datapath/actions.c

   1 /*
   2  * Copyright (c) 2007-2014 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/skbuff.h>
  22 #include <linux/in.h>
  23 #include <linux/ip.h>
  24 #include <linux/openvswitch.h>
  25 #include <linux/sctp.h>
  26 #include <linux/tcp.h>
  27 #include <linux/udp.h>
  28 #include <linux/in6.h>
  29 #include <linux/if_arp.h>
  30 #include <linux/if_vlan.h>
  31 #include <net/ip.h>
  32 #include <net/ipv6.h>
  33 #include <net/checksum.h>
  34 #include <net/dsfield.h>
  35 #include <net/mpls.h>
  36 #include <net/sctp/checksum.h>
  37
  38 #include "datapath.h"
  39 #include "gso.h"
  40 #include "vlan.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 #define EXEC_ACTIONS_LEVEL_LIMIT 4   /* limit used to detect packet
  65                                       * looping by the network stack
  66                                       */
  67 static DEFINE_PER_CPU(int, exec_actions_level);
  68
  69 static void action_fifo_init(struct action_fifo *fifo)
  70 {
  71         fifo->head = 0;
  72         fifo->tail = 0;
  73 }
  74
  75 static bool action_fifo_is_empty(const struct action_fifo *fifo)
  76 {
  77         return (fifo->head == fifo->tail);
  78 }
  79
  80 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
  81 {
  82         if (action_fifo_is_empty(fifo))
  83                 return NULL;
  84
  85         return &fifo->fifo[fifo->tail++];
  86 }
  87
  88 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
  89 {
  90         if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
  91                 return NULL;
  92
  93         return &fifo->fifo[fifo->head++];
  94 }
  95
  96 /* Return queue entry if fifo is not full */
  97 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
  98                                                     const struct sw_flow_key *key,
  99                                                     const struct nlattr *attr)
 100 {
 101         struct action_fifo *fifo;
 102         struct deferred_action *da;
 103
 104         fifo = this_cpu_ptr(action_fifos);
 105         da = action_fifo_put(fifo);
 106         if (da) {
 107                 da->skb = skb;
 108                 da->actions = attr;
 109                 da->pkt_key = *key;
 110         }
 111
 112         return da;
 113 }
 114
 115 static void invalidate_flow_key(struct sw_flow_key *key)
 116 {
 117         key->eth.type = htons(0);
 118 }
 119
 120 static bool is_flow_key_valid(const struct sw_flow_key *key)
 121 {
 122         return !!key->eth.type;
 123 }
 124
 125 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 126                      const struct ovs_action_push_mpls *mpls)
 127 {
 128         __be32 *new_mpls_lse;
 129         struct ethhdr *hdr;
 130
 131         /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
 132         if (skb_encapsulation(skb))
 133                 return -ENOTSUPP;
 134
 135         if (skb_cow_head(skb, MPLS_HLEN) < 0)
 136                 return -ENOMEM;
 137
 138         skb_push(skb, MPLS_HLEN);
 139         memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
 140                 skb->mac_len);
 141         skb_reset_mac_header(skb);
 142
 143         new_mpls_lse = (__be32 *)skb_mpls_header(skb);
 144         *new_mpls_lse = mpls->mpls_lse;
 145
 146         if (skb->ip_summed == CHECKSUM_COMPLETE)
 147                 skb->csum = csum_add(skb->csum, csum_partial(new_mpls_lse,
 148                                                              MPLS_HLEN, 0));
 149
 150         hdr = eth_hdr(skb);
 151         hdr->h_proto = mpls->mpls_ethertype;
 152         if (!ovs_skb_get_inner_protocol(skb))
 153                 ovs_skb_set_inner_protocol(skb, skb->protocol);
 154         skb->protocol = mpls->mpls_ethertype;
 155
 156         invalidate_flow_key(key);
 157         return 0;
 158 }
 159
 160 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 161                     const __be16 ethertype)
 162 {
 163         struct ethhdr *hdr;
 164         int err;
 165
 166         err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
 167         if (unlikely(err))
 168                 return err;
 169
 170         if (skb->ip_summed == CHECKSUM_COMPLETE)
 171                 skb->csum = csum_sub(skb->csum,
 172                                      csum_partial(skb_mpls_header(skb),
 173                                                   MPLS_HLEN, 0));
 174
 175         memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
 176                 skb->mac_len);
 177
 178         __skb_pull(skb, MPLS_HLEN);
 179         skb_reset_mac_header(skb);
 180
 181         /* skb_mpls_header() is used to locate the ethertype
 182          * field correctly in the presence of VLAN tags.
