net/openvswitch/actions.c

   1 /*
   2  * Copyright (c) 2007-2014 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/skbuff.h>
  22 #include <linux/in.h>
  23 #include <linux/ip.h>
  24 #include <linux/openvswitch.h>
  25 #include <linux/sctp.h>
  26 #include <linux/tcp.h>
  27 #include <linux/udp.h>
  28 #include <linux/in6.h>
  29 #include <linux/if_arp.h>
  30 #include <linux/if_vlan.h>
  31 #include <net/ip.h>
  32 #include <net/ipv6.h>
  33 #include <net/checksum.h>
  34 #include <net/dsfield.h>
  35 #include <net/sctp/checksum.h>
  36
  37 #include "datapath.h"
  38 #include "flow.h"
  39 #include "vport.h"
  40
  41 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
  42                               struct sw_flow_key *key,
  43                               const struct nlattr *attr, int len);
  44
  45 struct deferred_action {
  46         struct sk_buff *skb;
  47         const struct nlattr *actions;
  48
  49         /* Store pkt_key clone when creating deferred action. */
  50         struct sw_flow_key pkt_key;
  51 };
  52
  53 #define DEFERRED_ACTION_FIFO_SIZE 10
  54 struct action_fifo {
  55         int head;
  56         int tail;
  57         /* Deferred action fifo queue storage. */
  58         struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
  59 };
  60
  61 static struct action_fifo __percpu *action_fifos;
  62 static DEFINE_PER_CPU(int, exec_actions_level);
  63
  64 static void action_fifo_init(struct action_fifo *fifo)
  65 {
  66         fifo->head = 0;
  67         fifo->tail = 0;
  68 }
  69
  70 static bool action_fifo_is_empty(struct action_fifo *fifo)
  71 {
  72         return (fifo->head == fifo->tail);
  73 }
  74
  75 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
  76 {
  77         if (action_fifo_is_empty(fifo))
  78                 return NULL;
  79
  80         return &fifo->fifo[fifo->tail++];
  81 }
  82
  83 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
  84 {
  85         if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
  86                 return NULL;
  87
  88         return &fifo->fifo[fifo->head++];
  89 }
  90
  91 /* Return true if fifo is not full */
  92 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
  93                                                     struct sw_flow_key *key,
  94                                                     const struct nlattr *attr)
  95 {
  96         struct action_fifo *fifo;
  97         struct deferred_action *da;
  98
  99         fifo = this_cpu_ptr(action_fifos);
 100         da = action_fifo_put(fifo);
 101         if (da) {
 102                 da->skb = skb;
 103                 da->actions = attr;
 104                 da->pkt_key = *key;
 105         }
 106
 107         return da;
 108 }
 109
 110 static int make_writable(struct sk_buff *skb, int write_len)
 111 {
 112         if (!pskb_may_pull(skb, write_len))
 113                 return -ENOMEM;
 114
 115         if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
 116                 return 0;
 117
 118         return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
 119 }
 120
 121 /* remove VLAN header from packet and update csum accordingly. */
 122 static int __pop_vlan_tci(struct sk_buff *skb, __be16 *current_tci)
 123 {
 124         struct vlan_hdr *vhdr;
 125         int err;
 126
 127         err = make_writable(skb, VLAN_ETH_HLEN);
 128         if (unlikely(err))
 129                 return err;
 130
 131         if (skb->ip_summed == CHECKSUM_COMPLETE)
 132                 skb->csum = csum_sub(skb->csum, csum_partial(skb->data
 133                                         + (2 * ETH_ALEN), VLAN_HLEN, 0));
 134
 135         vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
 136         *current_tci = vhdr->h_vlan_TCI;
 137
 138         memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
 139         __skb_pull(skb, VLAN_HLEN);
 140
 141         vlan_set_encap_proto(skb, vhdr);
 142         skb->mac_header += VLAN_HLEN;
 143         if (skb_network_offset(skb) < ETH_HLEN)
 144                 skb_set_network_header(skb, ETH_HLEN);
 145         skb_reset_mac_len(skb);
 146
 147         return 0;
 148 }
 149
 150 static int pop_vlan(struct sk_buff *skb)
 151 {
 152         __be16 tci;
 153         int err;
 154
 155         if (likely(vlan_tx_tag_present(skb))) {
 156                 skb->vlan_tci = 0;
 157         } else {
 158                 if (unlikely(skb->protocol != htons(ETH_P_8021Q) ||
 159                              skb->len < VLAN_ETH_HLEN))
 160                         return 0;
 161
 162                 err = __pop_vlan_tci(skb, &tci);
 163                 if (err)
