datapath/actions.c

   1 /*
   2  * Copyright (c) 2007-2015 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/netfilter_ipv6.h>
  26 #include <linux/sctp.h>
  27 #include <linux/tcp.h>
  28 #include <linux/udp.h>
  29 #include <linux/in6.h>
  30 #include <linux/if_arp.h>
  31 #include <linux/if_vlan.h>
  32
  33 #include <net/dst.h>
  34 #include <net/ip.h>
  35 #include <net/ipv6.h>
  36 #include <net/checksum.h>
  37 #include <net/dsfield.h>
  38 #include <net/mpls.h>
  39 #include <net/sctp/checksum.h>
  40
  41 #include "datapath.h"
  42 #include "conntrack.h"
  43 #include "gso.h"
  44 #include "vlan.h"
  45 #include "vport.h"
  46
  47 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
  48                               struct sw_flow_key *key,
  49                               const struct nlattr *attr, int len);
  50
  51 struct deferred_action {
  52         struct sk_buff *skb;
  53         const struct nlattr *actions;
  54
  55         /* Store pkt_key clone when creating deferred action. */
  56         struct sw_flow_key pkt_key;
  57 };
  58
  59 #define MAX_L2_LEN      (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
  60 struct ovs_frag_data {
  61         unsigned long dst;
  62         struct vport *vport;
  63         struct ovs_skb_cb cb;
  64         __be16 inner_protocol;
  65         __u16 vlan_tci;
  66         __be16 vlan_proto;
  67         unsigned int l2_len;
  68         u8 l2_data[MAX_L2_LEN];
  69 };
  70
  71 #if LINUX_VERSION_CODE > KERNEL_VERSION(3,9,0)
  72 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
  73 #endif
  74
  75 #define DEFERRED_ACTION_FIFO_SIZE 10
  76 struct action_fifo {
  77         int head;
  78         int tail;
  79         /* Deferred action fifo queue storage. */
  80         struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
  81 };
  82
  83 static struct action_fifo __percpu *action_fifos;
  84 #define EXEC_ACTIONS_LEVEL_LIMIT 4   /* limit used to detect packet
  85                                       * looping by the network stack
  86                                       */
  87 static DEFINE_PER_CPU(int, exec_actions_level);
  88
  89 static void action_fifo_init(struct action_fifo *fifo)
  90 {
  91         fifo->head = 0;
  92         fifo->tail = 0;
  93 }
  94
  95 static bool action_fifo_is_empty(const struct action_fifo *fifo)
  96 {
  97         return (fifo->head == fifo->tail);
  98 }
  99
 100 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
 101 {
 102         if (action_fifo_is_empty(fifo))
 103                 return NULL;
 104
 105         return &fifo->fifo[fifo->tail++];
 106 }
 107
 108 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
 109 {
 110         if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
 111                 return NULL;
 112
 113         return &fifo->fifo[fifo->head++];
 114 }
 115
 116 /* Return queue entry if fifo is not full */
 117 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
 118                                                     const struct sw_flow_key *key,
 119                                                     const struct nlattr *attr)
 120 {
 121         struct action_fifo *fifo;
 122         struct deferred_action *da;
 123
 124         fifo = this_cpu_ptr(action_fifos);
 125         da = action_fifo_put(fifo);
 126         if (da) {
 127                 da->skb = skb;
 128                 da->actions = attr;
 129                 da->pkt_key = *key;
 130         }
 131
 132         return da;
 133 }
 134
 135 static void invalidate_flow_key(struct sw_flow_key *key)
 136 {
 137         key->eth.type = htons(0);
 138 }
 139
 140 static bool is_flow_key_valid(const struct sw_flow_key *key)
 141 {
 142         return !!key->eth.type;
 143 }
 144
 145 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 146                      const struct ovs_action_push_mpls *mpls)
 147 {
 148         __be32 *new_mpls_lse;
 149         struct ethhdr *hdr;
 150
 151         /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
 152         if (skb_encapsulation(skb))
 153                 return -ENOTSUPP;
 154
 155         if (skb_cow_head(skb, MPLS_HLEN) < 0)
 156                 return -ENOMEM;
 157
 158         skb_push(skb, MPLS_HLEN);
 159         memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
 160                 skb->mac_len);
 161         skb_reset_mac_header(skb);
 162
 163         new_mpls_lse = (__be32 *)skb_mpls_header(skb);
 164         *new_mpls_lse = mpls->mpls_lse;
 165
 166         if (skb->ip_summed == CHECKSUM_COMPLETE)
 167                 skb->csum = csum_add(skb->csum, csum_partial(new_mpls_lse,
 168                                                              MPLS_HLEN, 0));
 169
 170         hdr = eth_hdr(skb);
 171         hdr->h_proto = mpls->mpls_ethertype;
 172         if (!ovs_skb_get_inner_protocol(skb))
 173                 ovs_skb_set_inner_protocol(skb, skb->protocol);
 174         skb->protocol = mpls->mpls_ethertype;
 175
 176         invalidate_flow_key(key);
 177         return 0;
 178 }
 179
 180 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 181                     const __be16 ethertype)
 182 {
 183         struct ethhdr *hdr;
 184         int err;
 185
 186         err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
 187         if (unlikely(err))
 188                 return err;
 189
 190         skb_postpull_rcsum(skb, skb_mpls_header(skb), MPLS_HLEN);
 191
 192         memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
 193                 skb->mac_len);
 194
 195         __skb_pull(skb, MPLS_HLEN);
 196         skb_reset_mac_header(skb);
 197
 198         /* skb_mpls_header() is used to locate the ethertype
 199          * field correctly in the presence of VLAN tags.
