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 "vport.h"
  45
  46 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
  47                               struct sw_flow_key *key,
  48                               const struct nlattr *attr, int len);
  49
  50 struct deferred_action {
  51         struct sk_buff *skb;
  52         const struct nlattr *actions;
  53
  54         /* Store pkt_key clone when creating deferred action. */
  55         struct sw_flow_key pkt_key;
  56 };
  57
  58 #define MAX_L2_LEN      (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
  59 struct ovs_frag_data {
  60         unsigned long dst;
  61         struct vport *vport;
  62         struct ovs_gso_cb cb;
  63         __be16 inner_protocol;
  64         __u16 vlan_tci;
  65         __be16 vlan_proto;
  66         unsigned int l2_len;
  67         u8 l2_data[MAX_L2_LEN];
  68 };
  69
  70 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
  71
  72 #define DEFERRED_ACTION_FIFO_SIZE 10
  73 struct action_fifo {
  74         int head;
  75         int tail;
  76         /* Deferred action fifo queue storage. */
  77         struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
  78 };
  79
  80 static struct action_fifo __percpu *action_fifos;
  81 #define EXEC_ACTIONS_LEVEL_LIMIT 4   /* limit used to detect packet
  82                                       * looping by the network stack
  83                                       */
  84 static DEFINE_PER_CPU(int, exec_actions_level);
  85
  86 static void action_fifo_init(struct action_fifo *fifo)
  87 {
  88         fifo->head = 0;
  89         fifo->tail = 0;
  90 }
  91
  92 static bool action_fifo_is_empty(const struct action_fifo *fifo)
  93 {
  94         return (fifo->head == fifo->tail);
  95 }
  96
  97 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
  98 {
  99         if (action_fifo_is_empty(fifo))
 100                 return NULL;
 101
 102         return &fifo->fifo[fifo->tail++];
 103 }
 104
 105 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
 106 {
 107         if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
 108                 return NULL;
 109
 110         return &fifo->fifo[fifo->head++];
 111 }
 112
 113 /* Return queue entry if fifo is not full */
 114 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
 115                                                     const struct sw_flow_key *key,
 116                                                     const struct nlattr *attr)
 117 {
 118         struct action_fifo *fifo;
 119         struct deferred_action *da;
 120
 121         fifo = this_cpu_ptr(action_fifos);
 122         da = action_fifo_put(fifo);
 123         if (da) {
 124                 da->skb = skb;
 125                 da->actions = attr;
 126                 da->pkt_key = *key;
 127         }
 128
 129         return da;
 130 }
 131
 132 static void invalidate_flow_key(struct sw_flow_key *key)
 133 {
 134         key->eth.type = htons(0);
 135 }
 136
 137 static bool is_flow_key_valid(const struct sw_flow_key *key)
 138 {
 139         return !!key->eth.type;
 140 }
 141
 142 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 143                      const struct ovs_action_push_mpls *mpls)
 144 {
 145         __be32 *new_mpls_lse;
 146         struct ethhdr *hdr;
 147
 148         /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
 149         if (skb->encapsulation)
 150                 return -ENOTSUPP;
 151
 152         if (skb_cow_head(skb, MPLS_HLEN) < 0)
 153                 return -ENOMEM;
 154
 155         skb_push(skb, MPLS_HLEN);
 156         memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
 157                 skb->mac_len);
 158         skb_reset_mac_header(skb);
 159
 160         new_mpls_lse = (__be32 *)skb_mpls_header(skb);
 161         *new_mpls_lse = mpls->mpls_lse;
 162
 163         skb_postpush_rcsum(skb, new_mpls_lse, MPLS_HLEN);
 164
 165         hdr = eth_hdr(skb);
 166         hdr->h_proto = mpls->mpls_ethertype;
 167         if (!ovs_skb_get_inner_protocol(skb))
 168                 ovs_skb_set_inner_protocol(skb, skb->protocol);
 169         skb->protocol = mpls->mpls_ethertype;
 170
 171         invalidate_flow_key(key);
 172         return 0;
 173 }
 174
 175 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 176                     const __be16 ethertype)
 177 {
 178         struct ethhdr *hdr;
 179         int err;
 180
 181         err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
 182         if (unlikely(err))
 183                 return err;
 184
 185         skb_postpull_rcsum(skb, skb_mpls_header(skb), MPLS_HLEN);
 186
 187         memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
 188                 skb->mac_len);
 189
 190         __skb_pull(skb, MPLS_HLEN);
 191         skb_reset_mac_header(skb);
 192
 193         /* skb_mpls_header() is used to locate the ethertype
 194          * field correctly in the presence of VLAN tags.
