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
  32 #include <net/ip.h>
  33 #include <net/ipv6.h>
  34 #include <net/checksum.h>
  35 #include <net/dsfield.h>
  36 #include <net/mpls.h>
  37 #include <net/sctp/checksum.h>
  38
  39 #include "datapath.h"
  40 #include "flow.h"
  41 #include "vport.h"
  42
  43 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
  44                               struct sw_flow_key *key,
  45                               const struct nlattr *attr, int len);
  46
  47 struct deferred_action {
  48         struct sk_buff *skb;
  49         const struct nlattr *actions;
  50
  51         /* Store pkt_key clone when creating deferred action. */
  52         struct sw_flow_key pkt_key;
  53 };
  54
  55 #define DEFERRED_ACTION_FIFO_SIZE 10
  56 struct action_fifo {
  57         int head;
  58         int tail;
  59         /* Deferred action fifo queue storage. */
  60         struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
  61 };
  62
  63 static struct action_fifo __percpu *action_fifos;
  64 static DEFINE_PER_CPU(int, exec_actions_level);
  65
  66 static void action_fifo_init(struct action_fifo *fifo)
  67 {
  68         fifo->head = 0;
  69         fifo->tail = 0;
  70 }
  71
  72 static bool action_fifo_is_empty(const struct action_fifo *fifo)
  73 {
  74         return (fifo->head == fifo->tail);
  75 }
  76
  77 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
  78 {
  79         if (action_fifo_is_empty(fifo))
  80                 return NULL;
  81
  82         return &fifo->fifo[fifo->tail++];
  83 }
  84
  85 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
  86 {
  87         if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
  88                 return NULL;
  89
  90         return &fifo->fifo[fifo->head++];
  91 }
  92
  93 /* Return true if fifo is not full */
  94 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
  95                                                     const struct sw_flow_key *key,
  96                                                     const struct nlattr *attr)
  97 {
  98         struct action_fifo *fifo;
  99         struct deferred_action *da;
 100
 101         fifo = this_cpu_ptr(action_fifos);
 102         da = action_fifo_put(fifo);
 103         if (da) {
 104                 da->skb = skb;
 105                 da->actions = attr;
 106                 da->pkt_key = *key;
 107         }
 108
 109         return da;
 110 }
 111
 112 static void invalidate_flow_key(struct sw_flow_key *key)
 113 {
 114         key->eth.type = htons(0);
 115 }
 116
 117 static bool is_flow_key_valid(const struct sw_flow_key *key)
 118 {
 119         return !!key->eth.type;
 120 }
 121
 122 static int make_writable(struct sk_buff *skb, int write_len)
 123 {
 124         if (!pskb_may_pull(skb, write_len))
 125                 return -ENOMEM;
 126
 127         if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
 128                 return 0;
 129
 130         return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
 131 }
 132
 133 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 134                      const struct ovs_action_push_mpls *mpls)
 135 {
 136         __be32 *new_mpls_lse;
 137         struct ethhdr *hdr;
 138
 139         /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
 140         if (skb->encapsulation)
 141                 return -ENOTSUPP;
 142
 143         if (skb_cow_head(skb, MPLS_HLEN) < 0)
 144                 return -ENOMEM;
 145
 146         skb_push(skb, MPLS_HLEN);
 147         memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
 148                 skb->mac_len);
 149         skb_reset_mac_header(skb);
 150
 151         new_mpls_lse = (__be32 *)skb_mpls_header(skb);
 152         *new_mpls_lse = mpls->mpls_lse;
 153
 154         if (skb->ip_summed == CHECKSUM_COMPLETE)
 155                 skb->csum = csum_add(skb->csum, csum_partial(new_mpls_lse,
 156                                                              MPLS_HLEN, 0));
 157
 158         hdr = eth_hdr(skb);
 159         hdr->h_proto = mpls->mpls_ethertype;
 160
 161         skb_set_inner_protocol(skb, skb->protocol);
 162         skb->protocol = mpls->mpls_ethertype;
 163
 164         invalidate_flow_key(key);
 165         return 0;
 166 }
 167
 168 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 169                     const __be16 ethertype)
 170 {
 171         struct ethhdr *hdr;
 172         int err;
 173
 174         err = make_writable(skb, skb->mac_len + MPLS_HLEN);
 175         if (unlikely(err))
 176                 return err;
 177
 178         if (skb->ip_summed == CHECKSUM_COMPLETE)
 179                 skb->csum = csum_sub(skb->csum,
 180                                      csum_partial(skb_mpls_header(skb),
 181                                                   MPLS_HLEN, 0));
 182
 183         memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
 184                 skb->mac_len);
 185
 186         __skb_pull(skb, MPLS_HLEN);
 187         skb_reset_mac_header(skb);
 188
 189         /* skb_mpls_header() is used to locate the ethertype
 190          * field correctly in the presence of VLAN tags.
