datapath/actions.c

   1 /*
   2  * Copyright (c) 2007-2015 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/skbuff.h>
  22 #include <linux/in.h>
  23 #include <linux/ip.h>
  24 #include <linux/openvswitch.h>
  25 #include <linux/netfilter_ipv6.h>
  26 #include <linux/sctp.h>
  27 #include <linux/tcp.h>
  28 #include <linux/udp.h>
  29 #include <linux/in6.h>
  30 #include <linux/if_arp.h>
  31 #include <linux/if_vlan.h>
  32
  33 #include <net/dst.h>
  34 #include <net/ip.h>
  35 #include <net/ipv6.h>
  36 #include <net/checksum.h>
  37 #include <net/dsfield.h>
  38 #include <net/mpls.h>
  39 #include <net/sctp/checksum.h>
  40
  41 #include "datapath.h"
  42 #include "conntrack.h"
  43 #include "gso.h"
  44 #include "vlan.h"
  45 #include "vport.h"
  46
  47 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
  48                               struct sw_flow_key *key,
  49                               const struct nlattr *attr, int len);
  50
  51 struct deferred_action {
  52         struct sk_buff *skb;
  53         const struct nlattr *actions;
  54
  55         /* Store pkt_key clone when creating deferred action. */
  56         struct sw_flow_key pkt_key;
  57 };
  58
  59 #define MAX_L2_LEN      (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
  60 struct ovs_frag_data {
  61         unsigned long dst;
  62         struct vport *vport;
  63         struct ovs_skb_cb cb;
  64         __be16 inner_protocol;
  65         __u16 vlan_tci;
  66         __be16 vlan_proto;
  67         unsigned int l2_len;
  68         u8 l2_data[MAX_L2_LEN];
  69 };
  70
  71 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
  72
  73 #define DEFERRED_ACTION_FIFO_SIZE 10
  74 struct action_fifo {
  75         int head;
  76         int tail;
  77         /* Deferred action fifo queue storage. */
  78         struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
  79 };
  80
  81 static struct action_fifo __percpu *action_fifos;
  82 #define EXEC_ACTIONS_LEVEL_LIMIT 4   /* limit used to detect packet
  83                                       * looping by the network stack
  84                                       */
  85 static DEFINE_PER_CPU(int, exec_actions_level);
  86
  87 static void action_fifo_init(struct action_fifo *fifo)
  88 {
  89         fifo->head = 0;
  90         fifo->tail = 0;
  91 }
  92
  93 static bool action_fifo_is_empty(const struct action_fifo *fifo)
  94 {
  95         return (fifo->head == fifo->tail);
  96 }
  97
  98 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
  99 {
 100         if (action_fifo_is_empty(fifo))
 101                 return NULL;
 102
 103         return &fifo->fifo[fifo->tail++];
 104 }
 105
 106 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
 107 {
 108         if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
 109                 return NULL;
 110
 111         return &fifo->fifo[fifo->head++];
 112 }
 113
 114 /* Return queue entry if fifo is not full */
 115 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
 116                                                     const struct sw_flow_key *key,
 117                                                     const struct nlattr *attr)
 118 {
 119         struct action_fifo *fifo;
 120         struct deferred_action *da;
 121
 122         fifo = this_cpu_ptr(action_fifos);
 123         da = action_fifo_put(fifo);
 124         if (da) {
 125                 da->skb = skb;
 126                 da->actions = attr;
 127                 da->pkt_key = *key;
 128         }
 129
 130         return da;
 131 }
 132
 133 static void invalidate_flow_key(struct sw_flow_key *key)
 134 {
 135         key->eth.type = htons(0);
 136 }
 137
 138 static bool is_flow_key_valid(const struct sw_flow_key *key)
 139 {
 140         return !!key->eth.type;
 141 }
 142
 143 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 144                      const struct ovs_action_push_mpls *mpls)
 145 {
 146         __be32 *new_mpls_lse;
 147         struct ethhdr *hdr;
 148
 149         /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
 150         if (skb_encapsulation(skb))
 151                 return -ENOTSUPP;
 152
 153         if (skb_cow_head(skb, MPLS_HLEN) < 0)
 154                 return -ENOMEM;
 155
 156         skb_push(skb, MPLS_HLEN);
 157         memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
 158                 skb->mac_len);
 159         skb_reset_mac_header(skb);
 160
 161         new_mpls_lse = (__be32 *)skb_mpls_header(skb);
 162         *new_mpls_lse = mpls->mpls_lse;
 163
 164         if (skb->ip_summed == CHECKSUM_COMPLETE)
 165                 skb->csum = csum_add(skb->csum, csum_partial(new_mpls_lse,
 166                                                              MPLS_HLEN, 0));
 167
 168         hdr = eth_hdr(skb);
 169         hdr->h_proto = mpls->mpls_ethertype;
 170         if (!ovs_skb_get_inner_protocol(skb))
 171                 ovs_skb_set_inner_protocol(skb, skb->protocol);
 172         skb->protocol = mpls->mpls_ethertype;
 173
 174         invalidate_flow_key(key);
 175         return 0;
 176 }
 177
 178 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 179                     const __be16 ethertype)
 180 {
 181         struct ethhdr *hdr;
 182         int err;
 183
 184         err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
 185         if (unlikely(err))
 186                 return err;
 187
 188         skb_postpull_rcsum(skb, skb_mpls_header(skb), MPLS_HLEN);
 189
 190         memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
 191                 skb->mac_len);
 192
 193         __skb_pull(skb, MPLS_HLEN);
 194         skb_reset_mac_header(skb);
 195
 196         /* skb_mpls_header() is used to locate the ethertype
 197          * field correctly in the presence of VLAN tags.
