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/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 #include "openvswitch_trace.h"
  34
  35 struct deferred_action {
  36         struct sk_buff *skb;
  37         const struct nlattr *actions;
  38         int actions_len;
  39
  40         /* Store pkt_key clone when creating deferred action. */
  41         struct sw_flow_key pkt_key;
  42 };
  43
  44 #define MAX_L2_LEN      (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
  45 struct ovs_frag_data {
  46         unsigned long dst;
  47         struct vport *vport;
  48         struct ovs_skb_cb cb;
  49         __be16 inner_protocol;
  50         u16 network_offset;     /* valid only for MPLS */
  51         u16 vlan_tci;
  52         __be16 vlan_proto;
  53         unsigned int l2_len;
  54         u8 mac_proto;
  55         u8 l2_data[MAX_L2_LEN];
  56 };
  57
  58 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
  59
  60 #define DEFERRED_ACTION_FIFO_SIZE 10
  61 #define OVS_RECURSION_LIMIT 5
  62 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
  63 struct action_fifo {
  64         int head;
  65         int tail;
  66         /* Deferred action fifo queue storage. */
  67         struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
  68 };
  69
  70 struct action_flow_keys {
  71         struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
  72 };
  73
  74 static struct action_fifo __percpu *action_fifos;
  75 static struct action_flow_keys __percpu *flow_keys;
  76 static DEFINE_PER_CPU(int, exec_actions_level);
  77
  78 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
  79  * space. Return NULL if out of key spaces.
  80  */
  81 static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
  82 {
  83         struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
  84         int level = this_cpu_read(exec_actions_level);
  85         struct sw_flow_key *key = NULL;
  86
  87         if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
  88                 key = &keys->key[level - 1];
  89                 *key = *key_;
  90         }
  91
  92         return key;
  93 }
  94
  95 static void action_fifo_init(struct action_fifo *fifo)
  96 {
  97         fifo->head = 0;
  98         fifo->tail = 0;
  99 }
 100
 101 static bool action_fifo_is_empty(const struct action_fifo *fifo)
 102 {
 103         return (fifo->head == fifo->tail);
 104 }
 105
 106 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
 107 {
 108         if (action_fifo_is_empty(fifo))
 109                 return NULL;
 110
 111         return &fifo->fifo[fifo->tail++];
 112 }
 113
 114 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
 115 {
 116         if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
 117                 return NULL;
 118
 119         return &fifo->fifo[fifo->head++];
 120 }
 121
 122 /* Return true if fifo is not full */
 123 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
 124                                     const struct sw_flow_key *key,
 125                                     const struct nlattr *actions,
 126                                     const int actions_len)
 127 {
 128         struct action_fifo *fifo;
 129         struct deferred_action *da;
 130
 131         fifo = this_cpu_ptr(action_fifos);
 132         da = action_fifo_put(fifo);
 133         if (da) {
 134                 da->skb = skb;
 135                 da->actions = actions;
 136                 da->actions_len = actions_len;
 137                 da->pkt_key = *key;
 138         }
 139
 140         return da;
 141 }
 142
 143 static void invalidate_flow_key(struct sw_flow_key *key)
 144 {
 145         key->mac_proto |= SW_FLOW_KEY_INVALID;
 146 }
 147
 148 static bool is_flow_key_valid(const struct sw_flow_key *key)
 149 {
 150         return !(key->mac_proto & SW_FLOW_KEY_INVALID);
 151 }
 152
 153 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
 154                          struct sw_flow_key *key,
 155                          u32 recirc_id,
 156                          const struct nlattr *actions, int len,
 157                          bool last, bool clone_flow_key);
 158
 159 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 160                               struct sw_flow_key *key,
 161                               const struct nlattr *attr, int len);
 162
 163 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 164                      __be32 mpls_lse, __be16 mpls_ethertype, __u16 mac_len)
 165 {
 166         int err;
 167
 168         err = skb_mpls_push(skb, mpls_lse, mpls_ethertype, mac_len, !!mac_len);
 169         if (err)
 170                 return err;
 171
 172         if (!mac_len)
 173                 key->mac_proto = MAC_PROTO_NONE;
 174
 175         invalidate_flow_key(key);
 176         return 0;
 177 }
 178
 179 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 180                     const __be16 ethertype)
 181 {
 182         int err;
 183
 184         err = skb_mpls_pop(skb, ethertype, skb->mac_len,
 185                            ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET);
 186         if (err)
 187                 return err;
 188
 189         if (ethertype == htons(ETH_P_TEB))
 190                 key->mac_proto = MAC_PROTO_ETHERNET;
 191
 192         invalidate_flow_key(key);
 193         return 0;
 194 }
 195
 196 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
 197                     const __be32 *mpls_lse, const __be32 *mask)
 198 {
 199         struct mpls_shim_hdr *stack;
 200         __be32 lse;
 201         int err;
 202
 203         if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
 204                 return -ENOMEM;
 205
 206         stack = mpls_hdr(skb);
 207         lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
 208         err = skb_mpls_update_lse(skb, lse);
 209         if (err)
 210                 return err;
 211
 212         flow_key->mpls.lse[0] = lse;
 213         return 0;
 214 }
 215
 216 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 217 {
 218         int err;
 219
 220         err = skb_vlan_pop(skb);
 221         if (skb_vlan_tag_present(skb)) {
 222                 invalidate_flow_key(key);
 223         } else {
 224                 key->eth.vlan.tci = 0;
 225                 key->eth.vlan.tpid = 0;
 226         }
 227         return err;
 228 }
 229
 230 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
 231                      const struct ovs_action_push_vlan *vlan)
 232 {
 233         if (skb_vlan_tag_present(skb)) {
 234                 invalidate_flow_key(key);
 235         } else {
 236                 key->eth.vlan.tci = vlan->vlan_tci;
 237                 key->eth.vlan.tpid = vlan->vlan_tpid;
 238         }
 239         return skb_vlan_push(skb, vlan->vlan_tpid,
 240                              ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
 241 }
 242
 243 /* 'src' is already properly masked. */
 244 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
 245 {
 246         u16 *dst = (u16 *)dst_;
 247         const u16 *src = (const u16 *)src_;
 248         const u16 *mask = (const u16 *)mask_;
 249
 250         OVS_SET_MASKED(dst[0], src[0], mask[0]);
 251         OVS_SET_MASKED(dst[1], src[1], mask[1]);
 252         OVS_SET_MASKED(dst[2], src[2], mask[2]);
 253 }
 254
 255 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
 256                         const struct ovs_key_ethernet *key,
 257                         const struct ovs_key_ethernet *mask)
 258 {
 259         int err;
 260
 261         err = skb_ensure_writable(skb, ETH_HLEN);
 262         if (unlikely(err))
 263                 return err;
 264
 265         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 266
 267         ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
 268                                mask->eth_src);
 269         ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
 270                                mask->eth_dst);
 271
 272         skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 273
 274         ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
 275         ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
 276         return 0;
 277 }
 278
 279 /* pop_eth does not support VLAN packets as this action is never called
 280  * for them.