 183          */
 184         hdr = (struct ethhdr *)(skb_mpls_header(skb) - ETH_HLEN);
 185         hdr->h_proto = ethertype;
 186         if (eth_p_mpls(skb->protocol))
 187                 skb->protocol = ethertype;
 188
 189         invalidate_flow_key(key);
 190         return 0;
 191 }
 192
 193 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 194                     const __be32 *mpls_lse)
 195 {
 196         __be32 *stack;
 197         int err;
 198
 199         err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
 200         if (unlikely(err))
 201                 return err;
 202
 203         stack = (__be32 *)skb_mpls_header(skb);
 204         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 205                 __be32 diff[] = { ~(*stack), *mpls_lse };
 206                 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
 207                                           ~skb->csum);
 208         }
 209
 210         *stack = *mpls_lse;
 211         key->mpls.top_lse = *mpls_lse;
 212         return 0;
 213 }
 214
 215 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 216 {
 217         int err;
 218
 219         err = skb_vlan_pop(skb);
 220         if (skb_vlan_tag_present(skb))
 221                 invalidate_flow_key(key);
 222         else
 223                 key->eth.tci = 0;
 224
 225         return err;
 226 }
 227
 228 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
 229                      const struct ovs_action_push_vlan *vlan)
 230 {
 231         if (skb_vlan_tag_present(skb))
 232                 invalidate_flow_key(key);
 233         else
 234                 key->eth.tci = vlan->vlan_tci;
 235
 236         return skb_vlan_push(skb, vlan->vlan_tpid,
 237                              ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
 238 }
 239
 240 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *key,
 241                         const struct ovs_key_ethernet *eth_key)
 242 {
 243         int err;
 244         err = skb_ensure_writable(skb, ETH_HLEN);
 245         if (unlikely(err))
 246                 return err;
 247
 248         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 249
 250         ether_addr_copy(eth_hdr(skb)->h_source, eth_key->eth_src);
 251         ether_addr_copy(eth_hdr(skb)->h_dest, eth_key->eth_dst);
 252
 253         ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 254
 255         ether_addr_copy(key->eth.src, eth_key->eth_src);
 256         ether_addr_copy(key->eth.dst, eth_key->eth_dst);
 257         return 0;
 258 }
 259
 260 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 261                         __be32 *addr, __be32 new_addr)
 262 {
 263         int transport_len = skb->len - skb_transport_offset(skb);
 264
 265         if (nh->protocol == IPPROTO_TCP) {
 266                 if (likely(transport_len >= sizeof(struct tcphdr)))
 267                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 268                                                  *addr, new_addr, 1);
 269         } else if (nh->protocol == IPPROTO_UDP) {
 270                 if (likely(transport_len >= sizeof(struct udphdr))) {
 271                         struct udphdr *uh = udp_hdr(skb);
 272
 273                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 274                                 inet_proto_csum_replace4(&uh->check, skb,
 275                                                          *addr, new_addr, 1);
 276                                 if (!uh->check)
 277                                         uh->check = CSUM_MANGLED_0;
 278                         }
 279                 }
 280         }
 281
 282         csum_replace4(&nh->check, *addr, new_addr);
 283         skb_clear_hash(skb);
 284         *addr = new_addr;
 285 }
 286
 287 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 288                                  __be32 addr[4], const __be32 new_addr[4])
 289 {
 290         int transport_len = skb->len - skb_transport_offset(skb);
 291
 292         if (l4_proto == NEXTHDR_TCP) {
 293                 if (likely(transport_len >= sizeof(struct tcphdr)))
 294                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 295                                                   addr, new_addr, 1);
