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 "drop.h"
  31 #include "flow.h"
  32 #include "conntrack.h"
  33 #include "vport.h"
  34 #include "flow_netlink.h"
  35 #include "openvswitch_trace.h"
  36
  37 struct deferred_action {
  38         struct sk_buff *skb;
  39         const struct nlattr *actions;
  40         int actions_len;
  41
  42         /* Store pkt_key clone when creating deferred action. */
  43         struct sw_flow_key pkt_key;
  44 };
  45
  46 #define MAX_L2_LEN      (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
  47 struct ovs_frag_data {
  48         unsigned long dst;
  49         struct vport *vport;
  50         struct ovs_skb_cb cb;
  51         __be16 inner_protocol;
  52         u16 network_offset;     /* valid only for MPLS */
  53         u16 vlan_tci;
  54         __be16 vlan_proto;
  55         unsigned int l2_len;
  56         u8 mac_proto;
  57         u8 l2_data[MAX_L2_LEN];
  58 };
  59
  60 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
  61
  62 #define DEFERRED_ACTION_FIFO_SIZE 10
  63 #define OVS_RECURSION_LIMIT 5
  64 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
  65 struct action_fifo {
  66         int head;
  67         int tail;
  68         /* Deferred action fifo queue storage. */
  69         struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
  70 };
  71
  72 struct action_flow_keys {
  73         struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
  74 };
  75
  76 static struct action_fifo __percpu *action_fifos;
  77 static struct action_flow_keys __percpu *flow_keys;
  78 static DEFINE_PER_CPU(int, exec_actions_level);
  79
  80 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
  81  * space. Return NULL if out of key spaces.
  82  */
  83 static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
  84 {
  85         struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
  86         int level = this_cpu_read(exec_actions_level);
  87         struct sw_flow_key *key = NULL;
  88
  89         if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
  90                 key = &keys->key[level - 1];
  91                 *key = *key_;
  92         }
  93
  94         return key;
  95 }
  96
  97 static void action_fifo_init(struct action_fifo *fifo)
  98 {
  99         fifo->head = 0;
 100         fifo->tail = 0;
 101 }
 102
 103 static bool action_fifo_is_empty(const struct action_fifo *fifo)
 104 {
 105         return (fifo->head == fifo->tail);
 106 }
 107
 108 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
 109 {
 110         if (action_fifo_is_empty(fifo))
 111                 return NULL;
 112
 113         return &fifo->fifo[fifo->tail++];
 114 }
 115
 116 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
 117 {
 118         if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
 119                 return NULL;
 120
 121         return &fifo->fifo[fifo->head++];
 122 }
 123
 124 /* Return true if fifo is not full */
 125 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
 126                                     const struct sw_flow_key *key,
 127                                     const struct nlattr *actions,
 128                                     const int actions_len)
 129 {
 130         struct action_fifo *fifo;
 131         struct deferred_action *da;
 132
 133         fifo = this_cpu_ptr(action_fifos);
 134         da = action_fifo_put(fifo);
 135         if (da) {
 136                 da->skb = skb;
 137                 da->actions = actions;
 138                 da->actions_len = actions_len;
 139                 da->pkt_key = *key;
 140         }
 141
 142         return da;
 143 }
 144
 145 static void invalidate_flow_key(struct sw_flow_key *key)
 146 {
 147         key->mac_proto |= SW_FLOW_KEY_INVALID;
 148 }
 149
 150 static bool is_flow_key_valid(const struct sw_flow_key *key)
 151 {
 152         return !(key->mac_proto & SW_FLOW_KEY_INVALID);
 153 }
 154
 155 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
 156                          struct sw_flow_key *key,
 157                          u32 recirc_id,
 158                          const struct nlattr *actions, int len,
 159                          bool last, bool clone_flow_key);
 160
 161 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 162                               struct sw_flow_key *key,
 163                               const struct nlattr *attr, int len);
 164
 165 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 166                      __be32 mpls_lse, __be16 mpls_ethertype, __u16 mac_len)
 167 {
 168         int err;
 169
 170         err = skb_mpls_push(skb, mpls_lse, mpls_ethertype, mac_len, !!mac_len);
 171         if (err)
 172                 return err;
 173
 174         if (!mac_len)
 175                 key->mac_proto = MAC_PROTO_NONE;
 176
 177         invalidate_flow_key(key);
 178         return 0;
 179 }
 180
 181 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 182                     const __be16 ethertype)
 183 {
 184         int err;
 185
 186         err = skb_mpls_pop(skb, ethertype, skb->mac_len,
 187                            ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET);
 188         if (err)
 189                 return err;
 190
 191         if (ethertype == htons(ETH_P_TEB))
 192                 key->mac_proto = MAC_PROTO_ETHERNET;
 193
 194         invalidate_flow_key(key);
 195         return 0;
 196 }
 197
 198 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
 199                     const __be32 *mpls_lse, const __be32 *mask)
 200 {
 201         struct mpls_shim_hdr *stack;
 202         __be32 lse;
 203         int err;
 204
 205         if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
 206                 return -ENOMEM;
 207
 208         stack = mpls_hdr(skb);
 209         lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
 210         err = skb_mpls_update_lse(skb, lse);
 211         if (err)
 212                 return err;
 213
 214         flow_key->mpls.lse[0] = lse;
 215         return 0;
 216 }
 217
 218 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 219 {
 220         int err;
 221
 222         err = skb_vlan_pop(skb);
 223         if (skb_vlan_tag_present(skb)) {
 224                 invalidate_flow_key(key);
 225         } else {
 226                 key->eth.vlan.tci = 0;
 227                 key->eth.vlan.tpid = 0;
 228         }
 229         return err;
 230 }
 231
 232 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
 233                      const struct ovs_action_push_vlan *vlan)
 234 {
 235         if (skb_vlan_tag_present(skb)) {
 236                 invalidate_flow_key(key);
 237         } else {
 238                 key->eth.vlan.tci = vlan->vlan_tci;
 239                 key->eth.vlan.tpid = vlan->vlan_tpid;
 240         }
 241         return skb_vlan_push(skb, vlan->vlan_tpid,
 242                              ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
 243 }
 244
 245 /* 'src' is already properly masked. */
 246 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
 247 {
 248         u16 *dst = (u16 *)dst_;
 249         const u16 *src = (const u16 *)src_;
 250         const u16 *mask = (const u16 *)mask_;
 251
 252         OVS_SET_MASKED(dst[0], src[0], mask[0]);
 253         OVS_SET_MASKED(dst[1], src[1], mask[1]);
 254         OVS_SET_MASKED(dst[2], src[2], mask[2]);
 255 }
 256
 257 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
 258                         const struct ovs_key_ethernet *key,
 259                         const struct ovs_key_ethernet *mask)
 260 {
 261         int err;
 262
 263         err = skb_ensure_writable(skb, ETH_HLEN);
 264         if (unlikely(err))
 265                 return err;
 266
 267         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 268
 269         ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
 270                                mask->eth_src);
 271         ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
 272                                mask->eth_dst);
 273
 274         skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 275
 276         ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
 277         ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
 278         return 0;
 279 }
 280
 281 /* pop_eth does not support VLAN packets as this action is never called
 282  * for them.
