datapath/actions.c

   1 /*
   2  * Copyright (c) 2007-2014 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/skbuff.h>
  22 #include <linux/in.h>
  23 #include <linux/ip.h>
  24 #include <linux/openvswitch.h>
  25 #include <linux/sctp.h>
  26 #include <linux/tcp.h>
  27 #include <linux/udp.h>
  28 #include <linux/in6.h>
  29 #include <linux/if_arp.h>
  30 #include <linux/if_vlan.h>
  31 #include <net/ip.h>
  32 #include <net/ipv6.h>
  33 #include <net/checksum.h>
  34 #include <net/dsfield.h>
  35 #include <net/mpls.h>
  36 #include <net/sctp/checksum.h>
  37
  38 #include "datapath.h"
  39 #include "gso.h"
  40 #include "vlan.h"
  41 #include "vport.h"
  42
  43 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
  44                               struct sw_flow_key *key,
  45                               const struct nlattr *attr, int len);
  46
  47 struct deferred_action {
  48         struct sk_buff *skb;
  49         const struct nlattr *actions;
  50
  51         /* Store pkt_key clone when creating deferred action. */
  52         struct sw_flow_key pkt_key;
  53 };
  54
  55 #define DEFERRED_ACTION_FIFO_SIZE 10
  56 struct action_fifo {
  57         int head;
  58         int tail;
  59         /* Deferred action fifo queue storage. */
  60         struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
  61 };
  62
  63 static struct action_fifo __percpu *action_fifos;
  64 #define EXEC_ACTIONS_LEVEL_LIMIT 4   /* limit used to detect packet
  65                                       * looping by the network stack
  66                                       */
  67 static DEFINE_PER_CPU(int, exec_actions_level);
  68
  69 static void action_fifo_init(struct action_fifo *fifo)
  70 {
  71         fifo->head = 0;
  72         fifo->tail = 0;
  73 }
  74
  75 static bool action_fifo_is_empty(const struct action_fifo *fifo)
  76 {
  77         return (fifo->head == fifo->tail);
  78 }
  79
  80 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
  81 {
  82         if (action_fifo_is_empty(fifo))
  83                 return NULL;
  84
  85         return &fifo->fifo[fifo->tail++];
  86 }
  87
  88 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
  89 {
  90         if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
  91                 return NULL;
  92
  93         return &fifo->fifo[fifo->head++];
  94 }
  95
  96 /* Return queue entry if fifo is not full */
  97 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
  98                                                     const struct sw_flow_key *key,
  99                                                     const struct nlattr *attr)
 100 {
 101         struct action_fifo *fifo;
 102         struct deferred_action *da;
 103
 104         fifo = this_cpu_ptr(action_fifos);
 105         da = action_fifo_put(fifo);
 106         if (da) {
 107                 da->skb = skb;
 108                 da->actions = attr;
 109                 da->pkt_key = *key;
 110         }
 111
 112         return da;
 113 }
 114
 115 static void invalidate_flow_key(struct sw_flow_key *key)
 116 {
 117         key->eth.type = htons(0);
 118 }
 119
 120 static bool is_flow_key_valid(const struct sw_flow_key *key)
 121 {
 122         return !!key->eth.type;
 123 }
 124
 125 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 126                      const struct ovs_action_push_mpls *mpls)
 127 {
 128         __be32 *new_mpls_lse;
 129         struct ethhdr *hdr;
 130
 131         /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
 132         if (skb_encapsulation(skb))
 133                 return -ENOTSUPP;
 134
 135         if (skb_cow_head(skb, MPLS_HLEN) < 0)
 136                 return -ENOMEM;
 137
 138         skb_push(skb, MPLS_HLEN);
 139         memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
 140                 skb->mac_len);
 141         skb_reset_mac_header(skb);
 142
 143         new_mpls_lse = (__be32 *)skb_mpls_header(skb);
 144         *new_mpls_lse = mpls->mpls_lse;
 145
 146         if (skb->ip_summed == CHECKSUM_COMPLETE)
 147                 skb->csum = csum_add(skb->csum, csum_partial(new_mpls_lse,
 148                                                              MPLS_HLEN, 0));
 149
 150         hdr = eth_hdr(skb);
 151         hdr->h_proto = mpls->mpls_ethertype;
 152         if (!ovs_skb_get_inner_protocol(skb))
 153                 ovs_skb_set_inner_protocol(skb, skb->protocol);
 154         skb->protocol = mpls->mpls_ethertype;
 155
 156         invalidate_flow_key(key);
 157         return 0;
 158 }
 159
 160 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 161                     const __be16 ethertype)
 162 {
 163         struct ethhdr *hdr;
 164         int err;
 165
 166         err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
 167         if (unlikely(err))
 168                 return err;
 169
 170         if (skb->ip_summed == CHECKSUM_COMPLETE)
 171                 skb->csum = csum_sub(skb->csum,
 172                                      csum_partial(skb_mpls_header(skb),
 173                                                   MPLS_HLEN, 0));
 174
 175         memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
 176                 skb->mac_len);
 177
 178         __skb_pull(skb, MPLS_HLEN);
 179         skb_reset_mac_header(skb);
 180
 181         /* skb_mpls_header() is used to locate the ethertype
 182          * field correctly in the presence of VLAN tags.