 191          */
 192         hdr = (struct ethhdr *)(skb_mpls_header(skb) - ETH_HLEN);
 193         hdr->h_proto = ethertype;
 194         if (eth_p_mpls(skb->protocol))
 195                 skb->protocol = ethertype;
 196
 197         invalidate_flow_key(key);
 198         return 0;
 199 }
 200
 201 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 202                     const __be32 *mpls_lse)
 203 {
 204         __be32 *stack;
 205         int err;
 206
 207         err = make_writable(skb, skb->mac_len + MPLS_HLEN);
 208         if (unlikely(err))
 209                 return err;
 210
 211         stack = (__be32 *)skb_mpls_header(skb);
 212         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 213                 __be32 diff[] = { ~(*stack), *mpls_lse };
 214                 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
 215                                           ~skb->csum);
 216         }
 217
 218         *stack = *mpls_lse;
 219         key->mpls.top_lse = *mpls_lse;
 220         return 0;
 221 }
 222
 223 /* remove VLAN header from packet and update csum accordingly. */
 224 static int __pop_vlan_tci(struct sk_buff *skb, __be16 *current_tci)
 225 {
 226         struct vlan_hdr *vhdr;
 227         int err;
 228
 229         err = make_writable(skb, VLAN_ETH_HLEN);
 230         if (unlikely(err))
 231                 return err;
 232
 233         if (skb->ip_summed == CHECKSUM_COMPLETE)
 234                 skb->csum = csum_sub(skb->csum, csum_partial(skb->data
 235                                         + (2 * ETH_ALEN), VLAN_HLEN, 0));
 236
 237         vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
 238         *current_tci = vhdr->h_vlan_TCI;
 239
 240         memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
 241         __skb_pull(skb, VLAN_HLEN);
 242
 243         vlan_set_encap_proto(skb, vhdr);
 244         skb->mac_header += VLAN_HLEN;
 245
 246         if (skb_network_offset(skb) < ETH_HLEN)
 247                 skb_set_network_header(skb, ETH_HLEN);
 248
 249         /* Update mac_len for subsequent MPLS actions */
 250         skb_reset_mac_len(skb);
 251         return 0;
 252 }
 253
 254 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 255 {
 256         __be16 tci;
 257         int err;
 258
 259         if (likely(vlan_tx_tag_present(skb))) {
 260                 skb->vlan_tci = 0;
 261         } else {
 262                 if (unlikely(skb->protocol != htons(ETH_P_8021Q) ||
 263                              skb->len < VLAN_ETH_HLEN))
 264                         return 0;
 265
 266                 err = __pop_vlan_tci(skb, &tci);
 267                 if (err)
 268                         return err;
 269         }
 270         /* move next vlan tag to hw accel tag */
 271         if (likely(skb->protocol != htons(ETH_P_8021Q) ||
 272                    skb->len < VLAN_ETH_HLEN)) {
 273                 key->eth.tci = 0;
 274                 return 0;
 275         }
 276
 277         invalidate_flow_key(key);
 278         err = __pop_vlan_tci(skb, &tci);
 279         if (unlikely(err))
 280                 return err;
 281
 282         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(tci));
 283         return 0;
 284 }
 285
 286 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
 287                      const struct ovs_action_push_vlan *vlan)
 288 {
 289         if (unlikely(vlan_tx_tag_present(skb))) {
 290                 u16 current_tag;
 291
 292                 /* push down current VLAN tag */
 293                 current_tag = vlan_tx_tag_get(skb);
 294
 295                 if (!__vlan_put_tag(skb, skb->vlan_proto, current_tag))
 296                         return -ENOMEM;
 297                 /* Update mac_len for subsequent MPLS actions */
 298                 skb->mac_len += VLAN_HLEN;
 299
 300                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 301                         skb->csum = csum_add(skb->csum, csum_partial(skb->data
