 899                 refdst_drop(orig_dst);
 900         } else {
 901                 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
 902                           ovs_vport_name(vport), ntohs(key->eth.type), mru,
 903                           vport->dev->mtu);
 904                 goto err;
 905         }
 906
 907         return;
 908 err:
 909         kfree_skb(skb);
 910 }
 911
 912 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
 913                       struct sw_flow_key *key)
 914 {
 915         struct vport *vport = ovs_vport_rcu(dp, out_port);
 916
 917         if (likely(vport && netif_carrier_ok(vport->dev))) {
 918                 u16 mru = OVS_CB(skb)->mru;
 919                 u32 cutlen = OVS_CB(skb)->cutlen;
 920
 921                 if (unlikely(cutlen > 0)) {
 922                         if (skb->len - cutlen > ovs_mac_header_len(key))
 923                                 pskb_trim(skb, skb->len - cutlen);
 924                         else
 925                                 pskb_trim(skb, ovs_mac_header_len(key));
 926                 }
 927
 928                 if (likely(!mru ||
 929                            (skb->len <= mru + vport->dev->hard_header_len))) {
 930                         ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
 931                 } else if (mru <= vport->dev->mtu) {
 932                         struct net *net = read_pnet(&dp->net);
 933
 934                         ovs_fragment(net, vport, skb, mru, key);
 935                 } else {
 936                         kfree_skb(skb);
 937                 }
 938         } else {
 939                 kfree_skb(skb);
 940         }
 941 }
 942
 943 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 944                             struct sw_flow_key *key, const struct nlattr *attr,
 945                             const struct nlattr *actions, int actions_len,
 946                             uint32_t cutlen)
 947 {
 948         struct dp_upcall_info upcall;
 949         const struct nlattr *a;
 950         int rem;
 951
 952         memset(&upcall, 0, sizeof(upcall));
 953         upcall.cmd = OVS_PACKET_CMD_ACTION;
 954         upcall.mru = OVS_CB(skb)->mru;
 955
 956         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 957              a = nla_next(a, &rem)) {
 958                 switch (nla_type(a)) {
 959                 case OVS_USERSPACE_ATTR_USERDATA:
 960                         upcall.userdata = a;
 961                         break;
 962
 963                 case OVS_USERSPACE_ATTR_PID:
 964                         if (dp->user_features &
 965                             OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
 966                                 upcall.portid =
 967                                   ovs_dp_get_upcall_portid(dp,
 968                                                            smp_processor_id());
 969                         else
 970                                 upcall.portid = nla_get_u32(a);
 971                         break;
 972
 973                 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
 974                         /* Get out tunnel info. */
 975                         struct vport *vport;
 976
 977                         vport = ovs_vport_rcu(dp, nla_get_u32(a));
 978                         if (vport) {
 979                                 int err;
 980
 981                                 err = dev_fill_metadata_dst(vport->dev, skb);
 982                                 if (!err)
 983                                         upcall.egress_tun_info = skb_tunnel_info(skb);
 984                         }
 985
 986                         break;
 987                 }
 988
 989                 case OVS_USERSPACE_ATTR_ACTIONS: {
 990                         /* Include actions. */
 991                         upcall.actions = actions;
 992                         upcall.actions_len = actions_len;
 993                         break;
 994                 }
 995
 996                 } /* End of switch. */
 997         }
 998
 999         return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
1000 }
1001
1002 static int dec_ttl_exception_handler(struct datapath *dp, struct sk_buff *skb,
1003                                      struct sw_flow_key *key,
1004                                      const struct nlattr *attr)
1005 {
1006         /* The first attribute is always 'OVS_DEC_TTL_ATTR_ACTION'. */
1007         struct nlattr *actions = nla_data(attr);
1008
1009         if (nla_len(actions))
1010                 return clone_execute(dp, skb, key, 0, nla_data(actions),
1011                                      nla_len(actions), true, false);
1012
1013         consume_skb(skb);
1014         return 0;
1015 }
1016
1017 /* When 'last' is true, sample() should always consume the 'skb'.
1018  * Otherwise, sample() should keep 'skb' intact regardless what
1019  * actions are executed within sample().