 283  */
 284 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
 285 {
 286         int err;
 287
 288         err = skb_eth_pop(skb);
 289         if (err)
 290                 return err;
 291
 292         /* safe right before invalidate_flow_key */
 293         key->mac_proto = MAC_PROTO_NONE;
 294         invalidate_flow_key(key);
 295         return 0;
 296 }
 297
 298 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
 299                     const struct ovs_action_push_eth *ethh)
 300 {
 301         int err;
 302
 303         err = skb_eth_push(skb, ethh->addresses.eth_dst,
 304                            ethh->addresses.eth_src);
 305         if (err)
 306                 return err;
 307
 308         /* safe right before invalidate_flow_key */
 309         key->mac_proto = MAC_PROTO_ETHERNET;
 310         invalidate_flow_key(key);
 311         return 0;
 312 }
 313
 314 static noinline_for_stack int push_nsh(struct sk_buff *skb,
 315                                        struct sw_flow_key *key,
 316                                        const struct nlattr *a)
 317 {
 318         u8 buffer[NSH_HDR_MAX_LEN];
 319         struct nshhdr *nh = (struct nshhdr *)buffer;
 320         int err;
 321
 322         err = nsh_hdr_from_nlattr(a, nh, NSH_HDR_MAX_LEN);
 323         if (err)
 324                 return err;
 325
 326         err = nsh_push(skb, nh);
 327         if (err)
 328                 return err;
 329
 330         /* safe right before invalidate_flow_key */
 331         key->mac_proto = MAC_PROTO_NONE;
 332         invalidate_flow_key(key);
 333         return 0;
 334 }
 335
 336 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
 337 {
 338         int err;
 339
 340         err = nsh_pop(skb);
 341         if (err)
 342                 return err;
 343
 344         /* safe right before invalidate_flow_key */
 345         if (skb->protocol == htons(ETH_P_TEB))
 346                 key->mac_proto = MAC_PROTO_ETHERNET;
 347         else
 348                 key->mac_proto = MAC_PROTO_NONE;
 349         invalidate_flow_key(key);
 350         return 0;
 351 }
 352
 353 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
 354                                   __be32 addr, __be32 new_addr)
 355 {
 356         int transport_len = skb->len - skb_transport_offset(skb);
 357
 358         if (nh->frag_off & htons(IP_OFFSET))
 359                 return;
 360
 361         if (nh->protocol == IPPROTO_TCP) {
 362                 if (likely(transport_len >= sizeof(struct tcphdr)))
 363                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 364                                                  addr, new_addr, true);
 365         } else if (nh->protocol == IPPROTO_UDP) {
 366                 if (likely(transport_len >= sizeof(struct udphdr))) {
 367                         struct udphdr *uh = udp_hdr(skb);
 368
 369                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 370                                 inet_proto_csum_replace4(&uh->check, skb,
 371                                                          addr, new_addr, true);
 372                                 if (!uh->check)
 373                                         uh->check = CSUM_MANGLED_0;
 374                         }
 375                 }
 376         }
 377 }
 378
 379 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 380                         __be32 *addr, __be32 new_addr)
 381 {
 382         update_ip_l4_checksum(skb, nh, *addr, new_addr);
 383         csum_replace4(&nh->check, *addr, new_addr);
 384         skb_clear_hash(skb);
 385         ovs_ct_clear(skb, NULL);
 386         *addr = new_addr;
 387 }
 388
 389 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 390                                  __be32 addr[4], const __be32 new_addr[4])
 391 {
 392         int transport_len = skb->len - skb_transport_offset(skb);
 393
 394         if (l4_proto == NEXTHDR_TCP) {
 395                 if (likely(transport_len >= sizeof(struct tcphdr)))
 396                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 397                                                   addr, new_addr, true);
 398         } else if (l4_proto == NEXTHDR_UDP) {
 399                 if (likely(transport_len >= sizeof(struct udphdr))) {
 400                         struct udphdr *uh = udp_hdr(skb);
 401
 402                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 403                                 inet_proto_csum_replace16(&uh->check, skb,
 404                                                           addr, new_addr, true);
 405                                 if (!uh->check)
 406                                         uh->check = CSUM_MANGLED_0;
 407                         }
 408                 }
 409         } else if (l4_proto == NEXTHDR_ICMP) {
 410                 if (likely(transport_len >= sizeof(struct icmp6hdr)))
 411                         inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 412                                                   skb, addr, new_addr, true);
 413         }
 414 }
 415
 416 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
 417                            const __be32 mask[4], __be32 masked[4])
 418 {
 419         masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
 420         masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
 421         masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
 422         masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
 423 }
 424
 425 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 426                           __be32 addr[4], const __be32 new_addr[4],
 427                           bool recalculate_csum)
 428 {
 429         if (recalculate_csum)
 430                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 431
 432         skb_clear_hash(skb);
 433         ovs_ct_clear(skb, NULL);
 434         memcpy(addr, new_addr, sizeof(__be32[4]));
 435 }
 436
 437 static void set_ipv6_dsfield(struct sk_buff *skb, struct ipv6hdr *nh, u8 ipv6_tclass, u8 mask)
 438 {
 439         u8 old_ipv6_tclass = ipv6_get_dsfield(nh);
 440
 441         ipv6_tclass = OVS_MASKED(old_ipv6_tclass, ipv6_tclass, mask);
 442
 443         if (skb->ip_summed == CHECKSUM_COMPLETE)
 444                 csum_replace(&skb->csum, (__force __wsum)(old_ipv6_tclass << 12),
 445                              (__force __wsum)(ipv6_tclass << 12));
 446
 447         ipv6_change_dsfield(nh, ~mask, ipv6_tclass);
 448 }
 449
 450 static void set_ipv6_fl(struct sk_buff *skb, struct ipv6hdr *nh, u32 fl, u32 mask)
 451 {
 452         u32 ofl;
 453
 454         ofl = nh->flow_lbl[0] << 16 |  nh->flow_lbl[1] << 8 |  nh->flow_lbl[2];
 455         fl = OVS_MASKED(ofl, fl, mask);
 456
 457         /* Bits 21-24 are always unmasked, so this retains their values. */
 458         nh->flow_lbl[0] = (u8)(fl >> 16);
 459         nh->flow_lbl[1] = (u8)(fl >> 8);
 460         nh->flow_lbl[2] = (u8)fl;
 461
 462         if (skb->ip_summed == CHECKSUM_COMPLETE)
 463                 csum_replace(&skb->csum, (__force __wsum)htonl(ofl), (__force __wsum)htonl(fl));
 464 }
 465
 466 static void set_ipv6_ttl(struct sk_buff *skb, struct ipv6hdr *nh, u8 new_ttl, u8 mask)
 467 {
 468         new_ttl = OVS_MASKED(nh->hop_limit, new_ttl, mask);
 469
 470         if (skb->ip_summed == CHECKSUM_COMPLETE)
 471                 csum_replace(&skb->csum, (__force __wsum)(nh->hop_limit << 8),
 472                              (__force __wsum)(new_ttl << 8));
 473         nh->hop_limit = new_ttl;
 474 }
 475
 476 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
 477                        u8 mask)
 478 {
 479         new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
 480
 481         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 482         nh->ttl = new_ttl;
 483 }
 484
 485 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
 486                     const struct ovs_key_ipv4 *key,
 487                     const struct ovs_key_ipv4 *mask)
 488 {
 489         struct iphdr *nh;
 490         __be32 new_addr;
 491         int err;
 492
 493         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 494                                   sizeof(struct iphdr));
 495         if (unlikely(err))
 496                 return err;
 497
 498         nh = ip_hdr(skb);
 499
 500         /* Setting an IP addresses is typically only a side effect of
 501          * matching on them in the current userspace implementation, so it
 502          * makes sense to check if the value actually changed.