 281  */
 282 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
 283 {
 284         int err;
 285
 286         err = skb_eth_pop(skb);
 287         if (err)
 288                 return err;
 289
 290         /* safe right before invalidate_flow_key */
 291         key->mac_proto = MAC_PROTO_NONE;
 292         invalidate_flow_key(key);
 293         return 0;
 294 }
 295
 296 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
 297                     const struct ovs_action_push_eth *ethh)
 298 {
 299         int err;
 300
 301         err = skb_eth_push(skb, ethh->addresses.eth_dst,
 302                            ethh->addresses.eth_src);
 303         if (err)
 304                 return err;
 305
 306         /* safe right before invalidate_flow_key */
 307         key->mac_proto = MAC_PROTO_ETHERNET;
 308         invalidate_flow_key(key);
 309         return 0;
 310 }
 311
 312 static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
 313                     const struct nshhdr *nh)
 314 {
 315         int err;
 316
 317         err = nsh_push(skb, nh);
 318         if (err)
 319                 return err;
 320
 321         /* safe right before invalidate_flow_key */
 322         key->mac_proto = MAC_PROTO_NONE;
 323         invalidate_flow_key(key);
 324         return 0;
 325 }
 326
 327 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
 328 {
 329         int err;
 330
 331         err = nsh_pop(skb);
 332         if (err)
 333                 return err;
 334
 335         /* safe right before invalidate_flow_key */
 336         if (skb->protocol == htons(ETH_P_TEB))
 337                 key->mac_proto = MAC_PROTO_ETHERNET;
 338         else
 339                 key->mac_proto = MAC_PROTO_NONE;
 340         invalidate_flow_key(key);
 341         return 0;
 342 }
 343
 344 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
 345                                   __be32 addr, __be32 new_addr)
 346 {
 347         int transport_len = skb->len - skb_transport_offset(skb);
 348
 349         if (nh->frag_off & htons(IP_OFFSET))
 350                 return;
 351
 352         if (nh->protocol == IPPROTO_TCP) {
 353                 if (likely(transport_len >= sizeof(struct tcphdr)))
 354                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 355                                                  addr, new_addr, true);
 356         } else if (nh->protocol == IPPROTO_UDP) {
 357                 if (likely(transport_len >= sizeof(struct udphdr))) {
 358                         struct udphdr *uh = udp_hdr(skb);
 359
 360                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 361                                 inet_proto_csum_replace4(&uh->check, skb,
 362                                                          addr, new_addr, true);
 363                                 if (!uh->check)
 364                                         uh->check = CSUM_MANGLED_0;
 365                         }
 366                 }
 367         }
 368 }
 369
 370 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 371                         __be32 *addr, __be32 new_addr)
 372 {
 373         update_ip_l4_checksum(skb, nh, *addr, new_addr);
 374         csum_replace4(&nh->check, *addr, new_addr);
 375         skb_clear_hash(skb);
 376         ovs_ct_clear(skb, NULL);
 377         *addr = new_addr;
 378 }
 379
 380 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 381                                  __be32 addr[4], const __be32 new_addr[4])
 382 {
 383         int transport_len = skb->len - skb_transport_offset(skb);
 384
 385         if (l4_proto == NEXTHDR_TCP) {
 386                 if (likely(transport_len >= sizeof(struct tcphdr)))
 387                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 388                                                   addr, new_addr, true);
 389         } else if (l4_proto == NEXTHDR_UDP) {
 390                 if (likely(transport_len >= sizeof(struct udphdr))) {
 391                         struct udphdr *uh = udp_hdr(skb);
 392
 393                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 394                                 inet_proto_csum_replace16(&uh->check, skb,
 395                                                           addr, new_addr, true);
 396                                 if (!uh->check)
 397                                         uh->check = CSUM_MANGLED_0;
 398                         }
 399                 }
 400         } else if (l4_proto == NEXTHDR_ICMP) {
 401                 if (likely(transport_len >= sizeof(struct icmp6hdr)))
 402                         inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 403                                                   skb, addr, new_addr, true);
 404         }
 405 }
 406
 407 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
 408                            const __be32 mask[4], __be32 masked[4])
 409 {
 410         masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
 411         masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
 412         masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
 413         masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
 414 }
 415
 416 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 417                           __be32 addr[4], const __be32 new_addr[4],
 418                           bool recalculate_csum)
 419 {
 420         if (recalculate_csum)
 421                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 422
 423         skb_clear_hash(skb);
 424         ovs_ct_clear(skb, NULL);
 425         memcpy(addr, new_addr, sizeof(__be32[4]));
 426 }
 427
 428 static void set_ipv6_dsfield(struct sk_buff *skb, struct ipv6hdr *nh, u8 ipv6_tclass, u8 mask)
 429 {
 430         u8 old_ipv6_tclass = ipv6_get_dsfield(nh);
 431
 432         ipv6_tclass = OVS_MASKED(old_ipv6_tclass, ipv6_tclass, mask);
 433
 434         if (skb->ip_summed == CHECKSUM_COMPLETE)
 435                 csum_replace(&skb->csum, (__force __wsum)(old_ipv6_tclass << 12),
 436                              (__force __wsum)(ipv6_tclass << 12));
 437
 438         ipv6_change_dsfield(nh, ~mask, ipv6_tclass);
 439 }
 440
 441 static void set_ipv6_fl(struct sk_buff *skb, struct ipv6hdr *nh, u32 fl, u32 mask)
 442 {
 443         u32 ofl;
 444
 445         ofl = nh->flow_lbl[0] << 16 |  nh->flow_lbl[1] << 8 |  nh->flow_lbl[2];
 446         fl = OVS_MASKED(ofl, fl, mask);
 447
 448         /* Bits 21-24 are always unmasked, so this retains their values. */
 449         nh->flow_lbl[0] = (u8)(fl >> 16);
 450         nh->flow_lbl[1] = (u8)(fl >> 8);
 451         nh->flow_lbl[2] = (u8)fl;
 452
 453         if (skb->ip_summed == CHECKSUM_COMPLETE)
 454                 csum_replace(&skb->csum, (__force __wsum)htonl(ofl), (__force __wsum)htonl(fl));
 455 }
 456
 457 static void set_ipv6_ttl(struct sk_buff *skb, struct ipv6hdr *nh, u8 new_ttl, u8 mask)
 458 {
 459         new_ttl = OVS_MASKED(nh->hop_limit, new_ttl, mask);
 460
 461         if (skb->ip_summed == CHECKSUM_COMPLETE)
 462                 csum_replace(&skb->csum, (__force __wsum)(nh->hop_limit << 8),
 463                              (__force __wsum)(new_ttl << 8));
 464         nh->hop_limit = new_ttl;
 465 }
 466
 467 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
 468                        u8 mask)
 469 {
 470         new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
 471
 472         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 473         nh->ttl = new_ttl;
 474 }
 475
 476 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
 477                     const struct ovs_key_ipv4 *key,
 478                     const struct ovs_key_ipv4 *mask)
 479 {
 480         struct iphdr *nh;
 481         __be32 new_addr;
 482         int err;
 483
 484         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 485                                   sizeof(struct iphdr));
 486         if (unlikely(err))
 487                 return err;
 488
 489         nh = ip_hdr(skb);
 490
 491         /* Setting an IP addresses is typically only a side effect of
 492          * matching on them in the current userspace implementation, so it
 493          * makes sense to check if the value actually changed.
