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 /* remove VLAN header from packet and update csum accordingly. */
 216 static int __pop_vlan_tci(struct sk_buff *skb, __be16 *current_tci)
 217 {
 218         struct vlan_hdr *vhdr;
 219         int err;
 220
 221         err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
 222         if (unlikely(err))
 223                 return err;
 224
 225         if (skb->ip_summed == CHECKSUM_COMPLETE)
 226                 skb->csum = csum_sub(skb->csum, csum_partial(skb->data
 227                                         + (2 * ETH_ALEN), VLAN_HLEN, 0));
 228
 229         vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
 230         *current_tci = vhdr->h_vlan_TCI;
 231
 232         memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
 233         __skb_pull(skb, VLAN_HLEN);
 234
 235         vlan_set_encap_proto(skb, vhdr);
 236         skb->mac_header += VLAN_HLEN;
 237         /* Update mac_len for subsequent MPLS actions */
 238         skb->mac_len -= VLAN_HLEN;
 239
 240         return 0;
 241 }
 242
 243 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 244 {
 245         __be16 tci;
 246         int err;
 247
 248         if (likely(vlan_tx_tag_present(skb))) {
 249                 vlan_set_tci(skb, 0);
 250         } else {
 251                 if (unlikely(skb->protocol != htons(ETH_P_8021Q) ||
 252                              skb->len < VLAN_ETH_HLEN))
 253                         return 0;
 254
 255                 err = __pop_vlan_tci(skb, &tci);
 256                 if (err)
 257                         return err;
 258         }
 259         /* move next vlan tag to hw accel tag */
 260         if (likely(skb->protocol != htons(ETH_P_8021Q) ||
 261                    skb->len < VLAN_ETH_HLEN)) {
 262                 key->eth.tci = 0;
 263                 return 0;
 264         }
 265
 266         invalidate_flow_key(key);
 267         err = __pop_vlan_tci(skb, &tci);
 268         if (unlikely(err))
 269                 return err;
 270
 271         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(tci));
 272         return 0;
 273 }
 274
 275 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
 276                      const struct ovs_action_push_vlan *vlan)
 277 {
 278         if (unlikely(vlan_tx_tag_present(skb))) {
 279                 u16 current_tag;
 280
 281                 /* push down current VLAN tag */
 282                 current_tag = vlan_tx_tag_get(skb);
 283
 284                 if (!vlan_insert_tag_set_proto(skb, skb->vlan_proto, current_tag))
 285                         return -ENOMEM;
 286                 /* Update mac_len for subsequent MPLS actions */
 287                 skb->mac_len += VLAN_HLEN;
 288
 289                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 290                         skb->csum = csum_add(skb->csum, csum_partial(skb->data
 291                                         + (2 * ETH_ALEN), VLAN_HLEN, 0));
 292
 293                 invalidate_flow_key(key);
 294         } else {
 295                 key->eth.tci = vlan->vlan_tci;
 296         }
 297         __vlan_hwaccel_put_tag(skb, vlan->vlan_tpid, ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
 298         return 0;
 299 }
 300
 301 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *key,
 302                         const struct ovs_key_ethernet *eth_key)
 303 {
 304         int err;
 305         err = skb_ensure_writable(skb, ETH_HLEN);
 306         if (unlikely(err))
 307                 return err;
 308
 309         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 310
 311         ether_addr_copy(eth_hdr(skb)->h_source, eth_key->eth_src);
 312         ether_addr_copy(eth_hdr(skb)->h_dest, eth_key->eth_dst);
 313
 314         ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 315
 316         ether_addr_copy(key->eth.src, eth_key->eth_src);
 317         ether_addr_copy(key->eth.dst, eth_key->eth_dst);
 318         return 0;
 319 }
 320
 321 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 322                         __be32 *addr, __be32 new_addr)
 323 {
 324         int transport_len = skb->len - skb_transport_offset(skb);
 325
 326         if (nh->protocol == IPPROTO_TCP) {
 327                 if (likely(transport_len >= sizeof(struct tcphdr)))
 328                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 329                                                  *addr, new_addr, 1);
 330         } else if (nh->protocol == IPPROTO_UDP) {
 331                 if (likely(transport_len >= sizeof(struct udphdr))) {