 503          */
 504         if (mask->ipv4_src) {
 505                 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
 506
 507                 if (unlikely(new_addr != nh->saddr)) {
 508                         set_ip_addr(skb, nh, &nh->saddr, new_addr);
 509                         flow_key->ipv4.addr.src = new_addr;
 510                 }
 511         }
 512         if (mask->ipv4_dst) {
 513                 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
 514
 515                 if (unlikely(new_addr != nh->daddr)) {
 516                         set_ip_addr(skb, nh, &nh->daddr, new_addr);
 517                         flow_key->ipv4.addr.dst = new_addr;
 518                 }
 519         }
 520         if (mask->ipv4_tos) {
 521                 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
 522                 flow_key->ip.tos = nh->tos;
 523         }
 524         if (mask->ipv4_ttl) {
 525                 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
 526                 flow_key->ip.ttl = nh->ttl;
 527         }
 528
 529         return 0;
 530 }
 531
 532 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
 533 {
 534         return !!(addr[0] | addr[1] | addr[2] | addr[3]);
 535 }
 536
 537 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
 538                     const struct ovs_key_ipv6 *key,
 539                     const struct ovs_key_ipv6 *mask)
 540 {
 541         struct ipv6hdr *nh;
 542         int err;
 543
 544         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 545                                   sizeof(struct ipv6hdr));
 546         if (unlikely(err))
 547                 return err;
 548
 549         nh = ipv6_hdr(skb);
 550
 551         /* Setting an IP addresses is typically only a side effect of
 552          * matching on them in the current userspace implementation, so it
 553          * makes sense to check if the value actually changed.
 554          */
 555         if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
 556                 __be32 *saddr = (__be32 *)&nh->saddr;
 557                 __be32 masked[4];
 558
 559                 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
 560
 561                 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
 562                         set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
 563                                       true);
 564                         memcpy(&flow_key->ipv6.addr.src, masked,
 565                                sizeof(flow_key->ipv6.addr.src));
 566                 }
 567         }
 568         if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
 569                 unsigned int offset = 0;
 570                 int flags = IP6_FH_F_SKIP_RH;
 571                 bool recalc_csum = true;
 572                 __be32 *daddr = (__be32 *)&nh->daddr;
 573                 __be32 masked[4];
 574
 575                 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
 576
 577                 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
 578                         if (ipv6_ext_hdr(nh->nexthdr))
 579                                 recalc_csum = (ipv6_find_hdr(skb, &offset,
 580                                                              NEXTHDR_ROUTING,
 581                                                              NULL, &flags)
 582                                                != NEXTHDR_ROUTING);
 583
 584                         set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
 585                                       recalc_csum);
 586                         memcpy(&flow_key->ipv6.addr.dst, masked,
 587                                sizeof(flow_key->ipv6.addr.dst));
 588                 }
 589         }
 590         if (mask->ipv6_tclass) {
 591                 set_ipv6_dsfield(skb, nh, key->ipv6_tclass, mask->ipv6_tclass);
 592                 flow_key->ip.tos = ipv6_get_dsfield(nh);
 593         }
 594         if (mask->ipv6_label) {
 595                 set_ipv6_fl(skb, nh, ntohl(key->ipv6_label),
 596                             ntohl(mask->ipv6_label));
 597                 flow_key->ipv6.label =
 598                     *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 599         }
 600         if (mask->ipv6_hlimit) {
 601                 set_ipv6_ttl(skb, nh, key->ipv6_hlimit, mask->ipv6_hlimit);
 602                 flow_key->ip.ttl = nh->hop_limit;
 603         }
 604         return 0;
 605 }
 606
 607 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
 608                    const struct nlattr *a)
 609 {
 610         struct nshhdr *nh;
 611         size_t length;
 612         int err;
 613         u8 flags;
 614         u8 ttl;
 615         int i;
 616
 617         struct ovs_key_nsh key;
 618         struct ovs_key_nsh mask;
 619
 620         err = nsh_key_from_nlattr(a, &key, &mask);
 621         if (err)
 622                 return err;
 623
 624         /* Make sure the NSH base header is there */
 625         if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
 626                 return -ENOMEM;
 627
 628         nh = nsh_hdr(skb);
 629         length = nsh_hdr_len(nh);
 630
 631         /* Make sure the whole NSH header is there */
 632         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 633                                        length);
 634         if (unlikely(err))
 635                 return err;
 636
 637         nh = nsh_hdr(skb);
 638         skb_postpull_rcsum(skb, nh, length);
 639         flags = nsh_get_flags(nh);
 640         flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
 641         flow_key->nsh.base.flags = flags;
 642         ttl = nsh_get_ttl(nh);
 643         ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
 644         flow_key->nsh.base.ttl = ttl;
 645         nsh_set_flags_and_ttl(nh, flags, ttl);
 646         nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
 647                                   mask.base.path_hdr);
 648         flow_key->nsh.base.path_hdr = nh->path_hdr;
 649         switch (nh->mdtype) {
 650         case NSH_M_TYPE1:
 651                 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
 652                         nh->md1.context[i] =
 653                             OVS_MASKED(nh->md1.context[i], key.context[i],
 654                                        mask.context[i]);
 655                 }