 198          */
 199         hdr = (struct ethhdr *)(skb_mpls_header(skb) - ETH_HLEN);
 200         hdr->h_proto = ethertype;
 201         if (eth_p_mpls(skb->protocol))
 202                 skb->protocol = ethertype;
 203
 204         invalidate_flow_key(key);
 205         return 0;
 206 }
 207
 208 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
 209                     const __be32 *mpls_lse, const __be32 *mask)
 210 {
 211         __be32 *stack;
 212         __be32 lse;
 213         int err;
 214
 215         err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
 216         if (unlikely(err))
 217                 return err;
 218
 219         stack = (__be32 *)skb_mpls_header(skb);
 220         lse = OVS_MASKED(*stack, *mpls_lse, *mask);
 221         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 222                 __be32 diff[] = { ~(*stack), lse };
 223
 224                 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
 225                                           ~skb->csum);
 226         }
 227
 228         *stack = lse;
 229         flow_key->mpls.top_lse = lse;
 230         return 0;
 231 }
 232
 233 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 234 {
 235         int err;
 236
 237         err = skb_vlan_pop(skb);
 238         if (skb_vlan_tag_present(skb))
 239                 invalidate_flow_key(key);
 240         else
 241                 key->eth.tci = 0;
 242         return err;
 243 }
 244
 245 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
 246                      const struct ovs_action_push_vlan *vlan)
 247 {
 248         if (skb_vlan_tag_present(skb))
 249                 invalidate_flow_key(key);
 250         else
 251                 key->eth.tci = vlan->vlan_tci;
 252         return skb_vlan_push(skb, vlan->vlan_tpid,
 253                              ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
 254 }
 255
 256 /* 'src' is already properly masked. */
 257 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
 258 {
 259         u16 *dst = (u16 *)dst_;
 260         const u16 *src = (const u16 *)src_;
 261         const u16 *mask = (const u16 *)mask_;
 262
 263         OVS_SET_MASKED(dst[0], src[0], mask[0]);
 264         OVS_SET_MASKED(dst[1], src[1], mask[1]);
 265         OVS_SET_MASKED(dst[2], src[2], mask[2]);
 266 }
 267
 268 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
 269                         const struct ovs_key_ethernet *key,
 270                         const struct ovs_key_ethernet *mask)
 271 {
 272         int err;
 273
 274         err = skb_ensure_writable(skb, ETH_HLEN);
 275         if (unlikely(err))
 276                 return err;
 277
 278         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 279
 280         ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
 281                                mask->eth_src);
 282         ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
 283                                mask->eth_dst);
 284
 285         ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 286
 287         ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
 288         ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
 289         return 0;
 290 }
 291
 292 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
 293                                   __be32 addr, __be32 new_addr)
 294 {
 295         int transport_len = skb->len - skb_transport_offset(skb);
 296
 297         if (nh->frag_off & htons(IP_OFFSET))
 298                 return;
 299
 300         if (nh->protocol == IPPROTO_TCP) {
 301                 if (likely(transport_len >= sizeof(struct tcphdr)))
 302                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 303                                                  addr, new_addr, 1);
 304         } else if (nh->protocol == IPPROTO_UDP) {
 305                 if (likely(transport_len >= sizeof(struct udphdr))) {
 306                         struct udphdr *uh = udp_hdr(skb);
 307
 308                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 309                                 inet_proto_csum_replace4(&uh->check, skb,
 310                                                          addr, new_addr, 1);
 311                                 if (!uh->check)
 312                                         uh->check = CSUM_MANGLED_0;
 313                         }
 314                 }
 315         }
 316
 317 }
 318
 319 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 320                         __be32 *addr, __be32 new_addr)
 321 {
 322         update_ip_l4_checksum(skb, nh, *addr, new_addr);
 323         csum_replace4(&nh->check, *addr, new_addr);
 324         skb_clear_hash(skb);
 325         *addr = new_addr;
 326 }
 327
 328 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 329                                  __be32 addr[4], const __be32 new_addr[4])
 330 {
 331         int transport_len = skb->len - skb_transport_offset(skb);
 332
 333         if (l4_proto == NEXTHDR_TCP) {
 334                 if (likely(transport_len >= sizeof(struct tcphdr)))