 494          */
 495         if (mask->ipv4_src) {
 496                 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
 497
 498                 if (unlikely(new_addr != nh->saddr)) {
 499                         set_ip_addr(skb, nh, &nh->saddr, new_addr);
 500                         flow_key->ipv4.addr.src = new_addr;
 501                 }
 502         }
 503         if (mask->ipv4_dst) {
 504                 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
 505
 506                 if (unlikely(new_addr != nh->daddr)) {
 507                         set_ip_addr(skb, nh, &nh->daddr, new_addr);
 508                         flow_key->ipv4.addr.dst = new_addr;
 509                 }
 510         }
 511         if (mask->ipv4_tos) {
 512                 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
 513                 flow_key->ip.tos = nh->tos;
 514         }
 515         if (mask->ipv4_ttl) {
 516                 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
 517                 flow_key->ip.ttl = nh->ttl;
 518         }
 519
 520         return 0;
 521 }
 522
 523 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
 524 {
 525         return !!(addr[0] | addr[1] | addr[2] | addr[3]);
 526 }
 527
 528 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
 529                     const struct ovs_key_ipv6 *key,
 530                     const struct ovs_key_ipv6 *mask)
 531 {
 532         struct ipv6hdr *nh;
 533         int err;
 534
 535         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 536                                   sizeof(struct ipv6hdr));
 537         if (unlikely(err))
 538                 return err;
 539
 540         nh = ipv6_hdr(skb);
 541
 542         /* Setting an IP addresses is typically only a side effect of
 543          * matching on them in the current userspace implementation, so it
 544          * makes sense to check if the value actually changed.
 545          */
 546         if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
 547                 __be32 *saddr = (__be32 *)&nh->saddr;
 548                 __be32 masked[4];
 549
 550                 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
 551
 552                 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
 553                         set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
 554                                       true);
 555                         memcpy(&flow_key->ipv6.addr.src, masked,
 556                                sizeof(flow_key->ipv6.addr.src));
 557                 }
 558         }
 559         if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
 560                 unsigned int offset = 0;
 561                 int flags = IP6_FH_F_SKIP_RH;
 562                 bool recalc_csum = true;
 563                 __be32 *daddr = (__be32 *)&nh->daddr;
 564                 __be32 masked[4];
 565
 566                 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
 567
 568                 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
 569                         if (ipv6_ext_hdr(nh->nexthdr))
 570                                 recalc_csum = (ipv6_find_hdr(skb, &offset,
 571                                                              NEXTHDR_ROUTING,
 572                                                              NULL, &flags)
 573                                                != NEXTHDR_ROUTING);
 574
 575                         set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
 576                                       recalc_csum);
 577                         memcpy(&flow_key->ipv6.addr.dst, masked,
 578                                sizeof(flow_key->ipv6.addr.dst));
 579                 }
 580         }
 581         if (mask->ipv6_tclass) {
 582                 set_ipv6_dsfield(skb, nh, key->ipv6_tclass, mask->ipv6_tclass);
 583                 flow_key->ip.tos = ipv6_get_dsfield(nh);
 584         }
 585         if (mask->ipv6_label) {
 586                 set_ipv6_fl(skb, nh, ntohl(key->ipv6_label),
 587                             ntohl(mask->ipv6_label));
 588                 flow_key->ipv6.label =
 589                     *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 590         }
 591         if (mask->ipv6_hlimit) {
 592                 set_ipv6_ttl(skb, nh, key->ipv6_hlimit, mask->ipv6_hlimit);
 593                 flow_key->ip.ttl = nh->hop_limit;
 594         }
 595         return 0;
 596 }
 597
 598 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
 599                    const struct nlattr *a)
 600 {
 601         struct nshhdr *nh;
 602         size_t length;
 603         int err;
 604         u8 flags;
 605         u8 ttl;
 606         int i;
 607
 608         struct ovs_key_nsh key;
 609         struct ovs_key_nsh mask;
 610
 611         err = nsh_key_from_nlattr(a, &key, &mask);
 612         if (err)
 613                 return err;
 614
 615         /* Make sure the NSH base header is there */
 616         if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
 617                 return -ENOMEM;
 618
 619         nh = nsh_hdr(skb);
 620         length = nsh_hdr_len(nh);
 621
 622         /* Make sure the whole NSH header is there */
 623         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 624                                        length);
 625         if (unlikely(err))
 626                 return err;
 627
 628         nh = nsh_hdr(skb);
 629         skb_postpull_rcsum(skb, nh, length);
 630         flags = nsh_get_flags(nh);
 631         flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
 632         flow_key->nsh.base.flags = flags;
 633         ttl = nsh_get_ttl(nh);
 634         ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
 635         flow_key->nsh.base.ttl = ttl;
 636         nsh_set_flags_and_ttl(nh, flags, ttl);
 637         nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
 638                                   mask.base.path_hdr);
 639         flow_key->nsh.base.path_hdr = nh->path_hdr;
 640         switch (nh->mdtype) {
 641         case NSH_M_TYPE1:
 642                 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
 643                         nh->md1.context[i] =
 644                             OVS_MASKED(nh->md1.context[i], key.context[i],
 645                                        mask.context[i]);
 646                 }
