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                 skb = vlan_insert_tag_set_proto(skb, skb->vlan_proto,
 188                                                 current_tag);
 189                 if (!skb)
 190                         return -ENOMEM;
 191
 192                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 193                         skb->csum = csum_add(skb->csum, csum_partial(skb->data
 194                                         + (2 * ETH_ALEN), VLAN_HLEN, 0));
 195
 196         }
 197         __vlan_hwaccel_put_tag(skb, vlan->vlan_tpid, ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
 198         return 0;
 199 }
 200
 201 static int set_eth_addr(struct sk_buff *skb,
 202                         const struct ovs_key_ethernet *eth_key)
 203 {
 204         int err;
 205         err = make_writable(skb, ETH_HLEN);
 206         if (unlikely(err))
 207                 return err;
 208
 209         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 210
 211         ether_addr_copy(eth_hdr(skb)->h_source, eth_key->eth_src);
 212         ether_addr_copy(eth_hdr(skb)->h_dest, eth_key->eth_dst);
 213
 214         ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 215
 216         return 0;
 217 }
 218
 219 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 220                                 __be32 *addr, __be32 new_addr)
 221 {
 222         int transport_len = skb->len - skb_transport_offset(skb);
 223
 224         if (nh->protocol == IPPROTO_TCP) {
 225                 if (likely(transport_len >= sizeof(struct tcphdr)))
 226                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 227                                                  *addr, new_addr, 1);
 228         } else if (nh->protocol == IPPROTO_UDP) {
 229                 if (likely(transport_len >= sizeof(struct udphdr))) {
 230                         struct udphdr *uh = udp_hdr(skb);
 231
 232                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 233                                 inet_proto_csum_replace4(&uh->check, skb,
 234                                                          *addr, new_addr, 1);
 235                                 if (!uh->check)
 236                                         uh->check = CSUM_MANGLED_0;
 237                         }
 238                 }
 239         }
 240
 241         csum_replace4(&nh->check, *addr, new_addr);
 242         skb_clear_hash(skb);
 243         *addr = new_addr;
 244 }
 245
 246 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 247                                  __be32 addr[4], const __be32 new_addr[4])
 248 {
 249         int transport_len = skb->len - skb_transport_offset(skb);
 250
 251         if (l4_proto == NEXTHDR_TCP) {
 252                 if (likely(transport_len >= sizeof(struct tcphdr)))
 253                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 254                                                   addr, new_addr, 1);
 255         } else if (l4_proto == NEXTHDR_UDP) {
 256                 if (likely(transport_len >= sizeof(struct udphdr))) {
 257                         struct udphdr *uh = udp_hdr(skb);
 258
 259                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 260                                 inet_proto_csum_replace16(&uh->check, skb,
 261                                                           addr, new_addr, 1);
 262                                 if (!uh->check)
 263                                         uh->check = CSUM_MANGLED_0;
 264                         }
 265                 }
 266         } else if (l4_proto == NEXTHDR_ICMP) {
 267                 if (likely(transport_len >= sizeof(struct icmp6hdr)))
 268                         inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 269                                                   skb, addr, new_addr, 1);
 270         }
 271 }
 272
 273 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 274                           __be32 addr[4], const __be32 new_addr[4],
 275                           bool recalculate_csum)
 276 {
 277         if (recalculate_csum)
 278                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 279
 280         skb_clear_hash(skb);
 281         memcpy(addr, new_addr, sizeof(__be32[4]));
 282 }
 283
 284 static void set_ipv6_tc(struct ipv6hdr *nh, u8 tc)
 285 {
 286         nh->priority = tc >> 4;
 287         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0x0F) | ((tc & 0x0F) << 4);
 288 }
 289
