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
  32 #include <net/ip.h>
  33 #include <net/ipv6.h>
  34 #include <net/checksum.h>
  35 #include <net/dsfield.h>
  36 #include <net/mpls.h>
  37 #include <net/sctp/checksum.h>
  38
  39 #include "datapath.h"
  40 #include "gso.h"
  41 #include "vlan.h"
  42 #include "vport.h"
  43
  44 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
  45                               struct sw_flow_key *key,
  46                               const struct nlattr *attr, int len);
  47
  48 struct deferred_action {
  49         struct sk_buff *skb;
  50         const struct nlattr *actions;
  51
  52         /* Store pkt_key clone when creating deferred action. */
  53         struct sw_flow_key pkt_key;
  54 };
  55
  56 #define DEFERRED_ACTION_FIFO_SIZE 10
  57 struct action_fifo {
  58         int head;
  59         int tail;
  60         /* Deferred action fifo queue storage. */
  61         struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
  62 };
  63
  64 static struct action_fifo __percpu *action_fifos;
  65 #define EXEC_ACTIONS_LEVEL_LIMIT 4   /* limit used to detect packet
  66                                       * looping by the network stack
  67                                       */
  68 static DEFINE_PER_CPU(int, exec_actions_level);
  69
  70 static void action_fifo_init(struct action_fifo *fifo)
  71 {
  72         fifo->head = 0;
  73         fifo->tail = 0;
  74 }
  75
  76 static bool action_fifo_is_empty(const struct action_fifo *fifo)
  77 {
  78         return (fifo->head == fifo->tail);
  79 }
  80
  81 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
  82 {
  83         if (action_fifo_is_empty(fifo))
  84                 return NULL;
  85
  86         return &fifo->fifo[fifo->tail++];
  87 }
  88
  89 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
  90 {
  91         if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
  92                 return NULL;
  93
  94         return &fifo->fifo[fifo->head++];
  95 }
  96
  97 /* Return queue entry if fifo is not full */
  98 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
  99                                                     const struct sw_flow_key *key,
 100                                                     const struct nlattr *attr)
 101 {
 102         struct action_fifo *fifo;
 103         struct deferred_action *da;
 104
 105         fifo = this_cpu_ptr(action_fifos);
 106         da = action_fifo_put(fifo);
 107         if (da) {
 108                 da->skb = skb;
 109                 da->actions = attr;
 110                 da->pkt_key = *key;
 111         }
 112
 113         return da;
 114 }
 115
 116 static void invalidate_flow_key(struct sw_flow_key *key)
 117 {
 118         key->eth.type = htons(0);
 119 }
 120
 121 static bool is_flow_key_valid(const struct sw_flow_key *key)
 122 {
 123         return !!key->eth.type;
 124 }
 125
 126 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 127                      const struct ovs_action_push_mpls *mpls)
 128 {
 129         __be32 *new_mpls_lse;
 130         struct ethhdr *hdr;
 131
 132         /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
 133         if (skb_encapsulation(skb))
 134                 return -ENOTSUPP;
 135
 136         if (skb_cow_head(skb, MPLS_HLEN) < 0)
 137                 return -ENOMEM;
 138
 139         skb_push(skb, MPLS_HLEN);
 140         memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
 141                 skb->mac_len);
 142         skb_reset_mac_header(skb);
 143
 144         new_mpls_lse = (__be32 *)skb_mpls_header(skb);
 145         *new_mpls_lse = mpls->mpls_lse;
 146
 147         if (skb->ip_summed == CHECKSUM_COMPLETE)
 148                 skb->csum = csum_add(skb->csum, csum_partial(new_mpls_lse,
 149                                                              MPLS_HLEN, 0));
 150
 151         hdr = eth_hdr(skb);
 152         hdr->h_proto = mpls->mpls_ethertype;
 153         if (!ovs_skb_get_inner_protocol(skb))
 154                 ovs_skb_set_inner_protocol(skb, skb->protocol);
 155         skb->protocol = mpls->mpls_ethertype;
 156
 157         invalidate_flow_key(key);
 158         return 0;
 159 }
 160
 161 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 162                     const __be16 ethertype)
 163 {
 164         struct ethhdr *hdr;
 165         int err;
 166
 167         err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
 168         if (unlikely(err))
 169                 return err;
 170
 171         skb_postpull_rcsum(skb, skb_mpls_header(skb), MPLS_HLEN);
 172
 173         memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
 174                 skb->mac_len);
 175
 176         __skb_pull(skb, MPLS_HLEN);
 177         skb_reset_mac_header(skb);
 178
 179         /* skb_mpls_header() is used to locate the ethertype
 180          * field correctly in the presence of VLAN tags.