 335                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 336                                                   addr, new_addr, 1);
 337         } else if (l4_proto == NEXTHDR_UDP) {
 338                 if (likely(transport_len >= sizeof(struct udphdr))) {
 339                         struct udphdr *uh = udp_hdr(skb);
 340
 341                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 342                                 inet_proto_csum_replace16(&uh->check, skb,
 343                                                           addr, new_addr, 1);
 344                                 if (!uh->check)
 345                                         uh->check = CSUM_MANGLED_0;
 346                         }
 347                 }
 348         } else if (l4_proto == NEXTHDR_ICMP) {
 349                 if (likely(transport_len >= sizeof(struct icmp6hdr)))
 350                         inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 351                                                   skb, addr, new_addr, 1);
 352         }
 353 }
 354
 355 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
 356                            const __be32 mask[4], __be32 masked[4])
 357 {
 358         masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
 359         masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
 360         masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
 361         masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
 362 }
 363
 364 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 365                           __be32 addr[4], const __be32 new_addr[4],
 366                           bool recalculate_csum)
 367 {
 368         if (likely(recalculate_csum))
 369                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 370
 371         skb_clear_hash(skb);
 372         memcpy(addr, new_addr, sizeof(__be32[4]));
 373 }
 374
 375 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
 376 {
 377         /* Bits 21-24 are always unmasked, so this retains their values. */
 378         OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
 379         OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
 380         OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
 381 }
 382
 383 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
 384                        u8 mask)
 385 {
 386         new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
 387
 388         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 389         nh->ttl = new_ttl;
 390 }
 391
 392 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
 393                     const struct ovs_key_ipv4 *key,
 394                     const struct ovs_key_ipv4 *mask)
 395 {
 396         struct iphdr *nh;
 397         __be32 new_addr;
 398         int err;
 399
 400         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 401                                   sizeof(struct iphdr));
 402         if (unlikely(err))
 403                 return err;
 404
 405         nh = ip_hdr(skb);
 406
 407         /* Setting an IP addresses is typically only a side effect of
 408          * matching on them in the current userspace implementation, so it
 409          * makes sense to check if the value actually changed.
 410          */
 411         if (mask->ipv4_src) {
 412                 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
 413
 414                 if (unlikely(new_addr != nh->saddr)) {
 415                         set_ip_addr(skb, nh, &nh->saddr, new_addr);
 416                         flow_key->ipv4.addr.src = new_addr;
 417                 }
 418         }
 419         if (mask->ipv4_dst) {
 420                 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
 421
 422                 if (unlikely(new_addr != nh->daddr)) {
 423                         set_ip_addr(skb, nh, &nh->daddr, new_addr);
 424                         flow_key->ipv4.addr.dst = new_addr;
 425                 }
 426         }
 427         if (mask->ipv4_tos) {
 428                 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
 429                 flow_key->ip.tos = nh->tos;
 430         }
 431         if (mask->ipv4_ttl) {
 432                 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
 433                 flow_key->ip.ttl = nh->ttl;
 434         }
 435
 436         return 0;
 437 }
 438
 439 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
 440 {
 441         return !!(addr[0] | addr[1] | addr[2] | addr[3]);
 442 }
 443
 444 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
 445                     const struct ovs_key_ipv6 *key,
 446                     const struct ovs_key_ipv6 *mask)
 447 {
 448         struct ipv6hdr *nh;
 449         int err;
 450
 451         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 452                                   sizeof(struct ipv6hdr));
 453         if (unlikely(err))
 454                 return err;
 455
 456         nh = ipv6_hdr(skb);
 457
 458         /* Setting an IP addresses is typically only a side effect of
 459          * matching on them in the current userspace implementation, so it
 460          * makes sense to check if the value actually changed.