 183          */
 184         hdr = (struct ethhdr *)(skb_mpls_header(skb) - ETH_HLEN);
 185         hdr->h_proto = ethertype;
 186         if (eth_p_mpls(skb->protocol))
 187                 skb->protocol = ethertype;
 188
 189         invalidate_flow_key(key);
 190         return 0;
 191 }
 192
 193 /* 'KEY' must not have any bits set outside of the 'MASK' */
 194 #define MASKED(OLD, KEY, MASK) ((KEY) | ((OLD) & ~(MASK)))
 195 #define SET_MASKED(OLD, KEY, MASK) ((OLD) = MASKED(OLD, KEY, MASK))
 196
 197 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
 198                     const __be32 *mpls_lse, const __be32 *mask)
 199 {
 200         __be32 *stack;
 201         __be32 lse;
 202         int err;
 203
 204         err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
 205         if (unlikely(err))
 206                 return err;
 207
 208         stack = (__be32 *)skb_mpls_header(skb);
 209         lse = MASKED(*stack, *mpls_lse, *mask);
 210         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 211                 __be32 diff[] = { ~(*stack), lse };
 212
 213                 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
 214                                           ~skb->csum);
 215         }
 216
 217         *stack = lse;
 218         flow_key->mpls.top_lse = lse;
 219         return 0;
 220 }
 221
 222 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 223 {
 224         int err;
 225
 226         err = skb_vlan_pop(skb);
 227         if (skb_vlan_tag_present(skb))
 228                 invalidate_flow_key(key);
 229         else
 230                 key->eth.tci = 0;
 231
 232         return err;
 233 }
 234
 235 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
 236                      const struct ovs_action_push_vlan *vlan)
 237 {
 238         if (skb_vlan_tag_present(skb))
 239                 invalidate_flow_key(key);
 240         else
 241                 key->eth.tci = vlan->vlan_tci;
 242
 243         return skb_vlan_push(skb, vlan->vlan_tpid,
 244                              ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
 245 }
 246
 247 /* 'src' is already properly masked. */
 248 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
 249 {
 250         u16 *dst = (u16 *)dst_;
 251         const u16 *src = (const u16 *)src_;
 252         const u16 *mask = (const u16 *)mask_;
 253
 254         SET_MASKED(dst[0], src[0], mask[0]);
 255         SET_MASKED(dst[1], src[1], mask[1]);
 256         SET_MASKED(dst[2], src[2], mask[2]);
 257 }
 258
 259 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
 260                         const struct ovs_key_ethernet *key,
 261                         const struct ovs_key_ethernet *mask)
 262 {
 263         int err;
 264
 265         err = skb_ensure_writable(skb, ETH_HLEN);
 266         if (unlikely(err))
 267                 return err;
 268
 269         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 270
 271         ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
 272                                mask->eth_src);
 273         ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
 274                                mask->eth_dst);
 275
 276         ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 277
 278         ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
 279         ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
 280         return 0;
 281 }
 282
 283 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 284                         __be32 *addr, __be32 new_addr)
 285 {
 286         int transport_len = skb->len - skb_transport_offset(skb);
 287
 288         if (nh->protocol == IPPROTO_TCP) {
 289                 if (likely(transport_len >= sizeof(struct tcphdr)))
 290                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 291                                                  *addr, new_addr, 1);
 292         } else if (nh->protocol == IPPROTO_UDP) {
 293                 if (likely(transport_len >= sizeof(struct udphdr))) {
 294                         struct udphdr *uh = udp_hdr(skb);
 295