 647                 memcpy(flow_key->nsh.context, nh->md1.context,
 648                        sizeof(nh->md1.context));
 649                 break;
 650         case NSH_M_TYPE2:
 651                 memset(flow_key->nsh.context, 0,
 652                        sizeof(flow_key->nsh.context));
 653                 break;
 654         default:
 655                 return -EINVAL;
 656         }
 657         skb_postpush_rcsum(skb, nh, length);
 658         return 0;
 659 }
 660
 661 /* Must follow skb_ensure_writable() since that can move the skb data. */
 662 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 663                         __be16 new_port, __sum16 *check)
 664 {
 665         ovs_ct_clear(skb, NULL);
 666         inet_proto_csum_replace2(check, skb, *port, new_port, false);
 667         *port = new_port;
 668 }
 669
 670 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 671                    const struct ovs_key_udp *key,
 672                    const struct ovs_key_udp *mask)
 673 {
 674         struct udphdr *uh;
 675         __be16 src, dst;
 676         int err;
 677
 678         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 679                                   sizeof(struct udphdr));
 680         if (unlikely(err))
 681                 return err;
 682
 683         uh = udp_hdr(skb);
 684         /* Either of the masks is non-zero, so do not bother checking them. */
 685         src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
 686         dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
 687
 688         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 689                 if (likely(src != uh->source)) {
 690                         set_tp_port(skb, &uh->source, src, &uh->check);
 691                         flow_key->tp.src = src;
 692                 }
 693                 if (likely(dst != uh->dest)) {
 694                         set_tp_port(skb, &uh->dest, dst, &uh->check);
 695                         flow_key->tp.dst = dst;
 696                 }
 697
 698                 if (unlikely(!uh->check))
 699                         uh->check = CSUM_MANGLED_0;
 700         } else {
 701                 uh->source = src;
 702                 uh->dest = dst;
 703                 flow_key->tp.src = src;
 704                 flow_key->tp.dst = dst;
 705                 ovs_ct_clear(skb, NULL);
 706         }
 707
 708         skb_clear_hash(skb);
 709
 710         return 0;
 711 }
 712
 713 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 714                    const struct ovs_key_tcp *key,
 715                    const struct ovs_key_tcp *mask)
 716 {
 717         struct tcphdr *th;
 718         __be16 src, dst;
 719         int err;
 720
 721         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 722                                   sizeof(struct tcphdr));
 723         if (unlikely(err))
 724                 return err;
 725
 726         th = tcp_hdr(skb);
 727         src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
 728         if (likely(src != th->source)) {
 729                 set_tp_port(skb, &th->source, src, &th->check);
 730                 flow_key->tp.src = src;
 731         }
 732         dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
 733         if (likely(dst != th->dest)) {
 734                 set_tp_port(skb, &th->dest, dst, &th->check);
 735                 flow_key->tp.dst = dst;
 736         }
 737         skb_clear_hash(skb);
 738
 739         return 0;
 740 }
 741
 742 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 743                     const struct ovs_key_sctp *key,
 744                     const struct ovs_key_sctp *mask)
 745 {
 746         unsigned int sctphoff = skb_transport_offset(skb);
 747         struct sctphdr *sh;
 748         __le32 old_correct_csum, new_csum, old_csum;
 749         int err;
 750
 751         err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
 752         if (unlikely(err))
 753                 return err;
 754
 755         sh = sctp_hdr(skb);
 756         old_csum = sh->checksum;
 757         old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 758
 759         sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
 760         sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
 761
 762         new_csum = sctp_compute_cksum(skb, sctphoff);
 763
 764         /* Carry any checksum errors through. */
 765         sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 766
 767         skb_clear_hash(skb);
 768         ovs_ct_clear(skb, NULL);
 769
 770         flow_key->tp.src = sh->source;
 771         flow_key->tp.dst = sh->dest;
 772
 773         return 0;
 774 }
 775
 776 static int ovs_vport_output(struct net *net, struct sock *sk,
 777                             struct sk_buff *skb)
 778 {
 779         struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
 780         struct vport *vport = data->vport;
 781
 782         if (skb_cow_head(skb, data->l2_len) < 0) {
 783                 kfree_skb(skb);
 784                 return -ENOMEM;
 785         }
 786
 787         __skb_dst_copy(skb, data->dst);
 788         *OVS_CB(skb) = data->cb;
 789         skb->inner_protocol = data->inner_protocol;
 790         if (data->vlan_tci & VLAN_CFI_MASK)
 791                 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
 792         else
 793                 __vlan_hwaccel_clear_tag(skb);
 794
 795         /* Reconstruct the MAC header.  */
 796         skb_push(skb, data->l2_len);
 797         memcpy(skb->data, &data->l2_data, data->l2_len);
 798         skb_postpush_rcsum(skb, skb->data, data->l2_len);
 799         skb_reset_mac_header(skb);
 800
 801         if (eth_p_mpls(skb->protocol)) {
 802                 skb->inner_network_header = skb->network_header;
 803                 skb_set_network_header(skb, data->network_offset);
 804                 skb_reset_mac_len(skb);
 805         }
 806
 807         ovs_vport_send(vport, skb, data->mac_proto);
 808         return 0;
 809 }
 810
 811 static unsigned int
 812 ovs_dst_get_mtu(const struct dst_entry *dst)
 813 {
 814         return dst->dev->mtu;
 815 }
 816
 817 static struct dst_ops ovs_dst_ops = {
 818         .family = AF_UNSPEC,
 819         .mtu = ovs_dst_get_mtu,
 820 };
 821
 822 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
 823  * ovs_vport_output(), which is called once per fragmented packet.
