datapath/actions.c

   1 /*
   2  * Copyright (c) 2007-2012 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/tcp.h>
  26 #include <linux/udp.h>
  27 #include <linux/in6.h>
  28 #include <linux/if_arp.h>
  29 #include <linux/if_vlan.h>
  30 #include <net/ip.h>
  31 #include <net/ipv6.h>
  32 #include <net/checksum.h>
  33 #include <net/dsfield.h>
  34
  35 #include "checksum.h"
  36 #include "datapath.h"
  37 #include "vlan.h"
  38 #include "vport.h"
  39
  40 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
  41                               const struct nlattr *attr, int len,
  42                               struct ovs_key_ipv4_tunnel *tun_key, bool keep_skb);
  43
  44 static int make_writable(struct sk_buff *skb, int write_len)
  45 {
  46         if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
  47                 return 0;
  48
  49         return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
  50 }
  51
  52 /* remove VLAN header from packet and update csum accordingly. */
  53 static int __pop_vlan_tci(struct sk_buff *skb, __be16 *current_tci)
  54 {
  55         struct vlan_hdr *vhdr;
  56         int err;
  57
  58         err = make_writable(skb, VLAN_ETH_HLEN);
  59         if (unlikely(err))
  60                 return err;
  61
  62         if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
  63                 skb->csum = csum_sub(skb->csum, csum_partial(skb->data
  64                                         + ETH_HLEN, VLAN_HLEN, 0));
  65
  66         vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
  67         *current_tci = vhdr->h_vlan_TCI;
  68
  69         memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
  70         __skb_pull(skb, VLAN_HLEN);
  71
  72         vlan_set_encap_proto(skb, vhdr);
  73         skb->mac_header += VLAN_HLEN;
  74         skb_reset_mac_len(skb);
  75
  76         return 0;
  77 }
  78
  79 static int pop_vlan(struct sk_buff *skb)
  80 {
  81         __be16 tci;
  82         int err;
  83
  84         if (likely(vlan_tx_tag_present(skb))) {
  85                 vlan_set_tci(skb, 0);
  86         } else {
  87                 if (unlikely(skb->protocol != htons(ETH_P_8021Q) ||
  88                              skb->len < VLAN_ETH_HLEN))
  89                         return 0;
  90
  91                 err = __pop_vlan_tci(skb, &tci);
  92                 if (err)
  93                         return err;
  94         }
  95         /* move next vlan tag to hw accel tag */
  96         if (likely(skb->protocol != htons(ETH_P_8021Q) ||
  97                    skb->len < VLAN_ETH_HLEN))
  98                 return 0;
  99
 100         err = __pop_vlan_tci(skb, &tci);
 101         if (unlikely(err))
 102                 return err;
 103
 104         __vlan_hwaccel_put_tag(skb, ntohs(tci));
 105         return 0;
 106 }
 107
 108 static int push_vlan(struct sk_buff *skb, const struct ovs_action_push_vlan *vlan)
 109 {
 110         if (unlikely(vlan_tx_tag_present(skb))) {
 111                 u16 current_tag;
 112
 113                 /* push down current VLAN tag */
 114                 current_tag = vlan_tx_tag_get(skb);
 115
 116                 if (!__vlan_put_tag(skb, current_tag))
 117                         return -ENOMEM;
 118
 119                 if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
 120                         skb->csum = csum_add(skb->csum, csum_partial(skb->data
 121                                         + ETH_HLEN, VLAN_HLEN, 0));
 122
 123         }
 124         __vlan_hwaccel_put_tag(skb, ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
 125         return 0;
 126 }
 127
 128 static int set_eth_addr(struct sk_buff *skb,
 129                         const struct ovs_key_ethernet *eth_key)
 130 {
 131         int err;
 132         err = make_writable(skb, ETH_HLEN);
 133         if (unlikely(err))
 134                 return err;
 135
 136         memcpy(eth_hdr(skb)->h_source, eth_key->eth_src, ETH_ALEN);
 137         memcpy(eth_hdr(skb)->h_dest, eth_key->eth_dst, ETH_ALEN);
 138
 139         return 0;
 140 }
 141
 142 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 143                                 __be32 *addr, __be32 new_addr)
 144 {
 145         int transport_len = skb->len - skb_transport_offset(skb);
 146
 147         if (nh->protocol == IPPROTO_TCP) {
 148                 if (likely(transport_len >= sizeof(struct tcphdr)))
 149                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 150                                                  *addr, new_addr, 1);
 151         } else if (nh->protocol == IPPROTO_UDP) {
