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 int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
 315                     const struct nshhdr *nh)
 316 {
 317         int err;
 318
 319         err = nsh_push(skb, nh);
 320         if (err)
 321                 return err;
 322
 323         /* safe right before invalidate_flow_key */
 324         key->mac_proto = MAC_PROTO_NONE;
 325         invalidate_flow_key(key);
 326         return 0;
 327 }
 328
 329 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
 330 {
 331         int err;
 332
 333         err = nsh_pop(skb);
 334         if (err)
 335                 return err;
 336
 337         /* safe right before invalidate_flow_key */
 338         if (skb->protocol == htons(ETH_P_TEB))
 339                 key->mac_proto = MAC_PROTO_ETHERNET;
 340         else
 341                 key->mac_proto = MAC_PROTO_NONE;
 342         invalidate_flow_key(key);
 343         return 0;
 344 }
 345
 346 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
 347                                   __be32 addr, __be32 new_addr)
 348 {
 349         int transport_len = skb->len - skb_transport_offset(skb);
 350
 351         if (nh->frag_off & htons(IP_OFFSET))
 352                 return;
 353
 354         if (nh->protocol == IPPROTO_TCP) {
 355                 if (likely(transport_len >= sizeof(struct tcphdr)))
 356                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 357                                                  addr, new_addr, true);
 358         } else if (nh->protocol == IPPROTO_UDP) {
 359                 if (likely(transport_len >= sizeof(struct udphdr))) {
 360                         struct udphdr *uh = udp_hdr(skb);
 361
 362                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 363                                 inet_proto_csum_replace4(&uh->check, skb,
 364                                                          addr, new_addr, true);
 365                                 if (!uh->check)
 366                                         uh->check = CSUM_MANGLED_0;
 367                         }
 368                 }
 369         }
 370 }
 371
 372 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 373                         __be32 *addr, __be32 new_addr)
 374 {
 375         update_ip_l4_checksum(skb, nh, *addr, new_addr);
 376         csum_replace4(&nh->check, *addr, new_addr);
 377         skb_clear_hash(skb);
 378         ovs_ct_clear(skb, NULL);
 379         *addr = new_addr;
 380 }
 381
 382 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 383                                  __be32 addr[4], const __be32 new_addr[4])
 384 {
 385         int transport_len = skb->len - skb_transport_offset(skb);
 386
 387         if (l4_proto == NEXTHDR_TCP) {
 388                 if (likely(transport_len >= sizeof(struct tcphdr)))
 389                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 390                                                   addr, new_addr, true);
 391         } else if (l4_proto == NEXTHDR_UDP) {
 392                 if (likely(transport_len >= sizeof(struct udphdr))) {
 393                         struct udphdr *uh = udp_hdr(skb);
 394
 395                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 396                                 inet_proto_csum_replace16(&uh->check, skb,
 397                                                           addr, new_addr, true);
 398                                 if (!uh->check)
 399                                         uh->check = CSUM_MANGLED_0;
 400                         }
 401                 }
 402         } else if (l4_proto == NEXTHDR_ICMP) {
 403                 if (likely(transport_len >= sizeof(struct icmp6hdr)))
 404                         inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 405                                                   skb, addr, new_addr, true);
 406         }
 407 }
 408
 409 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
 410                            const __be32 mask[4], __be32 masked[4])
 411 {
 412         masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
 413         masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
 414         masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
 415         masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
 416 }
 417
 418 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 419                           __be32 addr[4], const __be32 new_addr[4],
 420                           bool recalculate_csum)
 421 {
 422         if (recalculate_csum)
 423                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 424
 425         skb_clear_hash(skb);
 426         ovs_ct_clear(skb, NULL);
 427         memcpy(addr, new_addr, sizeof(__be32[4]));
 428 }
 429
 430 static void set_ipv6_dsfield(struct sk_buff *skb, struct ipv6hdr *nh, u8 ipv6_tclass, u8 mask)
 431 {
 432         u8 old_ipv6_tclass = ipv6_get_dsfield(nh);
 433
 434         ipv6_tclass = OVS_MASKED(old_ipv6_tclass, ipv6_tclass, mask);
 435
 436         if (skb->ip_summed == CHECKSUM_COMPLETE)
 437                 csum_replace(&skb->csum, (__force __wsum)(old_ipv6_tclass << 12),
 438                              (__force __wsum)(ipv6_tclass << 12));
 439
 440         ipv6_change_dsfield(nh, ~mask, ipv6_tclass);
 441 }
 442
 443 static void set_ipv6_fl(struct sk_buff *skb, struct ipv6hdr *nh, u32 fl, u32 mask)
 444 {
 445         u32 ofl;
 446
 447         ofl = nh->flow_lbl[0] << 16 |  nh->flow_lbl[1] << 8 |  nh->flow_lbl[2];
 448         fl = OVS_MASKED(ofl, fl, mask);
 449
 450         /* Bits 21-24 are always unmasked, so this retains their values. */
 451         nh->flow_lbl[0] = (u8)(fl >> 16);
 452         nh->flow_lbl[1] = (u8)(fl >> 8);
 453         nh->flow_lbl[2] = (u8)fl;
 454
 455         if (skb->ip_summed == CHECKSUM_COMPLETE)
 456                 csum_replace(&skb->csum, (__force __wsum)htonl(ofl), (__force __wsum)htonl(fl));
 457 }
 458
 459 static void set_ipv6_ttl(struct sk_buff *skb, struct ipv6hdr *nh, u8 new_ttl, u8 mask)
 460 {
 461         new_ttl = OVS_MASKED(nh->hop_limit, new_ttl, mask);
 462
 463         if (skb->ip_summed == CHECKSUM_COMPLETE)
 464                 csum_replace(&skb->csum, (__force __wsum)(nh->hop_limit << 8),
 465                              (__force __wsum)(new_ttl << 8));
 466         nh->hop_limit = new_ttl;
 467 }
 468
 469 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
 470                        u8 mask)
 471 {
 472         new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
 473
 474         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 475         nh->ttl = new_ttl;
 476 }
 477
 478 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
 479                     const struct ovs_key_ipv4 *key,
 480                     const struct ovs_key_ipv4 *mask)
 481 {
 482         struct iphdr *nh;
 483         __be32 new_addr;
 484         int err;
 485
 486         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 487                                   sizeof(struct iphdr));
 488         if (unlikely(err))
 489                 return err;
 490
 491         nh = ip_hdr(skb);
 492
 493         /* Setting an IP addresses is typically only a side effect of
 494          * matching on them in the current userspace implementation, so it
 495          * makes sense to check if the value actually changed.