 296         } else if (l4_proto == NEXTHDR_UDP) {
 297                 if (likely(transport_len >= sizeof(struct udphdr))) {
 298                         struct udphdr *uh = udp_hdr(skb);
 299
 300                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 301                                 inet_proto_csum_replace16(&uh->check, skb,
 302                                                           addr, new_addr, 1);
 303                                 if (!uh->check)
 304                                         uh->check = CSUM_MANGLED_0;
 305                         }
 306                 }
 307         } else if (l4_proto == NEXTHDR_ICMP) {
 308                 if (likely(transport_len >= sizeof(struct icmp6hdr)))
 309                         inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 310                                                   skb, addr, new_addr, 1);
 311         }
 312 }
 313
 314 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 315                           __be32 addr[4], const __be32 new_addr[4],
 316                           bool recalculate_csum)
 317 {
 318         if (likely(recalculate_csum))
 319                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 320
 321         skb_clear_hash(skb);
 322         memcpy(addr, new_addr, sizeof(__be32[4]));
 323 }
 324
 325 static void set_ipv6_tc(struct ipv6hdr *nh, u8 tc)
 326 {
 327         nh->priority = tc >> 4;
 328         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0x0F) | ((tc & 0x0F) << 4);
 329 }
 330
 331 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl)
 332 {
 333         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0xF0) | (fl & 0x000F0000) >> 16;
 334         nh->flow_lbl[1] = (fl & 0x0000FF00) >> 8;
 335         nh->flow_lbl[2] = fl & 0x000000FF;
 336 }
 337
 338 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl)
 339 {
 340         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 341         nh->ttl = new_ttl;
 342 }
 343
 344 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *key,
 345                     const struct ovs_key_ipv4 *ipv4_key)
 346 {
 347         struct iphdr *nh;
 348         int err;
 349
 350         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 351                                   sizeof(struct iphdr));
 352         if (unlikely(err))
 353                 return err;
 354
 355         nh = ip_hdr(skb);
 356
 357         if (ipv4_key->ipv4_src != nh->saddr) {
 358                 set_ip_addr(skb, nh, &nh->saddr, ipv4_key->ipv4_src);
 359                 key->ipv4.addr.src = ipv4_key->ipv4_src;
 360         }
 361
 362         if (ipv4_key->ipv4_dst != nh->daddr) {
 363                 set_ip_addr(skb, nh, &nh->daddr, ipv4_key->ipv4_dst);
 364                 key->ipv4.addr.dst = ipv4_key->ipv4_dst;
 365         }
 366
 367         if (ipv4_key->ipv4_tos != nh->tos) {
 368                 ipv4_change_dsfield(nh, 0, ipv4_key->ipv4_tos);
 369                 key->ip.tos = nh->tos;
 370         }
 371
 372         if (ipv4_key->ipv4_ttl != nh->ttl) {
 373                 set_ip_ttl(skb, nh, ipv4_key->ipv4_ttl);
 374                 key->ip.ttl = ipv4_key->ipv4_ttl;
 375         }
 376
 377         return 0;
 378 }
 379
 380 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *key,
 381                     const struct ovs_key_ipv6 *ipv6_key)
 382 {
 383         struct ipv6hdr *nh;
 384         int err;
 385         __be32 *saddr;
 386         __be32 *daddr;
 387
 388         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 389                                   sizeof(struct ipv6hdr));
 390         if (unlikely(err))
 391                 return err;
 392
 393         nh = ipv6_hdr(skb);
 394         saddr = (__be32 *)&nh->saddr;
 395         daddr = (__be32 *)&nh->daddr;
 396
 397         if (memcmp(ipv6_key->ipv6_src, saddr, sizeof(ipv6_key->ipv6_src))) {
 398                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, saddr,
 399                               ipv6_key->ipv6_src, true);
 400                 memcpy(&key->ipv6.addr.src, ipv6_key->ipv6_src,
 401                        sizeof(ipv6_key->ipv6_src));
 402         }
 403