 200          */
 201         hdr = (struct ethhdr *)(skb_mpls_header(skb) - ETH_HLEN);
 202         hdr->h_proto = ethertype;
 203         if (eth_p_mpls(skb->protocol))
 204                 skb->protocol = ethertype;
 205
 206         invalidate_flow_key(key);
 207         return 0;
 208 }
 209
 210 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
 211                     const __be32 *mpls_lse, const __be32 *mask)
 212 {
 213         __be32 *stack;
 214         __be32 lse;
 215         int err;
 216
 217         err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
 218         if (unlikely(err))
 219                 return err;
 220
 221         stack = (__be32 *)skb_mpls_header(skb);
 222         lse = OVS_MASKED(*stack, *mpls_lse, *mask);
 223         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 224                 __be32 diff[] = { ~(*stack), lse };
 225
 226                 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
 227                                           ~skb->csum);
 228         }
 229
 230         *stack = lse;
 231         flow_key->mpls.top_lse = lse;
 232         return 0;
 233 }
 234
 235 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 236 {
 237         int err;
 238
 239         err = skb_vlan_pop(skb);
 240         if (skb_vlan_tag_present(skb))
 241                 invalidate_flow_key(key);
 242         else
 243                 key->eth.tci = 0;
 244         return err;
 245 }
 246
 247 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
 248                      const struct ovs_action_push_vlan *vlan)
 249 {
 250         if (skb_vlan_tag_present(skb))
 251                 invalidate_flow_key(key);
 252         else
 253                 key->eth.tci = vlan->vlan_tci;
 254         return skb_vlan_push(skb, vlan->vlan_tpid,
 255                              ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
 256 }
 257
 258 /* 'src' is already properly masked. */
 259 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
 260 {
 261         u16 *dst = (u16 *)dst_;
 262         const u16 *src = (const u16 *)src_;
 263         const u16 *mask = (const u16 *)mask_;
 264
 265         OVS_SET_MASKED(dst[0], src[0], mask[0]);
 266         OVS_SET_MASKED(dst[1], src[1], mask[1]);
 267         OVS_SET_MASKED(dst[2], src[2], mask[2]);
 268 }
 269
 270 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
 271                         const struct ovs_key_ethernet *key,
 272                         const struct ovs_key_ethernet *mask)
 273 {
 274         int err;
 275
 276         err = skb_ensure_writable(skb, ETH_HLEN);
 277         if (unlikely(err))
 278                 return err;
 279
 280         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 281
 282         ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
 283                                mask->eth_src);
 284         ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
 285                                mask->eth_dst);
 286
 287         ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 288
 289         ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
 290         ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
 291         return 0;
 292 }
 293
 294 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
 295                                   __be32 addr, __be32 new_addr)
 296 {
 297         int transport_len = skb->len - skb_transport_offset(skb);
 298
 299         if (nh->frag_off & htons(IP_OFFSET))
 300                 return;
 301
 302         if (nh->protocol == IPPROTO_TCP) {
 303                 if (likely(transport_len >= sizeof(struct tcphdr)))
 304                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 305                                                  addr, new_addr, 1);
 306         } else if (nh->protocol == IPPROTO_UDP) {
 307                 if (likely(transport_len >= sizeof(struct udphdr))) {
 308                         struct udphdr *uh = udp_hdr(skb);
 309
 310                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 311                                 inet_proto_csum_replace4(&uh->check, skb,
 312                                                          addr, new_addr, 1);
 313                                 if (!uh->check)
 314                                         uh->check = CSUM_MANGLED_0;
 315                         }
 316                 }
 317         }
 318
 319 }
 320
 321 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 322                         __be32 *addr, __be32 new_addr)
 323 {
 324         update_ip_l4_checksum(skb, nh, *addr, new_addr);
 325         csum_replace4(&nh->check, *addr, new_addr);
 326         skb_clear_hash(skb);
 327         *addr = new_addr;
 328 }
 329
 330 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 331                                  __be32 addr[4], const __be32 new_addr[4])
 332 {
 333         int transport_len = skb->len - skb_transport_offset(skb);
 334
 335         if (l4_proto == NEXTHDR_TCP) {
 336                 if (likely(transport_len >= sizeof(struct tcphdr)))