 496          */
 497         if (mask->ipv4_src) {
 498                 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
 499
 500                 if (unlikely(new_addr != nh->saddr)) {
 501                         set_ip_addr(skb, nh, &nh->saddr, new_addr);
 502                         flow_key->ipv4.addr.src = new_addr;
 503                 }
 504         }
 505         if (mask->ipv4_dst) {
 506                 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
 507
 508                 if (unlikely(new_addr != nh->daddr)) {
 509                         set_ip_addr(skb, nh, &nh->daddr, new_addr);
 510                         flow_key->ipv4.addr.dst = new_addr;
 511                 }
 512         }
 513         if (mask->ipv4_tos) {
 514                 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
 515                 flow_key->ip.tos = nh->tos;
 516         }
 517         if (mask->ipv4_ttl) {
 518                 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
 519                 flow_key->ip.ttl = nh->ttl;
 520         }
 521
 522         return 0;
 523 }
 524
 525 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
 526 {
 527         return !!(addr[0] | addr[1] | addr[2] | addr[3]);
 528 }
 529
 530 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
 531                     const struct ovs_key_ipv6 *key,
 532                     const struct ovs_key_ipv6 *mask)
 533 {
 534         struct ipv6hdr *nh;
 535         int err;
 536
 537         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 538                                   sizeof(struct ipv6hdr));
 539         if (unlikely(err))
 540                 return err;
 541
 542         nh = ipv6_hdr(skb);
 543
 544         /* Setting an IP addresses is typically only a side effect of
 545          * matching on them in the current userspace implementation, so it
 546          * makes sense to check if the value actually changed.
 547          */
 548         if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
 549                 __be32 *saddr = (__be32 *)&nh->saddr;
 550                 __be32 masked[4];
 551
 552                 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
 553
 554                 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
 555                         set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
 556                                       true);
 557                         memcpy(&flow_key->ipv6.addr.src, masked,
 558                                sizeof(flow_key->ipv6.addr.src));
 559                 }
 560         }
 561         if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
 562                 unsigned int offset = 0;
 563                 int flags = IP6_FH_F_SKIP_RH;
 564                 bool recalc_csum = true;
 565                 __be32 *daddr = (__be32 *)&nh->daddr;
 566                 __be32 masked[4];
 567
 568                 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
 569
 570                 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
 571                         if (ipv6_ext_hdr(nh->nexthdr))
 572                                 recalc_csum = (ipv6_find_hdr(skb, &offset,
 573                                                              NEXTHDR_ROUTING,
 574                                                              NULL, &flags)
 575                                                != NEXTHDR_ROUTING);
 576
 577                         set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
 578                                       recalc_csum);
 579                         memcpy(&flow_key->ipv6.addr.dst, masked,
 580                                sizeof(flow_key->ipv6.addr.dst));
 581                 }
 582         }
 583         if (mask->ipv6_tclass) {
 584                 set_ipv6_dsfield(skb, nh, key->ipv6_tclass, mask->ipv6_tclass);
 585                 flow_key->ip.tos = ipv6_get_dsfield(nh);
 586         }
 587         if (mask->ipv6_label) {
 588                 set_ipv6_fl(skb, nh, ntohl(key->ipv6_label),
 589                             ntohl(mask->ipv6_label));
 590                 flow_key->ipv6.label =
 591                     *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 592         }
 593         if (mask->ipv6_hlimit) {
 594                 set_ipv6_ttl(skb, nh, key->ipv6_hlimit, mask->ipv6_hlimit);
 595                 flow_key->ip.ttl = nh->hop_limit;
 596         }
 597         return 0;
 598 }
 599
 600 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
 601                    const struct nlattr *a)
 602 {
 603         struct nshhdr *nh;
 604         size_t length;
 605         int err;
 606         u8 flags;
 607         u8 ttl;
 608         int i;
 609
 610         struct ovs_key_nsh key;
 611         struct ovs_key_nsh mask;
 612
 613         err = nsh_key_from_nlattr(a, &key, &mask);
 614         if (err)
 615                 return err;
 616
 617         /* Make sure the NSH base header is there */
 618         if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
 619                 return -ENOMEM;
 620
 621         nh = nsh_hdr(skb);
 622         length = nsh_hdr_len(nh);
 623
 624         /* Make sure the whole NSH header is there */
 625         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 626                                        length);
 627         if (unlikely(err))
 628                 return err;
 629
 630         nh = nsh_hdr(skb);
 631         skb_postpull_rcsum(skb, nh, length);
 632         flags = nsh_get_flags(nh);
 633         flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
 634         flow_key->nsh.base.flags = flags;
 635         ttl = nsh_get_ttl(nh);
 636         ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
 637         flow_key->nsh.base.ttl = ttl;
 638         nsh_set_flags_and_ttl(nh, flags, ttl);
 639         nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
 640                                   mask.base.path_hdr);
 641         flow_key->nsh.base.path_hdr = nh->path_hdr;
 642         switch (nh->mdtype) {
 643         case NSH_M_TYPE1:
 644                 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
 645                         nh->md1.context[i] =
 646                             OVS_MASKED(nh->md1.context[i], key.context[i],
 647                                        mask.context[i]);
 648                 }