 181          */
 182         hdr = (struct ethhdr *)(skb_mpls_header(skb) - ETH_HLEN);
 183         hdr->h_proto = ethertype;
 184         if (eth_p_mpls(skb->protocol))
 185                 skb->protocol = ethertype;
 186
 187         invalidate_flow_key(key);
 188         return 0;
 189 }
 190
 191 /* 'KEY' must not have any bits set outside of the 'MASK' */
 192 #define MASKED(OLD, KEY, MASK) ((KEY) | ((OLD) & ~(MASK)))
 193 #define SET_MASKED(OLD, KEY, MASK) ((OLD) = MASKED(OLD, KEY, MASK))
 194
 195 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
 196                     const __be32 *mpls_lse, const __be32 *mask)
 197 {
 198         __be32 *stack;
 199         __be32 lse;
 200         int err;
 201
 202         err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
 203         if (unlikely(err))
 204                 return err;
 205
 206         stack = (__be32 *)skb_mpls_header(skb);
 207         lse = MASKED(*stack, *mpls_lse, *mask);
 208         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 209                 __be32 diff[] = { ~(*stack), lse };
 210
 211                 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
 212                                           ~skb->csum);
 213         }
 214
 215         *stack = lse;
 216         flow_key->mpls.top_lse = lse;
 217         return 0;
 218 }
 219
 220 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 221 {
 222         int err;
 223
 224         err = skb_vlan_pop(skb);
 225         if (skb_vlan_tag_present(skb))
 226                 invalidate_flow_key(key);
 227         else
 228                 key->eth.tci = 0;
 229         return err;
 230 }
 231
 232 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
 233                      const struct ovs_action_push_vlan *vlan)
 234 {
 235         if (skb_vlan_tag_present(skb))
 236                 invalidate_flow_key(key);
 237         else
 238                 key->eth.tci = vlan->vlan_tci;
 239         return skb_vlan_push(skb, vlan->vlan_tpid,
 240                              ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
 241 }
 242
 243 /* 'src' is already properly masked. */
 244 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
 245 {
 246         u16 *dst = (u16 *)dst_;
 247         const u16 *src = (const u16 *)src_;
 248         const u16 *mask = (const u16 *)mask_;
 249
 250         SET_MASKED(dst[0], src[0], mask[0]);
 251         SET_MASKED(dst[1], src[1], mask[1]);
 252         SET_MASKED(dst[2], src[2], mask[2]);
 253 }
 254
 255 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
 256                         const struct ovs_key_ethernet *key,
 257                         const struct ovs_key_ethernet *mask)
 258 {
 259         int err;
 260
 261         err = skb_ensure_writable(skb, ETH_HLEN);
 262         if (unlikely(err))
 263                 return err;
 264
 265         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 266
 267         ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
 268                                mask->eth_src);
 269         ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
 270                                mask->eth_dst);
 271
 272         ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 273
 274         ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
 275         ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
 276         return 0;
 277 }
 278
 279 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
 280                                   __be32 addr, __be32 new_addr)
 281 {
 282         int transport_len = skb->len - skb_transport_offset(skb);
 283
 284         if (nh->frag_off & htons(IP_OFFSET))
 285                 return;
 286
 287         if (nh->protocol == IPPROTO_TCP) {
 288                 if (likely(transport_len >= sizeof(struct tcphdr)))
 289                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 290                                                  addr, new_addr, 1);
 291         } else if (nh->protocol == IPPROTO_UDP) {
 292                 if (likely(transport_len >= sizeof(struct udphdr))) {
 293                         struct udphdr *uh = udp_hdr(skb);
 294
