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/sctp/checksum.h>
  36
  37 #include "datapath.h"
  38 #include "gso.h"
  39 #include "mpls.h"
  40 #include "vlan.h"
  41 #include "vport.h"
  42
  43 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
  44                               const struct nlattr *attr, int len);
  45
  46 static int make_writable(struct sk_buff *skb, int write_len)
  47 {
  48         if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
  49                 return 0;
  50
  51         return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
  52 }
  53
  54 /* The end of the mac header.
  55  *
  56  * For non-MPLS skbs this will correspond to the network header.
  57  * For MPLS skbs it will be before the network_header as the MPLS
  58  * label stack lies between the end of the mac header and the network
  59  * header. That is, for MPLS skbs the end of the mac header
  60  * is the top of the MPLS label stack.
  61  */
  62 static unsigned char *mac_header_end(const struct sk_buff *skb)
  63 {
  64         return skb_mac_header(skb) + skb->mac_len;
  65 }
  66
  67 static int push_mpls(struct sk_buff *skb,
  68                      const struct ovs_action_push_mpls *mpls)
  69 {
  70         __be32 *new_mpls_lse;
  71         struct ethhdr *hdr;
  72
  73         if (skb_cow_head(skb, MPLS_HLEN) < 0)
  74                 return -ENOMEM;
  75
  76         skb_push(skb, MPLS_HLEN);
  77         memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
  78                 skb->mac_len);
  79         skb_reset_mac_header(skb);
  80
  81         new_mpls_lse = (__be32 *)mac_header_end(skb);
  82         *new_mpls_lse = mpls->mpls_lse;
  83
  84         if (skb->ip_summed == CHECKSUM_COMPLETE)
  85                 skb->csum = csum_add(skb->csum, csum_partial(new_mpls_lse,
  86                                                              MPLS_HLEN, 0));
  87
  88         hdr = eth_hdr(skb);
  89         hdr->h_proto = mpls->mpls_ethertype;
  90         if (!ovs_skb_get_inner_protocol(skb))
  91                 ovs_skb_set_inner_protocol(skb, skb->protocol);
  92         skb->protocol = mpls->mpls_ethertype;
  93         return 0;
  94 }
  95
  96 static int pop_mpls(struct sk_buff *skb, const __be16 ethertype)
  97 {
  98         struct ethhdr *hdr;
  99         int err;
 100
 101         err = make_writable(skb, skb->mac_len + MPLS_HLEN);
 102         if (unlikely(err))
 103                 return err;
 104
 105         if (skb->ip_summed == CHECKSUM_COMPLETE)
 106                 skb->csum = csum_sub(skb->csum,
 107                                      csum_partial(mac_header_end(skb),
 108                                                   MPLS_HLEN, 0));
 109
 110         memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
 111                 skb->mac_len);
 112
 113         __skb_pull(skb, MPLS_HLEN);
 114         skb_reset_mac_header(skb);
 115
 116         /* mac_header_end() is used to locate the ethertype
 117          * field correctly in the presence of VLAN tags.