1020  */
1021 static int sample(struct datapath *dp, struct sk_buff *skb,
1022                   struct sw_flow_key *key, const struct nlattr *attr,
1023                   bool last)
1024 {
1025         struct nlattr *actions;
1026         struct nlattr *sample_arg;
1027         int rem = nla_len(attr);
1028         const struct sample_arg *arg;
1029         bool clone_flow_key;
1030
1031         /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
1032         sample_arg = nla_data(attr);
1033         arg = nla_data(sample_arg);
1034         actions = nla_next(sample_arg, &rem);
1035
1036         if ((arg->probability != U32_MAX) &&
1037             (!arg->probability || get_random_u32() > arg->probability)) {
1038                 if (last)
1039                         consume_skb(skb);
1040                 return 0;
1041         }
1042
1043         clone_flow_key = !arg->exec;
1044         return clone_execute(dp, skb, key, 0, actions, rem, last,
1045                              clone_flow_key);
1046 }
1047
1048 /* When 'last' is true, clone() should always consume the 'skb'.
1049  * Otherwise, clone() should keep 'skb' intact regardless what
1050  * actions are executed within clone().
1051  */
1052 static int clone(struct datapath *dp, struct sk_buff *skb,
1053                  struct sw_flow_key *key, const struct nlattr *attr,
1054                  bool last)
1055 {
1056         struct nlattr *actions;
1057         struct nlattr *clone_arg;
1058         int rem = nla_len(attr);
1059         bool dont_clone_flow_key;
1060
1061         /* The first action is always 'OVS_CLONE_ATTR_EXEC'. */
1062         clone_arg = nla_data(attr);
1063         dont_clone_flow_key = nla_get_u32(clone_arg);
1064         actions = nla_next(clone_arg, &rem);
1065
1066         return clone_execute(dp, skb, key, 0, actions, rem, last,
1067                              !dont_clone_flow_key);
1068 }
1069
1070 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1071                          const struct nlattr *attr)
1072 {
1073         struct ovs_action_hash *hash_act = nla_data(attr);
1074         u32 hash = 0;
1075
1076         if (hash_act->hash_alg == OVS_HASH_ALG_L4) {
1077                 /* OVS_HASH_ALG_L4 hasing type. */
1078                 hash = skb_get_hash(skb);
1079         } else if (hash_act->hash_alg == OVS_HASH_ALG_SYM_L4) {
1080                 /* OVS_HASH_ALG_SYM_L4 hashing type.  NOTE: this doesn't
1081                  * extend past an encapsulated header.
1082                  */
1083                 hash = __skb_get_hash_symmetric(skb);
1084         }
1085
1086         hash = jhash_1word(hash, hash_act->hash_basis);
1087         if (!hash)
1088                 hash = 0x1;
1089
1090         key->ovs_flow_hash = hash;
1091 }
1092
1093 static int execute_set_action(struct sk_buff *skb,
1094                               struct sw_flow_key *flow_key,
1095                               const struct nlattr *a)
1096 {
1097         /* Only tunnel set execution is supported without a mask. */
1098         if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1099                 struct ovs_tunnel_info *tun = nla_data(a);
1100
1101                 skb_dst_drop(skb);
1102                 dst_hold((struct dst_entry *)tun->tun_dst);
1103                 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1104                 return 0;
1105         }
1106
1107         return -EINVAL;
1108 }
1109
1110 /* Mask is at the midpoint of the data. */
1111 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1112
1113 static int execute_masked_set_action(struct sk_buff *skb,
1114                                      struct sw_flow_key *flow_key,
1115                                      const struct nlattr *a)
1116 {
1117         int err = 0;
1118
1119         switch (nla_type(a)) {
1120         case OVS_KEY_ATTR_PRIORITY:
1121                 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1122                                *get_mask(a, u32 *));
1123                 flow_key->phy.priority = skb->priority;
1124                 break;
1125
1126         case OVS_KEY_ATTR_SKB_MARK:
1127                 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1128                 flow_key->phy.skb_mark = skb->mark;
1129                 break;
1130
1131         case OVS_KEY_ATTR_TUNNEL_INFO:
1132                 /* Masked data not supported for tunnel. */
1133                 err = -EINVAL;
1134                 break;
1135
1136         case OVS_KEY_ATTR_ETHERNET:
1137                 err = set_eth_addr(skb, flow_key, nla_data(a),
1138                                    get_mask(a, struct ovs_key_ethernet *));
1139                 break;
1140
1141         case OVS_KEY_ATTR_NSH:
1142                 err = set_nsh(skb, flow_key, a);
1143                 break;