 824  */
 825 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
 826                          u16 orig_network_offset, u8 mac_proto)
 827 {
 828         unsigned int hlen = skb_network_offset(skb);
 829         struct ovs_frag_data *data;
 830
 831         data = this_cpu_ptr(&ovs_frag_data_storage);
 832         data->dst = skb->_skb_refdst;
 833         data->vport = vport;
 834         data->cb = *OVS_CB(skb);
 835         data->inner_protocol = skb->inner_protocol;
 836         data->network_offset = orig_network_offset;
 837         if (skb_vlan_tag_present(skb))
 838                 data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
 839         else
 840                 data->vlan_tci = 0;
 841         data->vlan_proto = skb->vlan_proto;
 842         data->mac_proto = mac_proto;
 843         data->l2_len = hlen;
 844         memcpy(&data->l2_data, skb->data, hlen);
 845
 846         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 847         skb_pull(skb, hlen);
 848 }
 849
 850 static void ovs_fragment(struct net *net, struct vport *vport,
 851                          struct sk_buff *skb, u16 mru,
 852                          struct sw_flow_key *key)
 853 {
 854         u16 orig_network_offset = 0;
 855
 856         if (eth_p_mpls(skb->protocol)) {
 857                 orig_network_offset = skb_network_offset(skb);
 858                 skb->network_header = skb->inner_network_header;
 859         }
 860
 861         if (skb_network_offset(skb) > MAX_L2_LEN) {
 862                 OVS_NLERR(1, "L2 header too long to fragment");
 863                 goto err;
 864         }
 865
 866         if (key->eth.type == htons(ETH_P_IP)) {
 867                 struct rtable ovs_rt = { 0 };
 868                 unsigned long orig_dst;
 869
 870                 prepare_frag(vport, skb, orig_network_offset,
 871                              ovs_key_mac_proto(key));
 872                 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
 873                          DST_OBSOLETE_NONE, DST_NOCOUNT);
 874                 ovs_rt.dst.dev = vport->dev;
 875
 876                 orig_dst = skb->_skb_refdst;
 877                 skb_dst_set_noref(skb, &ovs_rt.dst);
 878                 IPCB(skb)->frag_max_size = mru;
 879
 880                 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
 881                 refdst_drop(orig_dst);
 882         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 883                 unsigned long orig_dst;
 884                 struct rt6_info ovs_rt;
 885
 886                 prepare_frag(vport, skb, orig_network_offset,
 887                              ovs_key_mac_proto(key));
 888                 memset(&ovs_rt, 0, sizeof(ovs_rt));
 889                 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
 890                          DST_OBSOLETE_NONE, DST_NOCOUNT);
 891                 ovs_rt.dst.dev = vport->dev;
 892
 893                 orig_dst = skb->_skb_refdst;
 894                 skb_dst_set_noref(skb, &ovs_rt.dst);
 895                 IP6CB(skb)->frag_max_size = mru;
 896
 897                 ipv6_stub->ipv6_fragment(net, skb->sk, skb, ovs_vport_output);
 898                 refdst_drop(orig_dst);
 899         } else {
 900                 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
 901                           ovs_vport_name(vport), ntohs(key->eth.type), mru,
 902                           vport->dev->mtu);
 903                 goto err;
 904         }
 905
 906         return;
 907 err:
 908         kfree_skb(skb);
 909 }
 910
 911 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
 912                       struct sw_flow_key *key)
 913 {
 914         struct vport *vport = ovs_vport_rcu(dp, out_port);
 915
 916         if (likely(vport)) {
 917                 u16 mru = OVS_CB(skb)->mru;
 918                 u32 cutlen = OVS_CB(skb)->cutlen;
 919
 920                 if (unlikely(cutlen > 0)) {
 921                         if (skb->len - cutlen > ovs_mac_header_len(key))
 922                                 pskb_trim(skb, skb->len - cutlen);
 923                         else
 924                                 pskb_trim(skb, ovs_mac_header_len(key));
 925                 }
 926
 927                 if (likely(!mru ||
 928                            (skb->len <= mru + vport->dev->hard_header_len))) {
 929                         ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
 930                 } else if (mru <= vport->dev->mtu) {
 931                         struct net *net = read_pnet(&dp->net);
 932
 933                         ovs_fragment(net, vport, skb, mru, key);
 934                 } else {
 935                         kfree_skb(skb);
 936                 }
 937         } else {
 938                 kfree_skb(skb);
 939         }
 940 }
 941
 942 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 943                             struct sw_flow_key *key, const struct nlattr *attr,
 944                             const struct nlattr *actions, int actions_len,
 945                             uint32_t cutlen)
 946 {
 947         struct dp_upcall_info upcall;
 948         const struct nlattr *a;
 949         int rem;
 950
 951         memset(&upcall, 0, sizeof(upcall));
 952         upcall.cmd = OVS_PACKET_CMD_ACTION;
 953         upcall.mru = OVS_CB(skb)->mru;
 954
 955         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 956              a = nla_next(a, &rem)) {
 957                 switch (nla_type(a)) {
 958                 case OVS_USERSPACE_ATTR_USERDATA:
 959                         upcall.userdata = a;
 960                         break;
 961
 962                 case OVS_USERSPACE_ATTR_PID:
 963                         if (dp->user_features &
 964                             OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
 965                                 upcall.portid =
 966                                   ovs_dp_get_upcall_portid(dp,
 967                                                            smp_processor_id());
 968                         else
 969                                 upcall.portid = nla_get_u32(a);
 970                         break;
 971
 972                 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
 973                         /* Get out tunnel info. */
 974                         struct vport *vport;
 975
 976                         vport = ovs_vport_rcu(dp, nla_get_u32(a));
 977                         if (vport) {
 978                                 int err;
 979
 980                                 err = dev_fill_metadata_dst(vport->dev, skb);
 981                                 if (!err)
 982                                         upcall.egress_tun_info = skb_tunnel_info(skb);
 983                         }
 984
 985                         break;
 986                 }
 987
 988                 case OVS_USERSPACE_ATTR_ACTIONS: {
 989                         /* Include actions. */
 990                         upcall.actions = actions;
 991                         upcall.actions_len = actions_len;
 992                         break;
 993                 }
 994
 995                 } /* End of switch. */
 996         }
 997
 998         return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
 999 }
1000
1001 static int dec_ttl_exception_handler(struct datapath *dp, struct sk_buff *skb,
1002                                      struct sw_flow_key *key,
1003                                      const struct nlattr *attr)
1004 {
1005         /* The first attribute is always 'OVS_DEC_TTL_ATTR_ACTION'. */
1006         struct nlattr *actions = nla_data(attr);
1007
1008         if (nla_len(actions))
1009                 return clone_execute(dp, skb, key, 0, nla_data(actions),
1010                                      nla_len(actions), true, false);
1011
1012         consume_skb(skb);
1013         return 0;
1014 }
1015
1016 /* When 'last' is true, sample() should always consume the 'skb'.
1017  * Otherwise, sample() should keep 'skb' intact regardless what
1018  * actions are executed within sample().
1019  */
1020 static int sample(struct datapath *dp, struct sk_buff *skb,
1021                   struct sw_flow_key *key, const struct nlattr *attr,
1022                   bool last)
1023 {
1024         struct nlattr *actions;
1025         struct nlattr *sample_arg;
1026         int rem = nla_len(attr);
1027         const struct sample_arg *arg;
1028         bool clone_flow_key;
1029
1030         /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
1031         sample_arg = nla_data(attr);
1032         arg = nla_data(sample_arg);
1033         actions = nla_next(sample_arg, &rem);
1034
1035         if ((arg->probability != U32_MAX) &&
1036             (!arg->probability || prandom_u32() > arg->probability)) {
1037                 if (last)
1038                         consume_skb(skb);
1039                 return 0;
1040         }
1041
1042         clone_flow_key = !arg->exec;
1043         return clone_execute(dp, skb, key, 0, actions, rem, last,
1044                              clone_flow_key);
1045 }
1046
1047 /* When 'last' is true, clone() should always consume the 'skb'.
1048  * Otherwise, clone() should keep 'skb' intact regardless what
1049  * actions are executed within clone().