 290 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl)
 291 {
 292         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0xF0) | (fl & 0x000F0000) >> 16;
 293         nh->flow_lbl[1] = (fl & 0x0000FF00) >> 8;
 294         nh->flow_lbl[2] = fl & 0x000000FF;
 295 }
 296
 297 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl)
 298 {
 299         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 300         nh->ttl = new_ttl;
 301 }
 302
 303 static int set_ipv4(struct sk_buff *skb, const struct ovs_key_ipv4 *ipv4_key)
 304 {
 305         struct iphdr *nh;
 306         int err;
 307
 308         err = make_writable(skb, skb_network_offset(skb) +
 309                                  sizeof(struct iphdr));
 310         if (unlikely(err))
 311                 return err;
 312
 313         nh = ip_hdr(skb);
 314
 315         if (ipv4_key->ipv4_src != nh->saddr)
 316                 set_ip_addr(skb, nh, &nh->saddr, ipv4_key->ipv4_src);
 317
 318         if (ipv4_key->ipv4_dst != nh->daddr)
 319                 set_ip_addr(skb, nh, &nh->daddr, ipv4_key->ipv4_dst);
 320
 321         if (ipv4_key->ipv4_tos != nh->tos)
 322                 ipv4_change_dsfield(nh, 0, ipv4_key->ipv4_tos);
 323
 324         if (ipv4_key->ipv4_ttl != nh->ttl)
 325                 set_ip_ttl(skb, nh, ipv4_key->ipv4_ttl);
 326
 327         return 0;
 328 }
 329
 330 static int set_ipv6(struct sk_buff *skb, const struct ovs_key_ipv6 *ipv6_key)
 331 {
 332         struct ipv6hdr *nh;
 333         int err;
 334         __be32 *saddr;
 335         __be32 *daddr;
 336
 337         err = make_writable(skb, skb_network_offset(skb) +
 338                             sizeof(struct ipv6hdr));
 339         if (unlikely(err))
 340                 return err;
 341
 342         nh = ipv6_hdr(skb);
 343         saddr = (__be32 *)&nh->saddr;
 344         daddr = (__be32 *)&nh->daddr;
 345
 346         if (memcmp(ipv6_key->ipv6_src, saddr, sizeof(ipv6_key->ipv6_src)))
 347                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, saddr,
 348                               ipv6_key->ipv6_src, true);
 349
 350         if (memcmp(ipv6_key->ipv6_dst, daddr, sizeof(ipv6_key->ipv6_dst))) {
 351                 unsigned int offset = 0;
 352                 int flags = IP6_FH_F_SKIP_RH;
 353                 bool recalc_csum = true;
 354
 355                 if (ipv6_ext_hdr(nh->nexthdr))
 356                         recalc_csum = ipv6_find_hdr(skb, &offset,
 357                                                     NEXTHDR_ROUTING, NULL,
 358                                                     &flags) != NEXTHDR_ROUTING;
 359
 360                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, daddr,
 361                               ipv6_key->ipv6_dst, recalc_csum);
 362         }
 363
 364         set_ipv6_tc(nh, ipv6_key->ipv6_tclass);
 365         set_ipv6_fl(nh, ntohl(ipv6_key->ipv6_label));
 366         nh->hop_limit = ipv6_key->ipv6_hlimit;
 367
 368         return 0;
 369 }
 370
 371 /* Must follow make_writable() since that can move the skb data. */
 372 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 373                          __be16 new_port, __sum16 *check)
 374 {
 375         inet_proto_csum_replace2(check, skb, *port, new_port, 0);
 376         *port = new_port;
 377         skb_clear_hash(skb);
 378 }
 379
 380 static void set_udp_port(struct sk_buff *skb, __be16 *port, __be16 new_port)
 381 {
 382         struct udphdr *uh = udp_hdr(skb);
 383
 384         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 385                 set_tp_port(skb, port, new_port, &uh->check);
 386
 387                 if (!uh->check)
 388                         uh->check = CSUM_MANGLED_0;
 389         } else {
 390                 *port = new_port;
 391                 skb_clear_hash(skb);
 392         }
 393 }
 394
 395 static int set_udp(struct sk_buff *skb, const struct ovs_key_udp *udp_port_key)
 396 {
 397         struct udphdr *uh;
 398         int err;
 399
 400         err = make_writable(skb, skb_transport_offset(skb) +
 401                                  sizeof(struct udphdr));
 402         if (unlikely(err))
 403                 return err;
 404
 405         uh = udp_hdr(skb);
 406         if (udp_port_key->udp_src != uh->source)
 407                 set_udp_port(skb, &uh->source, udp_port_key->udp_src);
 408
 409         if (udp_port_key->udp_dst != uh->dest)
 410                 set_udp_port(skb, &uh->dest, udp_port_key->udp_dst);
 411
 412         return 0;
 413 }
 414