 118          */
 119         hdr = (struct ethhdr *)(mac_header_end(skb) - ETH_HLEN);
 120         hdr->h_proto = ethertype;
 121         if (eth_p_mpls(skb->protocol))
 122                 skb->protocol = ethertype;
 123         return 0;
 124 }
 125
 126 static int set_mpls(struct sk_buff *skb, const __be32 *mpls_lse)
 127 {
 128         __be32 *stack = (__be32 *)mac_header_end(skb);
 129         int err;
 130
 131         err = make_writable(skb, skb->mac_len + MPLS_HLEN);
 132         if (unlikely(err))
 133                 return err;
 134
 135         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 136                 __be32 diff[] = { ~(*stack), *mpls_lse };
 137                 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
 138                                           ~skb->csum);
 139         }
 140
 141         *stack = *mpls_lse;
 142
 143         return 0;
 144 }
 145
 146 /* remove VLAN header from packet and update csum accordingly. */
 147 static int __pop_vlan_tci(struct sk_buff *skb, __be16 *current_tci)
 148 {
 149         struct vlan_hdr *vhdr;
 150         int err;
 151
 152         err = make_writable(skb, VLAN_ETH_HLEN);
 153         if (unlikely(err))
 154                 return err;
 155
 156         if (skb->ip_summed == CHECKSUM_COMPLETE)
 157                 skb->csum = csum_sub(skb->csum, csum_partial(skb->data
 158                                         + (2 * ETH_ALEN), VLAN_HLEN, 0));
 159
 160         vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
 161         *current_tci = vhdr->h_vlan_TCI;
 162
 163         memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
 164         __skb_pull(skb, VLAN_HLEN);
 165
 166         vlan_set_encap_proto(skb, vhdr);
 167         skb->mac_header += VLAN_HLEN;
 168         /* Update mac_len for subsequent MPLS actions */
 169         skb->mac_len -= VLAN_HLEN;
 170
 171         return 0;
 172 }
 173
 174 static int pop_vlan(struct sk_buff *skb)
 175 {
 176         __be16 tci;
 177         int err;
 178
 179         if (likely(vlan_tx_tag_present(skb))) {
 180                 vlan_set_tci(skb, 0);
 181         } else {
 182                 if (unlikely(skb->protocol != htons(ETH_P_8021Q) ||
 183                              skb->len < VLAN_ETH_HLEN))
 184                         return 0;
 185
 186                 err = __pop_vlan_tci(skb, &tci);
 187                 if (err)
 188                         return err;
 189         }
 190         /* move next vlan tag to hw accel tag */
 191         if (likely(skb->protocol != htons(ETH_P_8021Q) ||
 192                    skb->len < VLAN_ETH_HLEN))
 193                 return 0;
 194
 195         err = __pop_vlan_tci(skb, &tci);
 196         if (unlikely(err))
 197                 return err;
 198
 199         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(tci));
 200         return 0;
 201 }
 202
 203 static int push_vlan(struct sk_buff *skb, const struct ovs_action_push_vlan *vlan)
 204 {
 205         if (unlikely(vlan_tx_tag_present(skb))) {
 206                 u16 current_tag;
 207
 208                 /* push down current VLAN tag */
 209                 current_tag = vlan_tx_tag_get(skb);
 210
 211                 if (!__vlan_put_tag(skb, skb->vlan_proto, current_tag))
 212                         return -ENOMEM;
 213
 214                 /* Update mac_len for subsequent MPLS actions */
 215                 skb->mac_len += VLAN_HLEN;
 216
 217                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 218                         skb->csum = csum_add(skb->csum, csum_partial(skb->data
 219                                         + (2 * ETH_ALEN), VLAN_HLEN, 0));
 220
 221         }
 222         __vlan_hwaccel_put_tag(skb, vlan->vlan_tpid, ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
 223         return 0;
 224 }
 225
 226 static int set_eth_addr(struct sk_buff *skb,
 227                         const struct ovs_key_ethernet *eth_key)
 228 {
 229         int err;
 230         err = make_writable(skb, ETH_HLEN);
 231         if (unlikely(err))
 232                 return err;
 233
 234         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 235
 236         ether_addr_copy(eth_hdr(skb)->h_source, eth_key->eth_src);
 237         ether_addr_copy(eth_hdr(skb)->h_dest, eth_key->eth_dst);