 649                 memcpy(flow_key->nsh.context, nh->md1.context,
 650                        sizeof(nh->md1.context));
 651                 break;
 652         case NSH_M_TYPE2:
 653                 memset(flow_key->nsh.context, 0,
 654                        sizeof(flow_key->nsh.context));
 655                 break;
 656         default:
 657                 return -EINVAL;
 658         }
 659         skb_postpush_rcsum(skb, nh, length);
 660         return 0;
 661 }
 662
 663 /* Must follow skb_ensure_writable() since that can move the skb data. */
 664 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 665                         __be16 new_port, __sum16 *check)
 666 {
 667         ovs_ct_clear(skb, NULL);
 668         inet_proto_csum_replace2(check, skb, *port, new_port, false);
 669         *port = new_port;
 670 }
 671
 672 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 673                    const struct ovs_key_udp *key,
 674                    const struct ovs_key_udp *mask)
 675 {
 676         struct udphdr *uh;
 677         __be16 src, dst;
 678         int err;
 679
 680         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 681                                   sizeof(struct udphdr));
 682         if (unlikely(err))
 683                 return err;
 684
 685         uh = udp_hdr(skb);
 686         /* Either of the masks is non-zero, so do not bother checking them. */
 687         src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
 688         dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
 689
 690         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 691                 if (likely(src != uh->source)) {
 692                         set_tp_port(skb, &uh->source, src, &uh->check);
 693                         flow_key->tp.src = src;
 694                 }
 695                 if (likely(dst != uh->dest)) {
 696                         set_tp_port(skb, &uh->dest, dst, &uh->check);
 697                         flow_key->tp.dst = dst;
 698                 }
 699
 700                 if (unlikely(!uh->check))
 701                         uh->check = CSUM_MANGLED_0;
 702         } else {
 703                 uh->source = src;
 704                 uh->dest = dst;
 705                 flow_key->tp.src = src;
 706                 flow_key->tp.dst = dst;
 707                 ovs_ct_clear(skb, NULL);
 708         }
 709
 710         skb_clear_hash(skb);
 711
 712         return 0;
 713 }
 714
 715 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 716                    const struct ovs_key_tcp *key,
 717                    const struct ovs_key_tcp *mask)
 718 {
 719         struct tcphdr *th;
 720         __be16 src, dst;
 721         int err;
 722
 723         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 724                                   sizeof(struct tcphdr));
 725         if (unlikely(err))
 726                 return err;
 727
 728         th = tcp_hdr(skb);
 729         src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
 730         if (likely(src != th->source)) {
 731                 set_tp_port(skb, &th->source, src, &th->check);
 732                 flow_key->tp.src = src;
 733         }
 734         dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
 735         if (likely(dst != th->dest)) {
 736                 set_tp_port(skb, &th->dest, dst, &th->check);
 737                 flow_key->tp.dst = dst;
 738         }
 739         skb_clear_hash(skb);
 740
 741         return 0;
 742 }
 743
 744 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 745                     const struct ovs_key_sctp *key,
 746                     const struct ovs_key_sctp *mask)
 747 {
 748         unsigned int sctphoff = skb_transport_offset(skb);
 749         struct sctphdr *sh;
 750         __le32 old_correct_csum, new_csum, old_csum;
 751         int err;
 752
 753         err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
 754         if (unlikely(err))
 755                 return err;
 756
 757         sh = sctp_hdr(skb);
 758         old_csum = sh->checksum;
 759         old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 760
 761         sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
 762         sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
 763
 764         new_csum = sctp_compute_cksum(skb, sctphoff);
 765
 766         /* Carry any checksum errors through. */
 767         sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 768
 769         skb_clear_hash(skb);
 770         ovs_ct_clear(skb, NULL);
 771
 772         flow_key->tp.src = sh->source;
 773         flow_key->tp.dst = sh->dest;
 774
 775         return 0;
 776 }
 777
 778 static int ovs_vport_output(struct net *net, struct sock *sk,
 779                             struct sk_buff *skb)
 780 {
 781         struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
 782         struct vport *vport = data->vport;
 783
 784         if (skb_cow_head(skb, data->l2_len) < 0) {
 785                 kfree_skb_reason(skb, SKB_DROP_REASON_NOMEM);
 786                 return -ENOMEM;
 787         }
 788
 789         __skb_dst_copy(skb, data->dst);
 790         *OVS_CB(skb) = data->cb;
 791         skb->inner_protocol = data->inner_protocol;
 792         if (data->vlan_tci & VLAN_CFI_MASK)
 793                 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
 794         else
 795                 __vlan_hwaccel_clear_tag(skb);
 796
 797         /* Reconstruct the MAC header.  */
 798         skb_push(skb, data->l2_len);
 799         memcpy(skb->data, &data->l2_data, data->l2_len);
 800         skb_postpush_rcsum(skb, skb->data, data->l2_len);
 801         skb_reset_mac_header(skb);
 802
 803         if (eth_p_mpls(skb->protocol)) {
 804                 skb->inner_network_header = skb->network_header;
 805                 skb_set_network_header(skb, data->network_offset);
 806                 skb_reset_mac_len(skb);
 807         }
 808
 809         ovs_vport_send(vport, skb, data->mac_proto);
 810         return 0;
 811 }
 812
 813 static unsigned int
 814 ovs_dst_get_mtu(const struct dst_entry *dst)
 815 {
 816         return dst->dev->mtu;
 817 }
 818
 819 static struct dst_ops ovs_dst_ops = {
 820         .family = AF_UNSPEC,
 821         .mtu = ovs_dst_get_mtu,
 822 };
 823
 824 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
 825  * ovs_vport_output(), which is called once per fragmented packet.