 302                                         + (2 * ETH_ALEN), VLAN_HLEN, 0));
 303
 304                 invalidate_flow_key(key);
 305         } else {
 306                 key->eth.tci = vlan->vlan_tci;
 307         }
 308         __vlan_hwaccel_put_tag(skb, vlan->vlan_tpid, ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
 309         return 0;
 310 }
 311
 312 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *key,
 313                         const struct ovs_key_ethernet *eth_key)
 314 {
 315         int err;
 316         err = make_writable(skb, ETH_HLEN);
 317         if (unlikely(err))
 318                 return err;
 319
 320         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 321
 322         ether_addr_copy(eth_hdr(skb)->h_source, eth_key->eth_src);
 323         ether_addr_copy(eth_hdr(skb)->h_dest, eth_key->eth_dst);
 324
 325         ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 326
 327         ether_addr_copy(key->eth.src, eth_key->eth_src);
 328         ether_addr_copy(key->eth.dst, eth_key->eth_dst);
 329         return 0;
 330 }
 331
 332 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 333                         __be32 *addr, __be32 new_addr)
 334 {
 335         int transport_len = skb->len - skb_transport_offset(skb);
 336
 337         if (nh->protocol == IPPROTO_TCP) {
 338                 if (likely(transport_len >= sizeof(struct tcphdr)))
 339                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 340                                                  *addr, new_addr, 1);
 341         } else if (nh->protocol == IPPROTO_UDP) {
 342                 if (likely(transport_len >= sizeof(struct udphdr))) {
 343                         struct udphdr *uh = udp_hdr(skb);
 344
 345                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 346                                 inet_proto_csum_replace4(&uh->check, skb,
 347                                                          *addr, new_addr, 1);
 348                                 if (!uh->check)
 349                                         uh->check = CSUM_MANGLED_0;
 350                         }
 351                 }
 352         }
 353
 354         csum_replace4(&nh->check, *addr, new_addr);
 355         skb_clear_hash(skb);
 356         *addr = new_addr;
 357 }
 358
 359 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 360                                  __be32 addr[4], const __be32 new_addr[4])
 361 {
 362         int transport_len = skb->len - skb_transport_offset(skb);
 363
 364         if (l4_proto == IPPROTO_TCP) {
 365                 if (likely(transport_len >= sizeof(struct tcphdr)))
 366                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 367                                                   addr, new_addr, 1);
 368         } else if (l4_proto == IPPROTO_UDP) {
 369                 if (likely(transport_len >= sizeof(struct udphdr))) {
 370                         struct udphdr *uh = udp_hdr(skb);
 371
 372                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 373                                 inet_proto_csum_replace16(&uh->check, skb,
 374                                                           addr, new_addr, 1);
 375                                 if (!uh->check)
 376                                         uh->check = CSUM_MANGLED_0;
 377                         }
 378                 }
 379         }
 380 }
 381
 382 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 383                           __be32 addr[4], const __be32 new_addr[4],
 384                           bool recalculate_csum)
 385 {
 386         if (recalculate_csum)
 387                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 388
 389         skb_clear_hash(skb);
 390         memcpy(addr, new_addr, sizeof(__be32[4]));
 391 }
 392
 393 static void set_ipv6_tc(struct ipv6hdr *nh, u8 tc)
 394 {
 395         nh->priority = tc >> 4;
 396         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0x0F) | ((tc & 0x0F) << 4);
 397 }
 398