 238
 239         ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 240
 241         return 0;
 242 }
 243
 244 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 245                                 __be32 *addr, __be32 new_addr)
 246 {
 247         int transport_len = skb->len - skb_transport_offset(skb);
 248
 249         if (nh->protocol == IPPROTO_TCP) {
 250                 if (likely(transport_len >= sizeof(struct tcphdr)))
 251                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 252                                                  *addr, new_addr, 1);
 253         } else if (nh->protocol == IPPROTO_UDP) {
 254                 if (likely(transport_len >= sizeof(struct udphdr))) {
 255                         struct udphdr *uh = udp_hdr(skb);
 256
 257                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 258                                 inet_proto_csum_replace4(&uh->check, skb,
 259                                                          *addr, new_addr, 1);
 260                                 if (!uh->check)
 261                                         uh->check = CSUM_MANGLED_0;
 262                         }
 263                 }
 264         }
 265
 266         csum_replace4(&nh->check, *addr, new_addr);
 267         skb_clear_hash(skb);
 268         *addr = new_addr;
 269 }
 270
 271 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 272                                  __be32 addr[4], const __be32 new_addr[4])
 273 {
 274         int transport_len = skb->len - skb_transport_offset(skb);
 275
 276         if (l4_proto == IPPROTO_TCP) {
 277                 if (likely(transport_len >= sizeof(struct tcphdr)))
 278                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 279                                                   addr, new_addr, 1);
 280         } else if (l4_proto == IPPROTO_UDP) {
 281                 if (likely(transport_len >= sizeof(struct udphdr))) {
 282                         struct udphdr *uh = udp_hdr(skb);
 283
 284                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 285                                 inet_proto_csum_replace16(&uh->check, skb,
 286                                                           addr, new_addr, 1);
 287                                 if (!uh->check)
 288                                         uh->check = CSUM_MANGLED_0;
 289                         }
 290                 }
 291         }
 292 }
 293
 294 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 295                           __be32 addr[4], const __be32 new_addr[4],
 296                           bool recalculate_csum)
 297 {
 298         if (recalculate_csum)
 299                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 300
 301         skb_clear_hash(skb);
 302         memcpy(addr, new_addr, sizeof(__be32[4]));
 303 }
 304
 305 static void set_ipv6_tc(struct ipv6hdr *nh, u8 tc)
 306 {
 307         nh->priority = tc >> 4;
 308         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0x0F) | ((tc & 0x0F) << 4);
 309 }
 310
 311 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl)
 312 {
 313         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0xF0) | (fl & 0x000F0000) >> 16;
 314         nh->flow_lbl[1] = (fl & 0x0000FF00) >> 8;
 315         nh->flow_lbl[2] = fl & 0x000000FF;
 316 }
 317
 318 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl)
 319 {
 320         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 321         nh->ttl = new_ttl;
 322 }
 323
 324 static int set_ipv4(struct sk_buff *skb, const struct ovs_key_ipv4 *ipv4_key)
 325 {
 326         struct iphdr *nh;
 327         int err;
 328
 329         err = make_writable(skb, skb_network_offset(skb) +
 330                                  sizeof(struct iphdr));
 331         if (unlikely(err))
 332                 return err;
 333
 334         nh = ip_hdr(skb);
 335
 336         if (ipv4_key->ipv4_src != nh->saddr)
 337                 set_ip_addr(skb, nh, &nh->saddr, ipv4_key->ipv4_src);
 338
 339         if (ipv4_key->ipv4_dst != nh->daddr)
 340                 set_ip_addr(skb, nh, &nh->daddr, ipv4_key->ipv4_dst);
 341
 342         if (ipv4_key->ipv4_tos != nh->tos)
 343                 ipv4_change_dsfield(nh, 0, ipv4_key->ipv4_tos);
 344