 195          */
 196         hdr = (struct ethhdr *)(skb_mpls_header(skb) - ETH_HLEN);
 197         hdr->h_proto = ethertype;
 198         if (eth_p_mpls(skb->protocol))
 199                 skb->protocol = ethertype;
 200
 201         invalidate_flow_key(key);
 202         return 0;
 203 }
 204
 205 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
 206                     const __be32 *mpls_lse, const __be32 *mask)
 207 {
 208         __be32 *stack;
 209         __be32 lse;
 210         int err;
 211
 212         err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
 213         if (unlikely(err))
 214                 return err;
 215
 216         stack = (__be32 *)skb_mpls_header(skb);
 217         lse = OVS_MASKED(*stack, *mpls_lse, *mask);
 218         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 219                 __be32 diff[] = { ~(*stack), lse };
 220
 221                 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
 222                                           ~skb->csum);
 223         }
 224
 225         *stack = lse;
 226         flow_key->mpls.top_lse = lse;
 227         return 0;
 228 }
 229
 230 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 231 {
 232         int err;
 233
 234         err = skb_vlan_pop(skb);
 235         if (skb_vlan_tag_present(skb))
 236                 invalidate_flow_key(key);
 237         else
 238                 key->eth.tci = 0;
 239         return err;
 240 }
 241
 242 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
 243                      const struct ovs_action_push_vlan *vlan)
 244 {
 245         if (skb_vlan_tag_present(skb))
 246                 invalidate_flow_key(key);
 247         else
 248                 key->eth.tci = vlan->vlan_tci;
 249         return skb_vlan_push(skb, vlan->vlan_tpid,
 250                              ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
 251 }
 252
 253 /* 'src' is already properly masked. */
 254 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
 255 {
 256         u16 *dst = (u16 *)dst_;
 257         const u16 *src = (const u16 *)src_;
 258         const u16 *mask = (const u16 *)mask_;
 259
 260         OVS_SET_MASKED(dst[0], src[0], mask[0]);
 261         OVS_SET_MASKED(dst[1], src[1], mask[1]);
 262         OVS_SET_MASKED(dst[2], src[2], mask[2]);
 263 }
 264
 265 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
 266                         const struct ovs_key_ethernet *key,
 267                         const struct ovs_key_ethernet *mask)
 268 {
 269         int err;
 270
 271         err = skb_ensure_writable(skb, ETH_HLEN);
 272         if (unlikely(err))
 273                 return err;
 274
 275         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 276
 277         ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
 278                                mask->eth_src);
 279         ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
 280                                mask->eth_dst);
 281
 282         skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 283
 284         ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
 285         ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
 286         return 0;
 287 }
 288
 289 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
 290                                   __be32 addr, __be32 new_addr)
 291 {
 292         int transport_len = skb->len - skb_transport_offset(skb);
 293
 294         if (nh->frag_off & htons(IP_OFFSET))
 295                 return;
 296
 297         if (nh->protocol == IPPROTO_TCP) {
 298                 if (likely(transport_len >= sizeof(struct tcphdr)))
 299                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 300                                                  addr, new_addr, true);
 301         } else if (nh->protocol == IPPROTO_UDP) {
 302                 if (likely(transport_len >= sizeof(struct udphdr))) {
 303                         struct udphdr *uh = udp_hdr(skb);
 304
 305                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 306                                 inet_proto_csum_replace4(&uh->check, skb,
 307                                                          addr, new_addr, true);
 308                                 if (!uh->check)
 309                                         uh->check = CSUM_MANGLED_0;
 310                         }
 311                 }
 312         }
 313
 314 }
 315
 316 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 317                         __be32 *addr, __be32 new_addr)
 318 {
 319         update_ip_l4_checksum(skb, nh, *addr, new_addr);
 320         csum_replace4(&nh->check, *addr, new_addr);
 321         skb_clear_hash(skb);
 322         *addr = new_addr;
 323 }
 324
 325 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 326                                  __be32 addr[4], const __be32 new_addr[4])
 327 {
 328         int transport_len = skb->len - skb_transport_offset(skb);
 329
 330         if (l4_proto == NEXTHDR_TCP) {
 331                 if (likely(transport_len >= sizeof(struct tcphdr)))
 332                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 333                                                   addr, new_addr, true);
 334         } else if (l4_proto == NEXTHDR_UDP) {
 335                 if (likely(transport_len >= sizeof(struct udphdr))) {
 336                         struct udphdr *uh = udp_hdr(skb);
 337
 338                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 339                                 inet_proto_csum_replace16(&uh->check, skb,
 340                                                           addr, new_addr, true);
 341                                 if (!uh->check)
 342                                         uh->check = CSUM_MANGLED_0;
 343                         }
 344                 }
 345         } else if (l4_proto == NEXTHDR_ICMP) {
 346                 if (likely(transport_len >= sizeof(struct icmp6hdr)))
 347                         inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 348                                                   skb, addr, new_addr, true);
 349         }
 350 }
 351
 352 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
 353                            const __be32 mask[4], __be32 masked[4])
 354 {
 355         masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
 356         masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
 357         masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
 358         masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
 359 }
 360
 361 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 362                           __be32 addr[4], const __be32 new_addr[4],
 363                           bool recalculate_csum)
 364 {
 365         if (likely(recalculate_csum))
 366                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 367
 368         skb_clear_hash(skb);
 369         memcpy(addr, new_addr, sizeof(__be32[4]));
 370 }
 371
 372 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
 373 {
 374         /* Bits 21-24 are always unmasked, so this retains their values. */
 375         OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
 376         OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
 377         OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
 378 }
 379
 380 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
 381                        u8 mask)
 382 {
 383         new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
 384
 385         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 386         nh->ttl = new_ttl;
 387 }
 388
 389 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
 390                     const struct ovs_key_ipv4 *key,
 391                     const struct ovs_key_ipv4 *mask)
 392 {
 393         struct iphdr *nh;
 394         __be32 new_addr;
 395         int err;
 396
 397         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 398                                   sizeof(struct iphdr));
 399         if (unlikely(err))
 400                 return err;
 401
 402         nh = ip_hdr(skb);
 403
 404         /* Setting an IP addresses is typically only a side effect of
 405          * matching on them in the current userspace implementation, so it
 406          * makes sense to check if the value actually changed.