 164                         return err;
 165         }
 166         /* move next vlan tag to hw accel tag */
 167         if (likely(skb->protocol != htons(ETH_P_8021Q) ||
 168                    skb->len < VLAN_ETH_HLEN))
 169                 return 0;
 170
 171         err = __pop_vlan_tci(skb, &tci);
 172         if (unlikely(err))
 173                 return err;
 174
 175         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(tci));
 176         return 0;
 177 }
 178
 179 static int push_vlan(struct sk_buff *skb, const struct ovs_action_push_vlan *vlan)
 180 {
 181         if (unlikely(vlan_tx_tag_present(skb))) {
 182                 u16 current_tag;
 183
 184                 /* push down current VLAN tag */
 185                 current_tag = vlan_tx_tag_get(skb);
 186
 187                 if (!__vlan_put_tag(skb, skb->vlan_proto, current_tag))
 188                         return -ENOMEM;
 189
 190                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 191                         skb->csum = csum_add(skb->csum, csum_partial(skb->data
 192                                         + (2 * ETH_ALEN), VLAN_HLEN, 0));
 193
 194         }
 195         __vlan_hwaccel_put_tag(skb, vlan->vlan_tpid, ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
 196         return 0;
 197 }
 198
 199 static int set_eth_addr(struct sk_buff *skb,
 200                         const struct ovs_key_ethernet *eth_key)
 201 {
 202         int err;
 203         err = make_writable(skb, ETH_HLEN);
 204         if (unlikely(err))
 205                 return err;
 206
 207         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 208
 209         ether_addr_copy(eth_hdr(skb)->h_source, eth_key->eth_src);
 210         ether_addr_copy(eth_hdr(skb)->h_dest, eth_key->eth_dst);
 211
 212         ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 213
 214         return 0;
 215 }
 216
 217 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 218                                 __be32 *addr, __be32 new_addr)
 219 {
 220         int transport_len = skb->len - skb_transport_offset(skb);
 221
 222         if (nh->protocol == IPPROTO_TCP) {
 223                 if (likely(transport_len >= sizeof(struct tcphdr)))
 224                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 225                                                  *addr, new_addr, 1);
 226         } else if (nh->protocol == IPPROTO_UDP) {
 227                 if (likely(transport_len >= sizeof(struct udphdr))) {
 228                         struct udphdr *uh = udp_hdr(skb);
 229
 230                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 231                                 inet_proto_csum_replace4(&uh->check, skb,
 232                                                          *addr, new_addr, 1);
 233                                 if (!uh->check)
 234                                         uh->check = CSUM_MANGLED_0;
 235                         }
 236                 }
 237         }
 238
 239         csum_replace4(&nh->check, *addr, new_addr);
 240         skb_clear_hash(skb);
 241         *addr = new_addr;
 242 }
 243
 244 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 245                                  __be32 addr[4], const __be32 new_addr[4])
 246 {
 247         int transport_len = skb->len - skb_transport_offset(skb);
 248
 249         if (l4_proto == IPPROTO_TCP) {
 250                 if (likely(transport_len >= sizeof(struct tcphdr)))
 251                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 252                                                   addr, new_addr, 1);
 253         } else if (l4_proto == IPPROTO_UDP) {
 254                 if (likely(transport_len >= sizeof(struct udphdr))) {
 255                         struct udphdr *uh = udp_hdr(skb);
 256
 257                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 258                                 inet_proto_csum_replace16(&uh->check, skb,
 259                                                           addr, new_addr, 1);
 260                                 if (!uh->check)
 261                                         uh->check = CSUM_MANGLED_0;
 262                         }
 263                 }
 264         }
 265 }
 266
 267 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 268                           __be32 addr[4], const __be32 new_addr[4],
 269                           bool recalculate_csum)
 270 {
 271         if (recalculate_csum)
 272                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 273
 274         skb_clear_hash(skb);
 275         memcpy(addr, new_addr, sizeof(__be32[4]));