 332                         struct udphdr *uh = udp_hdr(skb);
 333
 334                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 335                                 inet_proto_csum_replace4(&uh->check, skb,
 336                                                          *addr, new_addr, 1);
 337                                 if (!uh->check)
 338                                         uh->check = CSUM_MANGLED_0;
 339                         }
 340                 }
 341         }
 342
 343         csum_replace4(&nh->check, *addr, new_addr);
 344         skb_clear_hash(skb);
 345         *addr = new_addr;
 346 }
 347
 348 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 349                                  __be32 addr[4], const __be32 new_addr[4])
 350 {
 351         int transport_len = skb->len - skb_transport_offset(skb);
 352
 353         if (l4_proto == NEXTHDR_TCP) {
 354                 if (likely(transport_len >= sizeof(struct tcphdr)))
 355                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 356                                                   addr, new_addr, 1);
 357         } else if (l4_proto == NEXTHDR_UDP) {
 358                 if (likely(transport_len >= sizeof(struct udphdr))) {
 359                         struct udphdr *uh = udp_hdr(skb);
 360
 361                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 362                                 inet_proto_csum_replace16(&uh->check, skb,
 363                                                           addr, new_addr, 1);
 364                                 if (!uh->check)
 365                                         uh->check = CSUM_MANGLED_0;
 366                         }
 367                 }
 368         } else if (l4_proto == NEXTHDR_ICMP) {
 369                 if (likely(transport_len >= sizeof(struct icmp6hdr)))
 370                         inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 371                                                   skb, addr, new_addr, 1);
 372         }
 373 }
 374
 375 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 376                           __be32 addr[4], const __be32 new_addr[4],
 377                           bool recalculate_csum)
 378 {
 379         if (likely(recalculate_csum))
 380                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 381
 382         skb_clear_hash(skb);
 383         memcpy(addr, new_addr, sizeof(__be32[4]));
 384 }
 385
 386 static void set_ipv6_tc(struct ipv6hdr *nh, u8 tc)
 387 {
 388         nh->priority = tc >> 4;
 389         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0x0F) | ((tc & 0x0F) << 4);
 390 }
 391
 392 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl)
 393 {
 394         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0xF0) | (fl & 0x000F0000) >> 16;
 395         nh->flow_lbl[1] = (fl & 0x0000FF00) >> 8;
 396         nh->flow_lbl[2] = fl & 0x000000FF;
 397 }
 398
 399 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl)
 400 {
 401         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 402         nh->ttl = new_ttl;
 403 }
 404
 405 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *key,
 406                     const struct ovs_key_ipv4 *ipv4_key)
 407 {
 408         struct iphdr *nh;
 409         int err;
 410
 411         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 412                                   sizeof(struct iphdr));
 413         if (unlikely(err))
 414                 return err;
 415
 416         nh = ip_hdr(skb);
 417
 418         if (ipv4_key->ipv4_src != nh->saddr) {
 419                 set_ip_addr(skb, nh, &nh->saddr, ipv4_key->ipv4_src);
 420                 key->ipv4.addr.src = ipv4_key->ipv4_src;
 421         }
 422
 423         if (ipv4_key->ipv4_dst != nh->daddr) {
 424                 set_ip_addr(skb, nh, &nh->daddr, ipv4_key->ipv4_dst);
 425                 key->ipv4.addr.dst = ipv4_key->ipv4_dst;
 426         }
 427
 428         if (ipv4_key->ipv4_tos != nh->tos) {
 429                 ipv4_change_dsfield(nh, 0, ipv4_key->ipv4_tos);
 430                 key->ip.tos = nh->tos;
 431         }
 432
 433         if (ipv4_key->ipv4_ttl != nh->ttl) {
 434                 set_ip_ttl(skb, nh, ipv4_key->ipv4_ttl);
 435                 key->ip.ttl = ipv4_key->ipv4_ttl;
 436         }
 437
 438         return 0;
 439 }
 440
 441 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *key,
 442                     const struct ovs_key_ipv6 *ipv6_key)
 443 {
 444         struct ipv6hdr *nh;
 445         int err;
 446         __be32 *saddr;