 399 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl)
 400 {
 401         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0xF0) | (fl & 0x000F0000) >> 16;
 402         nh->flow_lbl[1] = (fl & 0x0000FF00) >> 8;
 403         nh->flow_lbl[2] = fl & 0x000000FF;
 404 }
 405
 406 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl)
 407 {
 408         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 409         nh->ttl = new_ttl;
 410 }
 411
 412 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *key,
 413                     const struct ovs_key_ipv4 *ipv4_key)
 414 {
 415         struct iphdr *nh;
 416         int err;
 417
 418         err = make_writable(skb, skb_network_offset(skb) +
 419                                  sizeof(struct iphdr));
 420         if (unlikely(err))
 421                 return err;
 422
 423         nh = ip_hdr(skb);
 424
 425         if (ipv4_key->ipv4_src != nh->saddr) {
 426                 set_ip_addr(skb, nh, &nh->saddr, ipv4_key->ipv4_src);
 427                 key->ipv4.addr.src = ipv4_key->ipv4_src;
 428         }
 429
 430         if (ipv4_key->ipv4_dst != nh->daddr) {
 431                 set_ip_addr(skb, nh, &nh->daddr, ipv4_key->ipv4_dst);
 432                 key->ipv4.addr.dst = ipv4_key->ipv4_dst;
 433         }
 434
 435         if (ipv4_key->ipv4_tos != nh->tos) {
 436                 ipv4_change_dsfield(nh, 0, ipv4_key->ipv4_tos);
 437                 key->ip.tos = nh->tos;
 438         }
 439
 440         if (ipv4_key->ipv4_ttl != nh->ttl) {
 441                 set_ip_ttl(skb, nh, ipv4_key->ipv4_ttl);
 442                 key->ip.ttl = ipv4_key->ipv4_ttl;
 443         }
 444
 445         return 0;
 446 }
 447
 448 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *key,
 449                     const struct ovs_key_ipv6 *ipv6_key)
 450 {
 451         struct ipv6hdr *nh;
 452         int err;
 453         __be32 *saddr;
 454         __be32 *daddr;
 455
 456         err = make_writable(skb, skb_network_offset(skb) +
 457                             sizeof(struct ipv6hdr));
 458         if (unlikely(err))
 459                 return err;
 460
 461         nh = ipv6_hdr(skb);
 462         saddr = (__be32 *)&nh->saddr;
 463         daddr = (__be32 *)&nh->daddr;
 464
 465         if (memcmp(ipv6_key->ipv6_src, saddr, sizeof(ipv6_key->ipv6_src))) {
 466                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, saddr,
 467                               ipv6_key->ipv6_src, true);
 468                 memcpy(&key->ipv6.addr.src, ipv6_key->ipv6_src,
 469                        sizeof(ipv6_key->ipv6_src));
 470         }
 471
 472         if (memcmp(ipv6_key->ipv6_dst, daddr, sizeof(ipv6_key->ipv6_dst))) {
 473                 unsigned int offset = 0;
 474                 int flags = IP6_FH_F_SKIP_RH;
 475                 bool recalc_csum = true;
 476
 477                 if (ipv6_ext_hdr(nh->nexthdr))
 478                         recalc_csum = ipv6_find_hdr(skb, &offset,
 479                                                     NEXTHDR_ROUTING, NULL,
 480                                                     &flags) != NEXTHDR_ROUTING;
 481
 482                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, daddr,
 483                               ipv6_key->ipv6_dst, recalc_csum);
 484                 memcpy(&key->ipv6.addr.dst, ipv6_key->ipv6_dst,
 485                        sizeof(ipv6_key->ipv6_dst));
 486         }
 487
 488         set_ipv6_tc(nh, ipv6_key->ipv6_tclass);
 489         key->ip.tos = ipv6_get_dsfield(nh);
 490
 491         set_ipv6_fl(nh, ntohl(ipv6_key->ipv6_label));
 492         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 493
 494         nh->hop_limit = ipv6_key->ipv6_hlimit;
 495         key->ip.ttl = ipv6_key->ipv6_hlimit;
 496         return 0;
 497 }
 498
 499 /* Must follow make_writable() since that can move the skb data. */
 500 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 501                          __be16 new_port, __sum16 *check)
 502 {
