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                               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 make_writable(struct sk_buff *skb, int write_len)
 126 {
 127         if (!pskb_may_pull(skb, write_len))
 128                 return -ENOMEM;
 129
 130         if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
 131                 return 0;
 132
 133         return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
 134 }
 135
 136 /* The end of the mac header.
 137  *
 138  * For non-MPLS skbs this will correspond to the network header.
 139  * For MPLS skbs it will be before the network_header as the MPLS
 140  * label stack lies between the end of the mac header and the network
 141  * header. That is, for MPLS skbs the end of the mac header
 142  * is the top of the MPLS label stack.
 143  */
 144 static unsigned char *mac_header_end(const struct sk_buff *skb)
 145 {
 146         return skb_mac_header(skb) + skb->mac_len;
 147 }
 148
 149 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 150                      const struct ovs_action_push_mpls *mpls)
 151 {
 152         __be32 *new_mpls_lse;
 153         struct ethhdr *hdr;
 154
 155         if (skb_cow_head(skb, MPLS_HLEN) < 0)
 156                 return -ENOMEM;
 157
 158         skb_push(skb, MPLS_HLEN);
 159         memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
 160                 skb->mac_len);
 161         skb_reset_mac_header(skb);
 162
 163         new_mpls_lse = (__be32 *)mac_header_end(skb);
 164         *new_mpls_lse = mpls->mpls_lse;
 165
 166         if (skb->ip_summed == CHECKSUM_COMPLETE)
 167                 skb->csum = csum_add(skb->csum, csum_partial(new_mpls_lse,
 168                                                              MPLS_HLEN, 0));
 169
 170         hdr = eth_hdr(skb);
 171         hdr->h_proto = mpls->mpls_ethertype;
 172         if (!ovs_skb_get_inner_protocol(skb))
 173                 ovs_skb_set_inner_protocol(skb, skb->protocol);
 174         skb->protocol = mpls->mpls_ethertype;
 175         invalidate_flow_key(key);
 176         return 0;
 177 }
 178
 179 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 180                     const __be16 ethertype)
 181 {
 182         struct ethhdr *hdr;
 183         int err;
 184
 185         err = make_writable(skb, skb->mac_len + MPLS_HLEN);
 186         if (unlikely(err))
 187                 return err;
 188
 189         if (skb->ip_summed == CHECKSUM_COMPLETE)
 190                 skb->csum = csum_sub(skb->csum,
 191                                      csum_partial(mac_header_end(skb),
 192                                                   MPLS_HLEN, 0));
 193
 194         memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
 195                 skb->mac_len);
 196
 197         __skb_pull(skb, MPLS_HLEN);
 198         skb_reset_mac_header(skb);
 199
 200         /* mac_header_end() is used to locate the ethertype
 201          * field correctly in the presence of VLAN tags.
 202          */
 203         hdr = (struct ethhdr *)(mac_header_end(skb) - ETH_HLEN);
 204         hdr->h_proto = ethertype;
 205         if (eth_p_mpls(skb->protocol))
 206                 skb->protocol = ethertype;
 207         invalidate_flow_key(key);
 208         return 0;
 209 }
 210
 211 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 212                     const __be32 *mpls_lse)
 213 {
 214         __be32 *stack = (__be32 *)mac_header_end(skb);
 215         int err;
 216
 217         err = make_writable(skb, skb->mac_len + MPLS_HLEN);
 218         if (unlikely(err))
 219                 return err;
 220
 221         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 222                 __be32 diff[] = { ~(*stack), *mpls_lse };
 223                 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
 224                                           ~skb->csum);
 225         }
 226
 227         *stack = *mpls_lse;
 228         key->mpls.top_lse = *mpls_lse;
 229         return 0;
 230 }
 231
 232 /* remove VLAN header from packet and update csum accordingly. */
 233 static int __pop_vlan_tci(struct sk_buff *skb, __be16 *current_tci)
 234 {
 235         struct vlan_hdr *vhdr;
 236         int err;
 237
 238         err = make_writable(skb, VLAN_ETH_HLEN);
 239         if (unlikely(err))
 240                 return err;
 241
 242         if (skb->ip_summed == CHECKSUM_COMPLETE)
 243                 skb->csum = csum_sub(skb->csum, csum_partial(skb->data