 295                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 296                                 inet_proto_csum_replace4(&uh->check, skb,
 297                                                          addr, new_addr, 1);
 298                                 if (!uh->check)
 299                                         uh->check = CSUM_MANGLED_0;
 300                         }
 301                 }
 302         }
 303
 304 }
 305
 306 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 307                         __be32 *addr, __be32 new_addr)
 308 {
 309         update_ip_l4_checksum(skb, nh, *addr, new_addr);
 310         csum_replace4(&nh->check, *addr, new_addr);
 311         skb_clear_hash(skb);
 312         *addr = new_addr;
 313 }
 314
 315 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 316                                  __be32 addr[4], const __be32 new_addr[4])
 317 {
 318         int transport_len = skb->len - skb_transport_offset(skb);
 319
 320         if (l4_proto == NEXTHDR_TCP) {
 321                 if (likely(transport_len >= sizeof(struct tcphdr)))
 322                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 323                                                   addr, new_addr, 1);
 324         } else if (l4_proto == NEXTHDR_UDP) {
 325                 if (likely(transport_len >= sizeof(struct udphdr))) {
 326                         struct udphdr *uh = udp_hdr(skb);
 327
 328                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 329                                 inet_proto_csum_replace16(&uh->check, skb,
 330                                                           addr, new_addr, 1);
 331                                 if (!uh->check)
 332                                         uh->check = CSUM_MANGLED_0;
 333                         }
 334                 }
 335         } else if (l4_proto == NEXTHDR_ICMP) {
 336                 if (likely(transport_len >= sizeof(struct icmp6hdr)))
 337                         inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 338                                                   skb, addr, new_addr, 1);
 339         }
 340 }
 341
 342 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
 343                            const __be32 mask[4], __be32 masked[4])
 344 {
 345         masked[0] = MASKED(old[0], addr[0], mask[0]);
 346         masked[1] = MASKED(old[1], addr[1], mask[1]);
 347         masked[2] = MASKED(old[2], addr[2], mask[2]);
 348         masked[3] = MASKED(old[3], addr[3], mask[3]);
 349 }
 350
 351 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 352                           __be32 addr[4], const __be32 new_addr[4],
 353                           bool recalculate_csum)
 354 {
 355         if (likely(recalculate_csum))
 356                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 357
 358         skb_clear_hash(skb);
 359         memcpy(addr, new_addr, sizeof(__be32[4]));
 360 }
 361
 362 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
 363 {
 364         /* Bits 21-24 are always unmasked, so this retains their values. */
 365         SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
 366         SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
 367         SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
 368 }
 369
 370 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
 371                        u8 mask)
 372 {
 373         new_ttl = MASKED(nh->ttl, new_ttl, mask);
 374
 375         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 376         nh->ttl = new_ttl;
 377 }
 378
 379 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
 380                     const struct ovs_key_ipv4 *key,
 381                     const struct ovs_key_ipv4 *mask)
 382 {
 383         struct iphdr *nh;
 384         __be32 new_addr;
 385         int err;
 386
 387         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 388                                   sizeof(struct iphdr));
 389         if (unlikely(err))
 390                 return err;
 391
 392         nh = ip_hdr(skb);
 393
 394         /* Setting an IP addresses is typically only a side effect of
 395          * matching on them in the current userspace implementation, so it
 396          * makes sense to check if the value actually changed.