 276 }
 277
 278 static void set_ipv6_tc(struct ipv6hdr *nh, u8 tc)
 279 {
 280         nh->priority = tc >> 4;
 281         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0x0F) | ((tc & 0x0F) << 4);
 282 }
 283
 284 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl)
 285 {
 286         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0xF0) | (fl & 0x000F0000) >> 16;
 287         nh->flow_lbl[1] = (fl & 0x0000FF00) >> 8;
 288         nh->flow_lbl[2] = fl & 0x000000FF;
 289 }
 290
 291 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl)
 292 {
 293         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 294         nh->ttl = new_ttl;
 295 }
 296
 297 static int set_ipv4(struct sk_buff *skb, const struct ovs_key_ipv4 *ipv4_key)
 298 {
 299         struct iphdr *nh;
 300         int err;
 301
 302         err = make_writable(skb, skb_network_offset(skb) +
 303                                  sizeof(struct iphdr));
 304         if (unlikely(err))
 305                 return err;
 306
 307         nh = ip_hdr(skb);
 308
 309         if (ipv4_key->ipv4_src != nh->saddr)
 310                 set_ip_addr(skb, nh, &nh->saddr, ipv4_key->ipv4_src);
 311
 312         if (ipv4_key->ipv4_dst != nh->daddr)
 313                 set_ip_addr(skb, nh, &nh->daddr, ipv4_key->ipv4_dst);
 314
 315         if (ipv4_key->ipv4_tos != nh->tos)
 316                 ipv4_change_dsfield(nh, 0, ipv4_key->ipv4_tos);
 317
 318         if (ipv4_key->ipv4_ttl != nh->ttl)
 319                 set_ip_ttl(skb, nh, ipv4_key->ipv4_ttl);
 320
 321         return 0;
 322 }
 323
 324 static int set_ipv6(struct sk_buff *skb, const struct ovs_key_ipv6 *ipv6_key)
 325 {
 326         struct ipv6hdr *nh;
 327         int err;
 328         __be32 *saddr;
 329         __be32 *daddr;
 330
 331         err = make_writable(skb, skb_network_offset(skb) +
 332                             sizeof(struct ipv6hdr));
 333         if (unlikely(err))
 334                 return err;
 335
 336         nh = ipv6_hdr(skb);
 337         saddr = (__be32 *)&nh->saddr;
 338         daddr = (__be32 *)&nh->daddr;
 339
 340         if (memcmp(ipv6_key->ipv6_src, saddr, sizeof(ipv6_key->ipv6_src)))
 341                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, saddr,
 342                               ipv6_key->ipv6_src, true);
 343
 344         if (memcmp(ipv6_key->ipv6_dst, daddr, sizeof(ipv6_key->ipv6_dst))) {
 345                 unsigned int offset = 0;
 346                 int flags = IP6_FH_F_SKIP_RH;
 347                 bool recalc_csum = true;
 348
 349                 if (ipv6_ext_hdr(nh->nexthdr))
 350                         recalc_csum = ipv6_find_hdr(skb, &offset,
 351                                                     NEXTHDR_ROUTING, NULL,
 352                                                     &flags) != NEXTHDR_ROUTING;
 353
 354                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, daddr,
 355                               ipv6_key->ipv6_dst, recalc_csum);
 356         }
 357
 358         set_ipv6_tc(nh, ipv6_key->ipv6_tclass);
 359         set_ipv6_fl(nh, ntohl(ipv6_key->ipv6_label));
 360         nh->hop_limit = ipv6_key->ipv6_hlimit;
 361
 362         return 0;
 363 }
 364
 365 /* Must follow make_writable() since that can move the skb data. */
 366 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 367                          __be16 new_port, __sum16 *check)
 368 {
 369         inet_proto_csum_replace2(check, skb, *port, new_port, 0);
 370         *port = new_port;
 371         skb_clear_hash(skb);
 372 }
 373
 374 static void set_udp_port(struct sk_buff *skb, __be16 *port, __be16 new_port)
 375 {
 376         struct udphdr *uh = udp_hdr(skb);
 377
 378         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 379                 set_tp_port(skb, port, new_port, &uh->check);
 380
 381                 if (!uh->check)
 382                         uh->check = CSUM_MANGLED_0;
 383         } else {
 384                 *port = new_port;
 385                 skb_clear_hash(skb);
 386         }
 387 }
 388
 389 static int set_udp(struct sk_buff *skb, const struct ovs_key_udp *udp_port_key)
 390 {
 391         struct udphdr *uh;
 392         int err;
 393
 394         err = make_writable(skb, skb_transport_offset(skb) +
 395                                  sizeof(struct udphdr));
 396         if (unlikely(err))
 397                 return err;
 398
 399         uh = udp_hdr(skb);
 400         if (udp_port_key->udp_src != uh->source)
 401                 set_udp_port(skb, &uh->source, udp_port_key->udp_src);
 402
 403         if (udp_port_key->udp_dst != uh->dest)
