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