 415 static int set_tcp(struct sk_buff *skb, const struct ovs_key_tcp *tcp_port_key)
 416 {
 417         struct tcphdr *th;
 418         int err;
 419
 420         err = make_writable(skb, skb_transport_offset(skb) +
 421                                  sizeof(struct tcphdr));
 422         if (unlikely(err))
 423                 return err;
 424
 425         th = tcp_hdr(skb);
 426         if (tcp_port_key->tcp_src != th->source)
 427                 set_tp_port(skb, &th->source, tcp_port_key->tcp_src, &th->check);
 428
 429         if (tcp_port_key->tcp_dst != th->dest)
 430                 set_tp_port(skb, &th->dest, tcp_port_key->tcp_dst, &th->check);
 431
 432         return 0;
 433 }
 434
 435 static int set_sctp(struct sk_buff *skb,
 436                      const struct ovs_key_sctp *sctp_port_key)
 437 {
 438         struct sctphdr *sh;
 439         int err;
 440         unsigned int sctphoff = skb_transport_offset(skb);
 441
 442         err = make_writable(skb, sctphoff + sizeof(struct sctphdr));
 443         if (unlikely(err))
 444                 return err;
 445
 446         sh = sctp_hdr(skb);
 447         if (sctp_port_key->sctp_src != sh->source ||
 448             sctp_port_key->sctp_dst != sh->dest) {
 449                 __le32 old_correct_csum, new_csum, old_csum;
 450
 451                 old_csum = sh->checksum;
 452                 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 453
 454                 sh->source = sctp_port_key->sctp_src;
 455                 sh->dest = sctp_port_key->sctp_dst;
 456
 457                 new_csum = sctp_compute_cksum(skb, sctphoff);
 458
 459                 /* Carry any checksum errors through. */
 460                 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 461
 462                 skb_clear_hash(skb);
 463         }
 464
 465         return 0;
 466 }
 467
 468 static int do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
 469 {
 470         struct vport *vport;
 471
 472         if (unlikely(!skb))
 473                 return -ENOMEM;
 474
 475         vport = ovs_vport_rcu(dp, out_port);
 476         if (unlikely(!vport)) {
 477                 kfree_skb(skb);
 478                 return -ENODEV;
 479         }
 480
 481         ovs_vport_send(vport, skb);
 482         return 0;
 483 }
 484
 485 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 486                             struct sw_flow_key *key, const struct nlattr *attr)
 487 {
 488         struct dp_upcall_info upcall;
 489         const struct nlattr *a;
 490         int rem;
 491
 492         upcall.cmd = OVS_PACKET_CMD_ACTION;
 493         upcall.key = key;
 494         upcall.userdata = NULL;
 495         upcall.portid = 0;
 496
 497         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 498                  a = nla_next(a, &rem)) {
 499                 switch (nla_type(a)) {
 500                 case OVS_USERSPACE_ATTR_USERDATA:
 501                         upcall.userdata = a;
 502                         break;
 503
 504                 case OVS_USERSPACE_ATTR_PID:
 505                         upcall.portid = nla_get_u32(a);
 506                         break;
 507                 }
 508         }
 509
 510         return ovs_dp_upcall(dp, skb, &upcall);
 511 }
 512
 513 static bool last_action(const struct nlattr *a, int rem)
 514 {
 515         return a->nla_len == rem;
 516 }
 517
 518 static int sample(struct datapath *dp, struct sk_buff *skb,
 519                   struct sw_flow_key *key, const struct nlattr *attr)
 520 {
 521         const struct nlattr *acts_list = NULL;
 522         const struct nlattr *a;
 523         int rem;
 524
 525         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 526                  a = nla_next(a, &rem)) {
 527                 switch (nla_type(a)) {
 528                 case OVS_SAMPLE_ATTR_PROBABILITY:
 529                         if (prandom_u32() >= nla_get_u32(a))
 530                                 return 0;
 531                         break;
 532
 533                 case OVS_SAMPLE_ATTR_ACTIONS:
 534                         acts_list = a;
 535                         break;
 536                 }
 537         }
 538
 539         rem = nla_len(acts_list);
 540         a = nla_data(acts_list);
 541
 542         /* Actions list is empty, do nothing */
 543         if (unlikely(!rem))
 544                 return 0;
 545
 546         /* The only known usage of sample action is having a single user-space
 547          * action. Treat this usage as a special case.