 447         __be32 *daddr;
 448
 449         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 450                                   sizeof(struct ipv6hdr));
 451         if (unlikely(err))
 452                 return err;
 453
 454         nh = ipv6_hdr(skb);
 455         saddr = (__be32 *)&nh->saddr;
 456         daddr = (__be32 *)&nh->daddr;
 457
 458         if (memcmp(ipv6_key->ipv6_src, saddr, sizeof(ipv6_key->ipv6_src))) {
 459                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, saddr,
 460                               ipv6_key->ipv6_src, true);
 461                 memcpy(&key->ipv6.addr.src, ipv6_key->ipv6_src,
 462                        sizeof(ipv6_key->ipv6_src));
 463         }
 464
 465         if (memcmp(ipv6_key->ipv6_dst, daddr, sizeof(ipv6_key->ipv6_dst))) {
 466                 unsigned int offset = 0;
 467                 int flags = IP6_FH_F_SKIP_RH;
 468                 bool recalc_csum = true;
 469
 470                 if (ipv6_ext_hdr(nh->nexthdr))
 471                         recalc_csum = ipv6_find_hdr(skb, &offset,
 472                                                     NEXTHDR_ROUTING, NULL,
 473                                                     &flags) != NEXTHDR_ROUTING;
 474
 475                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, daddr,
 476                               ipv6_key->ipv6_dst, recalc_csum);
 477                 memcpy(&key->ipv6.addr.dst, ipv6_key->ipv6_dst,
 478                        sizeof(ipv6_key->ipv6_dst));
 479         }
 480
 481         set_ipv6_tc(nh, ipv6_key->ipv6_tclass);
 482         key->ip.tos = ipv6_get_dsfield(nh);
 483
 484         set_ipv6_fl(nh, ntohl(ipv6_key->ipv6_label));
 485         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 486
 487         nh->hop_limit = ipv6_key->ipv6_hlimit;
 488         key->ip.ttl = ipv6_key->ipv6_hlimit;
 489         return 0;
 490 }
 491
 492 /* Must follow skb_ensure_writable() since that can move the skb data. */
 493 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 494                          __be16 new_port, __sum16 *check)
 495 {
 496         inet_proto_csum_replace2(check, skb, *port, new_port, 0);
 497         *port = new_port;
 498         skb_clear_hash(skb);
 499 }
 500
 501 static void set_udp_port(struct sk_buff *skb, __be16 *port, __be16 new_port)
 502 {
 503         struct udphdr *uh = udp_hdr(skb);
 504
 505         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 506                 set_tp_port(skb, port, new_port, &uh->check);
 507
 508                 if (!uh->check)
 509                         uh->check = CSUM_MANGLED_0;
 510         } else {
 511                 *port = new_port;
 512                 skb_clear_hash(skb);
 513         }
 514 }
 515
 516 static int set_udp(struct sk_buff *skb, struct sw_flow_key *key,
 517                    const struct ovs_key_udp *udp_port_key)
 518 {
 519         struct udphdr *uh;
 520         int err;
 521
 522         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 523                                   sizeof(struct udphdr));
 524         if (unlikely(err))
 525                 return err;
 526
 527         uh = udp_hdr(skb);
 528         if (udp_port_key->udp_src != uh->source) {
 529                 set_udp_port(skb, &uh->source, udp_port_key->udp_src);
 530                 key->tp.src = udp_port_key->udp_src;
 531         }
 532
 533         if (udp_port_key->udp_dst != uh->dest) {
 534                 set_udp_port(skb, &uh->dest, udp_port_key->udp_dst);
 535                 key->tp.dst = udp_port_key->udp_dst;
 536         }
 537
 538         return 0;
 539 }
 540
 541 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *key,
 542                    const struct ovs_key_tcp *tcp_port_key)
 543 {
 544         struct tcphdr *th;
 545         int err;
 546
 547         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 548                                   sizeof(struct tcphdr));
 549         if (unlikely(err))
 550                 return err;
 551
 552         th = tcp_hdr(skb);
 553         if (tcp_port_key->tcp_src != th->source) {
 554                 set_tp_port(skb, &th->source, tcp_port_key->tcp_src, &th->check);
 555                 key->tp.src = tcp_port_key->tcp_src;
 556         }
 557
 558         if (tcp_port_key->tcp_dst != th->dest) {
 559                 set_tp_port(skb, &th->dest, tcp_port_key->tcp_dst, &th->check);
 560                 key->tp.dst = tcp_port_key->tcp_dst;
 561         }
 562
 563         return 0;
 564 }
 565
 566 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *key,