1050  */
1051 static int clone(struct datapath *dp, struct sk_buff *skb,
1052                  struct sw_flow_key *key, const struct nlattr *attr,
1053                  bool last)
1054 {
1055         struct nlattr *actions;
1056         struct nlattr *clone_arg;
1057         int rem = nla_len(attr);
1058         bool dont_clone_flow_key;
1059
1060         /* The first action is always 'OVS_CLONE_ATTR_EXEC'. */
1061         clone_arg = nla_data(attr);
1062         dont_clone_flow_key = nla_get_u32(clone_arg);
1063         actions = nla_next(clone_arg, &rem);
1064
1065         return clone_execute(dp, skb, key, 0, actions, rem, last,
1066                              !dont_clone_flow_key);
1067 }
1068
1069 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1070                          const struct nlattr *attr)
1071 {
1072         struct ovs_action_hash *hash_act = nla_data(attr);
1073         u32 hash = 0;
1074
1075         /* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
1076         hash = skb_get_hash(skb);
1077         hash = jhash_1word(hash, hash_act->hash_basis);
1078         if (!hash)
1079                 hash = 0x1;
1080
1081         key->ovs_flow_hash = hash;
1082 }
1083
1084 static int execute_set_action(struct sk_buff *skb,
1085                               struct sw_flow_key *flow_key,
1086                               const struct nlattr *a)
1087 {
1088         /* Only tunnel set execution is supported without a mask. */
1089         if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1090                 struct ovs_tunnel_info *tun = nla_data(a);
1091
1092                 skb_dst_drop(skb);
1093                 dst_hold((struct dst_entry *)tun->tun_dst);
1094                 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1095                 return 0;
1096         }
1097
1098         return -EINVAL;
1099 }
1100
1101 /* Mask is at the midpoint of the data. */
1102 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1103
1104 static int execute_masked_set_action(struct sk_buff *skb,
1105                                      struct sw_flow_key *flow_key,
1106                                      const struct nlattr *a)
1107 {
1108         int err = 0;
1109
1110         switch (nla_type(a)) {
1111         case OVS_KEY_ATTR_PRIORITY:
1112                 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1113                                *get_mask(a, u32 *));
1114                 flow_key->phy.priority = skb->priority;
1115                 break;
1116
1117         case OVS_KEY_ATTR_SKB_MARK:
1118                 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1119                 flow_key->phy.skb_mark = skb->mark;
1120                 break;
1121
1122         case OVS_KEY_ATTR_TUNNEL_INFO:
1123                 /* Masked data not supported for tunnel. */
1124                 err = -EINVAL;
1125                 break;
1126
1127         case OVS_KEY_ATTR_ETHERNET:
1128                 err = set_eth_addr(skb, flow_key, nla_data(a),
1129                                    get_mask(a, struct ovs_key_ethernet *));
1130                 break;
1131
1132         case OVS_KEY_ATTR_NSH:
1133                 err = set_nsh(skb, flow_key, a);
1134                 break;
1135
1136         case OVS_KEY_ATTR_IPV4:
1137                 err = set_ipv4(skb, flow_key, nla_data(a),
1138                                get_mask(a, struct ovs_key_ipv4 *));
1139                 break;
1140
1141         case OVS_KEY_ATTR_IPV6:
1142                 err = set_ipv6(skb, flow_key, nla_data(a),
1143                                get_mask(a, struct ovs_key_ipv6 *));
1144                 break;
1145
1146         case OVS_KEY_ATTR_TCP:
1147                 err = set_tcp(skb, flow_key, nla_data(a),
1148                               get_mask(a, struct ovs_key_tcp *));
1149                 break;
1150
1151         case OVS_KEY_ATTR_UDP:
1152                 err = set_udp(skb, flow_key, nla_data(a),
1153                               get_mask(a, struct ovs_key_udp *));
1154                 break;
1155
1156         case OVS_KEY_ATTR_SCTP:
1157                 err = set_sctp(skb, flow_key, nla_data(a),
1158                                get_mask(a, struct ovs_key_sctp *));
1159                 break;
1160
1161         case OVS_KEY_ATTR_MPLS:
1162                 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1163                                                                     __be32 *));
1164                 break;
1165
1166         case OVS_KEY_ATTR_CT_STATE:
1167         case OVS_KEY_ATTR_CT_ZONE:
1168         case OVS_KEY_ATTR_CT_MARK:
1169         case OVS_KEY_ATTR_CT_LABELS:
1170         case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1171         case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1172                 err = -EINVAL;
1173                 break;
1174         }
1175
1176         return err;
1177 }
1178
1179 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1180                           struct sw_flow_key *key,
1181                           const struct nlattr *a, bool last)
1182 {
1183         u32 recirc_id;
1184
1185         if (!is_flow_key_valid(key)) {
1186                 int err;
1187
1188                 err = ovs_flow_key_update(skb, key);
1189                 if (err)
1190                         return err;
1191         }
1192         BUG_ON(!is_flow_key_valid(key));
1193
1194         recirc_id = nla_get_u32(a);
1195         return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1196 }
1197
1198 static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1199                                  struct sw_flow_key *key,
1200                                  const struct nlattr *attr, bool last)
1201 {
1202         struct ovs_skb_cb *ovs_cb = OVS_CB(skb);
1203         const struct nlattr *actions, *cpl_arg;
1204         int len, max_len, rem = nla_len(attr);
1205         const struct check_pkt_len_arg *arg;
1206         bool clone_flow_key;
1207
1208         /* The first netlink attribute in 'attr' is always
1209          * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1210          */
1211         cpl_arg = nla_data(attr);
1212         arg = nla_data(cpl_arg);
1213
1214         len = ovs_cb->mru ? ovs_cb->mru + skb->mac_len : skb->len;
1215         max_len = arg->pkt_len;
1216
1217         if ((skb_is_gso(skb) && skb_gso_validate_mac_len(skb, max_len)) ||
1218             len <= max_len) {