 296                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 297                                 inet_proto_csum_replace4(&uh->check, skb,
 298                                                          *addr, new_addr, 1);
 299                                 if (!uh->check)
 300                                         uh->check = CSUM_MANGLED_0;
 301                         }
 302                 }
 303         }
 304
 305         csum_replace4(&nh->check, *addr, new_addr);
 306         skb_clear_hash(skb);
 307         *addr = new_addr;
 308 }
 309
 310 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 311                                  __be32 addr[4], const __be32 new_addr[4])
 312 {
 313         int transport_len = skb->len - skb_transport_offset(skb);
 314
 315         if (l4_proto == NEXTHDR_TCP) {
 316                 if (likely(transport_len >= sizeof(struct tcphdr)))
 317                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 318                                                   addr, new_addr, 1);
 319         } else if (l4_proto == NEXTHDR_UDP) {
 320                 if (likely(transport_len >= sizeof(struct udphdr))) {
 321                         struct udphdr *uh = udp_hdr(skb);
 322
 323                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 324                                 inet_proto_csum_replace16(&uh->check, skb,
 325                                                           addr, new_addr, 1);
 326                                 if (!uh->check)
 327                                         uh->check = CSUM_MANGLED_0;
 328                         }
 329                 }
 330         } else if (l4_proto == NEXTHDR_ICMP) {
 331                 if (likely(transport_len >= sizeof(struct icmp6hdr)))
 332                         inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 333                                                   skb, addr, new_addr, 1);
 334         }
 335 }
 336
 337 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
 338                            const __be32 mask[4], __be32 masked[4])
 339 {
 340         masked[0] = MASKED(old[0], addr[0], mask[0]);
 341         masked[1] = MASKED(old[1], addr[1], mask[1]);
 342         masked[2] = MASKED(old[2], addr[2], mask[2]);
 343         masked[3] = MASKED(old[3], addr[3], mask[3]);
 344 }
 345
 346 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 347                           __be32 addr[4], const __be32 new_addr[4],
 348                           bool recalculate_csum)
 349 {
 350         if (likely(recalculate_csum))
 351                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 352
 353         skb_clear_hash(skb);
 354         memcpy(addr, new_addr, sizeof(__be32[4]));
 355 }
 356
 357 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
 358 {
 359         /* Bits 21-24 are always unmasked, so this retains their values. */
 360         SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
 361         SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
 362         SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
 363 }
 364
 365 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
 366                        u8 mask)
 367 {
 368         new_ttl = MASKED(nh->ttl, new_ttl, mask);
 369
 370         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 371         nh->ttl = new_ttl;
 372 }
 373
 374 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
 375                     const struct ovs_key_ipv4 *key,
 376                     const struct ovs_key_ipv4 *mask)
 377 {
 378         struct iphdr *nh;
 379         __be32 new_addr;
 380         int err;
 381
 382         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 383                                   sizeof(struct iphdr));
 384         if (unlikely(err))
 385                 return err;
 386
 387         nh = ip_hdr(skb);
 388
 389         /* Setting an IP addresses is typically only a side effect of
 390          * matching on them in the current userspace implementation, so it
 391          * makes sense to check if the value actually changed.