 404         if (memcmp(ipv6_key->ipv6_dst, daddr, sizeof(ipv6_key->ipv6_dst))) {
 405                 unsigned int offset = 0;
 406                 int flags = IP6_FH_F_SKIP_RH;
 407                 bool recalc_csum = true;
 408
 409                 if (ipv6_ext_hdr(nh->nexthdr))
 410                         recalc_csum = ipv6_find_hdr(skb, &offset,
 411                                                     NEXTHDR_ROUTING, NULL,
 412                                                     &flags) != NEXTHDR_ROUTING;
 413
 414                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, daddr,
 415                               ipv6_key->ipv6_dst, recalc_csum);
 416                 memcpy(&key->ipv6.addr.dst, ipv6_key->ipv6_dst,
 417                        sizeof(ipv6_key->ipv6_dst));
 418         }
 419
 420         set_ipv6_tc(nh, ipv6_key->ipv6_tclass);
 421         key->ip.tos = ipv6_get_dsfield(nh);
 422
 423         set_ipv6_fl(nh, ntohl(ipv6_key->ipv6_label));
 424         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 425
 426         nh->hop_limit = ipv6_key->ipv6_hlimit;
 427         key->ip.ttl = ipv6_key->ipv6_hlimit;
 428         return 0;
 429 }
 430
 431 /* Must follow skb_ensure_writable() since that can move the skb data. */
 432 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 433                          __be16 new_port, __sum16 *check)
 434 {
 435         inet_proto_csum_replace2(check, skb, *port, new_port, 0);
 436         *port = new_port;
 437         skb_clear_hash(skb);
 438 }
 439
 440 static void set_udp_port(struct sk_buff *skb, __be16 *port, __be16 new_port)
 441 {
 442         struct udphdr *uh = udp_hdr(skb);
 443
 444         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 445                 set_tp_port(skb, port, new_port, &uh->check);
 446
 447                 if (!uh->check)
 448                         uh->check = CSUM_MANGLED_0;
 449         } else {
 450                 *port = new_port;
 451                 skb_clear_hash(skb);
 452         }
 453 }
 454
 455 static int set_udp(struct sk_buff *skb, struct sw_flow_key *key,
 456                    const struct ovs_key_udp *udp_port_key)
 457 {
 458         struct udphdr *uh;
 459         int err;
 460
 461         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 462                                   sizeof(struct udphdr));
 463         if (unlikely(err))
 464                 return err;
 465
 466         uh = udp_hdr(skb);
 467         if (udp_port_key->udp_src != uh->source) {
 468                 set_udp_port(skb, &uh->source, udp_port_key->udp_src);
 469                 key->tp.src = udp_port_key->udp_src;
 470         }
 471
 472         if (udp_port_key->udp_dst != uh->dest) {
 473                 set_udp_port(skb, &uh->dest, udp_port_key->udp_dst);
 474                 key->tp.dst = udp_port_key->udp_dst;
 475         }
 476
 477         return 0;
 478 }
 479
 480 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *key,
 481                    const struct ovs_key_tcp *tcp_port_key)
 482 {
 483         struct tcphdr *th;
 484         int err;
 485
 486         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 487                                   sizeof(struct tcphdr));
 488         if (unlikely(err))
 489                 return err;
 490
 491         th = tcp_hdr(skb);
 492         if (tcp_port_key->tcp_src != th->source) {
 493                 set_tp_port(skb, &th->source, tcp_port_key->tcp_src, &th->check);
 494                 key->tp.src = tcp_port_key->tcp_src;
 495         }
 496
 497         if (tcp_port_key->tcp_dst != th->dest) {
 498                 set_tp_port(skb, &th->dest, tcp_port_key->tcp_dst, &th->check);
 499                 key->tp.dst = tcp_port_key->tcp_dst;
 500         }
 501
 502         return 0;
 503 }
 504
 505 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *key,
 506                     const struct ovs_key_sctp *sctp_port_key)
 507 {
 508         struct sctphdr *sh;
 509         int err;
 510         unsigned int sctphoff = skb_transport_offset(skb);
 511
 512         err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
 513         if (unlikely(err))
 514                 return err;
 515
 516         sh = sctp_hdr(skb);