 337                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 338                                                   addr, new_addr, 1);
 339         } else if (l4_proto == NEXTHDR_UDP) {
 340                 if (likely(transport_len >= sizeof(struct udphdr))) {
 341                         struct udphdr *uh = udp_hdr(skb);
 342
 343                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 344                                 inet_proto_csum_replace16(&uh->check, skb,
 345                                                           addr, new_addr, 1);
 346                                 if (!uh->check)
 347                                         uh->check = CSUM_MANGLED_0;
 348                         }
 349                 }
 350         } else if (l4_proto == NEXTHDR_ICMP) {
 351                 if (likely(transport_len >= sizeof(struct icmp6hdr)))
 352                         inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 353                                                   skb, addr, new_addr, 1);
 354         }
 355 }
 356
 357 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
 358                            const __be32 mask[4], __be32 masked[4])
 359 {
 360         masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
 361         masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
 362         masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
 363         masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
 364 }
 365
 366 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 367                           __be32 addr[4], const __be32 new_addr[4],
 368                           bool recalculate_csum)
 369 {
 370         if (likely(recalculate_csum))
 371                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 372
 373         skb_clear_hash(skb);
 374         memcpy(addr, new_addr, sizeof(__be32[4]));
 375 }
 376
 377 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
 378 {
 379         /* Bits 21-24 are always unmasked, so this retains their values. */
 380         OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
 381         OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
 382         OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
 383 }
 384
 385 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
 386                        u8 mask)
 387 {
 388         new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
 389
 390         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 391         nh->ttl = new_ttl;
 392 }
 393
 394 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
 395                     const struct ovs_key_ipv4 *key,
 396                     const struct ovs_key_ipv4 *mask)
 397 {
 398         struct iphdr *nh;
 399         __be32 new_addr;
 400         int err;
 401
 402         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 403                                   sizeof(struct iphdr));
 404         if (unlikely(err))
 405                 return err;
 406
 407         nh = ip_hdr(skb);
 408
 409         /* Setting an IP addresses is typically only a side effect of
 410          * matching on them in the current userspace implementation, so it
 411          * makes sense to check if the value actually changed.
 412          */
 413         if (mask->ipv4_src) {
 414                 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
 415
 416                 if (unlikely(new_addr != nh->saddr)) {
 417                         set_ip_addr(skb, nh, &nh->saddr, new_addr);
 418                         flow_key->ipv4.addr.src = new_addr;
 419                 }
 420         }
 421         if (mask->ipv4_dst) {
 422                 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
 423
 424                 if (unlikely(new_addr != nh->daddr)) {
 425                         set_ip_addr(skb, nh, &nh->daddr, new_addr);
 426                         flow_key->ipv4.addr.dst = new_addr;
 427                 }
 428         }
 429         if (mask->ipv4_tos) {
 430                 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
 431                 flow_key->ip.tos = nh->tos;
 432         }
 433         if (mask->ipv4_ttl) {
 434                 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
 435                 flow_key->ip.ttl = nh->ttl;
 436         }
 437
 438         return 0;
 439 }
 440
 441 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
 442 {
 443         return !!(addr[0] | addr[1] | addr[2] | addr[3]);
 444 }
 445
 446 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
 447                     const struct ovs_key_ipv6 *key,
 448                     const struct ovs_key_ipv6 *mask)
 449 {
 450         struct ipv6hdr *nh;
 451         int err;
 452
 453         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 454                                   sizeof(struct ipv6hdr));
 455         if (unlikely(err))
 456                 return err;
 457
 458         nh = ipv6_hdr(skb);
 459
 460         /* Setting an IP addresses is typically only a side effect of
 461          * matching on them in the current userspace implementation, so it
 462          * makes sense to check if the value actually changed.