 345         if (ipv4_key->ipv4_ttl != nh->ttl)
 346                 set_ip_ttl(skb, nh, ipv4_key->ipv4_ttl);
 347
 348         return 0;
 349 }
 350
 351 static int set_ipv6(struct sk_buff *skb, const struct ovs_key_ipv6 *ipv6_key)
 352 {
 353         struct ipv6hdr *nh;
 354         int err;
 355         __be32 *saddr;
 356         __be32 *daddr;
 357
 358         err = make_writable(skb, skb_network_offset(skb) +
 359                             sizeof(struct ipv6hdr));
 360         if (unlikely(err))
 361                 return err;
 362
 363         nh = ipv6_hdr(skb);
 364         saddr = (__be32 *)&nh->saddr;
 365         daddr = (__be32 *)&nh->daddr;
 366
 367         if (memcmp(ipv6_key->ipv6_src, saddr, sizeof(ipv6_key->ipv6_src)))
 368                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, saddr,
 369                               ipv6_key->ipv6_src, true);
 370
 371         if (memcmp(ipv6_key->ipv6_dst, daddr, sizeof(ipv6_key->ipv6_dst))) {
 372                 unsigned int offset = 0;
 373                 int flags = OVS_IP6T_FH_F_SKIP_RH;
 374                 bool recalc_csum = true;
 375
 376                 if (ipv6_ext_hdr(nh->nexthdr))
 377                         recalc_csum = ipv6_find_hdr(skb, &offset,
 378                                                     NEXTHDR_ROUTING, NULL,
 379                                                     &flags) != NEXTHDR_ROUTING;
 380
 381                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, daddr,
 382                               ipv6_key->ipv6_dst, recalc_csum);
 383         }
 384
 385         set_ipv6_tc(nh, ipv6_key->ipv6_tclass);
 386         set_ipv6_fl(nh, ntohl(ipv6_key->ipv6_label));
 387         nh->hop_limit = ipv6_key->ipv6_hlimit;
 388
 389         return 0;
 390 }
 391
 392 /* Must follow make_writable() since that can move the skb data. */
 393 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 394                          __be16 new_port, __sum16 *check)
 395 {
 396         inet_proto_csum_replace2(check, skb, *port, new_port, 0);
 397         *port = new_port;
 398         skb_clear_hash(skb);
 399 }
 400
 401 static void set_udp_port(struct sk_buff *skb, __be16 *port, __be16 new_port)
 402 {
 403         struct udphdr *uh = udp_hdr(skb);
 404
 405         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 406                 set_tp_port(skb, port, new_port, &uh->check);
 407
 408                 if (!uh->check)
 409                         uh->check = CSUM_MANGLED_0;
 410         } else {
 411                 *port = new_port;
 412                 skb_clear_hash(skb);
 413         }
 414 }
 415
 416 static int set_udp(struct sk_buff *skb, const struct ovs_key_udp *udp_port_key)
 417 {
 418         struct udphdr *uh;
 419         int err;
 420
 421         err = make_writable(skb, skb_transport_offset(skb) +
 422                                  sizeof(struct udphdr));
 423         if (unlikely(err))
 424                 return err;
 425
 426         uh = udp_hdr(skb);
 427         if (udp_port_key->udp_src != uh->source)
 428                 set_udp_port(skb, &uh->source, udp_port_key->udp_src);
 429
 430         if (udp_port_key->udp_dst != uh->dest)
 431                 set_udp_port(skb, &uh->dest, udp_port_key->udp_dst);
 432
 433         return 0;
 434 }
 435
 436 static int set_tcp(struct sk_buff *skb, const struct ovs_key_tcp *tcp_port_key)
 437 {
 438         struct tcphdr *th;
 439         int err;
 440
 441         err = make_writable(skb, skb_transport_offset(skb) +
 442                                  sizeof(struct tcphdr));
 443         if (unlikely(err))
 444                 return err;
 445
 446         th = tcp_hdr(skb);
 447         if (tcp_port_key->tcp_src != th->source)
 448                 set_tp_port(skb, &th->source, tcp_port_key->tcp_src, &th->check);
 449
 450         if (tcp_port_key->tcp_dst != th->dest)
 451                 set_tp_port(skb, &th->dest, tcp_port_key->tcp_dst, &th->check);
 452
 453         return 0;
 454 }
 455
 456 static int set_sctp(struct sk_buff *skb,
 457                      const struct ovs_key_sctp *sctp_port_key)
 458 {
 459         struct sctphdr *sh;
 460         int err;
 461         unsigned int sctphoff = skb_transport_offset(skb);
 462
 463         err = make_writable(skb, sctphoff + sizeof(struct sctphdr));