 826  */
 827 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
 828                          u16 orig_network_offset, u8 mac_proto)
 829 {
 830         unsigned int hlen = skb_network_offset(skb);
 831         struct ovs_frag_data *data;
 832
 833         data = this_cpu_ptr(&ovs_frag_data_storage);
 834         data->dst = skb->_skb_refdst;
 835         data->vport = vport;
 836         data->cb = *OVS_CB(skb);
 837         data->inner_protocol = skb->inner_protocol;
 838         data->network_offset = orig_network_offset;
 839         if (skb_vlan_tag_present(skb))
 840                 data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
 841         else
 842                 data->vlan_tci = 0;
 843         data->vlan_proto = skb->vlan_proto;
 844         data->mac_proto = mac_proto;
 845         data->l2_len = hlen;
 846         memcpy(&data->l2_data, skb->data, hlen);
 847
 848         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 849         skb_pull(skb, hlen);
 850 }
 851
 852 static void ovs_fragment(struct net *net, struct vport *vport,
 853                          struct sk_buff *skb, u16 mru,
 854                          struct sw_flow_key *key)
 855 {
 856         enum ovs_drop_reason reason;
 857         u16 orig_network_offset = 0;
 858
 859         if (eth_p_mpls(skb->protocol)) {
 860                 orig_network_offset = skb_network_offset(skb);
 861                 skb->network_header = skb->inner_network_header;
 862         }
 863
 864         if (skb_network_offset(skb) > MAX_L2_LEN) {
 865                 OVS_NLERR(1, "L2 header too long to fragment");
 866                 reason = OVS_DROP_FRAG_L2_TOO_LONG;
 867                 goto err;
 868         }
 869
 870         if (key->eth.type == htons(ETH_P_IP)) {
 871                 struct rtable ovs_rt = { 0 };
 872                 unsigned long orig_dst;
 873
 874                 prepare_frag(vport, skb, orig_network_offset,
 875                              ovs_key_mac_proto(key));
 876                 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
 877                          DST_OBSOLETE_NONE, DST_NOCOUNT);
 878                 ovs_rt.dst.dev = vport->dev;
 879
 880                 orig_dst = skb->_skb_refdst;
 881                 skb_dst_set_noref(skb, &ovs_rt.dst);
 882                 IPCB(skb)->frag_max_size = mru;
 883
 884                 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
 885                 refdst_drop(orig_dst);
 886         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 887                 unsigned long orig_dst;
 888                 struct rt6_info ovs_rt;
 889
 890                 prepare_frag(vport, skb, orig_network_offset,
 891                              ovs_key_mac_proto(key));
 892                 memset(&ovs_rt, 0, sizeof(ovs_rt));
 893                 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
 894                          DST_OBSOLETE_NONE, DST_NOCOUNT);
 895                 ovs_rt.dst.dev = vport->dev;
 896
 897                 orig_dst = skb->_skb_refdst;
 898                 skb_dst_set_noref(skb, &ovs_rt.dst);
 899                 IP6CB(skb)->frag_max_size = mru;
 900
 901                 ipv6_stub->ipv6_fragment(net, skb->sk, skb, ovs_vport_output);
 902                 refdst_drop(orig_dst);
 903         } else {
 904                 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
 905                           ovs_vport_name(vport), ntohs(key->eth.type), mru,
 906                           vport->dev->mtu);
 907                 reason = OVS_DROP_FRAG_INVALID_PROTO;
 908                 goto err;
 909         }
 910
 911         return;
 912 err:
 913         ovs_kfree_skb_reason(skb, reason);
 914 }
 915
 916 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
 917                       struct sw_flow_key *key)
 918 {
 919         struct vport *vport = ovs_vport_rcu(dp, out_port);
 920
 921         if (likely(vport && netif_carrier_ok(vport->dev))) {
 922                 u16 mru = OVS_CB(skb)->mru;
 923                 u32 cutlen = OVS_CB(skb)->cutlen;
 924
 925                 if (unlikely(cutlen > 0)) {
 926                         if (skb->len - cutlen > ovs_mac_header_len(key))
 927                                 pskb_trim(skb, skb->len - cutlen);
 928                         else
 929                                 pskb_trim(skb, ovs_mac_header_len(key));
 930                 }
 931
 932                 if (likely(!mru ||
 933                            (skb->len <= mru + vport->dev->hard_header_len))) {
 934                         ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
 935                 } else if (mru <= vport->dev->mtu) {
 936                         struct net *net = read_pnet(&dp->net);
 937
 938                         ovs_fragment(net, vport, skb, mru, key);
 939                 } else {
 940                         kfree_skb_reason(skb, SKB_DROP_REASON_PKT_TOO_BIG);
 941                 }
 942         } else {
 943                 kfree_skb_reason(skb, SKB_DROP_REASON_DEV_READY);
 944         }
 945 }
 946
 947 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 948                             struct sw_flow_key *key, const struct nlattr *attr,
 949                             const struct nlattr *actions, int actions_len,
 950                             uint32_t cutlen)
 951 {
 952         struct dp_upcall_info upcall;
 953         const struct nlattr *a;
 954         int rem;
 955
 956         memset(&upcall, 0, sizeof(upcall));
 957         upcall.cmd = OVS_PACKET_CMD_ACTION;
 958         upcall.mru = OVS_CB(skb)->mru;
 959
 960         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 961              a = nla_next(a, &rem)) {
 962                 switch (nla_type(a)) {
 963                 case OVS_USERSPACE_ATTR_USERDATA:
 964                         upcall.userdata = a;
 965                         break;
 966
 967                 case OVS_USERSPACE_ATTR_PID:
 968                         if (dp->user_features &
 969                             OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
 970                                 upcall.portid =
 971                                   ovs_dp_get_upcall_portid(dp,
 972                                                            smp_processor_id());
 973                         else
 974                                 upcall.portid = nla_get_u32(a);
 975                         break;
 976
 977                 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
 978                         /* Get out tunnel info. */
 979                         struct vport *vport;
 980
 981                         vport = ovs_vport_rcu(dp, nla_get_u32(a));
 982                         if (vport) {
 983                                 int err;
 984
 985                                 err = dev_fill_metadata_dst(vport->dev, skb);
 986                                 if (!err)
 987                                         upcall.egress_tun_info = skb_tunnel_info(skb);
 988                         }
 989
 990                         break;
 991                 }
 992
 993                 case OVS_USERSPACE_ATTR_ACTIONS: {
 994                         /* Include actions. */
 995                         upcall.actions = actions;
 996                         upcall.actions_len = actions_len;
 997                         break;
 998                 }
 999
1000                 } /* End of switch. */
1001         }
1002
1003         return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
1004 }
1005
1006 static int dec_ttl_exception_handler(struct datapath *dp, struct sk_buff *skb,
1007                                      struct sw_flow_key *key,
1008                                      const struct nlattr *attr)
1009 {
1010         /* The first attribute is always 'OVS_DEC_TTL_ATTR_ACTION'. */
1011         struct nlattr *actions = nla_data(attr);
1012
1013         if (nla_len(actions))
1014                 return clone_execute(dp, skb, key, 0, nla_data(actions),
1015                                      nla_len(actions), true, false);
1016
1017         ovs_kfree_skb_reason(skb, OVS_DROP_IP_TTL);
1018         return 0;
1019 }
1020
1021 /* When 'last' is true, sample() should always consume the 'skb'.