 463          */
 464         if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
 465                 __be32 *saddr = (__be32 *)&nh->saddr;
 466                 __be32 masked[4];
 467
 468                 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
 469
 470                 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
 471                         set_ipv6_addr(skb, key->ipv6_proto, saddr, masked,
 472                                       true);
 473                         memcpy(&flow_key->ipv6.addr.src, masked,
 474                                sizeof(flow_key->ipv6.addr.src));
 475                 }
 476         }
 477         if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
 478                 unsigned int offset = 0;
 479                 int flags = IP6_FH_F_SKIP_RH;
 480                 bool recalc_csum = true;
 481                 __be32 *daddr = (__be32 *)&nh->daddr;
 482                 __be32 masked[4];
 483
 484                 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
 485
 486                 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
 487                         if (ipv6_ext_hdr(nh->nexthdr))
 488                                 recalc_csum = (ipv6_find_hdr(skb, &offset,
 489                                                              NEXTHDR_ROUTING,
 490                                                              NULL, &flags)
 491                                                != NEXTHDR_ROUTING);
 492
 493                         set_ipv6_addr(skb, key->ipv6_proto, daddr, masked,
 494                                       recalc_csum);
 495                         memcpy(&flow_key->ipv6.addr.dst, masked,
 496                                sizeof(flow_key->ipv6.addr.dst));
 497                 }
 498         }
 499         if (mask->ipv6_tclass) {
 500                 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
 501                 flow_key->ip.tos = ipv6_get_dsfield(nh);
 502         }
 503         if (mask->ipv6_label) {
 504                 set_ipv6_fl(nh, ntohl(key->ipv6_label),
 505                             ntohl(mask->ipv6_label));
 506                 flow_key->ipv6.label =
 507                     *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 508         }
 509         if (mask->ipv6_hlimit) {
 510                 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
 511                                mask->ipv6_hlimit);
 512                 flow_key->ip.ttl = nh->hop_limit;
 513         }
 514         return 0;
 515 }
 516
 517 /* Must follow skb_ensure_writable() since that can move the skb data. */
 518 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 519                         __be16 new_port, __sum16 *check)
 520 {
 521         inet_proto_csum_replace2(check, skb, *port, new_port, 0);
 522         *port = new_port;
 523 }
 524
 525 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 526                    const struct ovs_key_udp *key,
 527                    const struct ovs_key_udp *mask)
 528 {
 529         struct udphdr *uh;
 530         __be16 src, dst;
 531         int err;
 532
 533         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 534                                   sizeof(struct udphdr));
 535         if (unlikely(err))
 536                 return err;
 537
 538         uh = udp_hdr(skb);
 539         /* Either of the masks is non-zero, so do not bother checking them. */
 540         src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
 541         dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
 542
 543         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 544                 if (likely(src != uh->source)) {
 545                         set_tp_port(skb, &uh->source, src, &uh->check);
 546                         flow_key->tp.src = src;
 547                 }
 548                 if (likely(dst != uh->dest)) {
 549                         set_tp_port(skb, &uh->dest, dst, &uh->check);
 550                         flow_key->tp.dst = dst;
 551                 }
 552
 553                 if (unlikely(!uh->check))
 554                         uh->check = CSUM_MANGLED_0;
 555         } else {
 556                 uh->source = src;
 557                 uh->dest = dst;
 558                 flow_key->tp.src = src;
 559                 flow_key->tp.dst = dst;
 560         }
 561
 562         skb_clear_hash(skb);
 563
 564         return 0;
 565 }
 566
 567 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 568                    const struct ovs_key_tcp *key,
 569                    const struct ovs_key_tcp *mask)
 570 {
 571         struct tcphdr *th;
 572         __be16 src, dst;
 573         int err;
 574
 575         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 576                                   sizeof(struct tcphdr));
 577         if (unlikely(err))
 578                 return err;
 579
 580         th = tcp_hdr(skb);
 581         src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
 582         if (likely(src != th->source)) {
 583                 set_tp_port(skb, &th->source, src, &th->check);
 584                 flow_key->tp.src = src;
 585         }
 586         dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
 587         if (likely(dst != th->dest)) {
 588                 set_tp_port(skb, &th->dest, dst, &th->check);
 589                 flow_key->tp.dst = dst;
 590         }
 591         skb_clear_hash(skb);
 592
 593         return 0;
 594 }
 595
 596 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 597                     const struct ovs_key_sctp *key,
 598                     const struct ovs_key_sctp *mask)
 599 {
 600         unsigned int sctphoff = skb_transport_offset(skb);
 601         struct sctphdr *sh;
 602         __le32 old_correct_csum, new_csum, old_csum;
 603         int err;
 604
 605         err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
 606         if (unlikely(err))
 607                 return err;
 608
 609         sh = sctp_hdr(skb);
 610         old_csum = sh->checksum;
 611         old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 612
 613         sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
 614         sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
 615
 616         new_csum = sctp_compute_cksum(skb, sctphoff);
 617
 618         /* Carry any checksum errors through. */
 619         sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 620
 621         skb_clear_hash(skb);
 622         flow_key->tp.src = sh->source;
 623         flow_key->tp.dst = sh->dest;
 624
 625         return 0;
 626 }
 627
 628 #if LINUX_VERSION_CODE > KERNEL_VERSION(3,9,0)
 629 static int ovs_vport_output(OVS_VPORT_OUTPUT_PARAMS)
 630 {
 631         struct ovs_frag_data *data = get_pcpu_ptr(ovs_frag_data_storage);
 632         struct vport *vport = data->vport;
 633
 634         if (skb_cow_head(skb, data->l2_len) < 0) {
 635                 kfree_skb(skb);
 636                 return -ENOMEM;
 637         }
 638
 639         __skb_dst_copy(skb, data->dst);
 640         *OVS_CB(skb) = data->cb;
 641         ovs_skb_set_inner_protocol(skb, data->inner_protocol);
 642         skb->vlan_tci = data->vlan_tci;
 643         skb->vlan_proto = data->vlan_proto;
 644
 645         /* Reconstruct the MAC header.  */
 646         skb_push(skb, data->l2_len);
 647         memcpy(skb->data, &data->l2_data, data->l2_len);
 648         ovs_skb_postpush_rcsum(skb, skb->data, data->l2_len);
 649         skb_reset_mac_header(skb);
 650
 651         ovs_vport_send(vport, skb);
 652         return 0;
 653 }
 654
 655 static unsigned int
 656 ovs_dst_get_mtu(const struct dst_entry *dst)
 657 {
 658         return dst->dev->mtu;
 659 }
 660
 661 static struct dst_ops ovs_dst_ops = {
 662         .family = AF_UNSPEC,
 663         .mtu = ovs_dst_get_mtu,
 664 };
 665
 666 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
 667  * ovs_vport_output(), which is called once per fragmented packet.