 461          */
 462         if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
 463                 __be32 *saddr = (__be32 *)&nh->saddr;
 464                 __be32 masked[4];
 465
 466                 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
 467
 468                 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
 469                         set_ipv6_addr(skb, key->ipv6_proto, saddr, masked,
 470                                       true);
 471                         memcpy(&flow_key->ipv6.addr.src, masked,
 472                                sizeof(flow_key->ipv6.addr.src));
 473                 }
 474         }
 475         if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
 476                 unsigned int offset = 0;
 477                 int flags = IP6_FH_F_SKIP_RH;
 478                 bool recalc_csum = true;
 479                 __be32 *daddr = (__be32 *)&nh->daddr;
 480                 __be32 masked[4];
 481
 482                 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
 483
 484                 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
 485                         if (ipv6_ext_hdr(nh->nexthdr))
 486                                 recalc_csum = (ipv6_find_hdr(skb, &offset,
 487                                                              NEXTHDR_ROUTING,
 488                                                              NULL, &flags)
 489                                                != NEXTHDR_ROUTING);
 490
 491                         set_ipv6_addr(skb, key->ipv6_proto, daddr, masked,
 492                                       recalc_csum);
 493                         memcpy(&flow_key->ipv6.addr.dst, masked,
 494                                sizeof(flow_key->ipv6.addr.dst));
 495                 }
 496         }
 497         if (mask->ipv6_tclass) {
 498                 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
 499                 flow_key->ip.tos = ipv6_get_dsfield(nh);
 500         }
 501         if (mask->ipv6_label) {
 502                 set_ipv6_fl(nh, ntohl(key->ipv6_label),
 503                             ntohl(mask->ipv6_label));
 504                 flow_key->ipv6.label =
 505                     *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 506         }
 507         if (mask->ipv6_hlimit) {
 508                 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
 509                                mask->ipv6_hlimit);
 510                 flow_key->ip.ttl = nh->hop_limit;
 511         }
 512         return 0;
 513 }
 514
 515 /* Must follow skb_ensure_writable() since that can move the skb data. */
 516 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 517                         __be16 new_port, __sum16 *check)
 518 {
 519         inet_proto_csum_replace2(check, skb, *port, new_port, 0);
 520         *port = new_port;
 521 }
 522
 523 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 524                    const struct ovs_key_udp *key,
 525                    const struct ovs_key_udp *mask)
 526 {
 527         struct udphdr *uh;
 528         __be16 src, dst;
 529         int err;
 530
 531         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 532                                   sizeof(struct udphdr));
 533         if (unlikely(err))
 534                 return err;
 535
 536         uh = udp_hdr(skb);
 537         /* Either of the masks is non-zero, so do not bother checking them. */
 538         src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
 539         dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
 540
 541         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 542                 if (likely(src != uh->source)) {
 543                         set_tp_port(skb, &uh->source, src, &uh->check);
 544                         flow_key->tp.src = src;
 545                 }
 546                 if (likely(dst != uh->dest)) {
 547                         set_tp_port(skb, &uh->dest, dst, &uh->check);
 548                         flow_key->tp.dst = dst;
 549                 }
 550
 551                 if (unlikely(!uh->check))
 552                         uh->check = CSUM_MANGLED_0;
 553         } else {
 554                 uh->source = src;
 555                 uh->dest = dst;
 556                 flow_key->tp.src = src;
 557                 flow_key->tp.dst = dst;
 558         }
 559
 560         skb_clear_hash(skb);
 561
 562         return 0;
 563 }
 564
 565 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 566                    const struct ovs_key_tcp *key,
 567                    const struct ovs_key_tcp *mask)
 568 {
 569         struct tcphdr *th;
 570         __be16 src, dst;
 571         int err;
 572
 573         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 574                                   sizeof(struct tcphdr));
 575         if (unlikely(err))
 576                 return err;
 577
 578         th = tcp_hdr(skb);
 579         src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
 580         if (likely(src != th->source)) {
 581                 set_tp_port(skb, &th->source, src, &th->check);
 582                 flow_key->tp.src = src;
 583         }
 584         dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
 585         if (likely(dst != th->dest)) {
 586                 set_tp_port(skb, &th->dest, dst, &th->check);
 587                 flow_key->tp.dst = dst;
 588         }
 589         skb_clear_hash(skb);
 590
 591         return 0;
 592 }
 593
 594 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 595                     const struct ovs_key_sctp *key,
 596                     const struct ovs_key_sctp *mask)
 597 {
 598         unsigned int sctphoff = skb_transport_offset(skb);
 599         struct sctphdr *sh;
 600         __le32 old_correct_csum, new_csum, old_csum;
 601         int err;
 602
 603         err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
 604         if (unlikely(err))
 605                 return err;
 606
 607         sh = sctp_hdr(skb);
 608         old_csum = sh->checksum;
 609         old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 610
 611         sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
 612         sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
 613
 614         new_csum = sctp_compute_cksum(skb, sctphoff);
 615
 616         /* Carry any checksum errors through. */
 617         sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 618
 619         skb_clear_hash(skb);
 620         flow_key->tp.src = sh->source;
 621         flow_key->tp.dst = sh->dest;
 622
 623         return 0;
 624 }
 625
 626 #if LINUX_VERSION_CODE > KERNEL_VERSION(3,9,0)
 627 static int ovs_vport_output(OVS_VPORT_OUTPUT_PARAMS)
 628 {
 629         struct ovs_frag_data *data = get_pcpu_ptr(ovs_frag_data_storage);
 630         struct vport *vport = data->vport;
 631
 632         if (skb_cow_head(skb, data->l2_len) < 0) {
 633                 kfree_skb(skb);
 634                 return -ENOMEM;
 635         }
 636
 637         __skb_dst_copy(skb, data->dst);
 638         *OVS_CB(skb) = data->cb;
 639         ovs_skb_set_inner_protocol(skb, data->inner_protocol);
 640         skb->vlan_tci = data->vlan_tci;
 641         skb->vlan_proto = data->vlan_proto;
 642
 643         /* Reconstruct the MAC header.  */
 644         skb_push(skb, data->l2_len);
 645         memcpy(skb->data, &data->l2_data, data->l2_len);
 646         ovs_skb_postpush_rcsum(skb, skb->data, data->l2_len);
 647         skb_reset_mac_header(skb);
 648
 649         ovs_vport_send(vport, skb);
 650         return 0;
 651 }
 652
 653 static unsigned int
 654 ovs_dst_get_mtu(const struct dst_entry *dst)
 655 {
 656         return dst->dev->mtu;
 657 }
 658
 659 static struct dst_ops ovs_dst_ops = {
 660         .family = AF_UNSPEC,
 661         .mtu = ovs_dst_get_mtu,
 662 };
 663
 664 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
 665  * ovs_vport_output(), which is called once per fragmented packet.