 407          */
 408         if (mask->ipv4_src) {
 409                 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
 410
 411                 if (unlikely(new_addr != nh->saddr)) {
 412                         set_ip_addr(skb, nh, &nh->saddr, new_addr);
 413                         flow_key->ipv4.addr.src = new_addr;
 414                 }
 415         }
 416         if (mask->ipv4_dst) {
 417                 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
 418
 419                 if (unlikely(new_addr != nh->daddr)) {
 420                         set_ip_addr(skb, nh, &nh->daddr, new_addr);
 421                         flow_key->ipv4.addr.dst = new_addr;
 422                 }
 423         }
 424         if (mask->ipv4_tos) {
 425                 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
 426                 flow_key->ip.tos = nh->tos;
 427         }
 428         if (mask->ipv4_ttl) {
 429                 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
 430                 flow_key->ip.ttl = nh->ttl;
 431         }
 432
 433         return 0;
 434 }
 435
 436 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
 437 {
 438         return !!(addr[0] | addr[1] | addr[2] | addr[3]);
 439 }
 440
 441 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
 442                     const struct ovs_key_ipv6 *key,
 443                     const struct ovs_key_ipv6 *mask)
 444 {
 445         struct ipv6hdr *nh;
 446         int err;
 447
 448         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 449                                   sizeof(struct ipv6hdr));
 450         if (unlikely(err))
 451                 return err;
 452
 453         nh = ipv6_hdr(skb);
 454
 455         /* Setting an IP addresses is typically only a side effect of
 456          * matching on them in the current userspace implementation, so it
 457          * makes sense to check if the value actually changed.
 458          */
 459         if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
 460                 __be32 *saddr = (__be32 *)&nh->saddr;
 461                 __be32 masked[4];
 462
 463                 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
 464
 465                 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
 466                         set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
 467                                       true);
 468                         memcpy(&flow_key->ipv6.addr.src, masked,
 469                                sizeof(flow_key->ipv6.addr.src));
 470                 }
 471         }
 472         if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
 473                 unsigned int offset = 0;
 474                 int flags = IP6_FH_F_SKIP_RH;
 475                 bool recalc_csum = true;
 476                 __be32 *daddr = (__be32 *)&nh->daddr;
 477                 __be32 masked[4];
 478
 479                 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
 480
 481                 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
 482                         if (ipv6_ext_hdr(nh->nexthdr))
 483                                 recalc_csum = (ipv6_find_hdr(skb, &offset,
 484                                                              NEXTHDR_ROUTING,
 485                                                              NULL, &flags)
 486                                                != NEXTHDR_ROUTING);
 487
 488                         set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
 489                                       recalc_csum);
 490                         memcpy(&flow_key->ipv6.addr.dst, masked,
 491                                sizeof(flow_key->ipv6.addr.dst));
 492                 }
 493         }
 494         if (mask->ipv6_tclass) {
 495                 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
 496                 flow_key->ip.tos = ipv6_get_dsfield(nh);
 497         }
 498         if (mask->ipv6_label) {
 499                 set_ipv6_fl(nh, ntohl(key->ipv6_label),
 500                             ntohl(mask->ipv6_label));
 501                 flow_key->ipv6.label =
 502                     *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 503         }
 504         if (mask->ipv6_hlimit) {
 505                 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
 506                                mask->ipv6_hlimit);
 507                 flow_key->ip.ttl = nh->hop_limit;
 508         }
 509         return 0;
 510 }
 511
 512 /* Must follow skb_ensure_writable() since that can move the skb data. */
 513 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 514                         __be16 new_port, __sum16 *check)
 515 {
 516         inet_proto_csum_replace2(check, skb, *port, new_port, false);
 517         *port = new_port;
 518 }
 519
 520 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 521                    const struct ovs_key_udp *key,