 244                                         + (2 * ETH_ALEN), VLAN_HLEN, 0));
 245
 246         vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
 247         *current_tci = vhdr->h_vlan_TCI;
 248
 249         memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
 250         __skb_pull(skb, VLAN_HLEN);
 251
 252         vlan_set_encap_proto(skb, vhdr);
 253         skb->mac_header += VLAN_HLEN;
 254         /* Update mac_len for subsequent MPLS actions */
 255         skb->mac_len -= VLAN_HLEN;
 256
 257         return 0;
 258 }
 259
 260 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 261 {
 262         __be16 tci;
 263         int err;
 264
 265         if (likely(vlan_tx_tag_present(skb))) {
 266                 vlan_set_tci(skb, 0);
 267         } else {
 268                 if (unlikely(skb->protocol != htons(ETH_P_8021Q) ||
 269                              skb->len < VLAN_ETH_HLEN))
 270                         return 0;
 271
 272                 err = __pop_vlan_tci(skb, &tci);
 273                 if (err)
 274                         return err;
 275         }
 276         /* move next vlan tag to hw accel tag */
 277         if (likely(skb->protocol != htons(ETH_P_8021Q) ||
 278                    skb->len < VLAN_ETH_HLEN)) {
 279                 key->eth.tci = 0;
 280                 return 0;
 281         }
 282
 283         invalidate_flow_key(key);
 284         err = __pop_vlan_tci(skb, &tci);
 285         if (unlikely(err))
 286                 return err;
 287
 288         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(tci));
 289         return 0;
 290 }
 291
 292 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
 293                      const struct ovs_action_push_vlan *vlan)
 294 {
 295         if (unlikely(vlan_tx_tag_present(skb))) {
 296                 u16 current_tag;
 297
 298                 /* push down current VLAN tag */
 299                 current_tag = vlan_tx_tag_get(skb);
 300
 301                 if (!__vlan_put_tag(skb, skb->vlan_proto, current_tag))
 302                         return -ENOMEM;
 303
 304                 /* Update mac_len for subsequent MPLS actions */
 305                 skb->mac_len += VLAN_HLEN;
 306
 307                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 308                         skb->csum = csum_add(skb->csum, csum_partial(skb->data
 309                                         + (2 * ETH_ALEN), VLAN_HLEN, 0));
 310
 311                 invalidate_flow_key(key);
 312         } else {
 313                 key->eth.tci = vlan->vlan_tci;
 314         }
 315         __vlan_hwaccel_put_tag(skb, vlan->vlan_tpid, ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
 316         return 0;
 317 }
 318
 319 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *key,
 320                         const struct ovs_key_ethernet *eth_key)
 321 {
 322         int err;
 323         err = make_writable(skb, ETH_HLEN);
 324         if (unlikely(err))
 325                 return err;
 326
 327         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 328
 329         ether_addr_copy(eth_hdr(skb)->h_source, eth_key->eth_src);
 330         ether_addr_copy(eth_hdr(skb)->h_dest, eth_key->eth_dst);
 331
 332         ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 333
 334         ether_addr_copy(key->eth.src, eth_key->eth_src);
 335         ether_addr_copy(key->eth.dst, eth_key->eth_dst);
 336         return 0;
 337 }
 338
 339 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 340                         __be32 *addr, __be32 new_addr)
 341 {
 342         int transport_len = skb->len - skb_transport_offset(skb);
 343
 344         if (nh->protocol == IPPROTO_TCP) {
 345                 if (likely(transport_len >= sizeof(struct tcphdr)))
 346                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 347                                                  *addr, new_addr, 1);
 348         } else if (nh->protocol == IPPROTO_UDP) {
 349                 if (likely(transport_len >= sizeof(struct udphdr))) {
 350                         struct udphdr *uh = udp_hdr(skb);
 351
 352                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 353                                 inet_proto_csum_replace4(&uh->check, skb,
 354                                                          *addr, new_addr, 1);
 355                                 if (!uh->check)
 356                                         uh->check = CSUM_MANGLED_0;