 666  */
 667 static void prepare_frag(struct vport *vport, struct sk_buff *skb)
 668 {
 669         unsigned int hlen = skb_network_offset(skb);
 670         struct ovs_frag_data *data;
 671
 672         data = get_pcpu_ptr(ovs_frag_data_storage);
 673         data->dst = (unsigned long) skb_dst(skb);
 674         data->vport = vport;
 675         data->cb = *OVS_CB(skb);
 676         data->inner_protocol = ovs_skb_get_inner_protocol(skb);
 677         data->vlan_tci = skb->vlan_tci;
 678         data->vlan_proto = skb->vlan_proto;
 679         data->l2_len = hlen;
 680         memcpy(&data->l2_data, skb->data, hlen);
 681
 682         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 683         skb_pull(skb, hlen);
 684 }
 685
 686 static void ovs_fragment(struct vport *vport, struct sk_buff *skb, u16 mru,
 687                          __be16 ethertype)
 688 {
 689         if (skb_network_offset(skb) > MAX_L2_LEN) {
 690                 OVS_NLERR(1, "L2 header too long to fragment");
 691                 return;
 692         }
 693
 694         if (ethertype == htons(ETH_P_IP)) {
 695                 struct dst_entry ovs_dst;
 696                 unsigned long orig_dst;
 697
 698                 prepare_frag(vport, skb);
 699                 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
 700                          DST_OBSOLETE_NONE, DST_NOCOUNT);
 701                 ovs_dst.dev = vport->dev;
 702
 703                 orig_dst = (unsigned long) skb_dst(skb);
 704                 skb_dst_set_noref(skb, &ovs_dst);
 705                 IPCB(skb)->frag_max_size = mru;
 706
 707                 ip_do_fragment(skb->sk, skb, ovs_vport_output);
 708                 refdst_drop(orig_dst);
 709         } else if (ethertype == htons(ETH_P_IPV6)) {
 710                 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
 711                 unsigned long orig_dst;
 712                 struct rt6_info ovs_rt;
 713
 714                 if (!v6ops) {
 715                         kfree_skb(skb);
 716                         return;
 717                 }
 718
 719                 prepare_frag(vport, skb);
 720                 memset(&ovs_rt, 0, sizeof(ovs_rt));
 721                 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
 722                          DST_OBSOLETE_NONE, DST_NOCOUNT);
 723                 ovs_rt.dst.dev = vport->dev;
 724
 725                 orig_dst = (unsigned long) skb_dst(skb);
 726                 skb_dst_set_noref(skb, &ovs_rt.dst);
 727                 IP6CB(skb)->frag_max_size = mru;
 728
 729                 v6ops->fragment(skb->sk, skb, ovs_vport_output);
 730                 refdst_drop(orig_dst);
 731         } else {
 732                 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
 733                           ovs_vport_name(vport), ntohs(ethertype), mru,
 734                           vport->dev->mtu);
 735                 kfree_skb(skb);
 736         }
 737 }
 738 #else /* <= 3.9 */
 739 static void ovs_fragment(struct vport *vport, struct sk_buff *skb, u16 mru,
 740                          __be16 ethertype)
 741 {
 742         WARN_ONCE(1, "Fragment unavailable ->%s: eth=%04x, MRU=%d, MTU=%d.",
 743                   ovs_vport_name(vport), ntohs(ethertype), mru,
 744                   vport->dev->mtu);
 745         kfree_skb(skb);
 746 }
 747 #endif
 748
 749 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
 750                       struct sw_flow_key *key)
 751 {
 752         struct vport *vport = ovs_vport_rcu(dp, out_port);
 753
 754         if (likely(vport)) {
 755                 u16 mru = OVS_CB(skb)->mru;
 756
 757                 if (likely(!mru || (skb->len <= mru + ETH_HLEN))) {
 758                         ovs_vport_send(vport, skb);
 759                 } else if (mru <= vport->dev->mtu) {
 760                         __be16 ethertype = key->eth.type;
 761
 762                         if (!is_flow_key_valid(key)) {
 763                                 if (eth_p_mpls(skb->protocol))
 764                                         ethertype = ovs_skb_get_inner_protocol(skb);
 765                                 else
 766                                         ethertype = vlan_get_protocol(skb);
 767                         }
 768
 769                         ovs_fragment(vport, skb, mru, ethertype);
 770                 } else {
 771                         OVS_NLERR(true, "Cannot fragment IP frames");
 772                         kfree_skb(skb);
 773                 }
 774         } else {
 775                 kfree_skb(skb);
 776         }
 777 }