 464         if (unlikely(err))
 465                 return err;
 466
 467         sh = sctp_hdr(skb);
 468         if (sctp_port_key->sctp_src != sh->source ||
 469             sctp_port_key->sctp_dst != sh->dest) {
 470                 __le32 old_correct_csum, new_csum, old_csum;
 471
 472                 old_csum = sh->checksum;
 473                 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 474
 475                 sh->source = sctp_port_key->sctp_src;
 476                 sh->dest = sctp_port_key->sctp_dst;
 477
 478                 new_csum = sctp_compute_cksum(skb, sctphoff);
 479
 480                 /* Carry any checksum errors through. */
 481                 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 482
 483                 skb_clear_hash(skb);
 484         }
 485
 486         return 0;
 487 }
 488
 489 static int do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
 490 {
 491         struct vport *vport;
 492
 493         if (unlikely(!skb))
 494                 return -ENOMEM;
 495
 496         vport = ovs_vport_rcu(dp, out_port);
 497         if (unlikely(!vport)) {
 498                 kfree_skb(skb);
 499                 return -ENODEV;
 500         }
 501
 502         ovs_vport_send(vport, skb);
 503         return 0;
 504 }
 505
 506 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 507                             const struct nlattr *attr)
 508 {
 509         struct dp_upcall_info upcall;
 510         const struct nlattr *a;
 511         int rem;
 512
 513         BUG_ON(!OVS_CB(skb)->pkt_key);
 514
 515         upcall.cmd = OVS_PACKET_CMD_ACTION;
 516         upcall.key = OVS_CB(skb)->pkt_key;
 517         upcall.userdata = NULL;
 518         upcall.portid = 0;
 519
 520         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 521                  a = nla_next(a, &rem)) {
 522                 switch (nla_type(a)) {
 523                 case OVS_USERSPACE_ATTR_USERDATA:
 524                         upcall.userdata = a;
 525                         break;
 526
 527                 case OVS_USERSPACE_ATTR_PID:
 528                         upcall.portid = nla_get_u32(a);
 529                         break;
 530                 }
 531         }
 532
 533         return ovs_dp_upcall(dp, skb, &upcall);
 534 }
 535
 536 static bool last_action(const struct nlattr *a, int rem)
 537 {
 538         return a->nla_len == rem;
 539 }
 540
 541 static int sample(struct datapath *dp, struct sk_buff *skb,
 542                   const struct nlattr *attr)
 543 {
 544         const struct nlattr *acts_list = NULL;
 545         const struct nlattr *a;
 546         struct sk_buff *sample_skb;
 547         int rem;
 548
 549         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 550                  a = nla_next(a, &rem)) {
 551                 switch (nla_type(a)) {
 552                 case OVS_SAMPLE_ATTR_PROBABILITY:
 553                         if (prandom_u32() >= nla_get_u32(a))
 554                                 return 0;
 555                         break;
 556
 557                 case OVS_SAMPLE_ATTR_ACTIONS:
 558                         acts_list = a;
 559                         break;
 560                 }
 561         }
 562
 563         rem = nla_len(acts_list);
 564         a = nla_data(acts_list);
 565
 566         /* Actions list is either empty or only contains a single user-space
 567          * action, the latter being a special case as it is the only known
 568          * usage of the sample action.
 569          * In these special cases don't clone the skb as there are no
 570          * side-effects in the nested actions.
 571          * Otherwise, clone in case the nested actions have side effects. */
 572         if (likely(rem == 0 ||
 573                    (nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
 574                     last_action(a, rem)))) {
 575                 sample_skb = skb;
 576                 skb_get(skb);
 577         } else {
 578                 sample_skb = skb_clone(skb, GFP_ATOMIC);
 579                 if (!sample_skb)
 580                         /* Skip the sample action when out of memory. */
 581                         return 0;
 582         }
 583
 584         /* Note that do_execute_actions() never consumes skb.