 392          */
 393         if (mask->ipv4_src) {
 394                 new_addr = MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
 395
 396                 if (unlikely(new_addr != nh->saddr)) {
 397                         set_ip_addr(skb, nh, &nh->saddr, new_addr);
 398                         flow_key->ipv4.addr.src = new_addr;
 399                 }
 400         }
 401         if (mask->ipv4_dst) {
 402                 new_addr = MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
 403
 404                 if (unlikely(new_addr != nh->daddr)) {
 405                         set_ip_addr(skb, nh, &nh->daddr, new_addr);
 406                         flow_key->ipv4.addr.dst = new_addr;
 407                 }
 408         }
 409         if (mask->ipv4_tos) {
 410                 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
 411                 flow_key->ip.tos = nh->tos;
 412         }
 413         if (mask->ipv4_ttl) {
 414                 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
 415                 flow_key->ip.ttl = nh->ttl;
 416         }
 417
 418         return 0;
 419 }
 420
 421 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
 422 {
 423         return !!(addr[0] | addr[1] | addr[2] | addr[3]);
 424 }
 425
 426 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
 427                     const struct ovs_key_ipv6 *key,
 428                     const struct ovs_key_ipv6 *mask)
 429 {
 430         struct ipv6hdr *nh;
 431         int err;
 432
 433         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 434                                   sizeof(struct ipv6hdr));
 435         if (unlikely(err))
 436                 return err;
 437
 438         nh = ipv6_hdr(skb);
 439
 440         /* Setting an IP addresses is typically only a side effect of
 441          * matching on them in the current userspace implementation, so it
 442          * makes sense to check if the value actually changed.
 443          */
 444         if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
 445                 __be32 *saddr = (__be32 *)&nh->saddr;
 446                 __be32 masked[4];
 447
 448                 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
 449
 450                 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
 451                         set_ipv6_addr(skb, key->ipv6_proto, saddr, masked,
 452                                       true);
 453                         memcpy(&flow_key->ipv6.addr.src, masked,
 454                                sizeof(flow_key->ipv6.addr.src));
 455                 }
 456         }
 457         if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
 458                 unsigned int offset = 0;
 459                 int flags = IP6_FH_F_SKIP_RH;
 460                 bool recalc_csum = true;
 461                 __be32 *daddr = (__be32 *)&nh->daddr;
 462                 __be32 masked[4];
 463
 464                 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
 465
 466                 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
 467                         if (ipv6_ext_hdr(nh->nexthdr))
 468                                 recalc_csum = (ipv6_find_hdr(skb, &offset,
 469                                                              NEXTHDR_ROUTING,
 470                                                              NULL, &flags)
 471                                                != NEXTHDR_ROUTING);
 472
 473                         set_ipv6_addr(skb, key->ipv6_proto, daddr, masked,
 474                                       recalc_csum);
 475                         memcpy(&flow_key->ipv6.addr.dst, masked,
 476                                sizeof(flow_key->ipv6.addr.dst));
 477                 }
 478         }
 479         if (mask->ipv6_tclass) {
 480                 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
 481                 flow_key->ip.tos = ipv6_get_dsfield(nh);
 482         }
 483         if (mask->ipv6_label) {
 484                 set_ipv6_fl(nh, ntohl(key->ipv6_label),
 485                             ntohl(mask->ipv6_label));
 486                 flow_key->ipv6.label =
 487                     *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 488         }
 489         if (mask->ipv6_hlimit) {
 490                 SET_MASKED(nh->hop_limit, key->ipv6_hlimit, mask->ipv6_hlimit);
 491                 flow_key->ip.ttl = nh->hop_limit;
 492         }
 493         return 0;
 494 }
 495
 496 /* Must follow skb_ensure_writable() since that can move the skb data. */
 497 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 498                         __be16 new_port, __sum16 *check)
 499 {
 500         inet_proto_csum_replace2(check, skb, *port, new_port, 0);
 501         *port = new_port;
 502 }
 503
 504 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 505                    const struct ovs_key_udp *key,
 506                    const struct ovs_key_udp *mask)
 507 {
 508         struct udphdr *uh;
 509         __be16 src, dst;
 510         int err;
 511
 512         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 513                                   sizeof(struct udphdr));