 522                    const struct ovs_key_udp *mask)
 523 {
 524         struct udphdr *uh;
 525         __be16 src, dst;
 526         int err;
 527
 528         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 529                                   sizeof(struct udphdr));
 530         if (unlikely(err))
 531                 return err;
 532
 533         uh = udp_hdr(skb);
 534         /* Either of the masks is non-zero, so do not bother checking them. */
 535         src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
 536         dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
 537
 538         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 539                 if (likely(src != uh->source)) {
 540                         set_tp_port(skb, &uh->source, src, &uh->check);
 541                         flow_key->tp.src = src;
 542                 }
 543                 if (likely(dst != uh->dest)) {
 544                         set_tp_port(skb, &uh->dest, dst, &uh->check);
 545                         flow_key->tp.dst = dst;
 546                 }
 547
 548                 if (unlikely(!uh->check))
 549                         uh->check = CSUM_MANGLED_0;
 550         } else {
 551                 uh->source = src;
 552                 uh->dest = dst;
 553                 flow_key->tp.src = src;
 554                 flow_key->tp.dst = dst;
 555         }
 556
 557         skb_clear_hash(skb);
 558
 559         return 0;
 560 }
 561
 562 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 563                    const struct ovs_key_tcp *key,
 564                    const struct ovs_key_tcp *mask)
 565 {
 566         struct tcphdr *th;
 567         __be16 src, dst;
 568         int err;
 569
 570         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 571                                   sizeof(struct tcphdr));
 572         if (unlikely(err))
 573                 return err;
 574
 575         th = tcp_hdr(skb);
 576         src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
 577         if (likely(src != th->source)) {
 578                 set_tp_port(skb, &th->source, src, &th->check);
 579                 flow_key->tp.src = src;
 580         }
 581         dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
 582         if (likely(dst != th->dest)) {
 583                 set_tp_port(skb, &th->dest, dst, &th->check);
 584                 flow_key->tp.dst = dst;
 585         }
 586         skb_clear_hash(skb);
 587
 588         return 0;
 589 }
 590
 591 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 592                     const struct ovs_key_sctp *key,
 593                     const struct ovs_key_sctp *mask)
 594 {
 595         unsigned int sctphoff = skb_transport_offset(skb);
 596         struct sctphdr *sh;
 597         __le32 old_correct_csum, new_csum, old_csum;
 598         int err;
 599
 600         err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
 601         if (unlikely(err))
 602                 return err;
 603
 604         sh = sctp_hdr(skb);
 605         old_csum = sh->checksum;
 606         old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 607
 608         sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
 609         sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
 610
 611         new_csum = sctp_compute_cksum(skb, sctphoff);
 612
 613         /* Carry any checksum errors through. */
 614         sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 615
 616         skb_clear_hash(skb);
 617         flow_key->tp.src = sh->source;
 618         flow_key->tp.dst = sh->dest;
 619
 620         return 0;
 621 }
 622
 623 static int ovs_vport_output(OVS_VPORT_OUTPUT_PARAMS)
 624 {
 625         struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
 626         struct vport *vport = data->vport;
 627
 628         if (skb_cow_head(skb, data->l2_len) < 0) {
 629                 kfree_skb(skb);
 630                 return -ENOMEM;
 631         }
 632
 633         __skb_dst_copy(skb, data->dst);
 634         *OVS_GSO_CB(skb) = data->cb;
 635         ovs_skb_set_inner_protocol(skb, data->inner_protocol);
 636         skb->vlan_tci = data->vlan_tci;
 637         skb->vlan_proto = data->vlan_proto;
 638
 639         /* Reconstruct the MAC header.  */
 640         skb_push(skb, data->l2_len);
 641         memcpy(skb->data, &data->l2_data, data->l2_len);
 642         skb_postpush_rcsum(skb, skb->data, data->l2_len);
 643         skb_reset_mac_header(skb);
 644
 645         ovs_vport_send(vport, skb);
 646         return 0;
 647 }
 648
 649 static unsigned int
 650 ovs_dst_get_mtu(const struct dst_entry *dst)
 651 {
 652         return dst->dev->mtu;
 653 }
 654
 655 static struct dst_ops ovs_dst_ops = {
 656         .family = AF_UNSPEC,
 657         .mtu = ovs_dst_get_mtu,
 658 };
 659
 660 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
 661  * ovs_vport_output(), which is called once per fragmented packet.