1144
1145         case OVS_KEY_ATTR_IPV4:
1146                 err = set_ipv4(skb, flow_key, nla_data(a),
1147                                get_mask(a, struct ovs_key_ipv4 *));
1148                 break;
1149
1150         case OVS_KEY_ATTR_IPV6:
1151                 err = set_ipv6(skb, flow_key, nla_data(a),
1152                                get_mask(a, struct ovs_key_ipv6 *));
1153                 break;
1154
1155         case OVS_KEY_ATTR_TCP:
1156                 err = set_tcp(skb, flow_key, nla_data(a),
1157                               get_mask(a, struct ovs_key_tcp *));
1158                 break;
1159
1160         case OVS_KEY_ATTR_UDP:
1161                 err = set_udp(skb, flow_key, nla_data(a),
1162                               get_mask(a, struct ovs_key_udp *));
1163                 break;
1164
1165         case OVS_KEY_ATTR_SCTP:
1166                 err = set_sctp(skb, flow_key, nla_data(a),
1167                                get_mask(a, struct ovs_key_sctp *));
1168                 break;
1169
1170         case OVS_KEY_ATTR_MPLS:
1171                 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1172                                                                     __be32 *));
1173                 break;
1174
1175         case OVS_KEY_ATTR_CT_STATE:
1176         case OVS_KEY_ATTR_CT_ZONE:
1177         case OVS_KEY_ATTR_CT_MARK:
1178         case OVS_KEY_ATTR_CT_LABELS:
1179         case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1180         case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1181                 err = -EINVAL;
1182                 break;
1183         }
1184
1185         return err;
1186 }
1187
1188 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1189                           struct sw_flow_key *key,
1190                           const struct nlattr *a, bool last)
1191 {
1192         u32 recirc_id;
1193
1194         if (!is_flow_key_valid(key)) {
1195                 int err;
1196
1197                 err = ovs_flow_key_update(skb, key);
1198                 if (err)
1199                         return err;
1200         }
1201         BUG_ON(!is_flow_key_valid(key));
1202
1203         recirc_id = nla_get_u32(a);
1204         return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1205 }
1206
1207 static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1208                                  struct sw_flow_key *key,
1209                                  const struct nlattr *attr, bool last)
1210 {
1211         struct ovs_skb_cb *ovs_cb = OVS_CB(skb);
1212         const struct nlattr *actions, *cpl_arg;
1213         int len, max_len, rem = nla_len(attr);
1214         const struct check_pkt_len_arg *arg;
1215         bool clone_flow_key;
1216
1217         /* The first netlink attribute in 'attr' is always
1218          * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1219          */
1220         cpl_arg = nla_data(attr);
1221         arg = nla_data(cpl_arg);
1222
1223         len = ovs_cb->mru ? ovs_cb->mru + skb->mac_len : skb->len;
1224         max_len = arg->pkt_len;
1225
1226         if ((skb_is_gso(skb) && skb_gso_validate_mac_len(skb, max_len)) ||
1227             len <= max_len) {
1228                 /* Second netlink attribute in 'attr' is always
1229                  * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1230                  */
1231                 actions = nla_next(cpl_arg, &rem);
1232                 clone_flow_key = !arg->exec_for_lesser_equal;
1233         } else {
1234                 /* Third netlink attribute in 'attr' is always
1235                  * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1236                  */
1237                 actions = nla_next(cpl_arg, &rem);
1238                 actions = nla_next(actions, &rem);
1239                 clone_flow_key = !arg->exec_for_greater;
1240         }
1241
1242         return clone_execute(dp, skb, key, 0, nla_data(actions),
1243                              nla_len(actions), last, clone_flow_key);
1244 }
1245
1246 static int execute_dec_ttl(struct sk_buff *skb, struct sw_flow_key *key)
1247 {
1248         int err;
1249
1250         if (skb->protocol == htons(ETH_P_IPV6)) {
1251                 struct ipv6hdr *nh;
1252
1253                 err = skb_ensure_writable(skb, skb_network_offset(skb) +
1254                                           sizeof(*nh));
1255                 if (unlikely(err))
1256                         return err;
1257
1258                 nh = ipv6_hdr(skb);
1259
1260                 if (nh->hop_limit <= 1)
1261                         return -EHOSTUNREACH;
1262
1263                 key->ip.ttl = --nh->hop_limit;
1264         } else if (skb->protocol == htons(ETH_P_IP)) {
1265                 struct iphdr *nh;
1266                 u8 old_ttl;