 503         inet_proto_csum_replace2(check, skb, *port, new_port, 0);
 504         *port = new_port;
 505         skb_clear_hash(skb);
 506 }
 507
 508 static void set_udp_port(struct sk_buff *skb, __be16 *port, __be16 new_port)
 509 {
 510         struct udphdr *uh = udp_hdr(skb);
 511
 512         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 513                 set_tp_port(skb, port, new_port, &uh->check);
 514
 515                 if (!uh->check)
 516                         uh->check = CSUM_MANGLED_0;
 517         } else {
 518                 *port = new_port;
 519                 skb_clear_hash(skb);
 520         }
 521 }
 522
 523 static int set_udp(struct sk_buff *skb, struct sw_flow_key *key,
 524                    const struct ovs_key_udp *udp_port_key)
 525 {
 526         struct udphdr *uh;
 527         int err;
 528
 529         err = make_writable(skb, skb_transport_offset(skb) +
 530                                  sizeof(struct udphdr));
 531         if (unlikely(err))
 532                 return err;
 533
 534         uh = udp_hdr(skb);
 535         if (udp_port_key->udp_src != uh->source) {
 536                 set_udp_port(skb, &uh->source, udp_port_key->udp_src);
 537                 key->tp.src = udp_port_key->udp_src;
 538         }
 539
 540         if (udp_port_key->udp_dst != uh->dest) {
 541                 set_udp_port(skb, &uh->dest, udp_port_key->udp_dst);
 542                 key->tp.dst = udp_port_key->udp_dst;
 543         }
 544
 545         return 0;
 546 }
 547
 548 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *key,
 549                    const struct ovs_key_tcp *tcp_port_key)
 550 {
 551         struct tcphdr *th;
 552         int err;
 553
 554         err = make_writable(skb, skb_transport_offset(skb) +
 555                                  sizeof(struct tcphdr));
 556         if (unlikely(err))
 557                 return err;
 558
 559         th = tcp_hdr(skb);
 560         if (tcp_port_key->tcp_src != th->source) {
 561                 set_tp_port(skb, &th->source, tcp_port_key->tcp_src, &th->check);
 562                 key->tp.src = tcp_port_key->tcp_src;
 563         }
 564
 565         if (tcp_port_key->tcp_dst != th->dest) {
 566                 set_tp_port(skb, &th->dest, tcp_port_key->tcp_dst, &th->check);
 567                 key->tp.dst = tcp_port_key->tcp_dst;
 568         }
 569
 570         return 0;
 571 }
 572
 573 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *key,
 574                     const struct ovs_key_sctp *sctp_port_key)
 575 {
 576         struct sctphdr *sh;
 577         int err;
 578         unsigned int sctphoff = skb_transport_offset(skb);
 579
 580         err = make_writable(skb, sctphoff + sizeof(struct sctphdr));
 581         if (unlikely(err))
 582                 return err;
 583
 584         sh = sctp_hdr(skb);
 585         if (sctp_port_key->sctp_src != sh->source ||
 586             sctp_port_key->sctp_dst != sh->dest) {
 587                 __le32 old_correct_csum, new_csum, old_csum;
 588
 589                 old_csum = sh->checksum;
 590                 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 591
 592                 sh->source = sctp_port_key->sctp_src;
 593                 sh->dest = sctp_port_key->sctp_dst;
 594
 595                 new_csum = sctp_compute_cksum(skb, sctphoff);
 596
 597                 /* Carry any checksum errors through. */
 598                 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 599
 600                 skb_clear_hash(skb);
 601                 key->tp.src = sctp_port_key->sctp_src;
 602                 key->tp.dst = sctp_port_key->sctp_dst;
 603         }
 604
 605         return 0;
 606 }
 607
 608 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
 609 {
 610         struct vport *vport = ovs_vport_rcu(dp, out_port);
 611
 612         if (likely(vport))
 613                 ovs_vport_send(vport, skb);
 614         else
 615                 kfree_skb(skb);
 616 }
 617