 662  */
 663 static void prepare_frag(struct vport *vport, struct sk_buff *skb)
 664 {
 665         unsigned int hlen = skb_network_offset(skb);
 666         struct ovs_frag_data *data;
 667
 668         data = this_cpu_ptr(&ovs_frag_data_storage);
 669         data->dst = (unsigned long) skb_dst(skb);
 670         data->vport = vport;
 671         data->cb = *OVS_GSO_CB(skb);
 672         data->inner_protocol = ovs_skb_get_inner_protocol(skb);
 673         data->vlan_tci = skb->vlan_tci;
 674         data->vlan_proto = skb->vlan_proto;
 675         data->l2_len = hlen;
 676         memcpy(&data->l2_data, skb->data, hlen);
 677
 678         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 679         skb_pull(skb, hlen);
 680 }
 681
 682 static void ovs_fragment(struct net *net, struct vport *vport,
 683                          struct sk_buff *skb, u16 mru, __be16 ethertype)
 684 {
 685         if (skb_network_offset(skb) > MAX_L2_LEN) {
 686                 OVS_NLERR(1, "L2 header too long to fragment");
 687                 goto err;
 688         }
 689
 690         if (ethertype == htons(ETH_P_IP)) {
 691                 struct dst_entry ovs_dst;
 692                 unsigned long orig_dst;
 693
 694                 prepare_frag(vport, skb);
 695                 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
 696                          DST_OBSOLETE_NONE, DST_NOCOUNT);
 697                 ovs_dst.dev = vport->dev;
 698
 699                 orig_dst = (unsigned long) skb_dst(skb);
 700                 skb_dst_set_noref(skb, &ovs_dst);
 701                 IPCB(skb)->frag_max_size = mru;
 702
 703                 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
 704                 refdst_drop(orig_dst);
 705         } else if (ethertype == htons(ETH_P_IPV6)) {
 706                 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
 707                 unsigned long orig_dst;
 708                 struct rt6_info ovs_rt;
 709
 710                 if (!v6ops) {
 711                         goto err;
 712                 }
 713
 714                 prepare_frag(vport, skb);
 715                 memset(&ovs_rt, 0, sizeof(ovs_rt));
 716                 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
 717                          DST_OBSOLETE_NONE, DST_NOCOUNT);
 718                 ovs_rt.dst.dev = vport->dev;
 719
 720                 orig_dst = (unsigned long) skb_dst(skb);
 721                 skb_dst_set_noref(skb, &ovs_rt.dst);
 722                 IP6CB(skb)->frag_max_size = mru;
 723
 724                 v6ops->fragment(skb->sk, skb, ovs_vport_output);
 725                 refdst_drop(orig_dst);
 726         } else {
 727                 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
 728                           ovs_vport_name(vport), ntohs(ethertype), mru,
 729                           vport->dev->mtu);
 730                 goto err;
 731         }
 732
 733         return;
 734 err:
 735         kfree_skb(skb);
 736 }
 737
 738 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
 739                       struct sw_flow_key *key)
 740 {
 741         struct vport *vport = ovs_vport_rcu(dp, out_port);
 742
 743         if (likely(vport)) {
 744                 u16 mru = OVS_CB(skb)->mru;
 745                 u32 cutlen = OVS_CB(skb)->cutlen;
 746
 747                 if (unlikely(cutlen > 0)) {
 748                         if (skb->len - cutlen > ETH_HLEN)
 749                                 pskb_trim(skb, skb->len - cutlen);
 750                         else
 751                                 pskb_trim(skb, ETH_HLEN);
 752                 }
 753
 754                 if (likely(!mru || (skb->len <= mru + ETH_HLEN))) {
 755                         ovs_vport_send(vport, skb);
 756                 } else if (mru <= vport->dev->mtu) {
 757                         struct net *net = ovs_dp_get_net(dp);
 758                         __be16 ethertype = key->eth.type;
 759
 760                         if (!is_flow_key_valid(key)) {
 761                                 if (eth_p_mpls(skb->protocol))
 762                                         ethertype = ovs_skb_get_inner_protocol(skb);
 763                                 else
 764                                         ethertype = vlan_get_protocol(skb);
 765                         }
 766
 767                         ovs_fragment(net, vport, skb, mru, ethertype);
 768                 } else {
 769                         OVS_NLERR(true, "Cannot fragment IP frames");
 770                         kfree_skb(skb);
 771                 }
 772         } else {
 773                 kfree_skb(skb);
 774         }
 775 }
 776
 777 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 778                             struct sw_flow_key *key, const struct nlattr *attr,
 779                             const struct nlattr *actions, int actions_len,
 780                             uint32_t cutlen)
 781 {
 782         struct dp_upcall_info upcall;
 783         const struct nlattr *a;
 784         int rem, err;
 785
 786         memset(&upcall, 0, sizeof(upcall));
 787         upcall.cmd = OVS_PACKET_CMD_ACTION;
 788         upcall.mru = OVS_CB(skb)->mru;
 789
 790         SKB_INIT_FILL_METADATA_DST(skb);
 791         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 792                  a = nla_next(a, &rem)) {
 793                 switch (nla_type(a)) {
 794                 case OVS_USERSPACE_ATTR_USERDATA:
 795                         upcall.userdata = a;
 796                         break;
 797
 798                 case OVS_USERSPACE_ATTR_PID:
 799                         upcall.portid = nla_get_u32(a);
 800                         break;
 801
 802                 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
 803                         /* Get out tunnel info. */
 804                         struct vport *vport;
 805
 806                         vport = ovs_vport_rcu(dp, nla_get_u32(a));
 807                         if (vport) {
 808                                 int err;