 585          * In the case where skb has been cloned above it is the clone that
 586          * is consumed.  Otherwise the skb_get(skb) call prevents
 587          * consumption by do_execute_actions(). Thus, it is safe to simply
 588          * return the error code and let the caller (also
 589          * do_execute_actions()) free skb on error. */
 590         return do_execute_actions(dp, sample_skb, a, rem);
 591 }
 592
 593 static void execute_hash(struct sk_buff *skb, const struct nlattr *attr)
 594 {
 595         struct sw_flow_key *key = OVS_CB(skb)->pkt_key;
 596         struct ovs_action_hash *hash_act = nla_data(attr);
 597         u32 hash = 0;
 598
 599         /* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
 600         hash = skb_get_hash(skb);
 601         hash = jhash_1word(hash, hash_act->hash_basis);
 602         if (!hash)
 603                 hash = 0x1;
 604
 605         key->ovs_flow_hash = hash;
 606 }
 607
 608 static int execute_set_action(struct sk_buff *skb,
 609                                  const struct nlattr *nested_attr)
 610 {
 611         int err = 0;
 612
 613         switch (nla_type(nested_attr)) {
 614         case OVS_KEY_ATTR_PRIORITY:
 615                 skb->priority = nla_get_u32(nested_attr);
 616                 break;
 617
 618         case OVS_KEY_ATTR_SKB_MARK:
 619                 skb->mark = nla_get_u32(nested_attr);
 620                 break;
 621
 622         case OVS_KEY_ATTR_TUNNEL_INFO:
 623                 OVS_CB(skb)->tun_info = nla_data(nested_attr);
 624                 break;
 625
 626         case OVS_KEY_ATTR_ETHERNET:
 627                 err = set_eth_addr(skb, nla_data(nested_attr));
 628                 break;
 629
 630         case OVS_KEY_ATTR_IPV4:
 631                 err = set_ipv4(skb, nla_data(nested_attr));
 632                 break;
 633
 634         case OVS_KEY_ATTR_IPV6:
 635                 err = set_ipv6(skb, nla_data(nested_attr));
 636                 break;
 637
 638         case OVS_KEY_ATTR_TCP:
 639                 err = set_tcp(skb, nla_data(nested_attr));
 640                 break;
 641
 642         case OVS_KEY_ATTR_UDP:
 643                 err = set_udp(skb, nla_data(nested_attr));
 644                 break;
 645
 646         case OVS_KEY_ATTR_SCTP:
 647                 err = set_sctp(skb, nla_data(nested_attr));
 648                 break;
 649
 650         case OVS_KEY_ATTR_MPLS:
 651                 err = set_mpls(skb, nla_data(nested_attr));
 652                 break;
 653         }
 654
 655         return err;
 656 }
 657
 658 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
 659                                  const struct nlattr *a)
 660 {
 661         struct sw_flow_key recirc_key;
 662         const struct vport *p = OVS_CB(skb)->input_vport;
 663         uint32_t hash = OVS_CB(skb)->pkt_key->ovs_flow_hash;
 664         int err;
 665
 666         err = ovs_flow_extract(skb, p->port_no, &recirc_key);
 667         if (err) {
 668                 kfree_skb(skb);
 669                 return err;
 670         }
 671
 672         recirc_key.ovs_flow_hash = hash;
 673         recirc_key.recirc_id = nla_get_u32(a);
 674
 675         ovs_dp_process_packet_with_key(skb, &recirc_key, true);
 676
 677         return 0;
 678 }
 679
 680 /* Execute a list of actions against 'skb'. */
 681 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 682                         const struct nlattr *attr, int len)
 683 {
 684         /* Every output action needs a separate clone of 'skb', but the common
 685          * case is just a single output action, so that doing a clone and
 686          * then freeing the original skbuff is wasteful.  So the following code
 687          * is slightly obscure just to avoid that. */
 688         int prev_port = -1;
 689         const struct nlattr *a;
 690         int rem;
 691
 692         for (a = attr, rem = len; rem > 0;
 693              a = nla_next(a, &rem)) {
 694                 int err = 0;
 695
 696                 if (prev_port != -1) {
 697                         do_output(dp, skb_clone(skb, GFP_ATOMIC), prev_port);
 698                         prev_port = -1;
 699                 }
 700
 701                 switch (nla_type(a)) {
 702                 case OVS_ACTION_ATTR_OUTPUT:
 703                         prev_port = nla_get_u32(a);
 704                         break;
 705
 706                 case OVS_ACTION_ATTR_USERSPACE:
 707                         output_userspace(dp, skb, a);
 708                         break;
 709
 710                 case OVS_ACTION_ATTR_HASH:
 711                         execute_hash(skb, a);