 397          */
 398         if (mask->ipv4_src) {
 399                 new_addr = MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
 400
 401                 if (unlikely(new_addr != nh->saddr)) {
 402                         set_ip_addr(skb, nh, &nh->saddr, new_addr);
 403                         flow_key->ipv4.addr.src = new_addr;
 404                 }
 405         }
 406         if (mask->ipv4_dst) {
 407                 new_addr = MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
 408
 409                 if (unlikely(new_addr != nh->daddr)) {
 410                         set_ip_addr(skb, nh, &nh->daddr, new_addr);
 411                         flow_key->ipv4.addr.dst = new_addr;
 412                 }
 413         }
 414         if (mask->ipv4_tos) {
 415                 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
 416                 flow_key->ip.tos = nh->tos;
 417         }
 418         if (mask->ipv4_ttl) {
 419                 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
 420                 flow_key->ip.ttl = nh->ttl;
 421         }
 422
 423         return 0;
 424 }
 425
 426 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
 427 {
 428         return !!(addr[0] | addr[1] | addr[2] | addr[3]);
 429 }
 430
 431 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
 432                     const struct ovs_key_ipv6 *key,
 433                     const struct ovs_key_ipv6 *mask)
 434 {
 435         struct ipv6hdr *nh;
 436         int err;
 437
 438         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 439                                   sizeof(struct ipv6hdr));
 440         if (unlikely(err))
 441                 return err;
 442
 443         nh = ipv6_hdr(skb);
 444
 445         /* Setting an IP addresses is typically only a side effect of
 446          * matching on them in the current userspace implementation, so it
 447          * makes sense to check if the value actually changed.
 448          */
 449         if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
 450                 __be32 *saddr = (__be32 *)&nh->saddr;
 451                 __be32 masked[4];
 452
 453                 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
 454
 455                 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
 456                         set_ipv6_addr(skb, key->ipv6_proto, saddr, masked,
 457                                       true);
 458                         memcpy(&flow_key->ipv6.addr.src, masked,
 459                                sizeof(flow_key->ipv6.addr.src));
 460                 }
 461         }
 462         if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
 463                 unsigned int offset = 0;
 464                 int flags = IP6_FH_F_SKIP_RH;
 465                 bool recalc_csum = true;
 466                 __be32 *daddr = (__be32 *)&nh->daddr;
 467                 __be32 masked[4];
 468
 469                 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
 470
 471                 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
 472                         if (ipv6_ext_hdr(nh->nexthdr))
 473                                 recalc_csum = (ipv6_find_hdr(skb, &offset,
 474                                                              NEXTHDR_ROUTING,
 475                                                              NULL, &flags)
 476                                                != NEXTHDR_ROUTING);
 477
 478                         set_ipv6_addr(skb, key->ipv6_proto, daddr, masked,
 479                                       recalc_csum);
 480                         memcpy(&flow_key->ipv6.addr.dst, masked,
 481                                sizeof(flow_key->ipv6.addr.dst));
 482                 }
 483         }
 484         if (mask->ipv6_tclass) {
 485                 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
 486                 flow_key->ip.tos = ipv6_get_dsfield(nh);
 487         }
 488         if (mask->ipv6_label) {
 489                 set_ipv6_fl(nh, ntohl(key->ipv6_label),
 490                             ntohl(mask->ipv6_label));
 491                 flow_key->ipv6.label =
 492                     *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 493         }
 494         if (mask->ipv6_hlimit) {
 495                 SET_MASKED(nh->hop_limit, key->ipv6_hlimit, mask->ipv6_hlimit);
 496                 flow_key->ip.ttl = nh->hop_limit;
 497         }
 498         return 0;
 499 }
 500
 501 /* Must follow skb_ensure_writable() since that can move the skb data. */
 502 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 503                         __be16 new_port, __sum16 *check)
 504 {
 505         inet_proto_csum_replace2(check, skb, *port, new_port, 0);
 506         *port = new_port;
 507 }
 508
 509 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 510                    const struct ovs_key_udp *key,
 511                    const struct ovs_key_udp *mask)
 512 {
 513         struct udphdr *uh;
 514         __be16 src, dst;
 515         int err;
 516
 517         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 518                                   sizeof(struct udphdr));
 519         if (unlikely(err))
 520                 return err;
 521
 522         uh = udp_hdr(skb);
 523         /* Either of the masks is non-zero, so do not bother checking them. */
 524         src = MASKED(uh->source, key->udp_src, mask->udp_src);
 525         dst = MASKED(uh->dest, key->udp_dst, mask->udp_dst);
 526
 527         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 528                 if (likely(src != uh->source)) {
 529                         set_tp_port(skb, &uh->source, src, &uh->check);
 530                         flow_key->tp.src = src;
 531                 }
 532                 if (likely(dst != uh->dest)) {
 533                         set_tp_port(skb, &uh->dest, dst, &uh->check);