 357                         }
 358                 }
 359         }
 360
 361         csum_replace4(&nh->check, *addr, new_addr);
 362         skb_clear_hash(skb);
 363         *addr = new_addr;
 364 }
 365
 366 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 367                                  __be32 addr[4], const __be32 new_addr[4])
 368 {
 369         int transport_len = skb->len - skb_transport_offset(skb);
 370
 371         if (l4_proto == NEXTHDR_TCP) {
 372                 if (likely(transport_len >= sizeof(struct tcphdr)))
 373                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 374                                                   addr, new_addr, 1);
 375         } else if (l4_proto == NEXTHDR_UDP) {
 376                 if (likely(transport_len >= sizeof(struct udphdr))) {
 377                         struct udphdr *uh = udp_hdr(skb);
 378
 379                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 380                                 inet_proto_csum_replace16(&uh->check, skb,
 381                                                           addr, new_addr, 1);
 382                                 if (!uh->check)
 383                                         uh->check = CSUM_MANGLED_0;
 384                         }
 385                 }
 386         } else if (l4_proto == NEXTHDR_ICMP) {
 387                 if (likely(transport_len >= sizeof(struct icmp6hdr)))
 388                         inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 389                                                   skb, addr, new_addr, 1);
 390         }
 391 }
 392
 393 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 394                           __be32 addr[4], const __be32 new_addr[4],
 395                           bool recalculate_csum)
 396 {
 397         if (likely(recalculate_csum))
 398                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 399
 400         skb_clear_hash(skb);
 401         memcpy(addr, new_addr, sizeof(__be32[4]));
 402 }
 403
 404 static void set_ipv6_tc(struct ipv6hdr *nh, u8 tc)
 405 {
 406         nh->priority = tc >> 4;
 407         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0x0F) | ((tc & 0x0F) << 4);
 408 }
 409
 410 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl)
 411 {
 412         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0xF0) | (fl & 0x000F0000) >> 16;
 413         nh->flow_lbl[1] = (fl & 0x0000FF00) >> 8;
 414         nh->flow_lbl[2] = fl & 0x000000FF;
 415 }
 416
 417 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl)
 418 {
 419         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 420         nh->ttl = new_ttl;
 421 }
 422
 423 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *key,
 424                     const struct ovs_key_ipv4 *ipv4_key)
 425 {
 426         struct iphdr *nh;
 427         int err;
 428
 429         err = make_writable(skb, skb_network_offset(skb) +
 430                                  sizeof(struct iphdr));
 431         if (unlikely(err))
 432                 return err;
 433
 434         nh = ip_hdr(skb);
 435
 436         if (ipv4_key->ipv4_src != nh->saddr) {
 437                 set_ip_addr(skb, nh, &nh->saddr, ipv4_key->ipv4_src);
 438                 key->ipv4.addr.src = ipv4_key->ipv4_src;
 439         }
 440
 441         if (ipv4_key->ipv4_dst != nh->daddr) {
 442                 set_ip_addr(skb, nh, &nh->daddr, ipv4_key->ipv4_dst);
 443                 key->ipv4.addr.dst = ipv4_key->ipv4_dst;
 444         }
 445
 446         if (ipv4_key->ipv4_tos != nh->tos) {
 447                 ipv4_change_dsfield(nh, 0, ipv4_key->ipv4_tos);
 448                 key->ip.tos = nh->tos;
 449         }
 450
 451         if (ipv4_key->ipv4_ttl != nh->ttl) {
 452                 set_ip_ttl(skb, nh, ipv4_key->ipv4_ttl);
 453                 key->ip.ttl = ipv4_key->ipv4_ttl;
 454         }
 455
 456         return 0;
 457 }
 458
 459 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *key,
 460                     const struct ovs_key_ipv6 *ipv6_key)
 461 {
 462         struct ipv6hdr *nh;
 463         int err;
 464         __be32 *saddr;
 465         __be32 *daddr;
 466
 467         err = make_writable(skb, skb_network_offset(skb) +
 468                             sizeof(struct ipv6hdr));
 469         if (unlikely(err))
 470                 return err;
 471
 472         nh = ipv6_hdr(skb);
 473         saddr = (__be32 *)&nh->saddr;
 474         daddr = (__be32 *)&nh->daddr;