1219                 /* Second netlink attribute in 'attr' is always
1220                  * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1221                  */
1222                 actions = nla_next(cpl_arg, &rem);
1223                 clone_flow_key = !arg->exec_for_lesser_equal;
1224         } else {
1225                 /* Third netlink attribute in 'attr' is always
1226                  * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1227                  */
1228                 actions = nla_next(cpl_arg, &rem);
1229                 actions = nla_next(actions, &rem);
1230                 clone_flow_key = !arg->exec_for_greater;
1231         }
1232
1233         return clone_execute(dp, skb, key, 0, nla_data(actions),
1234                              nla_len(actions), last, clone_flow_key);
1235 }
1236
1237 static int execute_dec_ttl(struct sk_buff *skb, struct sw_flow_key *key)
1238 {
1239         int err;
1240
1241         if (skb->protocol == htons(ETH_P_IPV6)) {
1242                 struct ipv6hdr *nh;
1243
1244                 err = skb_ensure_writable(skb, skb_network_offset(skb) +
1245                                           sizeof(*nh));
1246                 if (unlikely(err))
1247                         return err;
1248
1249                 nh = ipv6_hdr(skb);
1250
1251                 if (nh->hop_limit <= 1)
1252                         return -EHOSTUNREACH;
1253
1254                 key->ip.ttl = --nh->hop_limit;
1255         } else if (skb->protocol == htons(ETH_P_IP)) {
1256                 struct iphdr *nh;
1257                 u8 old_ttl;
1258
1259                 err = skb_ensure_writable(skb, skb_network_offset(skb) +
1260                                           sizeof(*nh));
1261                 if (unlikely(err))
1262                         return err;
1263
1264                 nh = ip_hdr(skb);
1265                 if (nh->ttl <= 1)
1266                         return -EHOSTUNREACH;
1267
1268                 old_ttl = nh->ttl--;
1269                 csum_replace2(&nh->check, htons(old_ttl << 8),
1270                               htons(nh->ttl << 8));
1271                 key->ip.ttl = nh->ttl;
1272         }
1273         return 0;
1274 }
1275
1276 /* Execute a list of actions against 'skb'. */
1277 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1278                               struct sw_flow_key *key,
1279                               const struct nlattr *attr, int len)
1280 {
1281         const struct nlattr *a;
1282         int rem;
1283
1284         for (a = attr, rem = len; rem > 0;
1285              a = nla_next(a, &rem)) {
1286                 int err = 0;
1287
1288                 if (trace_ovs_do_execute_action_enabled())
1289                         trace_ovs_do_execute_action(dp, skb, key, a, rem);
1290
1291                 switch (nla_type(a)) {
1292                 case OVS_ACTION_ATTR_OUTPUT: {
1293                         int port = nla_get_u32(a);
1294                         struct sk_buff *clone;
1295
1296                         /* Every output action needs a separate clone
1297                          * of 'skb', In case the output action is the
1298                          * last action, cloning can be avoided.
1299                          */
1300                         if (nla_is_last(a, rem)) {
1301                                 do_output(dp, skb, port, key);
1302                                 /* 'skb' has been used for output.
1303                                  */
1304                                 return 0;
1305                         }
1306
1307                         clone = skb_clone(skb, GFP_ATOMIC);
1308                         if (clone)
1309                                 do_output(dp, clone, port, key);
1310                         OVS_CB(skb)->cutlen = 0;
1311                         break;
1312                 }
1313
1314                 case OVS_ACTION_ATTR_TRUNC: {
1315                         struct ovs_action_trunc *trunc = nla_data(a);
1316
1317                         if (skb->len > trunc->max_len)
1318                                 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1319                         break;
1320                 }
1321
1322                 case OVS_ACTION_ATTR_USERSPACE:
1323                         output_userspace(dp, skb, key, a, attr,
1324                                                      len, OVS_CB(skb)->cutlen);
1325                         OVS_CB(skb)->cutlen = 0;
1326                         break;
1327
1328                 case OVS_ACTION_ATTR_HASH:
1329                         execute_hash(skb, key, a);
1330                         break;
1331
1332                 case OVS_ACTION_ATTR_PUSH_MPLS: {
1333                         struct ovs_action_push_mpls *mpls = nla_data(a);
1334
1335                         err = push_mpls(skb, key, mpls->mpls_lse,
1336                                         mpls->mpls_ethertype, skb->mac_len);
1337                         break;
1338                 }
1339                 case OVS_ACTION_ATTR_ADD_MPLS: {
1340                         struct ovs_action_add_mpls *mpls = nla_data(a);
1341                         __u16 mac_len = 0;
1342
1343                         if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK)
1344                                 mac_len = skb->mac_len;
1345
1346                         err = push_mpls(skb, key, mpls->mpls_lse,
1347                                         mpls->mpls_ethertype, mac_len);
1348                         break;
1349                 }
1350                 case OVS_ACTION_ATTR_POP_MPLS:
1351                         err = pop_mpls(skb, key, nla_get_be16(a));
1352                         break;
1353
1354                 case OVS_ACTION_ATTR_PUSH_VLAN:
1355                         err = push_vlan(skb, key, nla_data(a));
1356                         break;
1357
1358                 case OVS_ACTION_ATTR_POP_VLAN:
1359                         err = pop_vlan(skb, key);
1360                         break;
1361
1362                 case OVS_ACTION_ATTR_RECIRC: {
1363                         bool last = nla_is_last(a, rem);
1364
1365                         err = execute_recirc(dp, skb, key, a, last);
1366                         if (last) {
1367                                 /* If this is the last action, the skb has
1368                                  * been consumed or freed.
1369                                  * Return immediately.