1022  * Otherwise, sample() should keep 'skb' intact regardless what
1023  * actions are executed within sample().
1024  */
1025 static int sample(struct datapath *dp, struct sk_buff *skb,
1026                   struct sw_flow_key *key, const struct nlattr *attr,
1027                   bool last)
1028 {
1029         struct nlattr *actions;
1030         struct nlattr *sample_arg;
1031         int rem = nla_len(attr);
1032         const struct sample_arg *arg;
1033         bool clone_flow_key;
1034
1035         /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
1036         sample_arg = nla_data(attr);
1037         arg = nla_data(sample_arg);
1038         actions = nla_next(sample_arg, &rem);
1039
1040         if ((arg->probability != U32_MAX) &&
1041             (!arg->probability || get_random_u32() > arg->probability)) {
1042                 if (last)
1043                         ovs_kfree_skb_reason(skb, OVS_DROP_LAST_ACTION);
1044                 return 0;
1045         }
1046
1047         clone_flow_key = !arg->exec;
1048         return clone_execute(dp, skb, key, 0, actions, rem, last,
1049                              clone_flow_key);
1050 }
1051
1052 /* When 'last' is true, clone() should always consume the 'skb'.
1053  * Otherwise, clone() should keep 'skb' intact regardless what
1054  * actions are executed within clone().
1055  */
1056 static int clone(struct datapath *dp, struct sk_buff *skb,
1057                  struct sw_flow_key *key, const struct nlattr *attr,
1058                  bool last)
1059 {
1060         struct nlattr *actions;
1061         struct nlattr *clone_arg;
1062         int rem = nla_len(attr);
1063         bool dont_clone_flow_key;
1064
1065         /* The first action is always 'OVS_CLONE_ATTR_EXEC'. */
1066         clone_arg = nla_data(attr);
1067         dont_clone_flow_key = nla_get_u32(clone_arg);
1068         actions = nla_next(clone_arg, &rem);
1069
1070         return clone_execute(dp, skb, key, 0, actions, rem, last,
1071                              !dont_clone_flow_key);
1072 }
1073
1074 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1075                          const struct nlattr *attr)
1076 {
1077         struct ovs_action_hash *hash_act = nla_data(attr);
1078         u32 hash = 0;
1079
1080         if (hash_act->hash_alg == OVS_HASH_ALG_L4) {
1081                 /* OVS_HASH_ALG_L4 hasing type. */
1082                 hash = skb_get_hash(skb);
1083         } else if (hash_act->hash_alg == OVS_HASH_ALG_SYM_L4) {
1084                 /* OVS_HASH_ALG_SYM_L4 hashing type.  NOTE: this doesn't
1085                  * extend past an encapsulated header.
1086                  */
1087                 hash = __skb_get_hash_symmetric(skb);
1088         }
1089
1090         hash = jhash_1word(hash, hash_act->hash_basis);
1091         if (!hash)
1092                 hash = 0x1;
1093
1094         key->ovs_flow_hash = hash;
1095 }
1096
1097 static int execute_set_action(struct sk_buff *skb,
1098                               struct sw_flow_key *flow_key,
1099                               const struct nlattr *a)
1100 {
1101         /* Only tunnel set execution is supported without a mask. */
1102         if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1103                 struct ovs_tunnel_info *tun = nla_data(a);
1104
1105                 skb_dst_drop(skb);
1106                 dst_hold((struct dst_entry *)tun->tun_dst);
1107                 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1108                 return 0;
1109         }
1110
1111         return -EINVAL;
1112 }
1113
1114 /* Mask is at the midpoint of the data. */
1115 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1116
1117 static int execute_masked_set_action(struct sk_buff *skb,
1118                                      struct sw_flow_key *flow_key,
1119                                      const struct nlattr *a)
1120 {
1121         int err = 0;
1122
1123         switch (nla_type(a)) {
1124         case OVS_KEY_ATTR_PRIORITY:
1125                 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1126                                *get_mask(a, u32 *));
1127                 flow_key->phy.priority = skb->priority;
1128                 break;
1129
1130         case OVS_KEY_ATTR_SKB_MARK:
1131                 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1132                 flow_key->phy.skb_mark = skb->mark;
1133                 break;
1134
1135         case OVS_KEY_ATTR_TUNNEL_INFO:
1136                 /* Masked data not supported for tunnel. */
1137                 err = -EINVAL;
1138                 break;
1139
1140         case OVS_KEY_ATTR_ETHERNET:
1141                 err = set_eth_addr(skb, flow_key, nla_data(a),
1142                                    get_mask(a, struct ovs_key_ethernet *));
1143                 break;
1144
1145         case OVS_KEY_ATTR_NSH:
1146                 err = set_nsh(skb, flow_key, a);
1147                 break;
1148
1149         case OVS_KEY_ATTR_IPV4:
1150                 err = set_ipv4(skb, flow_key, nla_data(a),
1151                                get_mask(a, struct ovs_key_ipv4 *));
1152                 break;
1153
1154         case OVS_KEY_ATTR_IPV6:
1155                 err = set_ipv6(skb, flow_key, nla_data(a),
1156                                get_mask(a, struct ovs_key_ipv6 *));
1157                 break;
1158
1159         case OVS_KEY_ATTR_TCP:
1160                 err = set_tcp(skb, flow_key, nla_data(a),
1161                               get_mask(a, struct ovs_key_tcp *));
1162                 break;
1163
1164         case OVS_KEY_ATTR_UDP:
1165                 err = set_udp(skb, flow_key, nla_data(a),
1166                               get_mask(a, struct ovs_key_udp *));
1167                 break;
1168
1169         case OVS_KEY_ATTR_SCTP:
1170                 err = set_sctp(skb, flow_key, nla_data(a),
