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