 656                 memcpy(flow_key->nsh.context, nh->md1.context,
 657                        sizeof(nh->md1.context));
 658                 break;
 659         case NSH_M_TYPE2:
 660                 memset(flow_key->nsh.context, 0,
 661                        sizeof(flow_key->nsh.context));
 662                 break;
 663         default:
 664                 return -EINVAL;
 665         }
 666         skb_postpush_rcsum(skb, nh, length);
 667         return 0;
 668 }
 669
 670 /* Must follow skb_ensure_writable() since that can move the skb data. */
 671 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 672                         __be16 new_port, __sum16 *check)
 673 {
 674         ovs_ct_clear(skb, NULL);
 675         inet_proto_csum_replace2(check, skb, *port, new_port, false);
 676         *port = new_port;
 677 }
 678
 679 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 680                    const struct ovs_key_udp *key,
 681                    const struct ovs_key_udp *mask)
 682 {
 683         struct udphdr *uh;
 684         __be16 src, dst;
 685         int err;
 686
 687         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 688                                   sizeof(struct udphdr));
 689         if (unlikely(err))
 690                 return err;
 691
 692         uh = udp_hdr(skb);
 693         /* Either of the masks is non-zero, so do not bother checking them. */
 694         src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
 695         dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
 696
 697         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 698                 if (likely(src != uh->source)) {
 699                         set_tp_port(skb, &uh->source, src, &uh->check);
 700                         flow_key->tp.src = src;
 701                 }
 702                 if (likely(dst != uh->dest)) {
 703                         set_tp_port(skb, &uh->dest, dst, &uh->check);
 704                         flow_key->tp.dst = dst;
 705                 }
 706
 707                 if (unlikely(!uh->check))
 708                         uh->check = CSUM_MANGLED_0;
 709         } else {
 710                 uh->source = src;
 711                 uh->dest = dst;
 712                 flow_key->tp.src = src;
 713                 flow_key->tp.dst = dst;
 714                 ovs_ct_clear(skb, NULL);
 715         }
 716
 717         skb_clear_hash(skb);
 718
 719         return 0;
 720 }
 721
 722 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 723                    const struct ovs_key_tcp *key,
 724                    const struct ovs_key_tcp *mask)
 725 {
 726         struct tcphdr *th;
 727         __be16 src, dst;
 728         int err;
 729
 730         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 731                                   sizeof(struct tcphdr));
 732         if (unlikely(err))
 733                 return err;
 734
 735         th = tcp_hdr(skb);
 736         src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
 737         if (likely(src != th->source)) {
 738                 set_tp_port(skb, &th->source, src, &th->check);
 739                 flow_key->tp.src = src;
 740         }
 741         dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
 742         if (likely(dst != th->dest)) {
 743                 set_tp_port(skb, &th->dest, dst, &th->check);
 744                 flow_key->tp.dst = dst;
 745         }
 746         skb_clear_hash(skb);
 747
 748         return 0;
 749 }
 750
 751 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 752                     const struct ovs_key_sctp *key,
 753                     const struct ovs_key_sctp *mask)
 754 {
 755         unsigned int sctphoff = skb_transport_offset(skb);
 756         struct sctphdr *sh;
 757         __le32 old_correct_csum, new_csum, old_csum;
 758         int err;
 759
 760         err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
 761         if (unlikely(err))
 762                 return err;
 763
 764         sh = sctp_hdr(skb);
 765         old_csum = sh->checksum;
 766         old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 767
 768         sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
 769         sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
 770
 771         new_csum = sctp_compute_cksum(skb, sctphoff);
 772
 773         /* Carry any checksum errors through. */
 774         sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 775
 776         skb_clear_hash(skb);
 777         ovs_ct_clear(skb, NULL);
 778
 779         flow_key->tp.src = sh->source;
 780         flow_key->tp.dst = sh->dest;
 781
 782         return 0;
 783 }
 784
 785 static int ovs_vport_output(struct net *net, struct sock *sk,
 786                             struct sk_buff *skb)
 787 {
 788         struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
 789         struct vport *vport = data->vport;
 790
 791         if (skb_cow_head(skb, data->l2_len) < 0) {
 792                 kfree_skb_reason(skb, SKB_DROP_REASON_NOMEM);
 793                 return -ENOMEM;
 794         }
 795
 796         __skb_dst_copy(skb, data->dst);
 797         *OVS_CB(skb) = data->cb;
 798         skb->inner_protocol = data->inner_protocol;
 799         if (data->vlan_tci & VLAN_CFI_MASK)
 800                 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
 801         else
 802                 __vlan_hwaccel_clear_tag(skb);
 803
 804         /* Reconstruct the MAC header.  */
 805         skb_push(skb, data->l2_len);
 806         memcpy(skb->data, &data->l2_data, data->l2_len);
 807         skb_postpush_rcsum(skb, skb->data, data->l2_len);
 808         skb_reset_mac_header(skb);
 809
 810         if (eth_p_mpls(skb->protocol)) {
 811                 skb->inner_network_header = skb->network_header;
 812                 skb_set_network_header(skb, data->network_offset);
 813                 skb_reset_mac_len(skb);
 814         }
 815
 816         ovs_vport_send(vport, skb, data->mac_proto);
 817         return 0;
 818 }
 819
 820 static unsigned int
 821 ovs_dst_get_mtu(const struct dst_entry *dst)
 822 {
 823         return dst->dev->mtu;
 824 }
 825
 826 static struct dst_ops ovs_dst_ops = {
 827         .family = AF_UNSPEC,
 828         .mtu = ovs_dst_get_mtu,
 829 };
 830
 831 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
 832  * ovs_vport_output(), which is called once per fragmented packet.