 517         if (sctp_port_key->sctp_src != sh->source ||
 518             sctp_port_key->sctp_dst != sh->dest) {
 519                 __le32 old_correct_csum, new_csum, old_csum;
 520
 521                 old_csum = sh->checksum;
 522                 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 523
 524                 sh->source = sctp_port_key->sctp_src;
 525                 sh->dest = sctp_port_key->sctp_dst;
 526
 527                 new_csum = sctp_compute_cksum(skb, sctphoff);
 528
 529                 /* Carry any checksum errors through. */
 530                 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 531
 532                 skb_clear_hash(skb);
 533                 key->tp.src = sctp_port_key->sctp_src;
 534                 key->tp.dst = sctp_port_key->sctp_dst;
 535         }
 536
 537         return 0;
 538 }
 539
 540 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
 541 {
 542         struct vport *vport = ovs_vport_rcu(dp, out_port);
 543
 544         if (likely(vport))
 545                 ovs_vport_send(vport, skb);
 546         else
 547                 kfree_skb(skb);
 548 }
 549
 550 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 551                             struct sw_flow_key *key, const struct nlattr *attr)
 552 {
 553         struct ovs_tunnel_info info;
 554         struct dp_upcall_info upcall;
 555         const struct nlattr *a;
 556         int rem;
 557
 558         upcall.cmd = OVS_PACKET_CMD_ACTION;
 559         upcall.userdata = NULL;
 560         upcall.portid = 0;
 561         upcall.egress_tun_info = NULL;
 562
 563         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 564                  a = nla_next(a, &rem)) {
 565                 switch (nla_type(a)) {
 566                 case OVS_USERSPACE_ATTR_USERDATA:
 567                         upcall.userdata = a;
 568                         break;
 569
 570                 case OVS_USERSPACE_ATTR_PID:
 571                         upcall.portid = nla_get_u32(a);
 572                         break;
 573
 574                 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
 575                         /* Get out tunnel info. */
 576                         struct vport *vport;
 577
 578                         vport = ovs_vport_rcu(dp, nla_get_u32(a));
 579                         if (vport) {
 580                                 int err;
 581
 582                                 err = ovs_vport_get_egress_tun_info(vport, skb,
 583                                                                     &info);
 584                                 if (!err)
 585                                         upcall.egress_tun_info = &info;
 586                         }
 587                         break;
 588                 }
 589
 590                 } /* End of switch. */
 591         }
 592
 593         return ovs_dp_upcall(dp, skb, key, &upcall);
 594 }
 595
 596 static int sample(struct datapath *dp, struct sk_buff *skb,
 597                   struct sw_flow_key *key, const struct nlattr *attr)
 598 {
 599         const struct nlattr *acts_list = NULL;
 600         const struct nlattr *a;
 601         int rem;
 602
 603         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 604                  a = nla_next(a, &rem)) {
 605                 switch (nla_type(a)) {
 606                 case OVS_SAMPLE_ATTR_PROBABILITY:
 607                         if (prandom_u32() >= nla_get_u32(a))
 608                                 return 0;
 609                         break;
 610
 611                 case OVS_SAMPLE_ATTR_ACTIONS:
 612                         acts_list = a;
 613                         break;
 614                 }
 615         }
 616
 617         rem = nla_len(acts_list);
 618         a = nla_data(acts_list);
 619
 620         /* Actions list is empty, do nothing */
 621         if (unlikely(!rem))
 622                 return 0;
 623
 624         /* The only known usage of sample action is having a single user-space
 625          * action. Treat this usage as a special case.
 626          * The output_userspace() should clone the skb to be sent to the
 627          * user space. This skb will be consumed by its caller.