 534                         flow_key->tp.dst = dst;
 535                 }
 536
 537                 if (unlikely(!uh->check))
 538                         uh->check = CSUM_MANGLED_0;
 539         } else {
 540                 uh->source = src;
 541                 uh->dest = dst;
 542                 flow_key->tp.src = src;
 543                 flow_key->tp.dst = dst;
 544         }
 545
 546         skb_clear_hash(skb);
 547
 548         return 0;
 549 }
 550
 551 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 552                    const struct ovs_key_tcp *key,
 553                    const struct ovs_key_tcp *mask)
 554 {
 555         struct tcphdr *th;
 556         __be16 src, dst;
 557         int err;
 558
 559         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 560                                   sizeof(struct tcphdr));
 561         if (unlikely(err))
 562                 return err;
 563
 564         th = tcp_hdr(skb);
 565         src = MASKED(th->source, key->tcp_src, mask->tcp_src);
 566         if (likely(src != th->source)) {
 567                 set_tp_port(skb, &th->source, src, &th->check);
 568                 flow_key->tp.src = src;
 569         }
 570         dst = MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
 571         if (likely(dst != th->dest)) {
 572                 set_tp_port(skb, &th->dest, dst, &th->check);
 573                 flow_key->tp.dst = dst;
 574         }
 575         skb_clear_hash(skb);
 576
 577         return 0;
 578 }
 579
 580 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 581                     const struct ovs_key_sctp *key,
 582                     const struct ovs_key_sctp *mask)
 583 {
 584         unsigned int sctphoff = skb_transport_offset(skb);
 585         struct sctphdr *sh;
 586         __le32 old_correct_csum, new_csum, old_csum;
 587         int err;
 588
 589         err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
 590         if (unlikely(err))
 591                 return err;
 592
 593         sh = sctp_hdr(skb);
 594         old_csum = sh->checksum;
 595         old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 596
 597         sh->source = MASKED(sh->source, key->sctp_src, mask->sctp_src);
 598         sh->dest = MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
 599
 600         new_csum = sctp_compute_cksum(skb, sctphoff);
 601
 602         /* Carry any checksum errors through. */
 603         sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 604
 605         skb_clear_hash(skb);
 606         flow_key->tp.src = sh->source;
 607         flow_key->tp.dst = sh->dest;
 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                             const struct nlattr *actions, int actions_len)
 625 {
 626         struct ovs_tunnel_info info;
 627         struct dp_upcall_info upcall;
 628         const struct nlattr *a;
 629         int rem;
 630
 631         memset(&upcall, 0, sizeof(upcall));
 632         upcall.cmd = OVS_PACKET_CMD_ACTION;
 633
 634         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 635                  a = nla_next(a, &rem)) {
 636                 switch (nla_type(a)) {
 637                 case OVS_USERSPACE_ATTR_USERDATA:
 638                         upcall.userdata = a;
 639                         break;
 640
 641                 case OVS_USERSPACE_ATTR_PID:
 642                         upcall.portid = nla_get_u32(a);
 643                         break;
 644
 645                 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
 646                         /* Get out tunnel info. */
 647                         struct vport *vport;
 648
 649                         vport = ovs_vport_rcu(dp, nla_get_u32(a));
 650                         if (vport) {
 651                                 int err;
 652
 653                                 err = ovs_vport_get_egress_tun_info(vport, skb,
 654                                                                     &info);
 655                                 if (!err)
 656                                         upcall.egress_tun_info = &info;
 657                         }
 658                         break;
 659                 }
 660
 661                 case OVS_USERSPACE_ATTR_ACTIONS: {
 662                         /* Include actions. */
 663                         upcall.actions = actions;
 664                         upcall.actions_len = actions_len;
 665                         break;
 666                 }
 667
 668                 } /* End of switch. */
 669         }
 670
 671         return ovs_dp_upcall(dp, skb, key, &upcall);
 672 }
 673
 674 static int sample(struct datapath *dp, struct sk_buff *skb,
 675                   struct sw_flow_key *key, const struct nlattr *attr,
 676                   const struct nlattr *actions, int actions_len)
 677 {
 678         const struct nlattr *acts_list = NULL;
 679         const struct nlattr *a;
 680         int rem;
 681
 682         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 683                  a = nla_next(a, &rem)) {
 684                 u32 probability;
 685
 686                 switch (nla_type(a)) {
 687                 case OVS_SAMPLE_ATTR_PROBABILITY:
 688                         probability = nla_get_u32(a);
 689                         if (!probability || prandom_u32() > probability)
 690                                 return 0;
 691                         break;
 692
 693                 case OVS_SAMPLE_ATTR_ACTIONS:
 694                         acts_list = a;
 695                         break;
 696                 }
 697         }
 698
 699         rem = nla_len(acts_list);
 700         a = nla_data(acts_list);
 701
 702         /* Actions list is empty, do nothing */
 703         if (unlikely(!rem))
 704                 return 0;
 705
 706         /* The only known usage of sample action is having a single user-space
 707          * action. Treat this usage as a special case.