 618 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 619                             struct sw_flow_key *key, const struct nlattr *attr)
 620 {
 621         struct ovs_tunnel_info info;
 622         struct dp_upcall_info upcall;
 623         const struct nlattr *a;
 624         int rem;
 625
 626         upcall.cmd = OVS_PACKET_CMD_ACTION;
 627         upcall.userdata = NULL;
 628         upcall.portid = 0;
 629         upcall.egress_tun_info = NULL;
 630
 631         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 632                  a = nla_next(a, &rem)) {
 633                 switch (nla_type(a)) {
 634                 case OVS_USERSPACE_ATTR_USERDATA:
 635                         upcall.userdata = a;
 636                         break;
 637
 638                 case OVS_USERSPACE_ATTR_PID:
 639                         upcall.portid = nla_get_u32(a);
 640                         break;
 641
 642                 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
 643                         /* Get out tunnel info. */
 644                         struct vport *vport;
 645
 646                         vport = ovs_vport_rcu(dp, nla_get_u32(a));
 647                         if (vport) {
 648                                 int err;
 649
 650                                 err = ovs_vport_get_egress_tun_info(vport, skb,
 651                                                                     &info);
 652                                 if (!err)
 653                                         upcall.egress_tun_info = &info;
 654                         }
 655                         break;
 656                 }
 657
 658                 } /* End of switch. */
 659         }
 660
 661         return ovs_dp_upcall(dp, skb, key, &upcall);
 662 }
 663
 664 static int sample(struct datapath *dp, struct sk_buff *skb,
 665                   struct sw_flow_key *key, const struct nlattr *attr)
 666 {
 667         const struct nlattr *acts_list = NULL;
 668         const struct nlattr *a;
 669         int rem;
 670
 671         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 672                  a = nla_next(a, &rem)) {
 673                 switch (nla_type(a)) {
 674                 case OVS_SAMPLE_ATTR_PROBABILITY:
 675                         if (prandom_u32() >= nla_get_u32(a))
 676                                 return 0;
 677                         break;
 678
 679                 case OVS_SAMPLE_ATTR_ACTIONS:
 680                         acts_list = a;
 681                         break;
 682                 }
 683         }
 684
 685         rem = nla_len(acts_list);
 686         a = nla_data(acts_list);
 687
 688         /* Actions list is empty, do nothing */
 689         if (unlikely(!rem))
 690                 return 0;
 691
 692         /* The only known usage of sample action is having a single user-space
 693          * action. Treat this usage as a special case.
 694          * The output_userspace() should clone the skb to be sent to the
 695          * user space. This skb will be consumed by its caller.
 696          */
 697         if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
 698                    nla_is_last(a, rem)))
 699                 return output_userspace(dp, skb, key, a);
 700
 701         skb = skb_clone(skb, GFP_ATOMIC);
 702         if (!skb)
 703                 /* Skip the sample action when out of memory. */
 704                 return 0;
 705
 706         if (!add_deferred_actions(skb, key, a)) {
 707                 if (net_ratelimit())
 708                         pr_warn("%s: deferred actions limit reached, dropping sample action\n",
 709                                 ovs_dp_name(dp));
 710
 711                 kfree_skb(skb);
 712         }
 713         return 0;
 714 }
 715
 716 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
 717                          const struct nlattr *attr)
 718 {
 719         struct ovs_action_hash *hash_act = nla_data(attr);
 720         u32 hash = 0;
 721
 722         /* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