1267
1268                 err = skb_ensure_writable(skb, skb_network_offset(skb) +
1269                                           sizeof(*nh));
1270                 if (unlikely(err))
1271                         return err;
1272
1273                 nh = ip_hdr(skb);
1274                 if (nh->ttl <= 1)
1275                         return -EHOSTUNREACH;
1276
1277                 old_ttl = nh->ttl--;
1278                 csum_replace2(&nh->check, htons(old_ttl << 8),
1279                               htons(nh->ttl << 8));
1280                 key->ip.ttl = nh->ttl;
1281         }
1282         return 0;
1283 }
1284
1285 /* Execute a list of actions against 'skb'. */
1286 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1287                               struct sw_flow_key *key,
1288                               const struct nlattr *attr, int len)
1289 {
1290         const struct nlattr *a;
1291         int rem;
1292
1293         for (a = attr, rem = len; rem > 0;
1294              a = nla_next(a, &rem)) {
1295                 int err = 0;
1296
1297                 if (trace_ovs_do_execute_action_enabled())
1298                         trace_ovs_do_execute_action(dp, skb, key, a, rem);
1299
1300                 switch (nla_type(a)) {
1301                 case OVS_ACTION_ATTR_OUTPUT: {
1302                         int port = nla_get_u32(a);
1303                         struct sk_buff *clone;
1304
1305                         /* Every output action needs a separate clone
1306                          * of 'skb', In case the output action is the
1307                          * last action, cloning can be avoided.
1308                          */
1309                         if (nla_is_last(a, rem)) {
1310                                 do_output(dp, skb, port, key);
1311                                 /* 'skb' has been used for output.
1312                                  */
1313                                 return 0;
1314                         }
1315
1316                         clone = skb_clone(skb, GFP_ATOMIC);
1317                         if (clone)
1318                                 do_output(dp, clone, port, key);
1319                         OVS_CB(skb)->cutlen = 0;
1320                         break;
1321                 }
1322
1323                 case OVS_ACTION_ATTR_TRUNC: {
1324                         struct ovs_action_trunc *trunc = nla_data(a);
1325
1326                         if (skb->len > trunc->max_len)
1327                                 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1328                         break;
1329                 }
1330
1331                 case OVS_ACTION_ATTR_USERSPACE:
1332                         output_userspace(dp, skb, key, a, attr,
1333                                                      len, OVS_CB(skb)->cutlen);
1334                         OVS_CB(skb)->cutlen = 0;
1335                         break;
1336
1337                 case OVS_ACTION_ATTR_HASH:
1338                         execute_hash(skb, key, a);
1339                         break;
1340
1341                 case OVS_ACTION_ATTR_PUSH_MPLS: {
1342                         struct ovs_action_push_mpls *mpls = nla_data(a);
1343
1344                         err = push_mpls(skb, key, mpls->mpls_lse,
1345                                         mpls->mpls_ethertype, skb->mac_len);
1346                         break;
1347                 }
1348                 case OVS_ACTION_ATTR_ADD_MPLS: {
1349                         struct ovs_action_add_mpls *mpls = nla_data(a);
1350                         __u16 mac_len = 0;
1351
1352                         if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK)
1353                                 mac_len = skb->mac_len;
1354
1355                         err = push_mpls(skb, key, mpls->mpls_lse,
1356                                         mpls->mpls_ethertype, mac_len);
1357                         break;
1358                 }
1359                 case OVS_ACTION_ATTR_POP_MPLS:
1360                         err = pop_mpls(skb, key, nla_get_be16(a));
1361                         break;
1362
1363                 case OVS_ACTION_ATTR_PUSH_VLAN:
1364                         err = push_vlan(skb, key, nla_data(a));
1365                         break;
1366
1367                 case OVS_ACTION_ATTR_POP_VLAN:
1368                         err = pop_vlan(skb, key);
1369                         break;
1370
1371                 case OVS_ACTION_ATTR_RECIRC: {
1372                         bool last = nla_is_last(a, rem);
1373
1374                         err = execute_recirc(dp, skb, key, a, last);
1375                         if (last) {
1376                                 /* If this is the last action, the skb has
1377                                  * been consumed or freed.