 628          */
 629         if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
 630                    nla_is_last(a, rem)))
 631                 return output_userspace(dp, skb, key, a);
 632
 633         skb = skb_clone(skb, GFP_ATOMIC);
 634         if (!skb)
 635                 /* Skip the sample action when out of memory. */
 636                 return 0;
 637
 638         if (!add_deferred_actions(skb, key, a)) {
 639                 if (net_ratelimit())
 640                         pr_warn("%s: deferred actions limit reached, dropping sample action\n",
 641                                 ovs_dp_name(dp));
 642
 643                 kfree_skb(skb);
 644         }
 645         return 0;
 646 }
 647
 648 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
 649                          const struct nlattr *attr)
 650 {
 651         struct ovs_action_hash *hash_act = nla_data(attr);
 652         u32 hash = 0;
 653
 654         /* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
 655         hash = skb_get_hash(skb);
 656         hash = jhash_1word(hash, hash_act->hash_basis);
 657         if (!hash)
 658                 hash = 0x1;
 659
 660         key->ovs_flow_hash = hash;
 661 }
 662
 663 static int execute_set_action(struct sk_buff *skb, struct sw_flow_key *key,
 664                               const struct nlattr *nested_attr)
 665 {
 666         int err = 0;
 667
 668         switch (nla_type(nested_attr)) {
 669         case OVS_KEY_ATTR_PRIORITY:
 670                 skb->priority = nla_get_u32(nested_attr);
 671                 key->phy.priority = skb->priority;
 672                 break;
 673
 674         case OVS_KEY_ATTR_SKB_MARK:
 675                 skb->mark = nla_get_u32(nested_attr);
 676                 key->phy.skb_mark = skb->mark;
 677                 break;
 678
 679         case OVS_KEY_ATTR_TUNNEL_INFO:
 680                 OVS_CB(skb)->egress_tun_info = nla_data(nested_attr);
 681                 break;
 682
 683         case OVS_KEY_ATTR_ETHERNET:
 684                 err = set_eth_addr(skb, key, nla_data(nested_attr));
 685                 break;
 686
 687         case OVS_KEY_ATTR_IPV4:
 688                 err = set_ipv4(skb, key, nla_data(nested_attr));
 689                 break;
 690
 691         case OVS_KEY_ATTR_IPV6:
 692                 err = set_ipv6(skb, key, nla_data(nested_attr));
 693                 break;
 694
 695         case OVS_KEY_ATTR_TCP:
 696                 err = set_tcp(skb, key, nla_data(nested_attr));
 697                 break;
 698
 699         case OVS_KEY_ATTR_UDP:
 700                 err = set_udp(skb, key, nla_data(nested_attr));
 701                 break;
 702
 703         case OVS_KEY_ATTR_SCTP:
 704                 err = set_sctp(skb, key, nla_data(nested_attr));
 705                 break;
 706
 707         case OVS_KEY_ATTR_MPLS:
 708                 err = set_mpls(skb, key, nla_data(nested_attr));
 709                 break;
 710         }
 711
 712         return err;
 713 }
 714
 715 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
 716                           struct sw_flow_key *key,
 717                           const struct nlattr *a, int rem)
 718 {
 719         struct deferred_action *da;
 720
 721         if (!is_flow_key_valid(key)) {
 722                 int err;
 723
 724                 err = ovs_flow_key_update(skb, key);
 725                 if (err)
 726                         return err;
 727         }
 728         BUG_ON(!is_flow_key_valid(key));
 729
 730         if (!nla_is_last(a, rem)) {
 731                 /* Recirc action is the not the last action
 732                  * of the action list, need to clone the skb.
 733                  */
 734                 skb = skb_clone(skb, GFP_ATOMIC);
 735
 736                 /* Skip the recirc action when out of memory, but
 737                  * continue on with the rest of the action list.
 738                  */
 739                 if (!skb)
 740                         return 0;
 741         }
 742
 743         da = add_deferred_actions(skb, key, NULL);
 744         if (da) {
 745                 da->pkt_key.recirc_id = nla_get_u32(a);
 746         } else {
 747                 kfree_skb(skb);
 748
 749                 if (net_ratelimit())
 750                         pr_warn("%s: deferred action limit reached, drop recirc action\n",
 751                                 ovs_dp_name(dp));
 752         }
 753
 754         return 0;
 755 }
 756
 757 /* Execute a list of actions against 'skb'. */
 758 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 759                               struct sw_flow_key *key,
 760                               const struct nlattr *attr, int len)
 761 {
 762         /* Every output action needs a separate clone of 'skb', but the common
 763          * case is just a single output action, so that doing a clone and
 764          * then freeing the original skbuff is wasteful.  So the following code
 765          * is slightly obscure just to avoid that.