 723         hash = skb_get_hash(skb);
 724         hash = jhash_1word(hash, hash_act->hash_basis);
 725         if (!hash)
 726                 hash = 0x1;
 727
 728         key->ovs_flow_hash = hash;
 729 }
 730
 731 static int execute_set_action(struct sk_buff *skb, struct sw_flow_key *key,
 732                               const struct nlattr *nested_attr)
 733 {
 734         int err = 0;
 735
 736         switch (nla_type(nested_attr)) {
 737         case OVS_KEY_ATTR_PRIORITY:
 738                 skb->priority = nla_get_u32(nested_attr);
 739                 key->phy.priority = skb->priority;
 740                 break;
 741
 742         case OVS_KEY_ATTR_SKB_MARK:
 743                 skb->mark = nla_get_u32(nested_attr);
 744                 key->phy.skb_mark = skb->mark;
 745                 break;
 746
 747         case OVS_KEY_ATTR_TUNNEL_INFO:
 748                 OVS_CB(skb)->egress_tun_info = nla_data(nested_attr);
 749                 break;
 750
 751         case OVS_KEY_ATTR_ETHERNET:
 752                 err = set_eth_addr(skb, key, nla_data(nested_attr));
 753                 break;
 754
 755         case OVS_KEY_ATTR_IPV4:
 756                 err = set_ipv4(skb, key, nla_data(nested_attr));
 757                 break;
 758
 759         case OVS_KEY_ATTR_IPV6:
 760                 err = set_ipv6(skb, key, nla_data(nested_attr));
 761                 break;
 762
 763         case OVS_KEY_ATTR_TCP:
 764                 err = set_tcp(skb, key, nla_data(nested_attr));
 765                 break;
 766
 767         case OVS_KEY_ATTR_UDP:
 768                 err = set_udp(skb, key, nla_data(nested_attr));
 769                 break;
 770
 771         case OVS_KEY_ATTR_SCTP:
 772                 err = set_sctp(skb, key, nla_data(nested_attr));
 773                 break;
 774
 775         case OVS_KEY_ATTR_MPLS:
 776                 err = set_mpls(skb, key, nla_data(nested_attr));
 777                 break;
 778         }
 779
 780         return err;
 781 }
 782
 783 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
 784                           struct sw_flow_key *key,
 785                           const struct nlattr *a, int rem)
 786 {
 787         struct deferred_action *da;
 788
 789         if (!is_flow_key_valid(key)) {
 790                 int err;
 791
 792                 err = ovs_flow_key_update(skb, key);
 793                 if (err)
 794                         return err;
 795         }
 796         BUG_ON(!is_flow_key_valid(key));
 797
 798         if (!nla_is_last(a, rem)) {
 799                 /* Recirc action is the not the last action
 800                  * of the action list, need to clone the skb.
 801                  */
 802                 skb = skb_clone(skb, GFP_ATOMIC);
 803
 804                 /* Skip the recirc action when out of memory, but
 805                  * continue on with the rest of the action list.
 806                  */
 807                 if (!skb)
 808                         return 0;
 809         }
 810
 811         da = add_deferred_actions(skb, key, NULL);
 812         if (da) {
 813                 da->pkt_key.recirc_id = nla_get_u32(a);
 814         } else {
 815                 kfree_skb(skb);
 816
 817                 if (net_ratelimit())
 818                         pr_warn("%s: deferred action limit reached, drop recirc action\n",
 819                                 ovs_dp_name(dp));
 820         }
 821
 822         return 0;
 823 }
 824
 825 /* Execute a list of actions against 'skb'. */
 826 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 827                               struct sw_flow_key *key,
 828                               const struct nlattr *attr, int len)
 829 {
 830         /* Every output action needs a separate clone of 'skb', but the common
 831          * case is just a single output action, so that doing a clone and
 832          * then freeing the original skbuff is wasteful.  So the following code
 833          * is slightly obscure just to avoid that.