1378                                  * Return immediately.
1379                                  */
1380                                 return err;
1381                         }
1382                         break;
1383                 }
1384
1385                 case OVS_ACTION_ATTR_SET:
1386                         err = execute_set_action(skb, key, nla_data(a));
1387                         break;
1388
1389                 case OVS_ACTION_ATTR_SET_MASKED:
1390                 case OVS_ACTION_ATTR_SET_TO_MASKED:
1391                         err = execute_masked_set_action(skb, key, nla_data(a));
1392                         break;
1393
1394                 case OVS_ACTION_ATTR_SAMPLE: {
1395                         bool last = nla_is_last(a, rem);
1396
1397                         err = sample(dp, skb, key, a, last);
1398                         if (last)
1399                                 return err;
1400
1401                         break;
1402                 }
1403
1404                 case OVS_ACTION_ATTR_CT:
1405                         if (!is_flow_key_valid(key)) {
1406                                 err = ovs_flow_key_update(skb, key);
1407                                 if (err)
1408                                         return err;
1409                         }
1410
1411                         err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1412                                              nla_data(a));
1413
1414                         /* Hide stolen IP fragments from user space. */
1415                         if (err)
1416                                 return err == -EINPROGRESS ? 0 : err;
1417                         break;
1418
1419                 case OVS_ACTION_ATTR_CT_CLEAR:
1420                         err = ovs_ct_clear(skb, key);
1421                         break;
1422
1423                 case OVS_ACTION_ATTR_PUSH_ETH:
1424                         err = push_eth(skb, key, nla_data(a));
1425                         break;
1426
1427                 case OVS_ACTION_ATTR_POP_ETH:
1428                         err = pop_eth(skb, key);
1429                         break;
1430
1431                 case OVS_ACTION_ATTR_PUSH_NSH: {
1432                         u8 buffer[NSH_HDR_MAX_LEN];
1433                         struct nshhdr *nh = (struct nshhdr *)buffer;
1434
1435                         err = nsh_hdr_from_nlattr(nla_data(a), nh,
1436                                                   NSH_HDR_MAX_LEN);
1437                         if (unlikely(err))
1438                                 break;
1439                         err = push_nsh(skb, key, nh);
1440                         break;
1441                 }
1442
1443                 case OVS_ACTION_ATTR_POP_NSH:
1444                         err = pop_nsh(skb, key);
1445                         break;
1446
1447                 case OVS_ACTION_ATTR_METER:
1448                         if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1449                                 consume_skb(skb);
1450                                 return 0;
1451                         }
1452                         break;
1453
1454                 case OVS_ACTION_ATTR_CLONE: {
1455                         bool last = nla_is_last(a, rem);
1456
1457                         err = clone(dp, skb, key, a, last);
1458                         if (last)
1459                                 return err;
1460
1461                         break;
1462                 }
1463
1464                 case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1465                         bool last = nla_is_last(a, rem);
1466
1467                         err = execute_check_pkt_len(dp, skb, key, a, last);
1468                         if (last)
1469                                 return err;
1470
1471                         break;
1472                 }
1473
1474                 case OVS_ACTION_ATTR_DEC_TTL:
1475                         err = execute_dec_ttl(skb, key);
1476                         if (err == -EHOSTUNREACH)
1477                                 return dec_ttl_exception_handler(dp, skb,
1478                                                                  key, a);
1479                         break;
1480                 }
1481
1482                 if (unlikely(err)) {
1483                         kfree_skb(skb);
1484                         return err;
1485                 }
1486         }
1487
1488         consume_skb(skb);
1489         return 0;
1490 }
1491
1492 /* Execute the actions on the clone of the packet. The effect of the
1493  * execution does not affect the original 'skb' nor the original 'key'.
1494  *
1495  * The execution may be deferred in case the actions can not be executed
1496  * immediately.
1497  */
1498 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1499                          struct sw_flow_key *key, u32 recirc_id,
1500                          const struct nlattr *actions, int len,
1501                          bool last, bool clone_flow_key)
1502 {
1503         struct deferred_action *da;
1504         struct sw_flow_key *clone;
1505
1506         skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1507         if (!skb) {
1508                 /* Out of memory, skip this action.
1509                  */
1510                 return 0;
1511         }
1512
1513         /* When clone_flow_key is false, the 'key' will not be change
1514          * by the actions, then the 'key' can be used directly.