1171                                get_mask(a, struct ovs_key_sctp *));
1172                 break;
1173
1174         case OVS_KEY_ATTR_MPLS:
1175                 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1176                                                                     __be32 *));
1177                 break;
1178
1179         case OVS_KEY_ATTR_CT_STATE:
1180         case OVS_KEY_ATTR_CT_ZONE:
1181         case OVS_KEY_ATTR_CT_MARK:
1182         case OVS_KEY_ATTR_CT_LABELS:
1183         case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1184         case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1185                 err = -EINVAL;
1186                 break;
1187         }
1188
1189         return err;
1190 }
1191
1192 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1193                           struct sw_flow_key *key,
1194                           const struct nlattr *a, bool last)
1195 {
1196         u32 recirc_id;
1197
1198         if (!is_flow_key_valid(key)) {
1199                 int err;
1200
1201                 err = ovs_flow_key_update(skb, key);
1202                 if (err)
1203                         return err;
1204         }
1205         BUG_ON(!is_flow_key_valid(key));
1206
1207         recirc_id = nla_get_u32(a);
1208         return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1209 }
1210
1211 static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1212                                  struct sw_flow_key *key,
1213                                  const struct nlattr *attr, bool last)
1214 {
1215         struct ovs_skb_cb *ovs_cb = OVS_CB(skb);
1216         const struct nlattr *actions, *cpl_arg;
1217         int len, max_len, rem = nla_len(attr);
1218         const struct check_pkt_len_arg *arg;
1219         bool clone_flow_key;
1220
1221         /* The first netlink attribute in 'attr' is always
1222          * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1223          */
1224         cpl_arg = nla_data(attr);
1225         arg = nla_data(cpl_arg);
1226
1227         len = ovs_cb->mru ? ovs_cb->mru + skb->mac_len : skb->len;
1228         max_len = arg->pkt_len;
1229
1230         if ((skb_is_gso(skb) && skb_gso_validate_mac_len(skb, max_len)) ||
1231             len <= max_len) {
1232                 /* Second netlink attribute in 'attr' is always
1233                  * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1234                  */
1235                 actions = nla_next(cpl_arg, &rem);
1236                 clone_flow_key = !arg->exec_for_lesser_equal;
1237         } else {
1238                 /* Third netlink attribute in 'attr' is always
1239                  * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1240                  */
1241                 actions = nla_next(cpl_arg, &rem);
1242                 actions = nla_next(actions, &rem);
1243                 clone_flow_key = !arg->exec_for_greater;
1244         }
1245
1246         return clone_execute(dp, skb, key, 0, nla_data(actions),
1247                              nla_len(actions), last, clone_flow_key);
1248 }
1249
1250 static int execute_dec_ttl(struct sk_buff *skb, struct sw_flow_key *key)
1251 {
1252         int err;
1253
1254         if (skb->protocol == htons(ETH_P_IPV6)) {
1255                 struct ipv6hdr *nh;
1256
1257                 err = skb_ensure_writable(skb, skb_network_offset(skb) +
1258                                           sizeof(*nh));
1259                 if (unlikely(err))
1260                         return err;
1261
1262                 nh = ipv6_hdr(skb);
1263
1264                 if (nh->hop_limit <= 1)
1265                         return -EHOSTUNREACH;
1266
1267                 key->ip.ttl = --nh->hop_limit;
1268         } else if (skb->protocol == htons(ETH_P_IP)) {
1269                 struct iphdr *nh;
1270                 u8 old_ttl;
1271
1272                 err = skb_ensure_writable(skb, skb_network_offset(skb) +
1273                                           sizeof(*nh));
1274                 if (unlikely(err))
1275                         return err;
1276
1277                 nh = ip_hdr(skb);
1278                 if (nh->ttl <= 1)
1279                         return -EHOSTUNREACH;
1280
1281                 old_ttl = nh->ttl--;
1282                 csum_replace2(&nh->check, htons(old_ttl << 8),
1283                               htons(nh->ttl << 8));
1284                 key->ip.ttl = nh->ttl;
1285         }
1286         return 0;
1287 }
1288
1289 /* Execute a list of actions against 'skb'. */
1290 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1291                               struct sw_flow_key *key,
1292                               const struct nlattr *attr, int len)
1293 {
1294         const struct nlattr *a;
1295         int rem;
1296
1297         for (a = attr, rem = len; rem > 0;
1298              a = nla_next(a, &rem)) {
1299                 int err = 0;
1300
1301                 if (trace_ovs_do_execute_action_enabled())
1302                         trace_ovs_do_execute_action(dp, skb, key, a, rem);
1303
1304                 /* Actions that rightfully have to consume the skb should do it
1305                  * and return directly.
1306                  */
1307                 switch (nla_type(a)) {
1308                 case OVS_ACTION_ATTR_OUTPUT: {
1309                         int port = nla_get_u32(a);
1310                         struct sk_buff *clone;
1311
1312                         /* Every output action needs a separate clone
1313                          * of 'skb', In case the output action is the
1314                          * last action, cloning can be avoided.
1315                          */
1316                         if (nla_is_last(a, rem)) {
1317                                 do_output(dp, skb, port, key);
1318                                 /* 'skb' has been used for output.