 834          */
 835         int prev_port = -1;
 836         const struct nlattr *a;
 837         int rem;
 838
 839         for (a = attr, rem = len; rem > 0;
 840              a = nla_next(a, &rem)) {
 841                 int err = 0;
 842
 843                 if (unlikely(prev_port != -1)) {
 844                         struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
 845
 846                         if (out_skb)
 847                                 do_output(dp, out_skb, prev_port);
 848
 849                         prev_port = -1;
 850                 }
 851
 852                 switch (nla_type(a)) {
 853                 case OVS_ACTION_ATTR_OUTPUT:
 854                         prev_port = nla_get_u32(a);
 855                         break;
 856
 857                 case OVS_ACTION_ATTR_USERSPACE:
 858                         output_userspace(dp, skb, key, a);
 859                         break;
 860
 861                 case OVS_ACTION_ATTR_HASH:
 862                         execute_hash(skb, key, a);
 863                         break;
 864
 865                 case OVS_ACTION_ATTR_PUSH_MPLS:
 866                         err = push_mpls(skb, key, nla_data(a));
 867                         break;
 868
 869                 case OVS_ACTION_ATTR_POP_MPLS:
 870                         err = pop_mpls(skb, key, nla_get_be16(a));
 871                         break;
 872
 873                 case OVS_ACTION_ATTR_PUSH_VLAN:
 874                         err = push_vlan(skb, key, nla_data(a));
 875                         if (unlikely(err)) /* skb already freed. */
 876                                 return err;
 877                         break;
 878
 879                 case OVS_ACTION_ATTR_POP_VLAN:
 880                         err = pop_vlan(skb, key);
 881                         break;
 882
 883                 case OVS_ACTION_ATTR_RECIRC:
 884                         err = execute_recirc(dp, skb, key, a, rem);
 885                         if (nla_is_last(a, rem)) {
 886                                 /* If this is the last action, the skb has
 887                                  * been consumed or freed.
 888                                  * Return immediately.
 889                                  */
 890                                 return err;
 891                         }
 892                         break;
 893
 894                 case OVS_ACTION_ATTR_SET:
 895                         err = execute_set_action(skb, key, nla_data(a));
 896                         break;
 897
 898                 case OVS_ACTION_ATTR_SAMPLE:
 899                         err = sample(dp, skb, key, a);
 900                         if (unlikely(err)) /* skb already freed. */
 901                                 return err;
 902                         break;
 903                 }
 904
 905                 if (unlikely(err)) {
 906                         kfree_skb(skb);
 907                         return err;
 908                 }
 909         }
 910
 911         if (prev_port != -1)
 912                 do_output(dp, skb, prev_port);
 913         else
 914                 consume_skb(skb);
 915
 916         return 0;
 917 }
 918
 919 static void process_deferred_actions(struct datapath *dp)
 920 {
 921         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
 922
 923         /* Do not touch the FIFO in case there is no deferred actions. */
 924         if (action_fifo_is_empty(fifo))
 925                 return;
 926
 927         /* Finishing executing all deferred actions. */
 928         do {
 929                 struct deferred_action *da = action_fifo_get(fifo);
 930                 struct sk_buff *skb = da->skb;
 931                 struct sw_flow_key *key = &da->pkt_key;
 932                 const struct nlattr *actions = da->actions;
 933
 934                 if (actions)
 935                         do_execute_actions(dp, skb, key, actions,
 936                                            nla_len(actions));
 937                 else
 938                         ovs_dp_process_packet(skb, key);
 939         } while (!action_fifo_is_empty(fifo));
 940
 941         /* Reset FIFO for the next packet.  */
 942         action_fifo_init(fifo);
 943 }
 944
 945 /* Execute a list of actions against 'skb'. */
 946 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
 947                         const struct sw_flow_actions *acts,
 948                         struct sw_flow_key *key)
 949 {
 950         int level = this_cpu_read(exec_actions_level);
 951         int err;
 952
 953         this_cpu_inc(exec_actions_level);
 954         OVS_CB(skb)->egress_tun_info = NULL;
 955         err = do_execute_actions(dp, skb, key,
 956                                  acts->actions, acts->actions_len);
 957
 958         if (!level)
 959                 process_deferred_actions(dp);
 960
 961         this_cpu_dec(exec_actions_level);
 962         return err;
 963 }
 964
 965 int action_fifos_init(void)
 966 {
 967         action_fifos = alloc_percpu(struct action_fifo);
 968         if (!action_fifos)
 969                 return -ENOMEM;
 970
 971         return 0;
 972 }
 973
 974 void action_fifos_exit(void)
 975 {
 976         free_percpu(action_fifos);
 977 }