 548          * The output_userspace() should clone the skb to be sent to the
 549          * user space. This skb will be consumed by its caller.
 550          */
 551         if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
 552                    last_action(a, rem)))
 553                 return output_userspace(dp, skb, key, a);
 554
 555         skb = skb_clone(skb, GFP_ATOMIC);
 556         if (!skb)
 557                 /* Skip the sample action when out of memory. */
 558                 return 0;
 559
 560         if (!add_deferred_actions(skb, key, a)) {
 561                 if (net_ratelimit())
 562                         pr_warn("%s: deferred actions limit reached, dropping sample action\n",
 563                                 ovs_dp_name(dp));
 564
 565                 kfree_skb(skb);
 566         }
 567         return 0;
 568 }
 569
 570 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
 571                          const struct nlattr *attr)
 572 {
 573         struct ovs_action_hash *hash_act = nla_data(attr);
 574         u32 hash = 0;
 575
 576         /* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
 577         hash = skb_get_hash(skb);
 578         hash = jhash_1word(hash, hash_act->hash_basis);
 579         if (!hash)
 580                 hash = 0x1;
 581
 582         key->ovs_flow_hash = hash;
 583 }
 584
 585 static int execute_set_action(struct sk_buff *skb,
 586                                  const struct nlattr *nested_attr)
 587 {
 588         int err = 0;
 589
 590         switch (nla_type(nested_attr)) {
 591         case OVS_KEY_ATTR_PRIORITY:
 592                 skb->priority = nla_get_u32(nested_attr);
 593                 break;
 594
 595         case OVS_KEY_ATTR_SKB_MARK:
 596                 skb->mark = nla_get_u32(nested_attr);
 597                 break;
 598
 599         case OVS_KEY_ATTR_TUNNEL_INFO:
 600                 OVS_CB(skb)->egress_tun_info = nla_data(nested_attr);
 601                 break;
 602
 603         case OVS_KEY_ATTR_ETHERNET:
 604                 err = set_eth_addr(skb, nla_data(nested_attr));
 605                 break;
 606
 607         case OVS_KEY_ATTR_IPV4:
 608                 err = set_ipv4(skb, nla_data(nested_attr));
 609                 break;
 610
 611         case OVS_KEY_ATTR_IPV6:
 612                 err = set_ipv6(skb, nla_data(nested_attr));
 613                 break;
 614
 615         case OVS_KEY_ATTR_TCP:
 616                 err = set_tcp(skb, nla_data(nested_attr));
 617                 break;
 618
 619         case OVS_KEY_ATTR_UDP:
 620                 err = set_udp(skb, nla_data(nested_attr));
 621                 break;
 622
 623         case OVS_KEY_ATTR_SCTP:
 624                 err = set_sctp(skb, nla_data(nested_attr));
 625                 break;
 626         }
 627
 628         return err;
 629 }
 630
 631 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
 632                           struct sw_flow_key *key,
 633                           const struct nlattr *a, int rem)
 634 {
 635         struct deferred_action *da;
 636         int err;
 637
 638         err = ovs_flow_key_update(skb, key);
 639         if (err)
 640                 return err;
 641
 642         if (!last_action(a, rem)) {
 643                 /* Recirc action is the not the last action
 644                  * of the action list, need to clone the skb.
 645                  */
 646                 skb = skb_clone(skb, GFP_ATOMIC);
 647
 648                 /* Skip the recirc action when out of memory, but
 649                  * continue on with the rest of the action list.