 404                 set_udp_port(skb, &uh->dest, udp_port_key->udp_dst);
 405
 406         return 0;
 407 }
 408
 409 static int set_tcp(struct sk_buff *skb, const struct ovs_key_tcp *tcp_port_key)
 410 {
 411         struct tcphdr *th;
 412         int err;
 413
 414         err = make_writable(skb, skb_transport_offset(skb) +
 415                                  sizeof(struct tcphdr));
 416         if (unlikely(err))
 417                 return err;
 418
 419         th = tcp_hdr(skb);
 420         if (tcp_port_key->tcp_src != th->source)
 421                 set_tp_port(skb, &th->source, tcp_port_key->tcp_src, &th->check);
 422
 423         if (tcp_port_key->tcp_dst != th->dest)
 424                 set_tp_port(skb, &th->dest, tcp_port_key->tcp_dst, &th->check);
 425
 426         return 0;
 427 }
 428
 429 static int set_sctp(struct sk_buff *skb,
 430                      const struct ovs_key_sctp *sctp_port_key)
 431 {
 432         struct sctphdr *sh;
 433         int err;
 434         unsigned int sctphoff = skb_transport_offset(skb);
 435
 436         err = make_writable(skb, sctphoff + sizeof(struct sctphdr));
 437         if (unlikely(err))
 438                 return err;
 439
 440         sh = sctp_hdr(skb);
 441         if (sctp_port_key->sctp_src != sh->source ||
 442             sctp_port_key->sctp_dst != sh->dest) {
 443                 __le32 old_correct_csum, new_csum, old_csum;
 444
 445                 old_csum = sh->checksum;
 446                 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 447
 448                 sh->source = sctp_port_key->sctp_src;
 449                 sh->dest = sctp_port_key->sctp_dst;
 450
 451                 new_csum = sctp_compute_cksum(skb, sctphoff);
 452
 453                 /* Carry any checksum errors through. */
 454                 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 455
 456                 skb_clear_hash(skb);
 457         }
 458
 459         return 0;
 460 }
 461
 462 static int do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
 463 {
 464         struct vport *vport;
 465
 466         if (unlikely(!skb))
 467                 return -ENOMEM;
 468
 469         vport = ovs_vport_rcu(dp, out_port);
 470         if (unlikely(!vport)) {
 471                 kfree_skb(skb);
 472                 return -ENODEV;
 473         }
 474
 475         ovs_vport_send(vport, skb);
 476         return 0;
 477 }
 478
 479 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 480                             struct sw_flow_key *key, const struct nlattr *attr)
 481 {
 482         struct dp_upcall_info upcall;
 483         const struct nlattr *a;
 484         int rem;
 485
 486         upcall.cmd = OVS_PACKET_CMD_ACTION;
 487         upcall.key = key;
 488         upcall.userdata = NULL;
 489         upcall.portid = 0;
 490
 491         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 492                  a = nla_next(a, &rem)) {
 493                 switch (nla_type(a)) {
 494                 case OVS_USERSPACE_ATTR_USERDATA:
 495                         upcall.userdata = a;
 496                         break;
 497
 498                 case OVS_USERSPACE_ATTR_PID:
 499                         upcall.portid = nla_get_u32(a);
 500                         break;
 501                 }
 502         }
 503
 504         return ovs_dp_upcall(dp, skb, &upcall);
 505 }
 506
 507 static bool last_action(const struct nlattr *a, int rem)
 508 {
 509         return a->nla_len == rem;
 510 }
 511
 512 static int sample(struct datapath *dp, struct sk_buff *skb,
 513                   struct sw_flow_key *key, const struct nlattr *attr)
 514 {
 515         const struct nlattr *acts_list = NULL;
 516         const struct nlattr *a;
 517         int rem;
 518
 519         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 520                  a = nla_next(a, &rem)) {
 521                 switch (nla_type(a)) {
 522                 case OVS_SAMPLE_ATTR_PROBABILITY:
 523                         if (prandom_u32() >= nla_get_u32(a))
 524                                 return 0;
 525                         break;
 526
 527                 case OVS_SAMPLE_ATTR_ACTIONS:
 528                         acts_list = a;
 529                         break;
 530                 }
 531         }
 532
 533         rem = nla_len(acts_list);
 534         a = nla_data(acts_list);
 535
 536         /* Actions list is empty, do nothing */
 537         if (unlikely(!rem))
 538                 return 0;
 539
 540         /* The only known usage of sample action is having a single user-space
 541          * action. Treat this usage as a special case.