 475
 476         if (memcmp(ipv6_key->ipv6_src, saddr, sizeof(ipv6_key->ipv6_src))) {
 477                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, saddr,
 478                               ipv6_key->ipv6_src, true);
 479                 memcpy(&key->ipv6.addr.src, ipv6_key->ipv6_src,
 480                        sizeof(ipv6_key->ipv6_src));
 481         }
 482
 483         if (memcmp(ipv6_key->ipv6_dst, daddr, sizeof(ipv6_key->ipv6_dst))) {
 484                 unsigned int offset = 0;
 485                 int flags = IP6_FH_F_SKIP_RH;
 486                 bool recalc_csum = true;
 487
 488                 if (ipv6_ext_hdr(nh->nexthdr))
 489                         recalc_csum = ipv6_find_hdr(skb, &offset,
 490                                                     NEXTHDR_ROUTING, NULL,
 491                                                     &flags) != NEXTHDR_ROUTING;
 492
 493                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, daddr,
 494                               ipv6_key->ipv6_dst, recalc_csum);
 495                 memcpy(&key->ipv6.addr.dst, ipv6_key->ipv6_dst,
 496                        sizeof(ipv6_key->ipv6_dst));
 497         }
 498
 499         set_ipv6_tc(nh, ipv6_key->ipv6_tclass);
 500         key->ip.tos = ipv6_get_dsfield(nh);
 501
 502         set_ipv6_fl(nh, ntohl(ipv6_key->ipv6_label));
 503         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 504
 505         nh->hop_limit = ipv6_key->ipv6_hlimit;
 506         key->ip.ttl = ipv6_key->ipv6_hlimit;
 507         return 0;
 508 }
 509
 510 /* Must follow make_writable() since that can move the skb data. */
 511 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 512                          __be16 new_port, __sum16 *check)
 513 {
 514         inet_proto_csum_replace2(check, skb, *port, new_port, 0);
 515         *port = new_port;
 516         skb_clear_hash(skb);
 517 }
 518
 519 static void set_udp_port(struct sk_buff *skb, __be16 *port, __be16 new_port)
 520 {
 521         struct udphdr *uh = udp_hdr(skb);
 522
 523         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 524                 set_tp_port(skb, port, new_port, &uh->check);
 525
 526                 if (!uh->check)
 527                         uh->check = CSUM_MANGLED_0;
 528         } else {
 529                 *port = new_port;
 530                 skb_clear_hash(skb);
 531         }
 532 }
 533
 534 static int set_udp(struct sk_buff *skb, struct sw_flow_key *key,
 535                    const struct ovs_key_udp *udp_port_key)
 536 {
 537         struct udphdr *uh;
 538         int err;
 539
 540         err = make_writable(skb, skb_transport_offset(skb) +
 541                                  sizeof(struct udphdr));
 542         if (unlikely(err))
 543                 return err;
 544
 545         uh = udp_hdr(skb);
 546         if (udp_port_key->udp_src != uh->source) {
 547                 set_udp_port(skb, &uh->source, udp_port_key->udp_src);
 548                 key->tp.src = udp_port_key->udp_src;
 549         }
 550
 551         if (udp_port_key->udp_dst != uh->dest) {
 552                 set_udp_port(skb, &uh->dest, udp_port_key->udp_dst);
 553                 key->tp.dst = udp_port_key->udp_dst;
 554         }
 555
 556         return 0;
 557 }
 558
 559 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *key,
 560                    const struct ovs_key_tcp *tcp_port_key)
 561 {
 562         struct tcphdr *th;
 563         int err;
 564
 565         err = make_writable(skb, skb_transport_offset(skb) +
 566                                  sizeof(struct tcphdr));
 567         if (unlikely(err))
 568                 return err;
 569
 570         th = tcp_hdr(skb);
 571         if (tcp_port_key->tcp_src != th->source) {
 572                 set_tp_port(skb, &th->source, tcp_port_key->tcp_src, &th->check);
 573                 key->tp.src = tcp_port_key->tcp_src;
 574         }
 575
 576         if (tcp_port_key->tcp_dst != th->dest) {
 577                 set_tp_port(skb, &th->dest, tcp_port_key->tcp_dst, &th->check);
 578                 key->tp.dst = tcp_port_key->tcp_dst;
 579         }
 580
 581         return 0;
 582 }
 583
 584 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *key,
 585                     const struct ovs_key_sctp *sctp_port_key)
 586 {
 587         struct sctphdr *sh;
 588         int err;
 589         unsigned int sctphoff = skb_transport_offset(skb);
 590