 833  */
 834 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
 835                          u16 orig_network_offset, u8 mac_proto)
 836 {
 837         unsigned int hlen = skb_network_offset(skb);
 838         struct ovs_frag_data *data;
 839
 840         data = this_cpu_ptr(&ovs_frag_data_storage);
 841         data->dst = skb->_skb_refdst;
 842         data->vport = vport;
 843         data->cb = *OVS_CB(skb);
 844         data->inner_protocol = skb->inner_protocol;
 845         data->network_offset = orig_network_offset;
 846         if (skb_vlan_tag_present(skb))
 847                 data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
 848         else
 849                 data->vlan_tci = 0;
 850         data->vlan_proto = skb->vlan_proto;
 851         data->mac_proto = mac_proto;
 852         data->l2_len = hlen;
 853         memcpy(&data->l2_data, skb->data, hlen);
 854
 855         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 856         skb_pull(skb, hlen);
 857 }
 858
 859 static void ovs_fragment(struct net *net, struct vport *vport,
 860                          struct sk_buff *skb, u16 mru,
 861                          struct sw_flow_key *key)
 862 {
 863         enum ovs_drop_reason reason;
 864         u16 orig_network_offset = 0;
 865
 866         if (eth_p_mpls(skb->protocol)) {
 867                 orig_network_offset = skb_network_offset(skb);
 868                 skb->network_header = skb->inner_network_header;
 869         }
 870
 871         if (skb_network_offset(skb) > MAX_L2_LEN) {
 872                 OVS_NLERR(1, "L2 header too long to fragment");
 873                 reason = OVS_DROP_FRAG_L2_TOO_LONG;
 874                 goto err;
 875         }
 876
 877         if (key->eth.type == htons(ETH_P_IP)) {
 878                 struct rtable ovs_rt = { 0 };
 879                 unsigned long orig_dst;
 880
 881                 prepare_frag(vport, skb, orig_network_offset,
 882                              ovs_key_mac_proto(key));
 883                 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL,
 884                          DST_OBSOLETE_NONE, DST_NOCOUNT);
 885                 ovs_rt.dst.dev = vport->dev;
 886
 887                 orig_dst = skb->_skb_refdst;
 888                 skb_dst_set_noref(skb, &ovs_rt.dst);
 889                 IPCB(skb)->frag_max_size = mru;
 890
 891                 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
 892                 refdst_drop(orig_dst);
 893         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 894                 unsigned long orig_dst;
 895                 struct rt6_info ovs_rt;
 896
 897                 prepare_frag(vport, skb, orig_network_offset,
 898                              ovs_key_mac_proto(key));
 899                 memset(&ovs_rt, 0, sizeof(ovs_rt));
 900                 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL,
 901                          DST_OBSOLETE_NONE, DST_NOCOUNT);
 902                 ovs_rt.dst.dev = vport->dev;
 903
 904                 orig_dst = skb->_skb_refdst;
 905                 skb_dst_set_noref(skb, &ovs_rt.dst);
 906                 IP6CB(skb)->frag_max_size = mru;
 907
 908                 ipv6_stub->ipv6_fragment(net, skb->sk, skb, ovs_vport_output);
 909                 refdst_drop(orig_dst);
 910         } else {
 911                 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
 912                           ovs_vport_name(vport), ntohs(key->eth.type), mru,
 913                           vport->dev->mtu);
 914                 reason = OVS_DROP_FRAG_INVALID_PROTO;
 915                 goto err;
 916         }
 917
 918         return;
 919 err:
 920         ovs_kfree_skb_reason(skb, reason);
 921 }
 922
 923 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
 924                       struct sw_flow_key *key)
 925 {
 926         struct vport *vport = ovs_vport_rcu(dp, out_port);
 927
 928         if (likely(vport && netif_carrier_ok(vport->dev))) {
 929                 u16 mru = OVS_CB(skb)->mru;
 930                 u32 cutlen = OVS_CB(skb)->cutlen;
 931
 932                 if (unlikely(cutlen > 0)) {
 933                         if (skb->len - cutlen > ovs_mac_header_len(key))
 934                                 pskb_trim(skb, skb->len - cutlen);
 935                         else
 936                                 pskb_trim(skb, ovs_mac_header_len(key));
 937                 }
 938
 939                 if (likely(!mru ||
 940                            (skb->len <= mru + vport->dev->hard_header_len))) {
 941                         ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
 942                 } else if (mru <= vport->dev->mtu) {
 943                         struct net *net = read_pnet(&dp->net);
 944
 945                         ovs_fragment(net, vport, skb, mru, key);
 946                 } else {
 947                         kfree_skb_reason(skb, SKB_DROP_REASON_PKT_TOO_BIG);
 948                 }
 949         } else {
 950                 kfree_skb_reason(skb, SKB_DROP_REASON_DEV_READY);
 951         }
 952 }
 953
 954 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 955                             struct sw_flow_key *key, const struct nlattr *attr,
 956                             const struct nlattr *actions, int actions_len,
 957                             uint32_t cutlen)
 958 {
 959         struct dp_upcall_info upcall;
 960         const struct nlattr *a;
 961         int rem;
 962
 963         memset(&upcall, 0, sizeof(upcall));
 964         upcall.cmd = OVS_PACKET_CMD_ACTION;
 965         upcall.mru = OVS_CB(skb)->mru;
 966
 967         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 968              a = nla_next(a, &rem)) {
 969                 switch (nla_type(a)) {
 970                 case OVS_USERSPACE_ATTR_USERDATA:
 971                         upcall.userdata = a;
 972                         break;
 973
 974                 case OVS_USERSPACE_ATTR_PID:
 975                         if (dp->user_features &
 976                             OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
 977                                 upcall.portid =
 978                                   ovs_dp_get_upcall_portid(dp,
 979                                                            smp_processor_id());
 980                         else
 981                                 upcall.portid = nla_get_u32(a);
 982                         break;
 983
 984                 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
 985                         /* Get out tunnel info. */
 986                         struct vport *vport;
 987
 988                         vport = ovs_vport_rcu(dp, nla_get_u32(a));
 989                         if (vport) {
 990                                 int err;
 991
 992                                 err = dev_fill_metadata_dst(vport->dev, skb);
 993                                 if (!err)
 994                                         upcall.egress_tun_info = skb_tunnel_info(skb);
 995                         }
 996
 997                         break;
 998                 }
 999
1000                 case OVS_USERSPACE_ATTR_ACTIONS: {
1001                         /* Include actions. */
1002                         upcall.actions = actions;
1003                         upcall.actions_len = actions_len;
1004                         break;
1005                 }
1006
1007                 } /* End of switch. */
1008         }
1009
1010         return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
1011 }
1012
1013 static int dec_ttl_exception_handler(struct datapath *dp, struct sk_buff *skb,
1014                                      struct sw_flow_key *key,
1015                                      const struct nlattr *attr)
1016 {
1017         /* The first attribute is always 'OVS_DEC_TTL_ATTR_ACTION'. */
1018         struct nlattr *actions = nla_data(attr);
1019
1020         if (nla_len(actions))
1021                 return clone_execute(dp, skb, key, 0, nla_data(actions),
1022                                      nla_len(actions), true, false);
1023
1024         ovs_kfree_skb_reason(skb, OVS_DROP_IP_TTL);
1025         return 0;
1026 }
1027
1028 /* When 'last' is true, sample() should always consume the 'skb'.
1029  * Otherwise, sample() should keep 'skb' intact regardless what
1030  * actions are executed within sample().
1031  */
1032 static int sample(struct datapath *dp, struct sk_buff *skb,
1033                   struct sw_flow_key *key, const struct nlattr *attr,
1034                   bool last)
1035 {
1036         struct nlattr *actions;
1037         struct nlattr *sample_arg;
1038         int rem = nla_len(attr);
1039         const struct sample_arg *arg;
1040         bool clone_flow_key;
1041
1042         /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
1043         sample_arg = nla_data(attr);
1044         arg = nla_data(sample_arg);
1045         actions = nla_next(sample_arg, &rem);
1046
1047         if ((arg->probability != U32_MAX) &&
1048             (!arg->probability || get_random_u32() > arg->probability)) {
1049                 if (last)
1050                         ovs_kfree_skb_reason(skb, OVS_DROP_LAST_ACTION);
1051                 return 0;
1052         }
1053
1054         clone_flow_key = !arg->exec;
1055         return clone_execute(dp, skb, key, 0, actions, rem, last,
1056                              clone_flow_key);
1057 }
1058
1059 /* When 'last' is true, clone() should always consume the 'skb'.
1060  * Otherwise, clone() should keep 'skb' intact regardless what
1061  * actions are executed within clone().