 778 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 779                             struct sw_flow_key *key, const struct nlattr *attr,
 780                             const struct nlattr *actions, int actions_len)
 781 {
 782         struct ip_tunnel_info info;
 783         struct dp_upcall_info upcall;
 784         const struct nlattr *a;
 785         int rem;
 786
 787         memset(&upcall, 0, sizeof(upcall));
 788         upcall.cmd = OVS_PACKET_CMD_ACTION;
 789         upcall.mru = OVS_CB(skb)->mru;
 790
 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                                 upcall.egress_tun_info = &info;
 811                                 err = ovs_vport_get_egress_tun_info(vport, skb,
 812                                                                     &upcall);
 813                                 if (err)
 814                                         upcall.egress_tun_info = NULL;
 815                         }
 816
 817                         break;
 818                 }
 819
 820                 case OVS_USERSPACE_ATTR_ACTIONS: {
 821                         /* Include actions. */
 822                         upcall.actions = actions;
 823                         upcall.actions_len = actions_len;
 824                         break;
 825                 }
 826
 827                 } /* End of switch. */
 828         }
 829
 830         return ovs_dp_upcall(dp, skb, key, &upcall);
 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
 841         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 842                  a = nla_next(a, &rem)) {
 843                 u32 probability;
 844
 845                 switch (nla_type(a)) {
 846                 case OVS_SAMPLE_ATTR_PROBABILITY:
 847                         probability = nla_get_u32(a);
 848                         if (!probability || prandom_u32() > probability)
 849                                 return 0;
 850                         break;
 851
 852                 case OVS_SAMPLE_ATTR_ACTIONS:
 853                         acts_list = a;
 854                         break;
 855                 }
 856         }
 857
 858         rem = nla_len(acts_list);
 859         a = nla_data(acts_list);
 860
 861         /* Actions list is empty, do nothing */
 862         if (unlikely(!rem))
 863                 return 0;
 864
 865         /* The only known usage of sample action is having a single user-space
 866          * action. Treat this usage as a special case.
 867          * The output_userspace() should clone the skb to be sent to the
 868          * user space. This skb will be consumed by its caller.
 869          */
 870         if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
 871                    nla_is_last(a, rem)))
 872                 return output_userspace(dp, skb, key, a, actions, actions_len);
 873
 874         skb = skb_clone(skb, GFP_ATOMIC);
 875         if (!skb)
 876                 /* Skip the sample action when out of memory. */
 877                 return 0;
 878
 879         if (!add_deferred_actions(skb, key, a)) {
 880                 if (net_ratelimit())
 881                         pr_warn("%s: deferred actions limit reached, dropping sample action\n",
 882                                 ovs_dp_name(dp));
 883
 884                 kfree_skb(skb);
 885         }
 886         return 0;
 887 }
 888
 889 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
 890                          const struct nlattr *attr)
 891 {
 892         struct ovs_action_hash *hash_act = nla_data(attr);
 893         u32 hash = 0;
 894
 895         /* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
 896         hash = skb_get_hash(skb);
 897         hash = jhash_1word(hash, hash_act->hash_basis);
 898         if (!hash)
 899                 hash = 0x1;
 900
 901         key->ovs_flow_hash = hash;
 902 }
 903
 904 static int execute_set_action(struct sk_buff *skb,
 905                               struct sw_flow_key *flow_key,
 906                               const struct nlattr *a)
 907 {
 908         /* Only tunnel set execution is supported without a mask. */
 909         if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
 910                 struct ovs_tunnel_info *tun = nla_data(a);
 911
 912                 ovs_skb_dst_drop(skb);
 913                 ovs_dst_hold((struct dst_entry *)tun->tun_dst);
 914                 ovs_skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
 915                 return 0;
 916         }
 917
 918         return -EINVAL;
 919 }
 920
 921 /* Mask is at the midpoint of the data. */
 922 #define get_mask(a, type) ((const type)nla_data(a) + 1)
 923
 924 static int execute_masked_set_action(struct sk_buff *skb,