 809
 810                                 err = dev_fill_metadata_dst(vport->dev, skb);
 811                                 if (!err)
 812                                         upcall.egress_tun_info = skb_tunnel_info(skb);
 813                         }
 814
 815                         break;
 816                 }
 817
 818                 case OVS_USERSPACE_ATTR_ACTIONS: {
 819                         /* Include actions. */
 820                         upcall.actions = actions;
 821                         upcall.actions_len = actions_len;
 822                         break;
 823                 }
 824
 825                 } /* End of switch. */
 826         }
 827
 828         err = ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
 829         SKB_RESTORE_FILL_METADATA_DST(skb);
 830         return err;
 831 }
 832
 833 static int sample(struct datapath *dp, struct sk_buff *skb,
 834                   struct sw_flow_key *key, const struct nlattr *attr,
 835                   const struct nlattr *actions, int actions_len)
 836 {
 837         const struct nlattr *acts_list = NULL;
 838         const struct nlattr *a;
 839         int rem;
 840         u32 cutlen = 0;
 841
 842         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 843                  a = nla_next(a, &rem)) {
 844                 u32 probability;
 845
 846                 switch (nla_type(a)) {
 847                 case OVS_SAMPLE_ATTR_PROBABILITY:
 848                         probability = nla_get_u32(a);
 849                         if (!probability || prandom_u32() > probability)
 850                                 return 0;
 851                         break;
 852
 853                 case OVS_SAMPLE_ATTR_ACTIONS:
 854                         acts_list = a;
 855                         break;
 856                 }
 857         }
 858
 859         rem = nla_len(acts_list);
 860         a = nla_data(acts_list);
 861
 862         /* Actions list is empty, do nothing */
 863         if (unlikely(!rem))
 864                 return 0;
 865
 866         /* The only known usage of sample action is having a single user-space
 867          * action, or having a truncate action followed by a single user-space
 868          * action. Treat this usage as a special case.
 869          * The output_userspace() should clone the skb to be sent to the
 870          * user space. This skb will be consumed by its caller.
 871          */
 872         if (unlikely(nla_type(a) == OVS_ACTION_ATTR_TRUNC)) {
 873                 struct ovs_action_trunc *trunc = nla_data(a);
 874
 875                 if (skb->len > trunc->max_len)
 876                         cutlen = skb->len - trunc->max_len;
 877
 878                 a = nla_next(a, &rem);
 879         }
 880
 881         if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
 882                    nla_is_last(a, rem)))
 883                 return output_userspace(dp, skb, key, a, actions,
 884                                         actions_len, cutlen);
 885
 886         skb = skb_clone(skb, GFP_ATOMIC);
 887         if (!skb)
 888                 /* Skip the sample action when out of memory. */
 889                 return 0;
 890
 891         if (!add_deferred_actions(skb, key, a)) {
 892                 if (net_ratelimit())
 893                         pr_warn("%s: deferred actions limit reached, dropping sample action\n",
 894                                 ovs_dp_name(dp));
 895
 896                 kfree_skb(skb);
 897         }
 898         return 0;
 899 }
 900
 901 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
 902                          const struct nlattr *attr)
 903 {
 904         struct ovs_action_hash *hash_act = nla_data(attr);
 905         u32 hash = 0;
 906
 907         /* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
 908         hash = skb_get_hash(skb);
 909         hash = jhash_1word(hash, hash_act->hash_basis);
 910         if (!hash)
 911                 hash = 0x1;
 912
 913         key->ovs_flow_hash = hash;
 914 }
 915
 916 static int execute_set_action(struct sk_buff *skb,
 917                               struct sw_flow_key *flow_key,
 918                               const struct nlattr *a)
 919 {
 920         /* Only tunnel set execution is supported without a mask. */
 921         if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
 922                 struct ovs_tunnel_info *tun = nla_data(a);
 923
 924                 ovs_skb_dst_drop(skb);
 925                 ovs_dst_hold((struct dst_entry *)tun->tun_dst);
 926                 ovs_skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
 927                 return 0;
 928         }
 929
 930         return -EINVAL;
 931 }
 932
 933 /* Mask is at the midpoint of the data. */
 934 #define get_mask(a, type) ((const type)nla_data(a) + 1)
 935
 936 static int execute_masked_set_action(struct sk_buff *skb,
 937                                      struct sw_flow_key *flow_key,
 938                                      const struct nlattr *a)
 939 {
 940         int err = 0;
 941
 942         switch (nla_type(a)) {
 943         case OVS_KEY_ATTR_PRIORITY:
 944                 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
 945                                *get_mask(a, u32 *));
 946                 flow_key->phy.priority = skb->priority;
 947                 break;
 948
 949         case OVS_KEY_ATTR_SKB_MARK:
 950                 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
 951                 flow_key->phy.skb_mark = skb->mark;
 952                 break;
 953
 954         case OVS_KEY_ATTR_TUNNEL_INFO:
 955                 /* Masked data not supported for tunnel. */
 956                 err = -EINVAL;
 957                 break;
 958
 959         case OVS_KEY_ATTR_ETHERNET:
 960                 err = set_eth_addr(skb, flow_key, nla_data(a),
 961                                    get_mask(a, struct ovs_key_ethernet *));
 962                 break;
 963
 964         case OVS_KEY_ATTR_IPV4:
 965                 err = set_ipv4(skb, flow_key, nla_data(a),