 514         if (unlikely(err))
 515                 return err;
 516
 517         uh = udp_hdr(skb);
 518         /* Either of the masks is non-zero, so do not bother checking them. */
 519         src = MASKED(uh->source, key->udp_src, mask->udp_src);
 520         dst = MASKED(uh->dest, key->udp_dst, mask->udp_dst);
 521
 522         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 523                 if (likely(src != uh->source)) {
 524                         set_tp_port(skb, &uh->source, src, &uh->check);
 525                         flow_key->tp.src = src;
 526                 }
 527                 if (likely(dst != uh->dest)) {
 528                         set_tp_port(skb, &uh->dest, dst, &uh->check);
 529                         flow_key->tp.dst = dst;
 530                 }
 531
 532                 if (unlikely(!uh->check))
 533                         uh->check = CSUM_MANGLED_0;
 534         } else {
 535                 uh->source = src;
 536                 uh->dest = dst;
 537                 flow_key->tp.src = src;
 538                 flow_key->tp.dst = dst;
 539         }
 540
 541         skb_clear_hash(skb);
 542
 543         return 0;
 544 }
 545
 546 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 547                    const struct ovs_key_tcp *key,
 548                    const struct ovs_key_tcp *mask)
 549 {
 550         struct tcphdr *th;
 551         __be16 src, dst;
 552         int err;
 553
 554         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 555                                   sizeof(struct tcphdr));
 556         if (unlikely(err))
 557                 return err;
 558
 559         th = tcp_hdr(skb);
 560
 561         src = MASKED(th->source, key->tcp_src, mask->tcp_src);
 562         if (likely(src != th->source)) {
 563                 set_tp_port(skb, &th->source, src, &th->check);
 564                 flow_key->tp.src = src;
 565         }
 566         dst = MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
 567         if (likely(dst != th->dest)) {
 568                 set_tp_port(skb, &th->dest, dst, &th->check);
 569                 flow_key->tp.dst = dst;
 570         }
 571         skb_clear_hash(skb);
 572
 573         return 0;
 574 }
 575
 576 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 577                     const struct ovs_key_sctp *key,
 578                     const struct ovs_key_sctp *mask)
 579 {
 580         unsigned int sctphoff = skb_transport_offset(skb);
 581         struct sctphdr *sh;
 582         __le32 old_correct_csum, new_csum, old_csum;
 583         int err;
 584
 585         err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
 586         if (unlikely(err))
 587                 return err;
 588
 589         sh = sctp_hdr(skb);
 590
 591         old_csum = sh->checksum;
 592         old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 593
 594         sh->source = MASKED(sh->source, key->sctp_src, mask->sctp_src);
 595         sh->dest = MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
 596
 597         new_csum = sctp_compute_cksum(skb, sctphoff);
 598
 599         /* Carry any checksum errors through. */
 600         sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 601
 602         skb_clear_hash(skb);
 603         flow_key->tp.src = sh->source;
 604         flow_key->tp.dst = sh->dest;
 605
 606         return 0;
 607 }
 608
 609 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
 610 {
 611         struct vport *vport = ovs_vport_rcu(dp, out_port);
 612
 613         if (likely(vport))
 614                 ovs_vport_send(vport, skb);
 615         else
 616                 kfree_skb(skb);
 617 }
 618
 619 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 620                             struct sw_flow_key *key, const struct nlattr *attr,
 621                             const struct nlattr *actions, int actions_len)
 622 {
 623         struct ovs_tunnel_info info;
 624         struct dp_upcall_info upcall;
 625         const struct nlattr *a;
 626         int rem;
 627
 628         memset(&upcall, 0, sizeof(upcall));
 629         upcall.cmd = OVS_PACKET_CMD_ACTION;
 630
 631         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 632                  a = nla_next(a, &rem)) {
 633                 switch (nla_type(a)) {
 634                 case OVS_USERSPACE_ATTR_USERDATA:
 635                         upcall.userdata = a;
 636                         break;
 637
 638                 case OVS_USERSPACE_ATTR_PID:
 639                         upcall.portid = nla_get_u32(a);
 640                         break;
 641
 642                 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
 643                         /* Get out tunnel info. */
 644                         struct vport *vport;
 645
 646                         vport = ovs_vport_rcu(dp, nla_get_u32(a));
 647                         if (vport) {
 648                                 int err;
 649
 650                                 err = ovs_vport_get_egress_tun_info(vport, skb,
 651                                                                     &info);