 567                     const struct ovs_key_sctp *sctp_port_key)
 568 {
 569         struct sctphdr *sh;
 570         int err;
 571         unsigned int sctphoff = skb_transport_offset(skb);
 572
 573         err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
 574         if (unlikely(err))
 575                 return err;
 576
 577         sh = sctp_hdr(skb);
 578         if (sctp_port_key->sctp_src != sh->source ||
 579             sctp_port_key->sctp_dst != sh->dest) {
 580                 __le32 old_correct_csum, new_csum, old_csum;
 581
 582                 old_csum = sh->checksum;
 583                 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 584
 585                 sh->source = sctp_port_key->sctp_src;
 586                 sh->dest = sctp_port_key->sctp_dst;
 587
 588                 new_csum = sctp_compute_cksum(skb, sctphoff);
 589
 590                 /* Carry any checksum errors through. */
 591                 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 592
 593                 skb_clear_hash(skb);
 594                 key->tp.src = sctp_port_key->sctp_src;
 595                 key->tp.dst = sctp_port_key->sctp_dst;
 596         }
 597
 598         return 0;
 599 }
 600
 601 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
 602 {
 603         struct vport *vport = ovs_vport_rcu(dp, out_port);
 604
 605         if (likely(vport))
 606                 ovs_vport_send(vport, skb);
 607         else
 608                 kfree_skb(skb);
 609 }
 610
 611 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 612                             struct sw_flow_key *key, const struct nlattr *attr)
 613 {
 614         struct ovs_tunnel_info info;
 615         struct dp_upcall_info upcall;
 616         const struct nlattr *a;
 617         int rem;
 618
 619         upcall.cmd = OVS_PACKET_CMD_ACTION;
 620         upcall.userdata = NULL;
 621         upcall.portid = 0;
 622         upcall.egress_tun_info = NULL;
 623
 624         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 625                  a = nla_next(a, &rem)) {
 626                 switch (nla_type(a)) {
 627                 case OVS_USERSPACE_ATTR_USERDATA:
 628                         upcall.userdata = a;
 629                         break;
 630
 631                 case OVS_USERSPACE_ATTR_PID:
 632                         upcall.portid = nla_get_u32(a);
 633                         break;
 634
 635                 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
 636                         /* Get out tunnel info. */
 637                         struct vport *vport;
 638
 639                         vport = ovs_vport_rcu(dp, nla_get_u32(a));
 640                         if (vport) {
 641                                 int err;
 642
 643                                 err = ovs_vport_get_egress_tun_info(vport, skb,
 644                                                                     &info);
 645                                 if (!err)
 646                                         upcall.egress_tun_info = &info;
 647                         }
 648                         break;
 649                 }
 650
 651                 } /* End of switch. */
 652         }
 653
 654         return ovs_dp_upcall(dp, skb, key, &upcall);
 655 }
 656
 657 static int sample(struct datapath *dp, struct sk_buff *skb,
 658                   struct sw_flow_key *key, const struct nlattr *attr)
 659 {
 660         const struct nlattr *acts_list = NULL;
 661         const struct nlattr *a;
 662         int rem;
 663
 664         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 665                  a = nla_next(a, &rem)) {
 666                 switch (nla_type(a)) {
 667                 case OVS_SAMPLE_ATTR_PROBABILITY:
 668                         if (prandom_u32() >= nla_get_u32(a))
 669                                 return 0;
 670                         break;
 671
 672                 case OVS_SAMPLE_ATTR_ACTIONS:
 673                         acts_list = a;
 674                         break;
 675                 }
 676         }
 677
 678         rem = nla_len(acts_list);
 679         a = nla_data(acts_list);
 680
 681         /* Actions list is empty, do nothing */
 682         if (unlikely(!rem))
 683                 return 0;
 684
 685         /* The only known usage of sample action is having a single user-space
 686          * action. Treat this usage as a special case.
 687          * The output_userspace() should clone the skb to be sent to the
 688          * user space. This skb will be consumed by its caller.