1515          * Otherwise, try to clone key from the next recursion level of
1516          * 'flow_keys'. If clone is successful, execute the actions
1517          * without deferring.
1518          */
1519         clone = clone_flow_key ? clone_key(key) : key;
1520         if (clone) {
1521                 int err = 0;
1522
1523                 if (actions) { /* Sample action */
1524                         if (clone_flow_key)
1525                                 __this_cpu_inc(exec_actions_level);
1526
1527                         err = do_execute_actions(dp, skb, clone,
1528                                                  actions, len);
1529
1530                         if (clone_flow_key)
1531                                 __this_cpu_dec(exec_actions_level);
1532                 } else { /* Recirc action */
1533                         clone->recirc_id = recirc_id;
1534                         ovs_dp_process_packet(skb, clone);
1535                 }
1536                 return err;
1537         }
1538
1539         /* Out of 'flow_keys' space. Defer actions */
1540         da = add_deferred_actions(skb, key, actions, len);
1541         if (da) {
1542                 if (!actions) { /* Recirc action */
1543                         key = &da->pkt_key;
1544                         key->recirc_id = recirc_id;
1545                 }
1546         } else {
1547                 /* Out of per CPU action FIFO space. Drop the 'skb' and
1548                  * log an error.
1549                  */
1550                 kfree_skb(skb);
1551
1552                 if (net_ratelimit()) {
1553                         if (actions) { /* Sample action */
1554                                 pr_warn("%s: deferred action limit reached, drop sample action\n",
1555                                         ovs_dp_name(dp));
1556                         } else {  /* Recirc action */
1557                                 pr_warn("%s: deferred action limit reached, drop recirc action (recirc_id=%#x)\n",
1558                                         ovs_dp_name(dp), recirc_id);
1559                         }
1560                 }
1561         }
1562         return 0;
1563 }
1564
1565 static void process_deferred_actions(struct datapath *dp)
1566 {
1567         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1568
1569         /* Do not touch the FIFO in case there is no deferred actions. */
1570         if (action_fifo_is_empty(fifo))
1571                 return;
1572
1573         /* Finishing executing all deferred actions. */
1574         do {
1575                 struct deferred_action *da = action_fifo_get(fifo);
1576                 struct sk_buff *skb = da->skb;
1577                 struct sw_flow_key *key = &da->pkt_key;
1578                 const struct nlattr *actions = da->actions;
1579                 int actions_len = da->actions_len;
1580
1581                 if (actions)
1582                         do_execute_actions(dp, skb, key, actions, actions_len);
1583                 else
1584                         ovs_dp_process_packet(skb, key);
1585         } while (!action_fifo_is_empty(fifo));
1586
1587         /* Reset FIFO for the next packet.  */
1588         action_fifo_init(fifo);
1589 }
1590
1591 /* Execute a list of actions against 'skb'. */
1592 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1593                         const struct sw_flow_actions *acts,
1594                         struct sw_flow_key *key)
1595 {
1596         int err, level;
1597
1598         level = __this_cpu_inc_return(exec_actions_level);
1599         if (unlikely(level > OVS_RECURSION_LIMIT)) {
1600                 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1601                                      ovs_dp_name(dp));
1602                 kfree_skb(skb);
1603                 err = -ENETDOWN;
1604                 goto out;
1605         }
1606
1607         OVS_CB(skb)->acts_origlen = acts->orig_len;
1608         err = do_execute_actions(dp, skb, key,
1609                                  acts->actions, acts->actions_len);
1610
1611         if (level == 1)
1612                 process_deferred_actions(dp);
1613
1614 out:
1615         __this_cpu_dec(exec_actions_level);
1616         return err;
1617 }
1618
1619 int action_fifos_init(void)
1620 {
1621         action_fifos = alloc_percpu(struct action_fifo);
1622         if (!action_fifos)
1623                 return -ENOMEM;
1624
1625         flow_keys = alloc_percpu(struct action_flow_keys);
1626         if (!flow_keys) {
1627                 free_percpu(action_fifos);
1628                 return -ENOMEM;
1629         }
1630
1631         return 0;
1632 }
1633
1634 void action_fifos_exit(void)
1635 {
1636         free_percpu(action_fifos);
1637         free_percpu(flow_keys);
1638 }