1319                                  */
1320                                 return 0;
1321                         }
1322
1323                         clone = skb_clone(skb, GFP_ATOMIC);
1324                         if (clone)
1325                                 do_output(dp, clone, port, key);
1326                         OVS_CB(skb)->cutlen = 0;
1327                         break;
1328                 }
1329
1330                 case OVS_ACTION_ATTR_TRUNC: {
1331                         struct ovs_action_trunc *trunc = nla_data(a);
1332
1333                         if (skb->len > trunc->max_len)
1334                                 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1335                         break;
1336                 }
1337
1338                 case OVS_ACTION_ATTR_USERSPACE:
1339                         output_userspace(dp, skb, key, a, attr,
1340                                                      len, OVS_CB(skb)->cutlen);
1341                         OVS_CB(skb)->cutlen = 0;
1342                         if (nla_is_last(a, rem)) {
1343                                 consume_skb(skb);
1344                                 return 0;
1345                         }
1346                         break;
1347
1348                 case OVS_ACTION_ATTR_HASH:
1349                         execute_hash(skb, key, a);
1350                         break;
1351
1352                 case OVS_ACTION_ATTR_PUSH_MPLS: {
1353                         struct ovs_action_push_mpls *mpls = nla_data(a);
1354
1355                         err = push_mpls(skb, key, mpls->mpls_lse,
1356                                         mpls->mpls_ethertype, skb->mac_len);
1357                         break;
1358                 }
1359                 case OVS_ACTION_ATTR_ADD_MPLS: {
1360                         struct ovs_action_add_mpls *mpls = nla_data(a);
1361                         __u16 mac_len = 0;
1362
1363                         if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK)
1364                                 mac_len = skb->mac_len;
1365
1366                         err = push_mpls(skb, key, mpls->mpls_lse,
1367                                         mpls->mpls_ethertype, mac_len);
1368                         break;
1369                 }
1370                 case OVS_ACTION_ATTR_POP_MPLS:
1371                         err = pop_mpls(skb, key, nla_get_be16(a));
1372                         break;
1373
1374                 case OVS_ACTION_ATTR_PUSH_VLAN:
1375                         err = push_vlan(skb, key, nla_data(a));
1376                         break;
1377
1378                 case OVS_ACTION_ATTR_POP_VLAN:
1379                         err = pop_vlan(skb, key);
1380                         break;
1381
1382                 case OVS_ACTION_ATTR_RECIRC: {
1383                         bool last = nla_is_last(a, rem);
1384
1385                         err = execute_recirc(dp, skb, key, a, last);
1386                         if (last) {
1387                                 /* If this is the last action, the skb has
1388                                  * been consumed or freed.
1389                                  * Return immediately.
1390                                  */
1391                                 return err;
1392                         }
1393                         break;
1394                 }
1395
1396                 case OVS_ACTION_ATTR_SET:
1397                         err = execute_set_action(skb, key, nla_data(a));
1398                         break;
1399
1400                 case OVS_ACTION_ATTR_SET_MASKED:
1401                 case OVS_ACTION_ATTR_SET_TO_MASKED:
1402                         err = execute_masked_set_action(skb, key, nla_data(a));
1403                         break;
1404
1405                 case OVS_ACTION_ATTR_SAMPLE: {
1406                         bool last = nla_is_last(a, rem);
1407
1408                         err = sample(dp, skb, key, a, last);
1409                         if (last)
1410                                 return err;
1411
1412                         break;
1413                 }
1414
1415                 case OVS_ACTION_ATTR_CT:
1416                         if (!is_flow_key_valid(key)) {
1417                                 err = ovs_flow_key_update(skb, key);
1418                                 if (err)
1419                                         return err;
1420                         }
1421
1422                         err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1423                                              nla_data(a));
1424
1425                         /* Hide stolen IP fragments from user space. */
1426                         if (err)
1427                                 return err == -EINPROGRESS ? 0 : err;
1428                         break;
1429
1430                 case OVS_ACTION_ATTR_CT_CLEAR:
1431                         err = ovs_ct_clear(skb, key);
1432                         break;
1433
1434                 case OVS_ACTION_ATTR_PUSH_ETH:
1435                         err = push_eth(skb, key, nla_data(a));
1436                         break;
1437
1438                 case OVS_ACTION_ATTR_POP_ETH:
1439                         err = pop_eth(skb, key);
1440                         break;
1441
1442                 case OVS_ACTION_ATTR_PUSH_NSH: {
1443                         u8 buffer[NSH_HDR_MAX_LEN];
1444                         struct nshhdr *nh = (struct nshhdr *)buffer;
1445
1446                         err = nsh_hdr_from_nlattr(nla_data(a), nh,
1447                                                   NSH_HDR_MAX_LEN);
1448                         if (unlikely(err))
1449                                 break;
1450                         err = push_nsh(skb, key, nh);
1451                         break;
1452                 }
1453
1454                 case OVS_ACTION_ATTR_POP_NSH:
1455                         err = pop_nsh(skb, key);
1456                         break;
1457
1458                 case OVS_ACTION_ATTR_METER:
1459                         if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1460                                 ovs_kfree_skb_reason(skb, OVS_DROP_METER);
1461                                 return 0;
1462                         }
1463                         break;
1464
1465                 case OVS_ACTION_ATTR_CLONE: {
1466                         bool last = nla_is_last(a, rem);
1467
1468                         err = clone(dp, skb, key, a, last);
1469                         if (last)
1470                                 return err;
1471
1472                         break;
1473                 }
1474
1475                 case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1476                         bool last = nla_is_last(a, rem);
1477
1478                         err = execute_check_pkt_len(dp, skb, key, a, last);
1479                         if (last)
1480                                 return err;
1481
1482                         break;
1483                 }
1484
1485                 case OVS_ACTION_ATTR_DEC_TTL:
1486                         err = execute_dec_ttl(skb, key);
1487                         if (err == -EHOSTUNREACH)
1488                                 return dec_ttl_exception_handler(dp, skb,
1489                                                                  key, a);
1490                         break;
1491
1492                 case OVS_ACTION_ATTR_DROP: {
1493                         enum ovs_drop_reason reason = nla_get_u32(a)
1494                                 ? OVS_DROP_EXPLICIT_WITH_ERROR
1495                                 : OVS_DROP_EXPLICIT;
1496
1497                         ovs_kfree_skb_reason(skb, reason);
1498                         return 0;
1499                 }
1500                 }
1501
1502                 if (unlikely(err)) {
1503                         ovs_kfree_skb_reason(skb, OVS_DROP_ACTION_ERROR);
1504                         return err;
1505                 }
1506         }
1507
1508         ovs_kfree_skb_reason(skb, OVS_DROP_LAST_ACTION);
1509         return 0;
1510 }
1511
1512 /* Execute the actions on the clone of the packet. The effect of the
1513  * execution does not affect the original 'skb' nor the original 'key'.