 650                  */
 651                 if (!skb)
 652                         return 0;
 653         }
 654
 655         da = add_deferred_actions(skb, key, NULL);
 656         if (da) {
 657                 da->pkt_key.recirc_id = nla_get_u32(a);
 658         } else {
 659                 kfree_skb(skb);
 660
 661                 if (net_ratelimit())
 662                         pr_warn("%s: deferred action limit reached, drop recirc action\n",
 663                                 ovs_dp_name(dp));
 664         }
 665
 666         return 0;
 667 }
 668
 669 /* Execute a list of actions against 'skb'. */
 670 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 671                               struct sw_flow_key *key,
 672                               const struct nlattr *attr, int len)
 673 {
 674         /* Every output action needs a separate clone of 'skb', but the common
 675          * case is just a single output action, so that doing a clone and
 676          * then freeing the original skbuff is wasteful.  So the following code
 677          * is slightly obscure just to avoid that. */
 678         int prev_port = -1;
 679         const struct nlattr *a;
 680         int rem;
 681
 682         for (a = attr, rem = len; rem > 0;
 683              a = nla_next(a, &rem)) {
 684                 int err = 0;
 685
 686                 if (prev_port != -1) {
 687                         do_output(dp, skb_clone(skb, GFP_ATOMIC), prev_port);
 688                         prev_port = -1;
 689                 }
 690
 691                 switch (nla_type(a)) {
 692                 case OVS_ACTION_ATTR_OUTPUT:
 693                         prev_port = nla_get_u32(a);
 694                         break;
 695
 696                 case OVS_ACTION_ATTR_USERSPACE:
 697                         output_userspace(dp, skb, key, a);
 698                         break;
 699
 700                 case OVS_ACTION_ATTR_HASH:
 701                         execute_hash(skb, key, a);
 702                         break;
 703
 704                 case OVS_ACTION_ATTR_PUSH_VLAN:
 705                         err = push_vlan(skb, nla_data(a));
 706                         if (unlikely(err)) /* skb already freed. */
 707                                 return err;
 708                         break;
 709
 710                 case OVS_ACTION_ATTR_POP_VLAN:
 711                         err = pop_vlan(skb);
 712                         break;
 713
 714                 case OVS_ACTION_ATTR_RECIRC:
 715                         err = execute_recirc(dp, skb, key, a, rem);
 716                         if (last_action(a, rem)) {
 717                                 /* If this is the last action, the skb has
 718                                  * been consumed or freed.
 719                                  * Return immediately.
 720                                  */
 721                                 return err;
 722                         }
 723                         break;
 724
 725                 case OVS_ACTION_ATTR_SET:
 726                         err = execute_set_action(skb, nla_data(a));
 727                         break;
 728
 729                 case OVS_ACTION_ATTR_SAMPLE:
 730                         err = sample(dp, skb, key, a);
 731                         break;
 732                 }
 733
 734                 if (unlikely(err)) {
 735                         kfree_skb(skb);
 736                         return err;
 737                 }
 738         }
 739
 740         if (prev_port != -1)
 741                 do_output(dp, skb, prev_port);
 742         else
 743                 consume_skb(skb);
 744
 745         return 0;
 746 }
 747
 748 static void process_deferred_actions(struct datapath *dp)
 749 {
 750         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
 751
 752         /* Do not touch the FIFO in case there is no deferred actions. */
 753         if (action_fifo_is_empty(fifo))
 754                 return;
 755
 756         /* Finishing executing all deferred actions. */
 757         do {
 758                 struct deferred_action *da = action_fifo_get(fifo);
 759                 struct sk_buff *skb = da->skb;
 760                 struct sw_flow_key *key = &da->pkt_key;
 761                 const struct nlattr *actions = da->actions;
 762
 763                 if (actions)
 764                         do_execute_actions(dp, skb, key, actions,
 765                                            nla_len(actions));
 766                 else
 767                         ovs_dp_process_packet(skb, key);
 768         } while (!action_fifo_is_empty(fifo));
 769
 770         /* Reset FIFO for the next packet.  */
 771         action_fifo_init(fifo);
 772 }
 773
 774 /* Execute a list of actions against 'skb'. */
 775 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
 776                         struct sw_flow_key *key)
 777 {
 778         int level = this_cpu_read(exec_actions_level);
 779         struct sw_flow_actions *acts;
 780         int err;
 781
 782         acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
 783
 784         this_cpu_inc(exec_actions_level);
 785         OVS_CB(skb)->egress_tun_info = NULL;
 786         err = do_execute_actions(dp, skb, key,
 787                                  acts->actions, acts->actions_len);
 788
 789         if (!level)
 790                 process_deferred_actions(dp);
 791
 792         this_cpu_dec(exec_actions_level);
 793         return err;
 794 }
 795
 796 int action_fifos_init(void)
 797 {
 798         action_fifos = alloc_percpu(struct action_fifo);
 799         if (!action_fifos)
 800                 return -ENOMEM;
 801
 802         return 0;
 803 }
 804
 805 void action_fifos_exit(void)
 806 {
 807         free_percpu(action_fifos);
 808 }