 712                         break;
 713
 714                 case OVS_ACTION_ATTR_PUSH_MPLS:
 715                         err = push_mpls(skb, nla_data(a));
 716                         break;
 717
 718                 case OVS_ACTION_ATTR_POP_MPLS:
 719                         err = pop_mpls(skb, nla_get_be16(a));
 720                         break;
 721
 722                 case OVS_ACTION_ATTR_PUSH_VLAN:
 723                         err = push_vlan(skb, nla_data(a));
 724                         if (unlikely(err)) /* skb already freed. */
 725                                 return err;
 726                         break;
 727
 728                 case OVS_ACTION_ATTR_POP_VLAN:
 729                         err = pop_vlan(skb);
 730                         break;
 731
 732                 case OVS_ACTION_ATTR_RECIRC: {
 733                         struct sk_buff *recirc_skb;
 734
 735                         if (last_action(a, rem))
 736                                 return execute_recirc(dp, skb, a);
 737
 738                         /* Recirc action is the not the last action
 739                          * of the action list. */
 740                         recirc_skb = skb_clone(skb, GFP_ATOMIC);
 741
 742                         /* Skip the recirc action when out of memory, but
 743                          * continue on with the rest of the action list. */
 744                         if (recirc_skb)
 745                                 err = execute_recirc(dp, recirc_skb, a);
 746
 747                         break;
 748                 }
 749
 750                 case OVS_ACTION_ATTR_SET:
 751                         err = execute_set_action(skb, nla_data(a));
 752                         break;
 753
 754                 case OVS_ACTION_ATTR_SAMPLE:
 755                         err = sample(dp, skb, a);
 756                         break;
 757                 }
 758
 759                 if (unlikely(err)) {
 760                         kfree_skb(skb);
 761                         return err;
 762                 }
 763         }
 764
 765         if (prev_port != -1)
 766                 do_output(dp, skb, prev_port);
 767         else
 768                 consume_skb(skb);
 769
 770         return 0;
 771 }
 772
 773 /* We limit the number of times that we pass into execute_actions()
 774  * to avoid blowing out the stack in the event that we have a loop.
 775  *
 776  * Each loop adds some (estimated) cost to the kernel stack.
 777  * The loop terminates when the max cost is exceeded.
 778  * */
 779 #define RECIRC_STACK_COST 1
 780 #define DEFAULT_STACK_COST 4
 781 /* Allow up to 4 regular services, and up to 3 recirculations */
 782 #define MAX_STACK_COST (DEFAULT_STACK_COST * 4 + RECIRC_STACK_COST * 3)
 783
 784 struct loop_counter {
 785         u8 stack_cost;          /* loop stack cost. */
 786         bool looping;           /* Loop detected? */
 787 };
 788
 789 static DEFINE_PER_CPU(struct loop_counter, loop_counters);
 790
 791 static int loop_suppress(struct datapath *dp, struct sw_flow_actions *actions)
 792 {
 793         if (net_ratelimit())
 794                 pr_warn("%s: flow loop detected, dropping\n",
 795                                 ovs_dp_name(dp));
 796         actions->actions_len = 0;
 797         return -ELOOP;
 798 }
 799
 800 /* Execute a list of actions against 'skb'. */
 801 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb, bool recirc)
 802 {
 803         struct sw_flow_actions *acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
 804         const u8 stack_cost = recirc ? RECIRC_STACK_COST : DEFAULT_STACK_COST;
 805         struct loop_counter *loop;
 806         int error;
 807
 808         /* Check whether we've looped too much. */
 809         loop = &__get_cpu_var(loop_counters);
 810         loop->stack_cost += stack_cost;
 811         if (unlikely(loop->stack_cost > MAX_STACK_COST))
 812                 loop->looping = true;
 813         if (unlikely(loop->looping)) {
 814                 error = loop_suppress(dp, acts);
 815                 kfree_skb(skb);
 816                 goto out_loop;
 817         }
 818
 819         OVS_CB(skb)->tun_info = NULL;
 820         error = do_execute_actions(dp, skb, acts->actions, acts->actions_len);
 821
 822         /* Check whether sub-actions looped too much. */
 823         if (unlikely(loop->looping))
 824                 error = loop_suppress(dp, acts);
 825
 826 out_loop:
 827         /* Decrement loop stack cost. */
 828         loop->stack_cost -= stack_cost;
 829         if (!loop->stack_cost)
 830                 loop->looping = false;
 831
 832         return error;
 833 }