 542          * The output_userspace() should clone the skb to be sent to the
 543          * user space. This skb will be consumed by its caller.
 544          */
 545         if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
 546                    last_action(a, rem)))
 547                 return output_userspace(dp, skb, key, a);
 548
 549         skb = skb_clone(skb, GFP_ATOMIC);
 550         if (!skb)
 551                 /* Skip the sample action when out of memory. */
 552                 return 0;
 553
 554         if (!add_deferred_actions(skb, key, a)) {
 555                 if (net_ratelimit())
 556                         pr_warn("%s: deferred actions limit reached, dropping sample action\n",
 557                                 ovs_dp_name(dp));
 558
 559                 kfree_skb(skb);
 560         }
 561         return 0;
 562 }
 563
 564 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
 565                          const struct nlattr *attr)
 566 {
 567         struct ovs_action_hash *hash_act = nla_data(attr);
 568         u32 hash = 0;
 569
 570         /* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
 571         hash = skb_get_hash(skb);
 572         hash = jhash_1word(hash, hash_act->hash_basis);
 573         if (!hash)
 574                 hash = 0x1;
 575
 576         key->ovs_flow_hash = hash;
 577 }
 578
 579 static int execute_set_action(struct sk_buff *skb,
 580                                  const struct nlattr *nested_attr)
 581 {
 582         int err = 0;
 583
 584         switch (nla_type(nested_attr)) {
 585         case OVS_KEY_ATTR_PRIORITY:
 586                 skb->priority = nla_get_u32(nested_attr);
 587                 break;
 588
 589         case OVS_KEY_ATTR_SKB_MARK:
 590                 skb->mark = nla_get_u32(nested_attr);
 591                 break;
 592
 593         case OVS_KEY_ATTR_TUNNEL_INFO:
 594                 OVS_CB(skb)->egress_tun_info = nla_data(nested_attr);
 595                 break;
 596
 597         case OVS_KEY_ATTR_ETHERNET:
 598                 err = set_eth_addr(skb, nla_data(nested_attr));
 599                 break;
 600
 601         case OVS_KEY_ATTR_IPV4:
 602                 err = set_ipv4(skb, nla_data(nested_attr));
 603                 break;
 604
 605         case OVS_KEY_ATTR_IPV6:
 606                 err = set_ipv6(skb, nla_data(nested_attr));
 607                 break;
 608
 609         case OVS_KEY_ATTR_TCP:
 610                 err = set_tcp(skb, nla_data(nested_attr));
 611                 break;
 612
 613         case OVS_KEY_ATTR_UDP:
 614                 err = set_udp(skb, nla_data(nested_attr));
 615                 break;
 616
 617         case OVS_KEY_ATTR_SCTP:
 618                 err = set_sctp(skb, nla_data(nested_attr));
 619                 break;
 620         }
 621
 622         return err;
 623 }
 624
 625 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
 626                           struct sw_flow_key *key,
 627                           const struct nlattr *a, int rem)
 628 {
 629         struct deferred_action *da;
 630         int err;
 631
 632         err = ovs_flow_key_update(skb, key);
 633         if (err)
 634                 return err;
 635
 636         if (!last_action(a, rem)) {
 637                 /* Recirc action is the not the last action
 638                  * of the action list, need to clone the skb.
 639                  */
 640                 skb = skb_clone(skb, GFP_ATOMIC);
 641
 642                 /* Skip the recirc action when out of memory, but
 643                  * continue on with the rest of the action list.