 668  */
 669 static void prepare_frag(struct vport *vport, struct sk_buff *skb)
 670 {
 671         unsigned int hlen = skb_network_offset(skb);
 672         struct ovs_frag_data *data;
 673
 674         data = get_pcpu_ptr(ovs_frag_data_storage);
 675         data->dst = (unsigned long) skb_dst(skb);
 676         data->vport = vport;
 677         data->cb = *OVS_CB(skb);
 678         data->inner_protocol = ovs_skb_get_inner_protocol(skb);
 679         data->vlan_tci = skb->vlan_tci;
 680         data->vlan_proto = skb->vlan_proto;
 681         data->l2_len = hlen;
 682         memcpy(&data->l2_data, skb->data, hlen);
 683
 684         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 685         skb_pull(skb, hlen);
 686 }
 687
 688 static void ovs_fragment(struct vport *vport, struct sk_buff *skb, u16 mru,
 689                          __be16 ethertype)
 690 {
 691         if (skb_network_offset(skb) > MAX_L2_LEN) {
 692                 OVS_NLERR(1, "L2 header too long to fragment");
 693                 goto err;
 694         }
 695
 696         if (ethertype == htons(ETH_P_IP)) {
 697                 struct dst_entry ovs_dst;
 698                 unsigned long orig_dst;
 699
 700                 prepare_frag(vport, skb);
 701                 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
 702                          DST_OBSOLETE_NONE, DST_NOCOUNT);
 703                 ovs_dst.dev = vport->dev;
 704
 705                 orig_dst = (unsigned long) skb_dst(skb);
 706                 skb_dst_set_noref(skb, &ovs_dst);
 707                 IPCB(skb)->frag_max_size = mru;
 708
 709                 ip_do_fragment(skb->sk, skb, ovs_vport_output);
 710                 refdst_drop(orig_dst);
 711         } else if (ethertype == htons(ETH_P_IPV6)) {
 712                 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
 713                 unsigned long orig_dst;
 714                 struct rt6_info ovs_rt;
 715
 716                 if (!v6ops) {
 717                         goto err;
 718                 }
 719
 720                 prepare_frag(vport, skb);
 721                 memset(&ovs_rt, 0, sizeof(ovs_rt));
 722                 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
 723                          DST_OBSOLETE_NONE, DST_NOCOUNT);
 724                 ovs_rt.dst.dev = vport->dev;
 725
 726                 orig_dst = (unsigned long) skb_dst(skb);
 727                 skb_dst_set_noref(skb, &ovs_rt.dst);
 728                 IP6CB(skb)->frag_max_size = mru;
 729
 730                 v6ops->fragment(skb->sk, skb, ovs_vport_output);
 731                 refdst_drop(orig_dst);
 732         } else {
 733                 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
 734                           ovs_vport_name(vport), ntohs(ethertype), mru,
 735                           vport->dev->mtu);
 736                 goto err;
 737         }
 738
 739         return;
 740 err:
 741         kfree_skb(skb);
 742 }
 743 #else /* <= 3.9 */
 744 static void ovs_fragment(struct vport *vport, struct sk_buff *skb, u16 mru,
 745                          __be16 ethertype)
 746 {
 747         WARN_ONCE(1, "Fragment unavailable ->%s: eth=%04x, MRU=%d, MTU=%d.",
 748                   ovs_vport_name(vport), ntohs(ethertype), mru,
 749                   vport->dev->mtu);
 750         kfree_skb(skb);
 751 }
 752 #endif
 753
 754 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
 755                       struct sw_flow_key *key)
 756 {
 757         struct vport *vport = ovs_vport_rcu(dp, out_port);
 758
 759         if (likely(vport)) {
 760                 u16 mru = OVS_CB(skb)->mru;
 761
 762                 if (likely(!mru || (skb->len <= mru + ETH_HLEN))) {
 763                         ovs_vport_send(vport, skb);
 764                 } else if (mru <= vport->dev->mtu) {
 765                         __be16 ethertype = key->eth.type;
 766
 767                         if (!is_flow_key_valid(key)) {
 768                                 if (eth_p_mpls(skb->protocol))
 769                                         ethertype = ovs_skb_get_inner_protocol(skb);
 770                                 else
 771                                         ethertype = vlan_get_protocol(skb);
 772                         }
 773
 774                         ovs_fragment(vport, skb, mru, ethertype);
 775                 } else {
 776                         OVS_NLERR(true, "Cannot fragment IP frames");
 777                         kfree_skb(skb);
 778                 }
 779         } else {
 780                 kfree_skb(skb);
 781         }
 782 }
 783 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 784                             struct sw_flow_key *key, const struct nlattr *attr,
 785                             const struct nlattr *actions, int actions_len)
 786 {
 787         struct ip_tunnel_info info;
 788         struct dp_upcall_info upcall;
 789         const struct nlattr *a;
 790         int rem;
 791