 708          * The output_userspace() should clone the skb to be sent to the
 709          * user space. This skb will be consumed by its caller.
 710          */
 711         if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
 712                    nla_is_last(a, rem)))
 713                 return output_userspace(dp, skb, key, a, actions, actions_len);
 714
 715         skb = skb_clone(skb, GFP_ATOMIC);
 716         if (!skb)
 717                 /* Skip the sample action when out of memory. */
 718                 return 0;
 719
 720         if (!add_deferred_actions(skb, key, a)) {
 721                 if (net_ratelimit())
 722                         pr_warn("%s: deferred actions limit reached, dropping sample action\n",
 723                                 ovs_dp_name(dp));
 724
 725                 kfree_skb(skb);
 726         }
 727         return 0;
 728 }
 729
 730 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
 731                          const struct nlattr *attr)
 732 {
 733         struct ovs_action_hash *hash_act = nla_data(attr);
 734         u32 hash = 0;
 735
 736         /* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
 737         hash = skb_get_hash(skb);
 738         hash = jhash_1word(hash, hash_act->hash_basis);
 739         if (!hash)
 740                 hash = 0x1;
 741
 742         key->ovs_flow_hash = hash;
 743 }
 744
 745 static int execute_set_action(struct sk_buff *skb,
 746                               struct sw_flow_key *flow_key,
 747                               const struct nlattr *a)
 748 {
 749         /* Only tunnel set execution is supported without a mask. */
 750         if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
 751                 OVS_CB(skb)->egress_tun_info = nla_data(a);
 752                 return 0;
 753         }
 754
 755         return -EINVAL;
 756 }
 757
 758 /* Mask is at the midpoint of the data. */
 759 #define get_mask(a, type) ((const type)nla_data(a) + 1)
 760
 761 static int execute_masked_set_action(struct sk_buff *skb,
 762                                      struct sw_flow_key *flow_key,
 763                                      const struct nlattr *a)
 764 {
 765         int err = 0;
 766
 767         switch (nla_type(a)) {
 768         case OVS_KEY_ATTR_PRIORITY:
 769                 SET_MASKED(skb->priority, nla_get_u32(a), *get_mask(a, u32 *));
 770                 flow_key->phy.priority = skb->priority;
 771                 break;
 772
 773         case OVS_KEY_ATTR_SKB_MARK:
 774                 SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
 775                 flow_key->phy.skb_mark = skb->mark;
 776                 break;
 777
 778         case OVS_KEY_ATTR_TUNNEL_INFO:
 779                 /* Masked data not supported for tunnel. */
 780                 err = -EINVAL;
 781                 break;
 782
 783         case OVS_KEY_ATTR_ETHERNET:
 784                 err = set_eth_addr(skb, flow_key, nla_data(a),
 785                                    get_mask(a, struct ovs_key_ethernet *));
 786                 break;
 787
 788         case OVS_KEY_ATTR_IPV4:
 789                 err = set_ipv4(skb, flow_key, nla_data(a),
 790                                get_mask(a, struct ovs_key_ipv4 *));
 791                 break;
 792
 793         case OVS_KEY_ATTR_IPV6:
 794                 err = set_ipv6(skb, flow_key, nla_data(a),
 795                                get_mask(a, struct ovs_key_ipv6 *));
 796                 break;
 797
 798         case OVS_KEY_ATTR_TCP:
 799                 err = set_tcp(skb, flow_key, nla_data(a),
 800                               get_mask(a, struct ovs_key_tcp *));
 801                 break;
 802
 803         case OVS_KEY_ATTR_UDP:
 804                 err = set_udp(skb, flow_key, nla_data(a),
 805                               get_mask(a, struct ovs_key_udp *));
 806                 break;
 807
 808         case OVS_KEY_ATTR_SCTP:
 809                 err = set_sctp(skb, flow_key, nla_data(a),
 810                                get_mask(a, struct ovs_key_sctp *));
 811                 break;
 812
 813         case OVS_KEY_ATTR_MPLS:
 814                 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