 925                                      struct sw_flow_key *flow_key,
 926                                      const struct nlattr *a)
 927 {
 928         int err = 0;
 929
 930         switch (nla_type(a)) {
 931         case OVS_KEY_ATTR_PRIORITY:
 932                 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
 933                                *get_mask(a, u32 *));
 934                 flow_key->phy.priority = skb->priority;
 935                 break;
 936
 937         case OVS_KEY_ATTR_SKB_MARK:
 938                 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
 939                 flow_key->phy.skb_mark = skb->mark;
 940                 break;
 941
 942         case OVS_KEY_ATTR_TUNNEL_INFO:
 943                 /* Masked data not supported for tunnel. */
 944                 err = -EINVAL;
 945                 break;
 946
 947         case OVS_KEY_ATTR_ETHERNET:
 948                 err = set_eth_addr(skb, flow_key, nla_data(a),
 949                                    get_mask(a, struct ovs_key_ethernet *));
 950                 break;
 951
 952         case OVS_KEY_ATTR_IPV4:
 953                 err = set_ipv4(skb, flow_key, nla_data(a),
 954                                get_mask(a, struct ovs_key_ipv4 *));
 955                 break;
 956
 957         case OVS_KEY_ATTR_IPV6:
 958                 err = set_ipv6(skb, flow_key, nla_data(a),
 959                                get_mask(a, struct ovs_key_ipv6 *));
 960                 break;
 961
 962         case OVS_KEY_ATTR_TCP:
 963                 err = set_tcp(skb, flow_key, nla_data(a),
 964                               get_mask(a, struct ovs_key_tcp *));
 965                 break;
 966
 967         case OVS_KEY_ATTR_UDP:
 968                 err = set_udp(skb, flow_key, nla_data(a),
 969                               get_mask(a, struct ovs_key_udp *));
 970                 break;
 971
 972         case OVS_KEY_ATTR_SCTP:
 973                 err = set_sctp(skb, flow_key, nla_data(a),
 974                                get_mask(a, struct ovs_key_sctp *));
 975                 break;
 976
 977         case OVS_KEY_ATTR_MPLS:
 978                 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
 979                                                                     __be32 *));
 980                 break;
 981
 982         case OVS_KEY_ATTR_CT_STATE:
 983         case OVS_KEY_ATTR_CT_ZONE:
 984         case OVS_KEY_ATTR_CT_MARK:
 985         case OVS_KEY_ATTR_CT_LABEL:
 986                 err = -EINVAL;
 987                 break;
 988         }
 989
 990         return err;
 991 }
 992
 993 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
 994                           struct sw_flow_key *key,
 995                           const struct nlattr *a, int rem)
 996 {
 997         struct deferred_action *da;
 998
 999         if (!is_flow_key_valid(key)) {
1000                 int err;
1001
1002                 err = ovs_flow_key_update(skb, key);
1003                 if (err)
1004                         return err;
1005         }
1006         BUG_ON(!is_flow_key_valid(key));
1007
1008         if (!nla_is_last(a, rem)) {
1009                 /* Recirc action is the not the last action
1010                  * of the action list, need to clone the skb.
1011                  */
1012                 skb = skb_clone(skb, GFP_ATOMIC);
1013
1014                 /* Skip the recirc action when out of memory, but
1015                  * continue on with the rest of the action list.
1016                  */
1017                 if (!skb)
1018                         return 0;
1019         }
1020
1021         da = add_deferred_actions(skb, key, NULL);
1022         if (da) {
1023                 da->pkt_key.recirc_id = nla_get_u32(a);
1024         } else {
1025                 kfree_skb(skb);
1026
1027                 if (net_ratelimit())
1028                         pr_warn("%s: deferred action limit reached, drop recirc action\n",
1029                                 ovs_dp_name(dp));
1030         }
1031
1032         return 0;
1033 }
1034
1035 /* Execute a list of actions against 'skb'. */
1036 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1037                               struct sw_flow_key *key,
1038                               const struct nlattr *attr, int len)
1039 {
1040         /* Every output action needs a separate clone of 'skb', but the common
1041          * case is just a single output action, so that doing a clone and
1042          * then freeing the original skbuff is wasteful.  So the following code
1043          * is slightly obscure just to avoid that.