 591         err = make_writable(skb, sctphoff + sizeof(struct sctphdr));
 592         if (unlikely(err))
 593                 return err;
 594
 595         sh = sctp_hdr(skb);
 596         if (sctp_port_key->sctp_src != sh->source ||
 597             sctp_port_key->sctp_dst != sh->dest) {
 598                 __le32 old_correct_csum, new_csum, old_csum;
 599
 600                 old_csum = sh->checksum;
 601                 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 602
 603                 sh->source = sctp_port_key->sctp_src;
 604                 sh->dest = sctp_port_key->sctp_dst;
 605
 606                 new_csum = sctp_compute_cksum(skb, sctphoff);
 607
 608                 /* Carry any checksum errors through. */
 609                 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 610
 611                 skb_clear_hash(skb);
 612                 key->tp.src = sctp_port_key->sctp_src;
 613                 key->tp.dst = sctp_port_key->sctp_dst;
 614         }
 615
 616         return 0;
 617 }
 618
 619 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
 620 {
 621         struct vport *vport = ovs_vport_rcu(dp, out_port);
 622
 623         if (likely(vport))
 624                 ovs_vport_send(vport, skb);
 625         else
 626                 kfree_skb(skb);
 627 }
 628
 629 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 630                             struct sw_flow_key *key, const struct nlattr *attr)
 631 {
 632         struct dp_upcall_info upcall;
 633         const struct nlattr *a;
 634         int rem;
 635         struct ovs_tunnel_info info;
 636
 637         upcall.cmd = OVS_PACKET_CMD_ACTION;
 638         upcall.userdata = NULL;
 639         upcall.portid = 0;
 640         upcall.egress_tun_info = NULL;
 641
 642         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 643                  a = nla_next(a, &rem)) {
 644                 switch (nla_type(a)) {
 645                 case OVS_USERSPACE_ATTR_USERDATA:
 646                         upcall.userdata = a;
 647                         break;
 648
 649                 case OVS_USERSPACE_ATTR_PID:
 650                         upcall.portid = nla_get_u32(a);
 651                         break;
 652
 653                 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
 654                         /* Get out tunnel info. */
 655                         struct vport *vport;
 656
 657                         vport = ovs_vport_rcu(dp, nla_get_u32(a));
 658                         if (vport) {
 659                                 int err;
 660
 661                                 err = ovs_vport_get_egress_tun_info(vport, skb,
 662                                                                     &info);
 663                                 if (!err)
 664                                         upcall.egress_tun_info = &info;
 665                         }
 666                         break;
 667                 }
 668
 669                 } /* End of switch. */
 670         }
 671
 672         return ovs_dp_upcall(dp, skb, key, &upcall);
 673 }
 674
 675 static bool last_action(const struct nlattr *a, int rem)
 676 {
 677         return a->nla_len == rem;
 678 }
 679
 680 static int sample(struct datapath *dp, struct sk_buff *skb,
 681                   struct sw_flow_key *key, const struct nlattr *attr)
 682 {
 683         const struct nlattr *acts_list = NULL;
 684         const struct nlattr *a;
 685         int rem;
 686
 687         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 688                  a = nla_next(a, &rem)) {
 689                 switch (nla_type(a)) {
 690                 case OVS_SAMPLE_ATTR_PROBABILITY:
 691                         if (prandom_u32() >= nla_get_u32(a))
 692                                 return 0;
 693                         break;
 694
 695                 case OVS_SAMPLE_ATTR_ACTIONS:
 696                         acts_list = a;
 697                         break;
 698                 }
 699         }
 700
 701         rem = nla_len(acts_list);
 702         a = nla_data(acts_list);
 703
 704         /* Actions list is empty, do nothing */
 705         if (unlikely(!rem))
 706                 return 0;
 707
 708         /* The only known usage of sample action is having a single user-space
 709          * action. Treat this usage as a special case.
 710          * The output_userspace() should clone the skb to be sent to the
 711          * user space. This skb will be consumed by its caller.