 815                                                                     __be32 *));
 816                 break;
 817         }
 818
 819         return err;
 820 }
 821
 822 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
 823                           struct sw_flow_key *key,
 824                           const struct nlattr *a, int rem)
 825 {
 826         struct deferred_action *da;
 827
 828         if (!is_flow_key_valid(key)) {
 829                 int err;
 830
 831                 err = ovs_flow_key_update(skb, key);
 832                 if (err)
 833                         return err;
 834         }
 835         BUG_ON(!is_flow_key_valid(key));
 836
 837         if (!nla_is_last(a, rem)) {
 838                 /* Recirc action is the not the last action
 839                  * of the action list, need to clone the skb.
 840                  */
 841                 skb = skb_clone(skb, GFP_ATOMIC);
 842
 843                 /* Skip the recirc action when out of memory, but
 844                  * continue on with the rest of the action list.
 845                  */
 846                 if (!skb)
 847                         return 0;
 848         }
 849
 850         da = add_deferred_actions(skb, key, NULL);
 851         if (da) {
 852                 da->pkt_key.recirc_id = nla_get_u32(a);
 853         } else {
 854                 kfree_skb(skb);
 855
 856                 if (net_ratelimit())
 857                         pr_warn("%s: deferred action limit reached, drop recirc action\n",
 858                                 ovs_dp_name(dp));
 859         }
 860
 861         return 0;
 862 }
 863
 864 /* Execute a list of actions against 'skb'. */
 865 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 866                               struct sw_flow_key *key,
 867                               const struct nlattr *attr, int len)
 868 {
 869         /* Every output action needs a separate clone of 'skb', but the common
 870          * case is just a single output action, so that doing a clone and
 871          * then freeing the original skbuff is wasteful.  So the following code
 872          * is slightly obscure just to avoid that.
 873          */
 874         int prev_port = -1;
 875         const struct nlattr *a;
 876         int rem;
 877
 878         for (a = attr, rem = len; rem > 0;
 879              a = nla_next(a, &rem)) {
 880                 int err = 0;
 881
 882                 if (unlikely(prev_port != -1)) {
 883                         struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
 884
 885                         if (out_skb)
 886                                 do_output(dp, out_skb, prev_port);
 887
 888                         prev_port = -1;
 889                 }
 890
 891                 switch (nla_type(a)) {
 892                 case OVS_ACTION_ATTR_OUTPUT:
 893                         prev_port = nla_get_u32(a);
 894                         break;
 895
 896                 case OVS_ACTION_ATTR_USERSPACE:
 897                         output_userspace(dp, skb, key, a, attr, len);
 898                         break;
 899
 900                 case OVS_ACTION_ATTR_HASH:
 901                         execute_hash(skb, key, a);
 902                         break;
 903
 904                 case OVS_ACTION_ATTR_PUSH_MPLS:
 905                         err = push_mpls(skb, key, nla_data(a));
 906                         break;
 907
 908                 case OVS_ACTION_ATTR_POP_MPLS:
 909                         err = pop_mpls(skb, key, nla_get_be16(a));
 910                         break;
 911
 912                 case OVS_ACTION_ATTR_PUSH_VLAN:
 913                         err = push_vlan(skb, key, nla_data(a));
 914                         break;
 915
 916                 case OVS_ACTION_ATTR_POP_VLAN:
 917                         err = pop_vlan(skb, key);
 918                         break;
 919
 920                 case OVS_ACTION_ATTR_RECIRC:
 921                         err = execute_recirc(dp, skb, key, a, rem);
 922                         if (nla_is_last(a, rem)) {
 923                                 /* If this is the last action, the skb has
 924                                  * been consumed or freed.