 644                  */
 645                 if (!skb)
 646                         return 0;
 647         }
 648
 649         da = add_deferred_actions(skb, key, NULL);
 650         if (da) {
 651                 da->pkt_key.recirc_id = nla_get_u32(a);
 652         } else {
 653                 kfree_skb(skb);
 654
 655                 if (net_ratelimit())
 656                         pr_warn("%s: deferred action limit reached, drop recirc action\n",
 657                                 ovs_dp_name(dp));
 658         }
 659
 660         return 0;
 661 }
 662
 663 /* Execute a list of actions against 'skb'. */
 664 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 665                               struct sw_flow_key *key,
 666                               const struct nlattr *attr, int len)
 667 {
 668         /* Every output action needs a separate clone of 'skb', but the common
 669          * case is just a single output action, so that doing a clone and
 670          * then freeing the original skbuff is wasteful.  So the following code
 671          * is slightly obscure just to avoid that. */
 672         int prev_port = -1;
 673         const struct nlattr *a;
 674         int rem;
 675
 676         for (a = attr, rem = len; rem > 0;
 677              a = nla_next(a, &rem)) {
 678                 int err = 0;
 679
 680                 if (prev_port != -1) {
 681                         do_output(dp, skb_clone(skb, GFP_ATOMIC), prev_port);
 682                         prev_port = -1;
 683                 }
 684
 685                 switch (nla_type(a)) {
 686                 case OVS_ACTION_ATTR_OUTPUT:
 687                         prev_port = nla_get_u32(a);
 688                         break;
 689
 690                 case OVS_ACTION_ATTR_USERSPACE:
 691                         output_userspace(dp, skb, key, a);
 692                         break;
 693
 694                 case OVS_ACTION_ATTR_HASH:
 695                         execute_hash(skb, key, a);
 696                         break;
 697
 698                 case OVS_ACTION_ATTR_PUSH_VLAN:
 699                         err = push_vlan(skb, nla_data(a));
 700                         if (unlikely(err)) /* skb already freed. */
 701                                 return err;
 702                         break;
 703
 704                 case OVS_ACTION_ATTR_POP_VLAN:
 705                         err = pop_vlan(skb);
 706                         break;
 707
 708                 case OVS_ACTION_ATTR_RECIRC:
 709                         err = execute_recirc(dp, skb, key, a, rem);
 710                         if (last_action(a, rem)) {
 711                                 /* If this is the last action, the skb has
 712                                  * been consumed or freed.
 713                                  * Return immediately.
 714                                  */
 715                                 return err;
 716                         }
 717                         break;
 718
 719                 case OVS_ACTION_ATTR_SET:
 720                         err = execute_set_action(skb, nla_data(a));
 721                         break;
 722
 723                 case OVS_ACTION_ATTR_SAMPLE:
 724                         err = sample(dp, skb, key, a);
 725                         if (unlikely(err)) /* skb already freed. */
 726                                 return err;
 727                         break;
 728                 }
 729
 730                 if (unlikely(err)) {
 731                         kfree_skb(skb);
 732                         return err;
 733                 }
 734         }
 735
 736         if (prev_port != -1)
 737                 do_output(dp, skb, prev_port);
 738         else
 739                 consume_skb(skb);
 740
 741         return 0;
 742 }
 743
 744 static void process_deferred_actions(struct datapath *dp)
 745 {
 746         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
 747
 748         /* Do not touch the FIFO in case there is no deferred actions. */
 749         if (action_fifo_is_empty(fifo))
 750                 return;
 751
 752         /* Finishing executing all deferred actions. */
 753         do {
 754                 struct deferred_action *da = action_fifo_get(fifo);
 755                 struct sk_buff *skb = da->skb;
 756                 struct sw_flow_key *key = &da->pkt_key;
 757                 const struct nlattr *actions = da->actions;
 758
 759                 if (actions)
 760                         do_execute_actions(dp, skb, key, actions,
 761                                            nla_len(actions));
 762                 else
 763                         ovs_dp_process_packet(skb, key);
 764         } while (!action_fifo_is_empty(fifo));
 765
 766         /* Reset FIFO for the next packet.  */
 767         action_fifo_init(fifo);
 768 }
 769
 770 /* Execute a list of actions against 'skb'. */
 771 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
 772                         struct sw_flow_key *key)
 773 {
 774         int level = this_cpu_read(exec_actions_level);
 775         struct sw_flow_actions *acts;
 776         int err;
 777
 778         acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
 779
 780         this_cpu_inc(exec_actions_level);
 781         OVS_CB(skb)->egress_tun_info = NULL;
 782         err = do_execute_actions(dp, skb, key,
 783                                  acts->actions, acts->actions_len);
 784
 785         if (!level)
 786                 process_deferred_actions(dp);
 787
 788         this_cpu_dec(exec_actions_level);
 789         return err;
 790 }
 791
 792 int action_fifos_init(void)
 793 {
 794         action_fifos = alloc_percpu(struct action_fifo);
 795         if (!action_fifos)
 796                 return -ENOMEM;
 797
 798         return 0;
 799 }
 800
 801 void action_fifos_exit(void)
 802 {
 803         free_percpu(action_fifos);
 804 }