 712          */
 713         if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
 714                    last_action(a, rem)))
 715                 return output_userspace(dp, skb, key, a);
 716
 717         skb = skb_clone(skb, GFP_ATOMIC);
 718         if (!skb)
 719                 /* Skip the sample action when out of memory. */
 720                 return 0;
 721
 722         if (!add_deferred_actions(skb, key, a)) {
 723                 if (net_ratelimit())
 724                         pr_warn("%s: deferred actions limit reached, dropping sample action\n",
 725                                 ovs_dp_name(dp));
 726
 727                 kfree_skb(skb);
 728         }
 729         return 0;
 730 }
 731
 732 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
 733                          const struct nlattr *attr)
 734 {
 735         struct ovs_action_hash *hash_act = nla_data(attr);
 736         u32 hash = 0;
 737
 738         /* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
 739         hash = skb_get_hash(skb);
 740         hash = jhash_1word(hash, hash_act->hash_basis);
 741         if (!hash)
 742                 hash = 0x1;
 743
 744         key->ovs_flow_hash = hash;
 745 }
 746
 747 static int execute_set_action(struct sk_buff *skb, struct sw_flow_key *key,
 748                               const struct nlattr *nested_attr)
 749 {
 750         int err = 0;
 751
 752         switch (nla_type(nested_attr)) {
 753         case OVS_KEY_ATTR_PRIORITY:
 754                 skb->priority = nla_get_u32(nested_attr);
 755                 key->phy.priority = skb->priority;
 756                 break;
 757
 758         case OVS_KEY_ATTR_SKB_MARK:
 759                 skb->mark = nla_get_u32(nested_attr);
 760                 key->phy.skb_mark = skb->mark;
 761                 break;
 762
 763         case OVS_KEY_ATTR_TUNNEL_INFO:
 764                 OVS_CB(skb)->egress_tun_info = nla_data(nested_attr);
 765                 break;
 766
 767         case OVS_KEY_ATTR_ETHERNET:
 768                 err = set_eth_addr(skb, key, nla_data(nested_attr));
 769                 break;
 770
 771         case OVS_KEY_ATTR_IPV4:
 772                 err = set_ipv4(skb, key, nla_data(nested_attr));
 773                 break;
 774
 775         case OVS_KEY_ATTR_IPV6:
 776                 err = set_ipv6(skb, key, nla_data(nested_attr));
 777                 break;
 778
 779         case OVS_KEY_ATTR_TCP:
 780                 err = set_tcp(skb, key, nla_data(nested_attr));
 781                 break;
 782
 783         case OVS_KEY_ATTR_UDP:
 784                 err = set_udp(skb, key, nla_data(nested_attr));
 785                 break;
 786
 787         case OVS_KEY_ATTR_SCTP:
 788                 err = set_sctp(skb, key, nla_data(nested_attr));
 789                 break;
 790
 791         case OVS_KEY_ATTR_MPLS:
 792                 err = set_mpls(skb, key, nla_data(nested_attr));
 793                 break;
 794         }
 795
 796         return err;
 797 }
 798
 799 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
 800                           struct sw_flow_key *key, const struct nlattr *a, int rem)
 801 {
 802         struct deferred_action *da;
 803
 804         if (!is_flow_key_valid(key)) {
 805                 int err;
 806
 807                 err = ovs_flow_key_update(skb, key);
 808                 if (err)
 809                         return err;
 810
 811         }
 812         BUG_ON(!is_flow_key_valid(key));
 813
 814         if (!last_action(a, rem)) {
 815                 /* Recirc action is the not the last action
 816                  * of the action list, need to clone the skb.
 817                  */
 818                 skb = skb_clone(skb, GFP_ATOMIC);
 819
 820                 /* Skip the recirc action when out of memory, but
 821                  * continue on with the rest of the action list.
 822                  */
 823                 if (!skb)
 824                         return 0;
 825         }
 826
 827         da = add_deferred_actions(skb, key, NULL);
 828         if (da) {
 829                 da->pkt_key.recirc_id = nla_get_u32(a);
 830         } else {
 831                 kfree_skb(skb);
 832
 833                 if (net_ratelimit())
 834                         pr_warn("%s: deferred action limit reached, drop recirc action\n",
 835                                 ovs_dp_name(dp));
 836         }
 837
 838         return 0;
 839 }
 840
 841 /* Execute a list of actions against 'skb'. */
 842 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 843                               struct sw_flow_key *key,
 844                               const struct nlattr *attr, int len)
 845 {
 846         /* Every output action needs a separate clone of 'skb', but the common
 847          * case is just a single output action, so that doing a clone and
 848          * then freeing the original skbuff is wasteful.  So the following code
 849          * is slightly obscure just to avoid that.