 966                                get_mask(a, struct ovs_key_ipv4 *));
 967                 break;
 968
 969         case OVS_KEY_ATTR_IPV6:
 970                 err = set_ipv6(skb, flow_key, nla_data(a),
 971                                get_mask(a, struct ovs_key_ipv6 *));
 972                 break;
 973
 974         case OVS_KEY_ATTR_TCP:
 975                 err = set_tcp(skb, flow_key, nla_data(a),
 976                               get_mask(a, struct ovs_key_tcp *));
 977                 break;
 978
 979         case OVS_KEY_ATTR_UDP:
 980                 err = set_udp(skb, flow_key, nla_data(a),
 981                               get_mask(a, struct ovs_key_udp *));
 982                 break;
 983
 984         case OVS_KEY_ATTR_SCTP:
 985                 err = set_sctp(skb, flow_key, nla_data(a),
 986                                get_mask(a, struct ovs_key_sctp *));
 987                 break;
 988
 989         case OVS_KEY_ATTR_MPLS:
 990                 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
 991                                                                     __be32 *));
 992                 break;
 993
 994         case OVS_KEY_ATTR_CT_STATE:
 995         case OVS_KEY_ATTR_CT_ZONE:
 996         case OVS_KEY_ATTR_CT_MARK:
 997         case OVS_KEY_ATTR_CT_LABELS:
 998                 err = -EINVAL;
 999                 break;
1000         }
1001
1002         return err;
1003 }
1004
1005 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1006                           struct sw_flow_key *key,
1007                           const struct nlattr *a, int rem)
1008 {
1009         struct deferred_action *da;
1010
1011         if (!is_flow_key_valid(key)) {
1012                 int err;
1013
1014                 err = ovs_flow_key_update(skb, key);
1015                 if (err)
1016                         return err;
1017         }
1018         BUG_ON(!is_flow_key_valid(key));
1019
1020         if (!nla_is_last(a, rem)) {
1021                 /* Recirc action is the not the last action
1022                  * of the action list, need to clone the skb.
1023                  */
1024                 skb = skb_clone(skb, GFP_ATOMIC);
1025
1026                 /* Skip the recirc action when out of memory, but
1027                  * continue on with the rest of the action list.
1028                  */
1029                 if (!skb)
1030                         return 0;
1031         }
1032
1033         da = add_deferred_actions(skb, key, NULL);
1034         if (da) {
1035                 da->pkt_key.recirc_id = nla_get_u32(a);
1036         } else {
1037                 kfree_skb(skb);
1038
1039                 if (net_ratelimit())
1040                         pr_warn("%s: deferred action limit reached, drop recirc action\n",
1041                                 ovs_dp_name(dp));
1042         }
1043
1044         return 0;
1045 }
1046
1047 /* Execute a list of actions against 'skb'. */
1048 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1049                               struct sw_flow_key *key,
1050                               const struct nlattr *attr, int len)
1051 {
1052         /* Every output action needs a separate clone of 'skb', but the common
1053          * case is just a single output action, so that doing a clone and
1054          * then freeing the original skbuff is wasteful.  So the following code
1055          * is slightly obscure just to avoid that.
1056          */
1057         int prev_port = -1;
1058         const struct nlattr *a;
1059         int rem;
1060
1061         for (a = attr, rem = len; rem > 0;
1062              a = nla_next(a, &rem)) {
1063                 int err = 0;
1064
1065                 if (unlikely(prev_port != -1)) {
1066                         struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
1067
1068                         if (out_skb)
1069                                 do_output(dp, out_skb, prev_port, key);
1070
1071                         OVS_CB(skb)->cutlen = 0;
1072                         prev_port = -1;
1073                 }
1074
1075                 switch (nla_type(a)) {
1076                 case OVS_ACTION_ATTR_OUTPUT:
1077                         prev_port = nla_get_u32(a);
1078                         break;
1079
1080                 case OVS_ACTION_ATTR_TRUNC: {
1081                         struct ovs_action_trunc *trunc = nla_data(a);
1082
1083                         if (skb->len > trunc->max_len)
1084                                 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1085                         break;
1086                 }
1087
1088                 case OVS_ACTION_ATTR_USERSPACE:
1089                         output_userspace(dp, skb, key, a, attr,
1090                                                      len, OVS_CB(skb)->cutlen);
1091                         OVS_CB(skb)->cutlen = 0;
1092                         break;
1093
1094                 case OVS_ACTION_ATTR_HASH:
1095                         execute_hash(skb, key, a);
1096                         break;
1097
1098                 case OVS_ACTION_ATTR_PUSH_MPLS:
1099                         err = push_mpls(skb, key, nla_data(a));
1100                         break;
1101
1102                 case OVS_ACTION_ATTR_POP_MPLS:
1103                         err = pop_mpls(skb, key, nla_get_be16(a));
1104                         break;
1105
1106                 case OVS_ACTION_ATTR_PUSH_VLAN:
1107                         err = push_vlan(skb, key, nla_data(a));
1108                         break;
1109
1110                 case OVS_ACTION_ATTR_POP_VLAN:
1111                         err = pop_vlan(skb, key);
1112                         break;
1113
1114                 case OVS_ACTION_ATTR_RECIRC:
1115                         err = execute_recirc(dp, skb, key, a, rem);
1116                         if (nla_is_last(a, rem)) {
1117                                 /* If this is the last action, the skb has
1118                                  * been consumed or freed.