1514  *
1515  * The execution may be deferred in case the actions can not be executed
1516  * immediately.
1517  */
1518 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1519                          struct sw_flow_key *key, u32 recirc_id,
1520                          const struct nlattr *actions, int len,
1521                          bool last, bool clone_flow_key)
1522 {
1523         struct deferred_action *da;
1524         struct sw_flow_key *clone;
1525
1526         skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1527         if (!skb) {
1528                 /* Out of memory, skip this action.
1529                  */
1530                 return 0;
1531         }
1532
1533         /* When clone_flow_key is false, the 'key' will not be change
1534          * by the actions, then the 'key' can be used directly.
1535          * Otherwise, try to clone key from the next recursion level of
1536          * 'flow_keys'. If clone is successful, execute the actions
1537          * without deferring.
1538          */
1539         clone = clone_flow_key ? clone_key(key) : key;
1540         if (clone) {
1541                 int err = 0;
1542
1543                 if (actions) { /* Sample action */
1544                         if (clone_flow_key)
1545                                 __this_cpu_inc(exec_actions_level);
1546
1547                         err = do_execute_actions(dp, skb, clone,
1548                                                  actions, len);
1549
1550                         if (clone_flow_key)
1551                                 __this_cpu_dec(exec_actions_level);
1552                 } else { /* Recirc action */
1553                         clone->recirc_id = recirc_id;
1554                         ovs_dp_process_packet(skb, clone);
1555                 }
1556                 return err;
1557         }
1558
1559         /* Out of 'flow_keys' space. Defer actions */
1560         da = add_deferred_actions(skb, key, actions, len);
1561         if (da) {
1562                 if (!actions) { /* Recirc action */
1563                         key = &da->pkt_key;
1564                         key->recirc_id = recirc_id;
1565                 }
1566         } else {
1567                 /* Out of per CPU action FIFO space. Drop the 'skb' and
1568                  * log an error.
1569                  */
1570                 ovs_kfree_skb_reason(skb, OVS_DROP_DEFERRED_LIMIT);
1571
1572                 if (net_ratelimit()) {
1573                         if (actions) { /* Sample action */
1574                                 pr_warn("%s: deferred action limit reached, drop sample action\n",
1575                                         ovs_dp_name(dp));
1576                         } else {  /* Recirc action */
1577                                 pr_warn("%s: deferred action limit reached, drop recirc action (recirc_id=%#x)\n",
1578                                         ovs_dp_name(dp), recirc_id);
1579                         }
1580                 }
1581         }
1582         return 0;
1583 }
1584
1585 static void process_deferred_actions(struct datapath *dp)
1586 {
1587         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1588
1589         /* Do not touch the FIFO in case there is no deferred actions. */
1590         if (action_fifo_is_empty(fifo))
1591                 return;
1592
1593         /* Finishing executing all deferred actions. */
1594         do {
1595                 struct deferred_action *da = action_fifo_get(fifo);
1596                 struct sk_buff *skb = da->skb;
1597                 struct sw_flow_key *key = &da->pkt_key;
1598                 const struct nlattr *actions = da->actions;
1599                 int actions_len = da->actions_len;
1600
1601                 if (actions)
1602                         do_execute_actions(dp, skb, key, actions, actions_len);
1603                 else
1604                         ovs_dp_process_packet(skb, key);
1605         } while (!action_fifo_is_empty(fifo));
1606
1607         /* Reset FIFO for the next packet.  */
1608         action_fifo_init(fifo);
1609 }
1610
1611 /* Execute a list of actions against 'skb'. */
1612 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1613                         const struct sw_flow_actions *acts,
1614                         struct sw_flow_key *key)
1615 {
1616         int err, level;
1617
1618         level = __this_cpu_inc_return(exec_actions_level);
1619         if (unlikely(level > OVS_RECURSION_LIMIT)) {
1620                 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1621                                      ovs_dp_name(dp));
1622                 ovs_kfree_skb_reason(skb, OVS_DROP_RECURSION_LIMIT);
1623                 err = -ENETDOWN;
1624                 goto out;
1625         }
1626
1627         OVS_CB(skb)->acts_origlen = acts->orig_len;
1628         err = do_execute_actions(dp, skb, key,
1629                                  acts->actions, acts->actions_len);
1630
1631         if (level == 1)
1632                 process_deferred_actions(dp);
1633
1634 out:
1635         __this_cpu_dec(exec_actions_level);
1636         return err;
1637 }
1638
1639 int action_fifos_init(void)
1640 {
1641         action_fifos = alloc_percpu(struct action_fifo);
1642         if (!action_fifos)
1643                 return -ENOMEM;
1644
1645         flow_keys = alloc_percpu(struct action_flow_keys);
1646         if (!flow_keys) {
1647                 free_percpu(action_fifos);
1648                 return -ENOMEM;
1649         }
1650
1651         return 0;
1652 }
1653
1654 void action_fifos_exit(void)
1655 {
1656         free_percpu(action_fifos);
1657         free_percpu(flow_keys);
1658 }