1044          */
1045         int prev_port = -1;
1046         const struct nlattr *a;
1047         int rem;
1048
1049         for (a = attr, rem = len; rem > 0;
1050              a = nla_next(a, &rem)) {
1051                 int err = 0;
1052
1053                 if (unlikely(prev_port != -1)) {
1054                         struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
1055
1056                         if (out_skb)
1057                                 do_output(dp, out_skb, prev_port, key);
1058
1059                         prev_port = -1;
1060                 }
1061
1062                 switch (nla_type(a)) {
1063                 case OVS_ACTION_ATTR_OUTPUT:
1064                         prev_port = nla_get_u32(a);
1065                         break;
1066
1067                 case OVS_ACTION_ATTR_USERSPACE:
1068                         output_userspace(dp, skb, key, a, attr, len);
1069                         break;
1070
1071                 case OVS_ACTION_ATTR_HASH:
1072                         execute_hash(skb, key, a);
1073                         break;
1074
1075                 case OVS_ACTION_ATTR_PUSH_MPLS:
1076                         err = push_mpls(skb, key, nla_data(a));
1077                         break;
1078
1079                 case OVS_ACTION_ATTR_POP_MPLS:
1080                         err = pop_mpls(skb, key, nla_get_be16(a));
1081                         break;
1082
1083                 case OVS_ACTION_ATTR_PUSH_VLAN:
1084                         err = push_vlan(skb, key, nla_data(a));
1085                         break;
1086
1087                 case OVS_ACTION_ATTR_POP_VLAN:
1088                         err = pop_vlan(skb, key);
1089                         break;
1090
1091                 case OVS_ACTION_ATTR_RECIRC:
1092                         err = execute_recirc(dp, skb, key, a, rem);
1093                         if (nla_is_last(a, rem)) {
1094                                 /* If this is the last action, the skb has
1095                                  * been consumed or freed.
1096                                  * Return immediately.
1097                                  */
1098                                 return err;
1099                         }
1100                         break;
1101
1102                 case OVS_ACTION_ATTR_SET:
1103                         err = execute_set_action(skb, key, nla_data(a));
1104                         break;
1105
1106                 case OVS_ACTION_ATTR_SET_MASKED:
1107                 case OVS_ACTION_ATTR_SET_TO_MASKED:
1108                         err = execute_masked_set_action(skb, key, nla_data(a));
1109                         break;
1110
1111                 case OVS_ACTION_ATTR_SAMPLE:
1112                         err = sample(dp, skb, key, a, attr, len);
1113                         break;
1114
1115                 case OVS_ACTION_ATTR_CT:
1116                         err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1117                                              nla_data(a));
1118
1119                         /* Hide stolen IP fragments from user space. */
1120                         if (err == -EINPROGRESS)
1121                                 return 0;
1122                         break;
1123                 }
1124
1125                 if (unlikely(err)) {
1126                         kfree_skb(skb);
1127                         return err;
1128                 }
1129         }
1130
1131         if (prev_port != -1)
1132                 do_output(dp, skb, prev_port, key);
1133         else
1134                 consume_skb(skb);
1135
1136         return 0;
1137 }
1138
1139 static void process_deferred_actions(struct datapath *dp)
1140 {
1141         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1142
1143         /* Do not touch the FIFO in case there is no deferred actions. */
1144         if (action_fifo_is_empty(fifo))
1145                 return;
1146
1147         /* Finishing executing all deferred actions. */
1148         do {
1149                 struct deferred_action *da = action_fifo_get(fifo);
1150                 struct sk_buff *skb = da->skb;
1151                 struct sw_flow_key *key = &da->pkt_key;
1152                 const struct nlattr *actions = da->actions;
1153
1154                 if (actions)
1155                         do_execute_actions(dp, skb, key, actions,
1156                                            nla_len(actions));
1157                 else
1158                         ovs_dp_process_packet(skb, key);
1159         } while (!action_fifo_is_empty(fifo));
1160
1161         /* Reset FIFO for the next packet.  */
1162         action_fifo_init(fifo);
1163 }
1164
1165 /* Execute a list of actions against 'skb'. */
1166 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1167                         const struct sw_flow_actions *acts,
1168                         struct sw_flow_key *key)
1169 {
1170         int level = this_cpu_read(exec_actions_level);
1171         int err;
1172
1173         if (unlikely(level >= EXEC_ACTIONS_LEVEL_LIMIT)) {
1174                 if (net_ratelimit())
1175                         pr_warn("%s: packet loop detected, dropping.\n",
1176                                 ovs_dp_name(dp));
1177
1178                 kfree_skb(skb);
1179                 return -ELOOP;
1180         }
1181
1182         this_cpu_inc(exec_actions_level);
1183         err = do_execute_actions(dp, skb, key,
1184                                  acts->actions, acts->actions_len);
1185
1186         if (!level)
1187                 process_deferred_actions(dp);
1188
1189         this_cpu_dec(exec_actions_level);
1190
1191         /* This return status currently does not reflect the errors
1192          * encounted during deferred actions execution. Probably needs to
1193          * be fixed in the future.
1194          */
1195         return err;
1196 }
1197
1198 int action_fifos_init(void)
1199 {
1200         action_fifos = alloc_percpu(struct action_fifo);
1201         if (!action_fifos)
1202                 return -ENOMEM;
1203
1204         return 0;
1205 }
1206
1207 void action_fifos_exit(void)
1208 {
1209         free_percpu(action_fifos);
1210 }