1119                                  * Return immediately.
1120                                  */
1121                                 return err;
1122                         }
1123                         break;
1124
1125                 case OVS_ACTION_ATTR_SET:
1126                         err = execute_set_action(skb, key, nla_data(a));
1127                         break;
1128
1129                 case OVS_ACTION_ATTR_SET_MASKED:
1130                 case OVS_ACTION_ATTR_SET_TO_MASKED:
1131                         err = execute_masked_set_action(skb, key, nla_data(a));
1132                         break;
1133
1134                 case OVS_ACTION_ATTR_SAMPLE:
1135                         err = sample(dp, skb, key, a, attr, len);
1136                         break;
1137
1138                 case OVS_ACTION_ATTR_CT:
1139                         if (!is_flow_key_valid(key)) {
1140                                 err = ovs_flow_key_update(skb, key);
1141                                 if (err)
1142                                         return err;
1143                         }
1144
1145                         err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1146                                              nla_data(a));
1147
1148                         /* Hide stolen IP fragments from user space. */
1149                         if (err)
1150                                 return err == -EINPROGRESS ? 0 : err;
1151                         break;
1152                 }
1153
1154                 if (unlikely(err)) {
1155                         kfree_skb(skb);
1156                         return err;
1157                 }
1158         }
1159
1160         if (prev_port != -1)
1161                 do_output(dp, skb, prev_port, key);
1162         else
1163                 consume_skb(skb);
1164
1165         return 0;
1166 }
1167
1168 static void process_deferred_actions(struct datapath *dp)
1169 {
1170         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1171
1172         /* Do not touch the FIFO in case there is no deferred actions. */
1173         if (action_fifo_is_empty(fifo))
1174                 return;
1175
1176         /* Finishing executing all deferred actions. */
1177         do {
1178                 struct deferred_action *da = action_fifo_get(fifo);
1179                 struct sk_buff *skb = da->skb;
1180                 struct sw_flow_key *key = &da->pkt_key;
1181                 const struct nlattr *actions = da->actions;
1182
1183                 if (actions)
1184                         do_execute_actions(dp, skb, key, actions,
1185                                            nla_len(actions));
1186                 else
1187                         ovs_dp_process_packet(skb, key);
1188         } while (!action_fifo_is_empty(fifo));
1189
1190         /* Reset FIFO for the next packet.  */
1191         action_fifo_init(fifo);
1192 }
1193
1194 /* Execute a list of actions against 'skb'. */
1195 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1196                         const struct sw_flow_actions *acts,
1197                         struct sw_flow_key *key)
1198 {
1199         int level = this_cpu_read(exec_actions_level);
1200         int err;
1201
1202         if (unlikely(level >= EXEC_ACTIONS_LEVEL_LIMIT)) {
1203                 if (net_ratelimit())
1204                         pr_warn("%s: packet loop detected, dropping.\n",
1205                                 ovs_dp_name(dp));
1206
1207                 kfree_skb(skb);
1208                 return -ELOOP;
1209         }
1210
1211         this_cpu_inc(exec_actions_level);
1212         err = do_execute_actions(dp, skb, key,
1213                                  acts->actions, acts->actions_len);
1214
1215         if (!level)
1216                 process_deferred_actions(dp);
1217
1218         this_cpu_dec(exec_actions_level);
1219
1220         /* This return status currently does not reflect the errors
1221          * encounted during deferred actions execution. Probably needs to
1222          * be fixed in the future.
1223          */
1224         return err;
1225 }
1226
1227 int action_fifos_init(void)
1228 {
1229         action_fifos = alloc_percpu(struct action_fifo);
1230         if (!action_fifos)
1231                 return -ENOMEM;
1232
1233         return 0;
1234 }
1235
1236 void action_fifos_exit(void)
1237 {
1238         free_percpu(action_fifos);
1239 }