1062  */
1063 static int clone(struct datapath *dp, struct sk_buff *skb,
1064                  struct sw_flow_key *key, const struct nlattr *attr,
1065                  bool last)
1066 {
1067         struct nlattr *actions;
1068         struct nlattr *clone_arg;
1069         int rem = nla_len(attr);
1070         bool dont_clone_flow_key;
1071
1072         /* The first action is always 'OVS_CLONE_ATTR_EXEC'. */
1073         clone_arg = nla_data(attr);
1074         dont_clone_flow_key = nla_get_u32(clone_arg);
1075         actions = nla_next(clone_arg, &rem);
1076
1077         return clone_execute(dp, skb, key, 0, actions, rem, last,
1078                              !dont_clone_flow_key);
1079 }
1080
1081 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1082                          const struct nlattr *attr)
1083 {
1084         struct ovs_action_hash *hash_act = nla_data(attr);
1085         u32 hash = 0;
1086
1087         if (hash_act->hash_alg == OVS_HASH_ALG_L4) {
1088                 /* OVS_HASH_ALG_L4 hasing type. */
1089                 hash = skb_get_hash(skb);
1090         } else if (hash_act->hash_alg == OVS_HASH_ALG_SYM_L4) {
1091                 /* OVS_HASH_ALG_SYM_L4 hashing type.  NOTE: this doesn't
1092                  * extend past an encapsulated header.
1093                  */
1094                 hash = __skb_get_hash_symmetric(skb);
1095         }
1096
1097         hash = jhash_1word(hash, hash_act->hash_basis);
1098         if (!hash)
1099                 hash = 0x1;
1100
1101         key->ovs_flow_hash = hash;
1102 }
1103
1104 static int execute_set_action(struct sk_buff *skb,
1105                               struct sw_flow_key *flow_key,
1106                               const struct nlattr *a)
1107 {
1108         /* Only tunnel set execution is supported without a mask. */
1109         if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1110                 struct ovs_tunnel_info *tun = nla_data(a);
1111
1112                 skb_dst_drop(skb);
1113                 dst_hold((struct dst_entry *)tun->tun_dst);
1114                 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1115                 return 0;
1116         }
1117
1118         return -EINVAL;
1119 }
1120
1121 /* Mask is at the midpoint of the data. */
1122 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1123
1124 static int execute_masked_set_action(struct sk_buff *skb,
1125                                      struct sw_flow_key *flow_key,
1126                                      const struct nlattr *a)
1127 {
1128         int err = 0;
1129
1130         switch (nla_type(a)) {
1131         case OVS_KEY_ATTR_PRIORITY:
1132                 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1133                                *get_mask(a, u32 *));
1134                 flow_key->phy.priority = skb->priority;
1135                 break;
1136
1137         case OVS_KEY_ATTR_SKB_MARK:
1138                 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1139                 flow_key->phy.skb_mark = skb->mark;
1140                 break;
1141
1142         case OVS_KEY_ATTR_TUNNEL_INFO:
1143                 /* Masked data not supported for tunnel. */
1144                 err = -EINVAL;
1145                 break;
1146
1147         case OVS_KEY_ATTR_ETHERNET:
1148                 err = set_eth_addr(skb, flow_key, nla_data(a),
1149                                    get_mask(a, struct ovs_key_ethernet *));
1150                 break;
1151
1152         case OVS_KEY_ATTR_NSH:
1153                 err = set_nsh(skb, flow_key, a);
1154                 break;
1155
1156         case OVS_KEY_ATTR_IPV4:
1157                 err = set_ipv4(skb, flow_key, nla_data(a),
1158                                get_mask(a, struct ovs_key_ipv4 *));
1159                 break;
1160
1161         case OVS_KEY_ATTR_IPV6:
1162                 err = set_ipv6(skb, flow_key, nla_data(a),
1163                                get_mask(a, struct ovs_key_ipv6 *));
1164                 break;
1165
1166         case OVS_KEY_ATTR_TCP:
1167                 err = set_tcp(skb, flow_key, nla_data(a),
1168                               get_mask(a, struct ovs_key_tcp *));
1169                 break;
1170
1171         case OVS_KEY_ATTR_UDP:
1172                 err = set_udp(skb, flow_key, nla_data(a),
1173                               get_mask(a, struct ovs_key_udp *));
1174                 break;
1175
1176         case OVS_KEY_ATTR_SCTP:
1177                 err = set_sctp(skb, flow_key, nla_data(a),
1178                                get_mask(a, struct ovs_key_sctp *));
1179                 break;
1180
1181         case OVS_KEY_ATTR_MPLS:
1182                 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1183                                                                     __be32 *));
1184                 break;
1185
1186         case OVS_KEY_ATTR_CT_STATE:
1187         case OVS_KEY_ATTR_CT_ZONE:
1188         case OVS_KEY_ATTR_CT_MARK:
1189         case OVS_KEY_ATTR_CT_LABELS:
1190         case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1191         case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1192                 err = -EINVAL;
1193                 break;
1194         }
1195
1196         return err;
1197 }
1198
1199 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1200                           struct sw_flow_key *key,
1201                           const struct nlattr *a, bool last)
1202 {
1203         u32 recirc_id;
1204
1205         if (!is_flow_key_valid(key)) {
1206                 int err;
1207
1208                 err = ovs_flow_key_update(skb, key);
1209                 if (err)
1210                         return err;
1211         }
1212         BUG_ON(!is_flow_key_valid(key));
1213
1214         recirc_id = nla_get_u32(a);
1215         return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1216 }
1217
1218 static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1219                                  struct sw_flow_key *key,
1220                                  const struct nlattr *attr, bool last)
1221 {
1222         struct ovs_skb_cb *ovs_cb = OVS_CB(skb);
1223         const struct nlattr *actions, *cpl_arg;
1224         int len, max_len, rem = nla_len(attr);
1225         const struct check_pkt_len_arg *arg;
1226         bool clone_flow_key;
1227
1228         /* The first netlink attribute in 'attr' is always
1229          * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1230          */
1231         cpl_arg = nla_data(attr);
1232         arg = nla_data(cpl_arg);
1233
1234         len = ovs_cb->mru ? ovs_cb->mru + skb->mac_len : skb->len;
1235         max_len = arg->pkt_len;
1236
1237         if ((skb_is_gso(skb) && skb_gso_validate_mac_len(skb, max_len)) ||
1238             len <= max_len) {
1239                 /* Second netlink attribute in 'attr' is always
1240                  * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1241                  */
1242                 actions = nla_next(cpl_arg, &rem);
1243                 clone_flow_key = !arg->exec_for_lesser_equal;
1244         } else {
1245                 /* Third netlink attribute in 'attr' is always
1246                  * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1247                  */
1248                 actions = nla_next(cpl_arg, &rem);
1249                 actions = nla_next(actions, &rem);
1250                 clone_flow_key = !arg->exec_for_greater;
1251         }
1252
1253         return clone_execute(dp, skb, key, 0, nla_data(actions),
1254                              nla_len(actions), last, clone_flow_key);
1255 }
1256
1257 static int execute_dec_ttl(struct sk_buff *skb, struct sw_flow_key *key)
1258 {
1259         int err;
1260
1261         if (skb->protocol == htons(ETH_P_IPV6)) {
1262                 struct ipv6hdr *nh;
1263
1264                 err = skb_ensure_writable(skb, skb_network_offset(skb) +
1265                                           sizeof(*nh));
1266                 if (unlikely(err))
1267                         return err;
1268
1269                 nh = ipv6_hdr(skb);
1270
1271                 if (nh->hop_limit <= 1)
1272                         return -EHOSTUNREACH;
1273
1274                 key->ip.ttl = --nh->hop_limit;
1275         } else if (skb->protocol == htons(ETH_P_IP)) {
1276                 struct iphdr *nh;
1277                 u8 old_ttl;
1278
1279                 err = skb_ensure_writable(skb, skb_network_offset(skb) +
1280                                           sizeof(*nh));
1281                 if (unlikely(err))
1282                         return err;
1283
1284                 nh = ip_hdr(skb);
1285                 if (nh->ttl <= 1)
1286                         return -EHOSTUNREACH;
1287
1288                 old_ttl = nh->ttl--;
1289                 csum_replace2(&nh->check, htons(old_ttl << 8),
1290                               htons(nh->ttl << 8));
1291                 key->ip.ttl = nh->ttl;
1292         }
1293         return 0;
1294 }
1295
1296 /* Execute a list of actions against 'skb'. */
1297 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1298                               struct sw_flow_key *key,
1299                               const struct nlattr *attr, int len)
1300 {
1301         const struct nlattr *a;
1302         int rem;
1303
1304         for (a = attr, rem = len; rem > 0;
1305              a = nla_next(a, &rem)) {
1306                 int err = 0;
1307
1308                 if (trace_ovs_do_execute_action_enabled())
1309                         trace_ovs_do_execute_action(dp, skb, key, a, rem);
1310
1311                 /* Actions that rightfully have to consume the skb should do it
1312                  * and return directly.
