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