 652                                 if (!err)
 653                                         upcall.egress_tun_info = &info;
 654                         }
 655                         break;
 656                 }
 657
 658                 case OVS_USERSPACE_ATTR_ACTIONS: {
 659                         /* Include actions. */
 660                         upcall.actions = actions;
 661                         upcall.actions_len = actions_len;
 662                         break;
 663                 }
 664
 665                 } /* End of switch. */
 666         }
 667
 668         return ovs_dp_upcall(dp, skb, key, &upcall);
 669 }
 670
 671 static int sample(struct datapath *dp, struct sk_buff *skb,
 672                   struct sw_flow_key *key, const struct nlattr *attr,
 673                   const struct nlattr *actions, int actions_len)
 674 {
 675         const struct nlattr *acts_list = NULL;
 676         const struct nlattr *a;
 677         int rem;
 678
 679         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 680                  a = nla_next(a, &rem)) {
 681                 switch (nla_type(a)) {
 682                 case OVS_SAMPLE_ATTR_PROBABILITY:
 683                         if (prandom_u32() >= nla_get_u32(a))
 684                                 return 0;
 685                         break;
 686
 687                 case OVS_SAMPLE_ATTR_ACTIONS:
 688                         acts_list = a;
 689                         break;
 690                 }
 691         }
 692
 693         rem = nla_len(acts_list);
 694         a = nla_data(acts_list);
 695
 696         /* Actions list is empty, do nothing */
 697         if (unlikely(!rem))
 698                 return 0;
 699
 700         /* The only known usage of sample action is having a single user-space
 701          * action. Treat this usage as a special case.
 702          * The output_userspace() should clone the skb to be sent to the
 703          * user space. This skb will be consumed by its caller.
 704          */
 705         if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
 706                    nla_is_last(a, rem)))
 707                 return output_userspace(dp, skb, key, a, actions, actions_len);
 708
 709         skb = skb_clone(skb, GFP_ATOMIC);
 710         if (!skb)
 711                 /* Skip the sample action when out of memory. */
 712                 return 0;
 713
 714         if (!add_deferred_actions(skb, key, a)) {
 715                 if (net_ratelimit())
 716                         pr_warn("%s: deferred actions limit reached, dropping sample action\n",
 717                                 ovs_dp_name(dp));
 718
 719                 kfree_skb(skb);
 720         }
 721         return 0;
 722 }
 723
 724 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
 725                          const struct nlattr *attr)
 726 {
 727         struct ovs_action_hash *hash_act = nla_data(attr);
 728         u32 hash = 0;
 729
 730         /* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
 731         hash = skb_get_hash(skb);
 732         hash = jhash_1word(hash, hash_act->hash_basis);
 733         if (!hash)
 734                 hash = 0x1;
 735
 736         key->ovs_flow_hash = hash;
 737 }
 738
 739 static int execute_set_action(struct sk_buff *skb,
 740                               struct sw_flow_key *flow_key,
 741                               const struct nlattr *a)
 742 {
 743         /* Only tunnel set execution is supported without a mask. */
 744         if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
 745                 OVS_CB(skb)->egress_tun_info = nla_data(a);
 746                 return 0;
 747         }
 748
 749         return -EINVAL;
 750
 751 }
 752
 753 /* Mask is at the midpoint of the data. */
 754 #define get_mask(a, type) ((const type)nla_data(a) + 1)
 755
 756 static int execute_masked_set_action(struct sk_buff *skb,
 757                                      struct sw_flow_key *flow_key,
 758                                      const struct nlattr *a)
 759 {
 760         int err = 0;
 761
 762         switch (nla_type(a)) {
 763         case OVS_KEY_ATTR_PRIORITY:
 764                 SET_MASKED(skb->priority, nla_get_u32(a), *get_mask(a, u32 *));
 765                 flow_key->phy.priority = skb->priority;
 766                 break;
 767
 768         case OVS_KEY_ATTR_SKB_MARK:
 769                 SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
 770                 flow_key->phy.skb_mark = skb->mark;
 771                 break;
 772
 773         case OVS_KEY_ATTR_TUNNEL_INFO:
 774                 /* Masked data not supported for tunnel. */
 775                 err = -EINVAL;
 776                 break;
 777
 778         case OVS_KEY_ATTR_ETHERNET:
 779                 err = set_eth_addr(skb, flow_key, nla_data(a),
 780                                    get_mask(a, struct ovs_key_ethernet *));
 781                 break;
 782
 783         case OVS_KEY_ATTR_IPV4:
 784                 err = set_ipv4(skb, flow_key, nla_data(a),
 785                                get_mask(a, struct ovs_key_ipv4 *));
 786                 break;