 689          */
 690         if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
 691                    nla_is_last(a, rem)))
 692                 return output_userspace(dp, skb, key, a);
 693
 694         skb = skb_clone(skb, GFP_ATOMIC);
 695         if (!skb)
 696                 /* Skip the sample action when out of memory. */
 697                 return 0;
 698
 699         if (!add_deferred_actions(skb, key, a)) {
 700                 if (net_ratelimit())
 701                         pr_warn("%s: deferred actions limit reached, dropping sample action\n",
 702                                 ovs_dp_name(dp));
 703
 704                 kfree_skb(skb);
 705         }
 706         return 0;
 707 }
 708
 709 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
 710                          const struct nlattr *attr)
 711 {
 712         struct ovs_action_hash *hash_act = nla_data(attr);
 713         u32 hash = 0;
 714
 715         /* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
 716         hash = skb_get_hash(skb);
 717         hash = jhash_1word(hash, hash_act->hash_basis);
 718         if (!hash)
 719                 hash = 0x1;
 720
 721         key->ovs_flow_hash = hash;
 722 }
 723
 724 static int execute_set_action(struct sk_buff *skb, struct sw_flow_key *key,
 725                               const struct nlattr *nested_attr)
 726 {
 727         int err = 0;
 728
 729         switch (nla_type(nested_attr)) {
 730         case OVS_KEY_ATTR_PRIORITY:
 731                 skb->priority = nla_get_u32(nested_attr);
 732                 key->phy.priority = skb->priority;
 733                 break;
 734
 735         case OVS_KEY_ATTR_SKB_MARK:
 736                 skb->mark = nla_get_u32(nested_attr);
 737                 key->phy.skb_mark = skb->mark;
 738                 break;
 739
 740         case OVS_KEY_ATTR_TUNNEL_INFO:
 741                 OVS_CB(skb)->egress_tun_info = nla_data(nested_attr);
 742                 break;
 743
 744         case OVS_KEY_ATTR_ETHERNET:
 745                 err = set_eth_addr(skb, key, nla_data(nested_attr));
 746                 break;
 747
 748         case OVS_KEY_ATTR_IPV4:
 749                 err = set_ipv4(skb, key, nla_data(nested_attr));
 750                 break;
 751
 752         case OVS_KEY_ATTR_IPV6:
 753                 err = set_ipv6(skb, key, nla_data(nested_attr));
 754                 break;
 755
 756         case OVS_KEY_ATTR_TCP:
 757                 err = set_tcp(skb, key, nla_data(nested_attr));
 758                 break;
 759
 760         case OVS_KEY_ATTR_UDP:
 761                 err = set_udp(skb, key, nla_data(nested_attr));
 762                 break;
 763
 764         case OVS_KEY_ATTR_SCTP:
 765                 err = set_sctp(skb, key, nla_data(nested_attr));
 766                 break;
 767
 768         case OVS_KEY_ATTR_MPLS:
 769                 err = set_mpls(skb, key, nla_data(nested_attr));
 770                 break;
 771         }
 772
 773         return err;
 774 }
 775
 776 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
 777                           struct sw_flow_key *key,
 778                           const struct nlattr *a, int rem)
 779 {
 780         struct deferred_action *da;
 781
 782         if (!is_flow_key_valid(key)) {
 783                 int err;
 784
 785                 err = ovs_flow_key_update(skb, key);
 786                 if (err)
 787                         return err;
 788         }
 789         BUG_ON(!is_flow_key_valid(key));
 790
 791         if (!nla_is_last(a, rem)) {
 792                 /* Recirc action is the not the last action
 793                  * of the action list, need to clone the skb.
 794                  */
 795                 skb = skb_clone(skb, GFP_ATOMIC);
 796
 797                 /* Skip the recirc action when out of memory, but
 798                  * continue on with the rest of the action list.
 799                  */
 800                 if (!skb)
 801                         return 0;
 802         }
 803
 804         da = add_deferred_actions(skb, key, NULL);
 805         if (da) {
 806                 da->pkt_key.recirc_id = nla_get_u32(a);
 807         } else {
 808                 kfree_skb(skb);
 809
 810                 if (net_ratelimit())
 811                         pr_warn("%s: deferred action limit reached, drop recirc action\n",
 812                                 ovs_dp_name(dp));
 813         }
 814
 815         return 0;
 816 }
 817
 818 /* Execute a list of actions against 'skb'. */
 819 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 820                               struct sw_flow_key *key,
 821                               const struct nlattr *attr, int len)
 822 {
 823         /* Every output action needs a separate clone of 'skb', but the common
 824          * case is just a single output action, so that doing a clone and
 825          * then freeing the original skbuff is wasteful.  So the following code
 826          * is slightly obscure just to avoid that.