 792         memset(&upcall, 0, sizeof(upcall));
 793         upcall.cmd = OVS_PACKET_CMD_ACTION;
 794         upcall.mru = OVS_CB(skb)->mru;
 795
 796         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 797                  a = nla_next(a, &rem)) {
 798                 switch (nla_type(a)) {
 799                 case OVS_USERSPACE_ATTR_USERDATA:
 800                         upcall.userdata = a;
 801                         break;
 802
 803                 case OVS_USERSPACE_ATTR_PID:
 804                         upcall.portid = nla_get_u32(a);
 805                         break;
 806
 807                 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
 808                         /* Get out tunnel info. */
 809                         struct vport *vport;
 810
 811                         vport = ovs_vport_rcu(dp, nla_get_u32(a));
 812                         if (vport) {
 813                                 int err;
 814
 815                                 upcall.egress_tun_info = &info;
 816                                 err = ovs_vport_get_egress_tun_info(vport, skb,
 817                                                                     &upcall);
 818                                 if (err)
 819                                         upcall.egress_tun_info = NULL;
 820                         }
 821
 822                         break;
 823                 }
 824
 825                 case OVS_USERSPACE_ATTR_ACTIONS: {
 826                         /* Include actions. */
 827                         upcall.actions = actions;
 828                         upcall.actions_len = actions_len;
 829                         break;
 830                 }
 831
 832                 } /* End of switch. */
 833         }
 834
 835         return ovs_dp_upcall(dp, skb, key, &upcall);
 836 }
 837
 838 static int sample(struct datapath *dp, struct sk_buff *skb,
 839                   struct sw_flow_key *key, const struct nlattr *attr,
 840                   const struct nlattr *actions, int actions_len)
 841 {
 842         const struct nlattr *acts_list = NULL;
 843         const struct nlattr *a;
 844         int rem;
 845
 846         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 847                  a = nla_next(a, &rem)) {
 848                 u32 probability;
 849
 850                 switch (nla_type(a)) {
 851                 case OVS_SAMPLE_ATTR_PROBABILITY:
 852                         probability = nla_get_u32(a);
 853                         if (!probability || prandom_u32() > probability)
 854                                 return 0;
 855                         break;
 856
 857                 case OVS_SAMPLE_ATTR_ACTIONS:
 858                         acts_list = a;
 859                         break;
 860                 }
 861         }
 862
 863         rem = nla_len(acts_list);
 864         a = nla_data(acts_list);
 865
 866         /* Actions list is empty, do nothing */
 867         if (unlikely(!rem))
 868                 return 0;
 869
 870         /* The only known usage of sample action is having a single user-space
 871          * action. Treat this usage as a special case.
 872          * The output_userspace() should clone the skb to be sent to the
 873          * user space. This skb will be consumed by its caller.
 874          */
 875         if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
 876                    nla_is_last(a, rem)))
 877                 return output_userspace(dp, skb, key, a, actions, actions_len);
 878
 879         skb = skb_clone(skb, GFP_ATOMIC);
 880         if (!skb)
 881                 /* Skip the sample action when out of memory. */
 882                 return 0;
 883
 884         if (!add_deferred_actions(skb, key, a)) {
 885                 if (net_ratelimit())
 886                         pr_warn("%s: deferred actions limit reached, dropping sample action\n",
 887                                 ovs_dp_name(dp));
 888
 889                 kfree_skb(skb);
 890         }
 891         return 0;
 892 }
 893
 894 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
 895                          const struct nlattr *attr)
 896 {
 897         struct ovs_action_hash *hash_act = nla_data(attr);
 898         u32 hash = 0;
 899
 900         /* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
 901         hash = skb_get_hash(skb);
 902         hash = jhash_1word(hash, hash_act->hash_basis);
 903         if (!hash)
 904                 hash = 0x1;
 905
 906         key->ovs_flow_hash = hash;
 907 }
 908
 909 static int execute_set_action(struct sk_buff *skb,
 910                               struct sw_flow_key *flow_key,
 911                               const struct nlattr *a)
 912 {
 913         /* Only tunnel set execution is supported without a mask. */
 914         if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
 915                 struct ovs_tunnel_info *tun = nla_data(a);
 916
 917                 ovs_skb_dst_drop(skb);
 918                 ovs_dst_hold((struct dst_entry *)tun->tun_dst);