 850          */
 851         int prev_port = -1;
 852         const struct nlattr *a;
 853         int rem;
 854
 855         for (a = attr, rem = len; rem > 0;
 856              a = nla_next(a, &rem)) {
 857                 int err = 0;
 858
 859                 if (unlikely(prev_port != -1)) {
 860                         struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
 861
 862                         if (out_skb)
 863                                 do_output(dp, out_skb, prev_port);
 864
 865                         prev_port = -1;
 866                 }
 867
 868                 switch (nla_type(a)) {
 869                 case OVS_ACTION_ATTR_OUTPUT:
 870                         prev_port = nla_get_u32(a);
 871                         break;
 872
 873                 case OVS_ACTION_ATTR_USERSPACE:
 874                         output_userspace(dp, skb, key, a);
 875                         break;
 876
 877                 case OVS_ACTION_ATTR_HASH:
 878                         execute_hash(skb, key, a);
 879                         break;
 880
 881                 case OVS_ACTION_ATTR_PUSH_MPLS:
 882                         err = push_mpls(skb, key, nla_data(a));
 883                         break;
 884
 885                 case OVS_ACTION_ATTR_POP_MPLS:
 886                         err = pop_mpls(skb, key, nla_get_be16(a));
 887                         break;
 888
 889                 case OVS_ACTION_ATTR_PUSH_VLAN:
 890                         err = push_vlan(skb, key, nla_data(a));
 891                         if (unlikely(err)) /* skb already freed. */
 892                                 return err;
 893                         break;
 894
 895                 case OVS_ACTION_ATTR_POP_VLAN:
 896                         err = pop_vlan(skb, key);
 897                         break;
 898
 899                 case OVS_ACTION_ATTR_RECIRC:
 900                         err = execute_recirc(dp, skb, key, a, rem);
 901                         if (last_action(a, rem)) {
 902                                 /* If this is the last action, the skb has
 903                                  * been consumed or freed.
 904                                  * Return immediately.
 905                                  */
 906                                 return err;
 907                         }
 908                         break;
 909
 910                 case OVS_ACTION_ATTR_SET:
 911                         err = execute_set_action(skb, key, nla_data(a));
 912                         break;
 913
 914                 case OVS_ACTION_ATTR_SAMPLE:
 915                         err = sample(dp, skb, key, a);
 916                         break;
 917                 }
 918
 919                 if (unlikely(err)) {
 920                         kfree_skb(skb);
 921                         return err;
 922                 }
 923         }
 924
 925         if (prev_port != -1)
 926                 do_output(dp, skb, prev_port);
 927         else
 928                 consume_skb(skb);
 929
 930         return 0;
 931 }
 932
 933 static void process_deferred_actions(struct datapath *dp)
 934 {
 935         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
 936
 937         /* Do not touch the FIFO in case there is no deferred actions. */
 938         if (action_fifo_is_empty(fifo))
 939                 return;
 940
 941         /* Finishing executing all deferred actions. */
 942         do {
 943                 struct deferred_action *da = action_fifo_get(fifo);
 944                 struct sk_buff *skb = da->skb;
 945                 const struct nlattr *actions = da->actions;
 946
 947                 if (actions)
 948                         do_execute_actions(dp, skb, &da->pkt_key, actions,
 949                                            nla_len(actions));
 950                 else
 951                         ovs_dp_process_packet(skb, &da->pkt_key);
 952         } while (!action_fifo_is_empty(fifo));
 953
 954         /* Reset FIFO for the next packet.  */
 955         action_fifo_init(fifo);
 956 }
 957
 958 /* Execute a list of actions against 'skb'. */
 959 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
 960                         struct sw_flow_key *key,
 961                         const struct sw_flow_actions *acts)
 962 {
 963         int level = this_cpu_read(exec_actions_level);
 964         int err;
 965
 966         if (unlikely(level >= EXEC_ACTIONS_LEVEL_LIMIT)) {
 967                 if (net_ratelimit())
 968                         pr_warn("%s: packet loop detected, dropping.\n",
 969                                 ovs_dp_name(dp));
 970
 971                 kfree_skb(skb);
 972                 return -ELOOP;
 973         }
 974
 975         this_cpu_inc(exec_actions_level);
 976
 977         err = do_execute_actions(dp, skb, key, acts->actions, acts->actions_len);
 978
 979         if (!level)
 980                 process_deferred_actions(dp);
 981
 982         this_cpu_dec(exec_actions_level);
 983
 984         /* This return status currently does not reflect the errors
 985          * encounted during deferred actions execution. Probably needs to
 986          * be fixed in the future.
 987          */
 988         return err;
 989 }
 990
 991 int action_fifos_init(void)
 992 {
 993         action_fifos = alloc_percpu(struct action_fifo);
 994         if (!action_fifos)
 995                 return -ENOMEM;
 996
 997         return 0;
 998 }
 999
1000 void action_fifos_exit(void)
1001 {
1002         free_percpu(action_fifos);
1003 }