 787
 788         case OVS_KEY_ATTR_IPV6:
 789                 err = set_ipv6(skb, flow_key, nla_data(a),
 790                                get_mask(a, struct ovs_key_ipv6 *));
 791                 break;
 792
 793         case OVS_KEY_ATTR_TCP:
 794                 err = set_tcp(skb, flow_key, nla_data(a),
 795                               get_mask(a, struct ovs_key_tcp *));
 796                 break;
 797
 798         case OVS_KEY_ATTR_UDP:
 799                 err = set_udp(skb, flow_key, nla_data(a),
 800                               get_mask(a, struct ovs_key_udp *));
 801                 break;
 802
 803         case OVS_KEY_ATTR_SCTP:
 804                 err = set_sctp(skb, flow_key, nla_data(a),
 805                                get_mask(a, struct ovs_key_sctp *));
 806                 break;
 807
 808         case OVS_KEY_ATTR_MPLS:
 809                 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
 810                                                                     __be32 *));
 811                 break;
 812         }
 813
 814         return err;
 815 }
 816
 817 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
 818                           struct sw_flow_key *key,
 819                           const struct nlattr *a, int rem)
 820 {
 821         struct deferred_action *da;
 822
 823         if (!is_flow_key_valid(key)) {
 824                 int err;
 825
 826                 err = ovs_flow_key_update(skb, key);
 827                 if (err)
 828                         return err;
 829         }
 830         BUG_ON(!is_flow_key_valid(key));
 831
 832         if (!nla_is_last(a, rem)) {
 833                 /* Recirc action is the not the last action
 834                  * of the action list, need to clone the skb.
 835                  */
 836                 skb = skb_clone(skb, GFP_ATOMIC);
 837
 838                 /* Skip the recirc action when out of memory, but
 839                  * continue on with the rest of the action list.
 840                  */
 841                 if (!skb)
 842                         return 0;
 843         }
 844
 845         da = add_deferred_actions(skb, key, NULL);
 846         if (da) {
 847                 da->pkt_key.recirc_id = nla_get_u32(a);
 848         } else {
 849                 kfree_skb(skb);
 850
 851                 if (net_ratelimit())
 852                         pr_warn("%s: deferred action limit reached, drop recirc action\n",
 853                                 ovs_dp_name(dp));
 854         }
 855
 856         return 0;
 857 }
 858
 859 /* Execute a list of actions against 'skb'. */
 860 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 861                               struct sw_flow_key *key,
 862                               const struct nlattr *attr, int len)
 863 {
 864         /* Every output action needs a separate clone of 'skb', but the common
 865          * case is just a single output action, so that doing a clone and
 866          * then freeing the original skbuff is wasteful.  So the following code
 867          * is slightly obscure just to avoid that.
 868          */
 869         int prev_port = -1;
 870         const struct nlattr *a;
 871         int rem;
 872
 873         for (a = attr, rem = len; rem > 0;
 874              a = nla_next(a, &rem)) {
 875                 int err = 0;
 876
 877                 if (unlikely(prev_port != -1)) {
 878                         struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
 879
 880                         if (out_skb)
 881                                 do_output(dp, out_skb, prev_port);
 882
 883                         prev_port = -1;
 884                 }
 885
 886                 switch (nla_type(a)) {
 887                 case OVS_ACTION_ATTR_OUTPUT:
 888                         prev_port = nla_get_u32(a);
 889                         break;
 890
 891                 case OVS_ACTION_ATTR_USERSPACE:
 892                         output_userspace(dp, skb, key, a, attr, len);
 893                         break;
 894
 895                 case OVS_ACTION_ATTR_HASH:
 896                         execute_hash(skb, key, a);
 897                         break;
 898
 899                 case OVS_ACTION_ATTR_PUSH_MPLS:
 900                         err = push_mpls(skb, key, nla_data(a));
 901                         break;
 902
 903                 case OVS_ACTION_ATTR_POP_MPLS:
 904                         err = pop_mpls(skb, key, nla_get_be16(a));
 905                         break;
 906
 907                 case OVS_ACTION_ATTR_PUSH_VLAN:
 908                         err = push_vlan(skb, key, nla_data(a));
 909                         break;
 910
 911                 case OVS_ACTION_ATTR_POP_VLAN:
 912                         err = pop_vlan(skb, key);
 913                         break;
 914
 915                 case OVS_ACTION_ATTR_RECIRC:
 916                         err = execute_recirc(dp, skb, key, a, rem);
 917                         if (nla_is_last(a, rem)) {
 918                                 /* If this is the last action, the skb has
 919                                  * been consumed or freed.