 919                 ovs_skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
 920                 return 0;
 921         }
 922
 923         return -EINVAL;
 924 }
 925
 926 /* Mask is at the midpoint of the data. */
 927 #define get_mask(a, type) ((const type)nla_data(a) + 1)
 928
 929 static int execute_masked_set_action(struct sk_buff *skb,
 930                                      struct sw_flow_key *flow_key,
 931                                      const struct nlattr *a)
 932 {
 933         int err = 0;
 934
 935         switch (nla_type(a)) {
 936         case OVS_KEY_ATTR_PRIORITY:
 937                 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
 938                                *get_mask(a, u32 *));
 939                 flow_key->phy.priority = skb->priority;
 940                 break;
 941
 942         case OVS_KEY_ATTR_SKB_MARK:
 943                 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
 944                 flow_key->phy.skb_mark = skb->mark;
 945                 break;
 946
 947         case OVS_KEY_ATTR_TUNNEL_INFO:
 948                 /* Masked data not supported for tunnel. */
 949                 err = -EINVAL;
 950                 break;
 951
 952         case OVS_KEY_ATTR_ETHERNET:
 953                 err = set_eth_addr(skb, flow_key, nla_data(a),
 954                                    get_mask(a, struct ovs_key_ethernet *));
 955                 break;
 956
 957         case OVS_KEY_ATTR_IPV4:
 958                 err = set_ipv4(skb, flow_key, nla_data(a),
 959                                get_mask(a, struct ovs_key_ipv4 *));
 960                 break;
 961
 962         case OVS_KEY_ATTR_IPV6:
 963                 err = set_ipv6(skb, flow_key, nla_data(a),
 964                                get_mask(a, struct ovs_key_ipv6 *));
 965                 break;
 966
 967         case OVS_KEY_ATTR_TCP:
 968                 err = set_tcp(skb, flow_key, nla_data(a),
 969                               get_mask(a, struct ovs_key_tcp *));
 970                 break;
 971
 972         case OVS_KEY_ATTR_UDP:
 973                 err = set_udp(skb, flow_key, nla_data(a),
 974                               get_mask(a, struct ovs_key_udp *));
 975                 break;
 976
 977         case OVS_KEY_ATTR_SCTP:
 978                 err = set_sctp(skb, flow_key, nla_data(a),
 979                                get_mask(a, struct ovs_key_sctp *));
 980                 break;
 981
 982         case OVS_KEY_ATTR_MPLS:
 983                 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
 984                                                                     __be32 *));
 985                 break;
 986
 987         case OVS_KEY_ATTR_CT_STATE:
 988         case OVS_KEY_ATTR_CT_ZONE:
 989         case OVS_KEY_ATTR_CT_MARK:
 990         case OVS_KEY_ATTR_CT_LABELS:
 991                 err = -EINVAL;
 992                 break;
 993         }
 994
 995         return err;
 996 }
 997
 998 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
 999                           struct sw_flow_key *key,
1000                           const struct nlattr *a, int rem)
1001 {
1002         struct deferred_action *da;
1003
1004         if (!is_flow_key_valid(key)) {
1005                 int err;
1006
1007                 err = ovs_flow_key_update(skb, key);
1008                 if (err)
1009                         return err;
1010         }
1011         BUG_ON(!is_flow_key_valid(key));
1012
1013         if (!nla_is_last(a, rem)) {
1014                 /* Recirc action is the not the last action
1015                  * of the action list, need to clone the skb.
1016                  */
1017                 skb = skb_clone(skb, GFP_ATOMIC);
1018
1019                 /* Skip the recirc action when out of memory, but
1020                  * continue on with the rest of the action list.
1021                  */
1022                 if (!skb)
1023                         return 0;
1024         }
1025
1026         da = add_deferred_actions(skb, key, NULL);
1027         if (da) {
1028                 da->pkt_key.recirc_id = nla_get_u32(a);
1029         } else {
1030                 kfree_skb(skb);
1031
1032                 if (net_ratelimit())
1033                         pr_warn("%s: deferred action limit reached, drop recirc action\n",
1034                                 ovs_dp_name(dp));
1035         }
1036
1037         return 0;
1038 }
1039
1040 /* Execute a list of actions against 'skb'. */
1041 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1042                               struct sw_flow_key *key,
1043                               const struct nlattr *attr, int len)
1044 {
1045         /* Every output action needs a separate clone of 'skb', but the common
1046          * case is just a single output action, so that doing a clone and
1047          * then freeing the original skbuff is wasteful.  So the following code
1048          * is slightly obscure just to avoid that.