 925                                  * Return immediately.
 926                                  */
 927                                 return err;
 928                         }
 929                         break;
 930
 931                 case OVS_ACTION_ATTR_SET:
 932                         err = execute_set_action(skb, key, nla_data(a));
 933                         break;
 934
 935                 case OVS_ACTION_ATTR_SET_MASKED:
 936                 case OVS_ACTION_ATTR_SET_TO_MASKED:
 937                         err = execute_masked_set_action(skb, key, nla_data(a));
 938                         break;
 939
 940                 case OVS_ACTION_ATTR_SAMPLE:
 941                         err = sample(dp, skb, key, a, attr, len);
 942                         break;
 943                 }
 944
 945                 if (unlikely(err)) {
 946                         kfree_skb(skb);
 947                         return err;
 948                 }
 949         }
 950
 951         if (prev_port != -1)
 952                 do_output(dp, skb, prev_port);
 953         else
 954                 consume_skb(skb);
 955
 956         return 0;
 957 }
 958
 959 static void process_deferred_actions(struct datapath *dp)
 960 {
 961         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
 962
 963         /* Do not touch the FIFO in case there is no deferred actions. */
 964         if (action_fifo_is_empty(fifo))
 965                 return;
 966
 967         /* Finishing executing all deferred actions. */
 968         do {
 969                 struct deferred_action *da = action_fifo_get(fifo);
 970                 struct sk_buff *skb = da->skb;
 971                 struct sw_flow_key *key = &da->pkt_key;
 972                 const struct nlattr *actions = da->actions;
 973
 974                 if (actions)
 975                         do_execute_actions(dp, skb, key, actions,
 976                                            nla_len(actions));
 977                 else
 978                         ovs_dp_process_packet(skb, key);
 979         } while (!action_fifo_is_empty(fifo));
 980
 981         /* Reset FIFO for the next packet.  */
 982         action_fifo_init(fifo);
 983 }
 984
 985 /* Execute a list of actions against 'skb'. */
 986 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
 987                         const struct sw_flow_actions *acts,
 988                         struct sw_flow_key *key)
 989 {
 990         int level = this_cpu_read(exec_actions_level);
 991         int err;
 992
 993         if (unlikely(level >= EXEC_ACTIONS_LEVEL_LIMIT)) {
 994                 if (net_ratelimit())
 995                         pr_warn("%s: packet loop detected, dropping.\n",
 996                                 ovs_dp_name(dp));
 997
 998                 kfree_skb(skb);
 999                 return -ELOOP;
1000         }
1001
1002         this_cpu_inc(exec_actions_level);
1003         err = do_execute_actions(dp, skb, key,
1004                                  acts->actions, acts->actions_len);
1005
1006         if (!level)
1007                 process_deferred_actions(dp);
1008
1009         this_cpu_dec(exec_actions_level);
1010
1011         /* This return status currently does not reflect the errors
1012          * encounted during deferred actions execution. Probably needs to
1013          * be fixed in the future.
1014          */
1015         return err;
1016 }
1017
1018 int action_fifos_init(void)
1019 {
1020         action_fifos = alloc_percpu(struct action_fifo);
1021         if (!action_fifos)
1022                 return -ENOMEM;
1023
1024         return 0;
1025 }
1026
1027 void action_fifos_exit(void)
1028 {
1029         free_percpu(action_fifos);
1030 }