1370                                  */
1371                                 return err;
1372                         }
1373                         break;
1374                 }
1375
1376                 case OVS_ACTION_ATTR_SET:
1377                         err = execute_set_action(skb, key, nla_data(a));
1378                         break;
1379
1380                 case OVS_ACTION_ATTR_SET_MASKED:
1381                 case OVS_ACTION_ATTR_SET_TO_MASKED:
1382                         err = execute_masked_set_action(skb, key, nla_data(a));
1383                         break;
1384
1385                 case OVS_ACTION_ATTR_SAMPLE: {
1386                         bool last = nla_is_last(a, rem);
1387
1388                         err = sample(dp, skb, key, a, last);
1389                         if (last)
1390                                 return err;
1391
1392                         break;
1393                 }
1394
1395                 case OVS_ACTION_ATTR_CT:
1396                         if (!is_flow_key_valid(key)) {
1397                                 err = ovs_flow_key_update(skb, key);
1398                                 if (err)
1399                                         return err;
1400                         }
1401
1402                         err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1403                                              nla_data(a));
1404
1405                         /* Hide stolen IP fragments from user space. */
1406                         if (err)
1407                                 return err == -EINPROGRESS ? 0 : err;
1408                         break;
1409
1410                 case OVS_ACTION_ATTR_CT_CLEAR:
1411                         err = ovs_ct_clear(skb, key);
1412                         break;
1413
1414                 case OVS_ACTION_ATTR_PUSH_ETH:
1415                         err = push_eth(skb, key, nla_data(a));
1416                         break;
1417
1418                 case OVS_ACTION_ATTR_POP_ETH:
1419                         err = pop_eth(skb, key);
1420                         break;
1421
1422                 case OVS_ACTION_ATTR_PUSH_NSH: {
1423                         u8 buffer[NSH_HDR_MAX_LEN];
1424                         struct nshhdr *nh = (struct nshhdr *)buffer;
1425
1426                         err = nsh_hdr_from_nlattr(nla_data(a), nh,
1427                                                   NSH_HDR_MAX_LEN);
1428                         if (unlikely(err))
1429                                 break;
1430                         err = push_nsh(skb, key, nh);
1431                         break;
1432                 }
1433
1434                 case OVS_ACTION_ATTR_POP_NSH:
1435                         err = pop_nsh(skb, key);
1436                         break;
1437
1438                 case OVS_ACTION_ATTR_METER:
1439                         if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1440                                 consume_skb(skb);
1441                                 return 0;
1442                         }
1443                         break;
1444
1445                 case OVS_ACTION_ATTR_CLONE: {
1446                         bool last = nla_is_last(a, rem);
1447
1448                         err = clone(dp, skb, key, a, last);
1449                         if (last)
1450                                 return err;
1451
1452                         break;
1453                 }
1454
1455                 case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1456                         bool last = nla_is_last(a, rem);
1457
1458                         err = execute_check_pkt_len(dp, skb, key, a, last);
1459                         if (last)
1460                                 return err;
1461
1462                         break;
1463                 }
1464
1465                 case OVS_ACTION_ATTR_DEC_TTL:
1466                         err = execute_dec_ttl(skb, key);
1467                         if (err == -EHOSTUNREACH)
1468                                 return dec_ttl_exception_handler(dp, skb,
1469                                                                  key, a);
1470                         break;
1471                 }
1472
1473                 if (unlikely(err)) {
1474                         kfree_skb(skb);
1475                         return err;
1476                 }
1477         }
1478
1479         consume_skb(skb);
1480         return 0;
1481 }
1482
1483 /* Execute the actions on the clone of the packet. The effect of the
1484  * execution does not affect the original 'skb' nor the original 'key'.
1485  *
1486  * The execution may be deferred in case the actions can not be executed
1487  * immediately.
1488  */
1489 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1490                          struct sw_flow_key *key, u32 recirc_id,
1491                          const struct nlattr *actions, int len,
1492                          bool last, bool clone_flow_key)
1493 {
1494         struct deferred_action *da;
1495         struct sw_flow_key *clone;
1496
1497         skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1498         if (!skb) {
1499                 /* Out of memory, skip this action.
1500                  */
1501                 return 0;
1502         }
1503
1504         /* When clone_flow_key is false, the 'key' will not be change
1505          * by the actions, then the 'key' can be used directly.
1506          * Otherwise, try to clone key from the next recursion level of
1507          * 'flow_keys'. If clone is successful, execute the actions
1508          * without deferring.
1509          */
1510         clone = clone_flow_key ? clone_key(key) : key;
1511         if (clone) {
1512                 int err = 0;
1513
1514                 if (actions) { /* Sample action */
1515                         if (clone_flow_key)
1516                                 __this_cpu_inc(exec_actions_level);
1517
1518                         err = do_execute_actions(dp, skb, clone,
1519                                                  actions, len);
1520
1521                         if (clone_flow_key)
1522                                 __this_cpu_dec(exec_actions_level);
1523                 } else { /* Recirc action */
1524                         clone->recirc_id = recirc_id;
1525                         ovs_dp_process_packet(skb, clone);
1526                 }
1527                 return err;
1528         }
1529
1530         /* Out of 'flow_keys' space. Defer actions */
1531         da = add_deferred_actions(skb, key, actions, len);
1532         if (da) {
1533                 if (!actions) { /* Recirc action */
1534                         key = &da->pkt_key;
1535                         key->recirc_id = recirc_id;
1536                 }
1537         } else {
1538                 /* Out of per CPU action FIFO space. Drop the 'skb' and
1539                  * log an error.
1540                  */
1541                 kfree_skb(skb);
1542
1543                 if (net_ratelimit()) {
1544                         if (actions) { /* Sample action */
1545                                 pr_warn("%s: deferred action limit reached, drop sample action\n",
1546                                         ovs_dp_name(dp));
1547                         } else {  /* Recirc action */
1548                                 pr_warn("%s: deferred action limit reached, drop recirc action (recirc_id=%#x)\n",
1549                                         ovs_dp_name(dp), recirc_id);
1550                         }
1551                 }
1552         }
1553         return 0;
1554 }
1555
1556 static void process_deferred_actions(struct datapath *dp)
1557 {
1558         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1559
1560         /* Do not touch the FIFO in case there is no deferred actions. */
1561         if (action_fifo_is_empty(fifo))
1562                 return;
1563
1564         /* Finishing executing all deferred actions. */
1565         do {
1566                 struct deferred_action *da = action_fifo_get(fifo);
1567                 struct sk_buff *skb = da->skb;
1568                 struct sw_flow_key *key = &da->pkt_key;
1569                 const struct nlattr *actions = da->actions;
1570                 int actions_len = da->actions_len;
1571
1572                 if (actions)
1573                         do_execute_actions(dp, skb, key, actions, actions_len);
1574                 else
1575                         ovs_dp_process_packet(skb, key);
1576         } while (!action_fifo_is_empty(fifo));
1577
1578         /* Reset FIFO for the next packet.  */
1579         action_fifo_init(fifo);
1580 }
1581
1582 /* Execute a list of actions against 'skb'. */
1583 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1584                         const struct sw_flow_actions *acts,
1585                         struct sw_flow_key *key)
1586 {
1587         int err, level;
1588
1589         level = __this_cpu_inc_return(exec_actions_level);
1590         if (unlikely(level > OVS_RECURSION_LIMIT)) {
1591                 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1592                                      ovs_dp_name(dp));
1593                 kfree_skb(skb);
1594                 err = -ENETDOWN;
1595                 goto out;
1596         }
1597
1598         OVS_CB(skb)->acts_origlen = acts->orig_len;
1599         err = do_execute_actions(dp, skb, key,
1600                                  acts->actions, acts->actions_len);
1601
1602         if (level == 1)
1603                 process_deferred_actions(dp);
1604
1605 out:
1606         __this_cpu_dec(exec_actions_level);
1607         return err;
1608 }
1609
1610 int action_fifos_init(void)
1611 {
1612         action_fifos = alloc_percpu(struct action_fifo);
1613         if (!action_fifos)
1614                 return -ENOMEM;
1615
1616         flow_keys = alloc_percpu(struct action_flow_keys);
1617         if (!flow_keys) {
1618                 free_percpu(action_fifos);
1619                 return -ENOMEM;
1620         }
1621
1622         return 0;
1623 }
1624
1625 void action_fifos_exit(void)
1626 {
1627         free_percpu(action_fifos);
1628         free_percpu(flow_keys);
1629 }