1049          */
1050         int prev_port = -1;
1051         const struct nlattr *a;
1052         int rem;
1053
1054         for (a = attr, rem = len; rem > 0;
1055              a = nla_next(a, &rem)) {
1056                 int err = 0;
1057
1058                 if (unlikely(prev_port != -1)) {
1059                         struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
1060
1061                         if (out_skb)
1062                                 do_output(dp, out_skb, prev_port, key);
1063
1064                         prev_port = -1;
1065                 }
1066
1067                 switch (nla_type(a)) {
1068                 case OVS_ACTION_ATTR_OUTPUT:
1069                         prev_port = nla_get_u32(a);
1070                         break;
1071
1072                 case OVS_ACTION_ATTR_USERSPACE:
1073                         output_userspace(dp, skb, key, a, attr, len);
1074                         break;
1075
1076                 case OVS_ACTION_ATTR_HASH:
1077                         execute_hash(skb, key, a);
1078                         break;
1079
1080                 case OVS_ACTION_ATTR_PUSH_MPLS:
1081                         err = push_mpls(skb, key, nla_data(a));
1082                         break;
1083
1084                 case OVS_ACTION_ATTR_POP_MPLS:
1085                         err = pop_mpls(skb, key, nla_get_be16(a));
1086                         break;
1087
1088                 case OVS_ACTION_ATTR_PUSH_VLAN:
1089                         err = push_vlan(skb, key, nla_data(a));
1090                         break;
1091
1092                 case OVS_ACTION_ATTR_POP_VLAN:
1093                         err = pop_vlan(skb, key);
1094                         break;
1095
1096                 case OVS_ACTION_ATTR_RECIRC:
1097                         err = execute_recirc(dp, skb, key, a, rem);
1098                         if (nla_is_last(a, rem)) {
1099                                 /* If this is the last action, the skb has
1100                                  * been consumed or freed.
1101                                  * Return immediately.
1102                                  */
1103                                 return err;
1104                         }
1105                         break;
1106
1107                 case OVS_ACTION_ATTR_SET:
1108                         err = execute_set_action(skb, key, nla_data(a));
1109                         break;
1110
1111                 case OVS_ACTION_ATTR_SET_MASKED:
1112                 case OVS_ACTION_ATTR_SET_TO_MASKED:
1113                         err = execute_masked_set_action(skb, key, nla_data(a));
1114                         break;
1115
1116                 case OVS_ACTION_ATTR_SAMPLE:
1117                         err = sample(dp, skb, key, a, attr, len);
1118                         break;
1119
1120                 case OVS_ACTION_ATTR_CT:
1121                         if (!is_flow_key_valid(key)) {
1122                                 err = ovs_flow_key_update(skb, key);
1123                                 if (err)
1124                                         return err;
1125                         }
1126
1127                         err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1128                                              nla_data(a));
1129
1130                         /* Hide stolen IP fragments from user space. */
1131                         if (err)
1132                                 return err == -EINPROGRESS ? 0 : err;
1133                         break;
1134                 }
1135
1136                 if (unlikely(err)) {
1137                         kfree_skb(skb);
1138                         return err;
1139                 }
1140         }
1141
1142         if (prev_port != -1)
1143                 do_output(dp, skb, prev_port, key);
1144         else
1145                 consume_skb(skb);
1146
1147         return 0;
1148 }
1149
1150 static void process_deferred_actions(struct datapath *dp)
1151 {
1152         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1153
1154         /* Do not touch the FIFO in case there is no deferred actions. */
1155         if (action_fifo_is_empty(fifo))
1156                 return;
1157
1158         /* Finishing executing all deferred actions. */
1159         do {
1160                 struct deferred_action *da = action_fifo_get(fifo);
1161                 struct sk_buff *skb = da->skb;
1162                 struct sw_flow_key *key = &da->pkt_key;
1163                 const struct nlattr *actions = da->actions;
1164
1165                 if (actions)
1166                         do_execute_actions(dp, skb, key, actions,
1167                                            nla_len(actions));
1168                 else
1169                         ovs_dp_process_packet(skb, key);
1170         } while (!action_fifo_is_empty(fifo));
1171
1172         /* Reset FIFO for the next packet.  */
1173         action_fifo_init(fifo);
1174 }
1175
1176 /* Execute a list of actions against 'skb'. */
1177 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1178                         const struct sw_flow_actions *acts,
1179                         struct sw_flow_key *key)
1180 {
1181         int level = this_cpu_read(exec_actions_level);
1182         int err;
1183
1184         if (unlikely(level >= EXEC_ACTIONS_LEVEL_LIMIT)) {
1185                 if (net_ratelimit())
1186                         pr_warn("%s: packet loop detected, dropping.\n",
1187                                 ovs_dp_name(dp));
1188
1189                 kfree_skb(skb);
1190                 return -ELOOP;
1191         }
1192
1193         this_cpu_inc(exec_actions_level);
1194         err = do_execute_actions(dp, skb, key,
1195                                  acts->actions, acts->actions_len);
1196
1197         if (!level)
1198                 process_deferred_actions(dp);
1199
1200         this_cpu_dec(exec_actions_level);
1201
1202         /* This return status currently does not reflect the errors
1203          * encounted during deferred actions execution. Probably needs to
1204          * be fixed in the future.
1205          */
1206         return err;
1207 }
1208
1209 int action_fifos_init(void)
1210 {
1211         action_fifos = alloc_percpu(struct action_fifo);
1212         if (!action_fifos)
1213                 return -ENOMEM;
1214
1215         return 0;
1216 }
1217
1218 void action_fifos_exit(void)
1219 {
1220         free_percpu(action_fifos);
1221 }