 152                 if (likely(transport_len >= sizeof(struct udphdr))) {
 153                         struct udphdr *uh = udp_hdr(skb);
 154
 155                         if (uh->check ||
 156                             get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
 157                                 inet_proto_csum_replace4(&uh->check, skb,
 158                                                          *addr, new_addr, 1);
 159                                 if (!uh->check)
 160                                         uh->check = CSUM_MANGLED_0;
 161                         }
 162                 }
 163         }
 164
 165         csum_replace4(&nh->check, *addr, new_addr);
 166         skb_clear_rxhash(skb);
 167         *addr = new_addr;
 168 }
 169
 170 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 171                                  __be32 addr[4], const __be32 new_addr[4])
 172 {
 173         int transport_len = skb->len - skb_transport_offset(skb);
 174
 175         if (l4_proto == IPPROTO_TCP) {
 176                 if (likely(transport_len >= sizeof(struct tcphdr)))
 177                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 178                                                   addr, new_addr, 1);
 179         } else if (l4_proto == IPPROTO_UDP) {
 180                 if (likely(transport_len >= sizeof(struct udphdr))) {
 181                         struct udphdr *uh = udp_hdr(skb);
 182
 183                         if (uh->check ||
 184                             get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
 185                                 inet_proto_csum_replace16(&uh->check, skb,
 186                                                           addr, new_addr, 1);
 187                                 if (!uh->check)
 188                                         uh->check = CSUM_MANGLED_0;
 189                         }
 190                 }
 191         }
 192 }
 193
 194 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 195                           __be32 addr[4], const __be32 new_addr[4],
 196                           bool recalculate_csum)
 197 {
 198         if (recalculate_csum)
 199                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 200
 201         skb_clear_rxhash(skb);
 202         memcpy(addr, new_addr, sizeof(__be32[4]));
 203 }
 204
 205 static void set_ipv6_tc(struct ipv6hdr *nh, u8 tc)
 206 {
 207         nh->priority = tc >> 4;
 208         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0x0F) | ((tc & 0x0F) << 4);
 209 }
 210
 211 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl)
 212 {
 213         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0xF0) | (fl & 0x000F0000) >> 16;
 214         nh->flow_lbl[1] = (fl & 0x0000FF00) >> 8;
 215         nh->flow_lbl[2] = fl & 0x000000FF;
 216 }
 217
 218 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl)
 219 {
 220         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 221         nh->ttl = new_ttl;
 222 }
 223
 224 static int set_ipv4(struct sk_buff *skb, const struct ovs_key_ipv4 *ipv4_key)
 225 {
 226         struct iphdr *nh;
 227         int err;
 228
 229         err = make_writable(skb, skb_network_offset(skb) +
 230                                  sizeof(struct iphdr));
 231         if (unlikely(err))
 232                 return err;
 233
 234         nh = ip_hdr(skb);
 235
 236         if (ipv4_key->ipv4_src != nh->saddr)
 237                 set_ip_addr(skb, nh, &nh->saddr, ipv4_key->ipv4_src);
 238
 239         if (ipv4_key->ipv4_dst != nh->daddr)
 240                 set_ip_addr(skb, nh, &nh->daddr, ipv4_key->ipv4_dst);
 241
 242         if (ipv4_key->ipv4_tos != nh->tos)
 243                 ipv4_change_dsfield(nh, 0, ipv4_key->ipv4_tos);
 244
 245         if (ipv4_key->ipv4_ttl != nh->ttl)
 246                 set_ip_ttl(skb, nh, ipv4_key->ipv4_ttl);
 247
 248         return 0;
 249 }
 250
 251 static int set_ipv6(struct sk_buff *skb, const struct ovs_key_ipv6 *ipv6_key)
 252 {
 253         struct ipv6hdr *nh;
 254         int err;
 255         __be32 *saddr;
 256         __be32 *daddr;
 257
 258         err = make_writable(skb, skb_network_offset(skb) +
 259                             sizeof(struct ipv6hdr));
 260         if (unlikely(err))
 261                 return err;
 262
 263         nh = ipv6_hdr(skb);
 264         saddr = (__be32 *)&nh->saddr;
 265         daddr = (__be32 *)&nh->daddr;
 266
 267         if (memcmp(ipv6_key->ipv6_src, saddr, sizeof(ipv6_key->ipv6_src)))
 268                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, saddr,
 269                               ipv6_key->ipv6_src, true);
 270