 920                                  * Return immediately.
 921                                  */
 922                                 return err;
 923                         }
 924                         break;
 925
 926                 case OVS_ACTION_ATTR_SET:
 927                         err = execute_set_action(skb, key, nla_data(a));
 928                         break;
 929
 930                 case OVS_ACTION_ATTR_SET_MASKED:
 931                 case OVS_ACTION_ATTR_SET_TO_MASKED:
 932                         err = execute_masked_set_action(skb, key, nla_data(a));
 933                         break;
 934
 935                 case OVS_ACTION_ATTR_SAMPLE:
 936                         err = sample(dp, skb, key, a, attr, len);
 937                         break;
 938                 }
 939
 940                 if (unlikely(err)) {
 941                         kfree_skb(skb);
 942                         return err;
 943                 }
 944         }
 945
 946         if (prev_port != -1)
 947                 do_output(dp, skb, prev_port);
 948         else
 949                 consume_skb(skb);
 950
 951         return 0;
 952 }
 953
 954 static void process_deferred_actions(struct datapath *dp)
 955 {
 956         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
 957
 958         /* Do not touch the FIFO in case there is no deferred actions. */
 959         if (action_fifo_is_empty(fifo))
 960                 return;
 961
 962         /* Finishing executing all deferred actions. */
 963         do {
 964                 struct deferred_action *da = action_fifo_get(fifo);
 965                 struct sk_buff *skb = da->skb;
 966                 struct sw_flow_key *key = &da->pkt_key;
 967                 const struct nlattr *actions = da->actions;
 968
 969                 if (actions)
 970                         do_execute_actions(dp, skb, key, actions,
 971                                            nla_len(actions));
 972                 else
 973                         ovs_dp_process_packet(skb, key);
 974         } while (!action_fifo_is_empty(fifo));
 975
 976         /* Reset FIFO for the next packet.  */
 977         action_fifo_init(fifo);
 978 }
 979
 980 /* Execute a list of actions against 'skb'. */
 981 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
 982                         const struct sw_flow_actions *acts,
 983                         struct sw_flow_key *key)
 984 {
 985         int level = this_cpu_read(exec_actions_level);
 986         int err;
 987
 988         if (unlikely(level >= EXEC_ACTIONS_LEVEL_LIMIT)) {
 989                 if (net_ratelimit())
 990                         pr_warn("%s: packet loop detected, dropping.\n",
 991                                 ovs_dp_name(dp));
 992
 993                 kfree_skb(skb);
 994                 return -ELOOP;
 995         }
 996
 997         this_cpu_inc(exec_actions_level);
 998         err = do_execute_actions(dp, skb, key,
 999                                  acts->actions, acts->actions_len);
1000
1001         if (!level)
1002                 process_deferred_actions(dp);
1003
1004         this_cpu_dec(exec_actions_level);
1005
1006         /* This return status currently does not reflect the errors
1007          * encounted during deferred actions execution. Probably needs to
1008          * be fixed in the future.
1009          */
1010         return err;
1011 }
1012
1013 int action_fifos_init(void)
1014 {
1015         action_fifos = alloc_percpu(struct action_fifo);
1016         if (!action_fifos)
1017                 return -ENOMEM;
1018
1019         return 0;
1020 }
1021
1022 void action_fifos_exit(void)
1023 {
1024         free_percpu(action_fifos);
1025 }