1313                  */
1314                 switch (nla_type(a)) {
1315                 case OVS_ACTION_ATTR_OUTPUT: {
1316                         int port = nla_get_u32(a);
1317                         struct sk_buff *clone;
1318
1319                         /* Every output action needs a separate clone
1320                          * of 'skb', In case the output action is the
1321                          * last action, cloning can be avoided.
1322                          */
1323                         if (nla_is_last(a, rem)) {
1324                                 do_output(dp, skb, port, key);
1325                                 /* 'skb' has been used for output.
1326                                  */
1327                                 return 0;
1328                         }
1329
1330                         clone = skb_clone(skb, GFP_ATOMIC);
1331                         if (clone)
1332                                 do_output(dp, clone, port, key);
1333                         OVS_CB(skb)->cutlen = 0;
1334                         break;
1335                 }
1336
1337                 case OVS_ACTION_ATTR_TRUNC: {
1338                         struct ovs_action_trunc *trunc = nla_data(a);
1339
1340                         if (skb->len > trunc->max_len)
1341                                 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1342                         break;
1343                 }
1344
1345                 case OVS_ACTION_ATTR_USERSPACE:
1346                         output_userspace(dp, skb, key, a, attr,
1347                                                      len, OVS_CB(skb)->cutlen);
1348                         OVS_CB(skb)->cutlen = 0;
1349                         if (nla_is_last(a, rem)) {
1350                                 consume_skb(skb);
1351                                 return 0;
1352                         }
1353                         break;
1354
1355                 case OVS_ACTION_ATTR_HASH:
1356                         execute_hash(skb, key, a);
1357                         break;
1358
1359                 case OVS_ACTION_ATTR_PUSH_MPLS: {
1360                         struct ovs_action_push_mpls *mpls = nla_data(a);
1361
1362                         err = push_mpls(skb, key, mpls->mpls_lse,
1363                                         mpls->mpls_ethertype, skb->mac_len);
1364                         break;
1365                 }
1366                 case OVS_ACTION_ATTR_ADD_MPLS: {
1367                         struct ovs_action_add_mpls *mpls = nla_data(a);
1368                         __u16 mac_len = 0;
1369
1370                         if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK)
1371                                 mac_len = skb->mac_len;
1372
1373                         err = push_mpls(skb, key, mpls->mpls_lse,
1374                                         mpls->mpls_ethertype, mac_len);
1375                         break;
1376                 }
1377                 case OVS_ACTION_ATTR_POP_MPLS:
1378                         err = pop_mpls(skb, key, nla_get_be16(a));
1379                         break;
1380
1381                 case OVS_ACTION_ATTR_PUSH_VLAN:
1382                         err = push_vlan(skb, key, nla_data(a));
1383                         break;
1384
1385                 case OVS_ACTION_ATTR_POP_VLAN:
1386                         err = pop_vlan(skb, key);
1387                         break;
1388
1389                 case OVS_ACTION_ATTR_RECIRC: {
1390                         bool last = nla_is_last(a, rem);
1391
1392                         err = execute_recirc(dp, skb, key, a, last);
1393                         if (last) {
1394                                 /* If this is the last action, the skb has
1395                                  * been consumed or freed.
1396                                  * Return immediately.