 827          */
 828         int prev_port = -1;
 829         const struct nlattr *a;
 830         int rem;
 831
 832         for (a = attr, rem = len; rem > 0;
 833              a = nla_next(a, &rem)) {
 834                 int err = 0;
 835
 836                 if (unlikely(prev_port != -1)) {
 837                         struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
 838
 839                         if (out_skb)
 840                                 do_output(dp, out_skb, prev_port);
 841
 842                         prev_port = -1;
 843                 }
 844
 845                 switch (nla_type(a)) {
 846                 case OVS_ACTION_ATTR_OUTPUT:
 847                         prev_port = nla_get_u32(a);
 848                         break;
 849
 850                 case OVS_ACTION_ATTR_USERSPACE:
 851                         output_userspace(dp, skb, key, a);
 852                         break;
 853
 854                 case OVS_ACTION_ATTR_HASH:
 855                         execute_hash(skb, key, a);
 856                         break;
 857
 858                 case OVS_ACTION_ATTR_PUSH_MPLS:
 859                         err = push_mpls(skb, key, nla_data(a));
 860                         break;
 861
 862                 case OVS_ACTION_ATTR_POP_MPLS:
 863                         err = pop_mpls(skb, key, nla_get_be16(a));
 864                         break;
 865
 866                 case OVS_ACTION_ATTR_PUSH_VLAN:
 867                         err = push_vlan(skb, key, nla_data(a));
 868                         if (unlikely(err)) /* skb already freed. */
 869                                 return err;
 870                         break;
 871
 872                 case OVS_ACTION_ATTR_POP_VLAN:
 873                         err = pop_vlan(skb, key);
 874                         break;
 875
 876                 case OVS_ACTION_ATTR_RECIRC:
 877                         err = execute_recirc(dp, skb, key, a, rem);
 878                         if (nla_is_last(a, rem)) {
 879                                 /* If this is the last action, the skb has
 880                                  * been consumed or freed.
 881                                  * Return immediately.
 882                                  */
 883                                 return err;
 884                         }
 885                         break;
 886
 887                 case OVS_ACTION_ATTR_SET:
 888                         err = execute_set_action(skb, key, nla_data(a));
 889                         break;
 890
 891                 case OVS_ACTION_ATTR_SAMPLE:
 892                         err = sample(dp, skb, key, a);
 893                         break;
 894                 }
 895
 896                 if (unlikely(err)) {
 897                         kfree_skb(skb);
 898                         return err;
 899                 }
 900         }
 901
 902         if (prev_port != -1)
 903                 do_output(dp, skb, prev_port);
 904         else
 905                 consume_skb(skb);
 906
 907         return 0;
 908 }
 909
 910 static void process_deferred_actions(struct datapath *dp)
 911 {
 912         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
 913
 914         /* Do not touch the FIFO in case there is no deferred actions. */
 915         if (action_fifo_is_empty(fifo))
 916                 return;
 917
 918         /* Finishing executing all deferred actions. */
 919         do {
 920                 struct deferred_action *da = action_fifo_get(fifo);
 921                 struct sk_buff *skb = da->skb;
 922                 struct sw_flow_key *key = &da->pkt_key;
 923                 const struct nlattr *actions = da->actions;
 924
 925                 if (actions)
 926                         do_execute_actions(dp, skb, key, actions,
 927                                            nla_len(actions));
 928                 else
 929                         ovs_dp_process_packet(skb, key);
 930         } while (!action_fifo_is_empty(fifo));
 931
 932         /* Reset FIFO for the next packet.  */
 933         action_fifo_init(fifo);
 934 }
 935
 936 /* Execute a list of actions against 'skb'. */
 937 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
 938                         const struct sw_flow_actions *acts,
 939                         struct sw_flow_key *key)
 940 {
 941         int level = this_cpu_read(exec_actions_level);
 942         int err;
 943
 944         if (unlikely(level >= EXEC_ACTIONS_LEVEL_LIMIT)) {
 945                 if (net_ratelimit())
 946                         pr_warn("%s: packet loop detected, dropping.\n",
 947                                 ovs_dp_name(dp));
 948
 949                 kfree_skb(skb);
 950                 return -ELOOP;
 951         }
 952
 953         this_cpu_inc(exec_actions_level);
 954         err = do_execute_actions(dp, skb, key,
 955                                  acts->actions, acts->actions_len);
 956
 957         if (!level)
 958                 process_deferred_actions(dp);
 959
 960         this_cpu_dec(exec_actions_level);
 961
 962         /* This return status currently does not reflect the errors
 963          * encounted during deferred actions execution. Probably needs to
 964          * be fixed in the future.
 965          */
 966         return err;
 967 }
 968
 969 int action_fifos_init(void)
 970 {
 971         action_fifos = alloc_percpu(struct action_fifo);
 972         if (!action_fifos)
 973                 return -ENOMEM;
 974
 975         return 0;
 976 }
 977
 978 void action_fifos_exit(void)
 979 {
 980         free_percpu(action_fifos);
 981 }