 766          */
 767         int prev_port = -1;
 768         const struct nlattr *a;
 769         int rem;
 770
 771         for (a = attr, rem = len; rem > 0;
 772              a = nla_next(a, &rem)) {
 773                 int err = 0;
 774
 775                 if (unlikely(prev_port != -1)) {
 776                         struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
 777
 778                         if (out_skb)
 779                                 do_output(dp, out_skb, prev_port);
 780
 781                         prev_port = -1;
 782                 }
 783
 784                 switch (nla_type(a)) {
 785                 case OVS_ACTION_ATTR_OUTPUT:
 786                         prev_port = nla_get_u32(a);
 787                         break;
 788
 789                 case OVS_ACTION_ATTR_USERSPACE:
 790                         output_userspace(dp, skb, key, a);
 791                         break;
 792
 793                 case OVS_ACTION_ATTR_HASH:
 794                         execute_hash(skb, key, a);
 795                         break;
 796
 797                 case OVS_ACTION_ATTR_PUSH_MPLS:
 798                         err = push_mpls(skb, key, nla_data(a));
 799                         break;
 800
 801                 case OVS_ACTION_ATTR_POP_MPLS:
 802                         err = pop_mpls(skb, key, nla_get_be16(a));
 803                         break;
 804
 805                 case OVS_ACTION_ATTR_PUSH_VLAN:
 806                         err = push_vlan(skb, key, nla_data(a));
 807                         break;
 808
 809                 case OVS_ACTION_ATTR_POP_VLAN:
 810                         err = pop_vlan(skb, key);
 811                         break;
 812
 813                 case OVS_ACTION_ATTR_RECIRC:
 814                         err = execute_recirc(dp, skb, key, a, rem);
 815                         if (nla_is_last(a, rem)) {
 816                                 /* If this is the last action, the skb has
 817                                  * been consumed or freed.
 818                                  * Return immediately.
 819                                  */
 820                                 return err;
 821                         }
 822                         break;
 823
 824                 case OVS_ACTION_ATTR_SET:
 825                         err = execute_set_action(skb, key, nla_data(a));
 826                         break;
 827
 828                 case OVS_ACTION_ATTR_SAMPLE:
 829                         err = sample(dp, skb, key, a);
 830                         break;
 831                 }
 832
 833                 if (unlikely(err)) {
 834                         kfree_skb(skb);
 835                         return err;
 836                 }
 837         }
 838
 839         if (prev_port != -1)
 840                 do_output(dp, skb, prev_port);
 841         else
 842                 consume_skb(skb);
 843
 844         return 0;
 845 }
 846
 847 static void process_deferred_actions(struct datapath *dp)
 848 {
 849         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
 850
 851         /* Do not touch the FIFO in case there is no deferred actions. */
 852         if (action_fifo_is_empty(fifo))
 853                 return;
 854
 855         /* Finishing executing all deferred actions. */
 856         do {
 857                 struct deferred_action *da = action_fifo_get(fifo);
 858                 struct sk_buff *skb = da->skb;
 859                 struct sw_flow_key *key = &da->pkt_key;
 860                 const struct nlattr *actions = da->actions;
 861
 862                 if (actions)
 863                         do_execute_actions(dp, skb, key, actions,
 864                                            nla_len(actions));
 865                 else
 866                         ovs_dp_process_packet(skb, key);
 867         } while (!action_fifo_is_empty(fifo));
 868
 869         /* Reset FIFO for the next packet.  */
 870         action_fifo_init(fifo);
 871 }
 872
 873 /* Execute a list of actions against 'skb'. */
 874 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
 875                         const struct sw_flow_actions *acts,
 876                         struct sw_flow_key *key)
 877 {
 878         int level = this_cpu_read(exec_actions_level);
 879         int err;
 880
 881         if (unlikely(level >= EXEC_ACTIONS_LEVEL_LIMIT)) {
 882                 if (net_ratelimit())
 883                         pr_warn("%s: packet loop detected, dropping.\n",
 884                                 ovs_dp_name(dp));
 885
 886                 kfree_skb(skb);
 887                 return -ELOOP;
 888         }
 889
 890         this_cpu_inc(exec_actions_level);
 891         err = do_execute_actions(dp, skb, key,
 892                                  acts->actions, acts->actions_len);
 893
 894         if (!level)
 895                 process_deferred_actions(dp);
 896
 897         this_cpu_dec(exec_actions_level);
 898
 899         /* This return status currently does not reflect the errors
 900          * encounted during deferred actions execution. Probably needs to
 901          * be fixed in the future.
 902          */
 903         return err;
 904 }
 905
 906 int action_fifos_init(void)
 907 {
 908         action_fifos = alloc_percpu(struct action_fifo);
 909         if (!action_fifos)
 910                 return -ENOMEM;
 911
 912         return 0;
 913 }
 914
 915 void action_fifos_exit(void)
 916 {
 917         free_percpu(action_fifos);
 918 }