1397                                  */
1398                                 return err;
1399                         }
1400                         break;
1401                 }
1402
1403                 case OVS_ACTION_ATTR_SET:
1404                         err = execute_set_action(skb, key, nla_data(a));
1405                         break;
1406
1407                 case OVS_ACTION_ATTR_SET_MASKED:
1408                 case OVS_ACTION_ATTR_SET_TO_MASKED:
1409                         err = execute_masked_set_action(skb, key, nla_data(a));
1410                         break;
1411
1412                 case OVS_ACTION_ATTR_SAMPLE: {
1413                         bool last = nla_is_last(a, rem);
1414
1415                         err = sample(dp, skb, key, a, last);
1416                         if (last)
1417                                 return err;
1418
1419                         break;
1420                 }
1421
1422                 case OVS_ACTION_ATTR_CT:
1423                         if (!is_flow_key_valid(key)) {
1424                                 err = ovs_flow_key_update(skb, key);
1425                                 if (err)
1426                                         return err;
1427                         }
1428
1429                         err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1430                                              nla_data(a));
1431
1432                         /* Hide stolen IP fragments from user space. */
1433                         if (err)
1434                                 return err == -EINPROGRESS ? 0 : err;
1435                         break;
1436
1437                 case OVS_ACTION_ATTR_CT_CLEAR:
1438                         err = ovs_ct_clear(skb, key);
1439                         break;
1440
1441                 case OVS_ACTION_ATTR_PUSH_ETH:
1442                         err = push_eth(skb, key, nla_data(a));
1443                         break;
1444
1445                 case OVS_ACTION_ATTR_POP_ETH:
1446                         err = pop_eth(skb, key);
1447                         break;
1448
1449                 case OVS_ACTION_ATTR_PUSH_NSH:
1450                         err = push_nsh(skb, key, nla_data(a));
1451                         break;
1452
1453                 case OVS_ACTION_ATTR_POP_NSH:
1454                         err = pop_nsh(skb, key);
1455                         break;
1456
1457                 case OVS_ACTION_ATTR_METER:
1458                         if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1459                                 ovs_kfree_skb_reason(skb, OVS_DROP_METER);
1460                                 return 0;
1461                         }
1462                         break;
1463
1464                 case OVS_ACTION_ATTR_CLONE: {
1465                         bool last = nla_is_last(a, rem);
1466
1467                         err = clone(dp, skb, key, a, last);
1468                         if (last)
1469                                 return err;
1470
1471                         break;
1472                 }
1473
1474                 case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1475                         bool last = nla_is_last(a, rem);
1476
1477                         err = execute_check_pkt_len(dp, skb, key, a, last);
1478                         if (last)
1479                                 return err;
1480
1481                         break;
1482                 }
1483
1484                 case OVS_ACTION_ATTR_DEC_TTL:
1485                         err = execute_dec_ttl(skb, key);
1486                         if (err == -EHOSTUNREACH)
1487                                 return dec_ttl_exception_handler(dp, skb,
1488                                                                  key, a);
1489                         break;
1490
1491                 case OVS_ACTION_ATTR_DROP: {
1492                         enum ovs_drop_reason reason = nla_get_u32(a)
1493                                 ? OVS_DROP_EXPLICIT_WITH_ERROR
1494                                 : OVS_DROP_EXPLICIT;
1495
1496                         ovs_kfree_skb_reason(skb, reason);
1497                         return 0;
1498                 }
1499                 }
1500
1501                 if (unlikely(err)) {
1502                         ovs_kfree_skb_reason(skb, OVS_DROP_ACTION_ERROR);
1503                         return err;
1504                 }
1505         }
1506
1507         ovs_kfree_skb_reason(skb, OVS_DROP_LAST_ACTION);
1508         return 0;
1509 }
1510
1511 /* Execute the actions on the clone of the packet. The effect of the
1512  * execution does not affect the original 'skb' nor the original 'key'.
1513  *
1514  * The execution may be deferred in case the actions can not be executed
1515  * immediately.
1516  */
1517 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1518                          struct sw_flow_key *key, u32 recirc_id,
1519                          const struct nlattr *actions, int len,
1520                          bool last, bool clone_flow_key)
1521 {
1522         struct deferred_action *da;
1523         struct sw_flow_key *clone;
1524
1525         skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1526         if (!skb) {
1527                 /* Out of memory, skip this action.
1528                  */
1529                 return 0;
1530         }
1531
1532         /* When clone_flow_key is false, the 'key' will not be change
1533          * by the actions, then the 'key' can be used directly.
1534          * Otherwise, try to clone key from the next recursion level of
1535          * 'flow_keys'. If clone is successful, execute the actions
1536          * without deferring.
1537          */
1538         clone = clone_flow_key ? clone_key(key) : key;
1539         if (clone) {
1540                 int err = 0;
1541
1542                 if (actions) { /* Sample action */
1543                         if (clone_flow_key)
1544                                 __this_cpu_inc(exec_actions_level);
1545
1546                         err = do_execute_actions(dp, skb, clone,
1547                                                  actions, len);
1548
1549                         if (clone_flow_key)
1550                                 __this_cpu_dec(exec_actions_level);
1551                 } else { /* Recirc action */
1552                         clone->recirc_id = recirc_id;
1553                         ovs_dp_process_packet(skb, clone);
1554                 }
1555                 return err;
1556         }
1557
1558         /* Out of 'flow_keys' space. Defer actions */
1559         da = add_deferred_actions(skb, key, actions, len);
1560         if (da) {
1561                 if (!actions) { /* Recirc action */
1562                         key = &da->pkt_key;
1563                         key->recirc_id = recirc_id;
1564                 }
1565         } else {
1566                 /* Out of per CPU action FIFO space. Drop the 'skb' and
1567                  * log an error.
1568                  */
1569                 ovs_kfree_skb_reason(skb, OVS_DROP_DEFERRED_LIMIT);
1570
1571                 if (net_ratelimit()) {
1572                         if (actions) { /* Sample action */
1573                                 pr_warn("%s: deferred action limit reached, drop sample action\n",
1574                                         ovs_dp_name(dp));
1575                         } else {  /* Recirc action */
1576                                 pr_warn("%s: deferred action limit reached, drop recirc action (recirc_id=%#x)\n",
1577                                         ovs_dp_name(dp), recirc_id);
1578                         }
1579                 }
1580         }
1581         return 0;
1582 }
1583
1584 static void process_deferred_actions(struct datapath *dp)
1585 {
1586         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1587
1588         /* Do not touch the FIFO in case there is no deferred actions. */
1589         if (action_fifo_is_empty(fifo))
1590                 return;
1591
1592         /* Finishing executing all deferred actions. */
1593         do {
1594                 struct deferred_action *da = action_fifo_get(fifo);
1595                 struct sk_buff *skb = da->skb;
1596                 struct sw_flow_key *key = &da->pkt_key;
1597                 const struct nlattr *actions = da->actions;
1598                 int actions_len = da->actions_len;
1599
1600                 if (actions)
1601                         do_execute_actions(dp, skb, key, actions, actions_len);
1602                 else
1603                         ovs_dp_process_packet(skb, key);
1604         } while (!action_fifo_is_empty(fifo));
1605
1606         /* Reset FIFO for the next packet.  */
1607         action_fifo_init(fifo);
1608 }
1609
1610 /* Execute a list of actions against 'skb'. */
1611 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1612                         const struct sw_flow_actions *acts,
1613                         struct sw_flow_key *key)
1614 {
1615         int err, level;
1616
1617         level = __this_cpu_inc_return(exec_actions_level);
1618         if (unlikely(level > OVS_RECURSION_LIMIT)) {
1619                 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1620                                      ovs_dp_name(dp));
1621                 ovs_kfree_skb_reason(skb, OVS_DROP_RECURSION_LIMIT);
1622                 err = -ENETDOWN;
1623                 goto out;
1624         }
1625
1626         OVS_CB(skb)->acts_origlen = acts->orig_len;
1627         err = do_execute_actions(dp, skb, key,
1628                                  acts->actions, acts->actions_len);
1629
1630         if (level == 1)
1631                 process_deferred_actions(dp);
1632
1633 out:
1634         __this_cpu_dec(exec_actions_level);
1635         return err;
1636 }
1637
1638 int action_fifos_init(void)
1639 {
1640         action_fifos = alloc_percpu(struct action_fifo);
1641         if (!action_fifos)
1642                 return -ENOMEM;
1643
1644         flow_keys = alloc_percpu(struct action_flow_keys);
1645         if (!flow_keys) {
1646                 free_percpu(action_fifos);
1647                 return -ENOMEM;
1648         }
1649
1650         return 0;
1651 }
1652
1653 void action_fifos_exit(void)
1654 {
1655         free_percpu(action_fifos);
1656         free_percpu(flow_keys);
1657 }