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