 271         if (memcmp(ipv6_key->ipv6_dst, daddr, sizeof(ipv6_key->ipv6_dst))) {
 272                 unsigned int offset = 0;
 273                 int flags = OVS_IP6T_FH_F_SKIP_RH;
 274                 bool recalc_csum = true;
 275
 276                 if (ipv6_ext_hdr(nh->nexthdr))
 277                         recalc_csum = ipv6_find_hdr(skb, &offset,
 278                                                     NEXTHDR_ROUTING, NULL,
 279                                                     &flags) != NEXTHDR_ROUTING;
 280
 281                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, daddr,
 282                               ipv6_key->ipv6_dst, recalc_csum);
 283         }
 284
 285         set_ipv6_tc(nh, ipv6_key->ipv6_tclass);
 286         set_ipv6_fl(nh, ntohl(ipv6_key->ipv6_label));
 287         nh->hop_limit = ipv6_key->ipv6_hlimit;
 288
 289         return 0;
 290 }
 291
 292 /* Must follow make_writable() since that can move the skb data. */
 293 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 294                          __be16 new_port, __sum16 *check)
 295 {
 296         inet_proto_csum_replace2(check, skb, *port, new_port, 0);
 297         *port = new_port;
 298         skb_clear_rxhash(skb);
 299 }
 300
 301 static void set_udp_port(struct sk_buff *skb, __be16 *port, __be16 new_port)
 302 {
 303         struct udphdr *uh = udp_hdr(skb);
 304
 305         if (uh->check && get_ip_summed(skb) != OVS_CSUM_PARTIAL) {
 306                 set_tp_port(skb, port, new_port, &uh->check);
 307
 308                 if (!uh->check)
 309                         uh->check = CSUM_MANGLED_0;
 310         } else {
 311                 *port = new_port;
 312                 skb_clear_rxhash(skb);
 313         }
 314 }
 315
 316 static int set_udp(struct sk_buff *skb, const struct ovs_key_udp *udp_port_key)
 317 {
 318         struct udphdr *uh;
 319         int err;
 320
 321         err = make_writable(skb, skb_transport_offset(skb) +
 322                                  sizeof(struct udphdr));
 323         if (unlikely(err))
 324                 return err;
 325
 326         uh = udp_hdr(skb);
 327         if (udp_port_key->udp_src != uh->source)
 328                 set_udp_port(skb, &uh->source, udp_port_key->udp_src);
 329
 330         if (udp_port_key->udp_dst != uh->dest)
 331                 set_udp_port(skb, &uh->dest, udp_port_key->udp_dst);
 332
 333         return 0;
 334 }
 335
 336 static int set_tcp(struct sk_buff *skb, const struct ovs_key_tcp *tcp_port_key)
 337 {
 338         struct tcphdr *th;
 339         int err;
 340
 341         err = make_writable(skb, skb_transport_offset(skb) +
 342                                  sizeof(struct tcphdr));
 343         if (unlikely(err))
 344                 return err;
 345
 346         th = tcp_hdr(skb);
 347         if (tcp_port_key->tcp_src != th->source)
 348                 set_tp_port(skb, &th->source, tcp_port_key->tcp_src, &th->check);
 349
 350         if (tcp_port_key->tcp_dst != th->dest)
 351                 set_tp_port(skb, &th->dest, tcp_port_key->tcp_dst, &th->check);
 352
 353         return 0;
 354 }
 355
 356 static int do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
 357 {
 358         struct vport *vport;
 359
 360         if (unlikely(!skb))
 361                 return -ENOMEM;
 362
 363         vport = ovs_vport_rcu(dp, out_port);
 364         if (unlikely(!vport)) {
 365                 kfree_skb(skb);
 366                 return -ENODEV;
 367         }
 368
 369         ovs_vport_send(vport, skb);
 370         return 0;
 371 }
 372
 373 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 374                             const struct nlattr *attr)
 375 {
 376         struct dp_upcall_info upcall;
 377         const struct nlattr *a;
 378         int rem;
 379
 380         upcall.cmd = OVS_PACKET_CMD_ACTION;
 381         upcall.key = &OVS_CB(skb)->flow->key;
 382         upcall.userdata = NULL;
 383         upcall.pid = 0;
 384
 385         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 386                  a = nla_next(a, &rem)) {
 387                 switch (nla_type(a)) {
 388                 case OVS_USERSPACE_ATTR_USERDATA:
 389                         upcall.userdata = a;
 390                         break;
 391
 392                 case OVS_USERSPACE_ATTR_PID:
 393                         upcall.pid = nla_get_u32(a);
 394                         break;
 395                 }
 396         }
 397
 398         return ovs_dp_upcall(dp, skb, &upcall);
 399 }
 400
 401 static int sample(struct datapath *dp, struct sk_buff *skb,
 402                   const struct nlattr *attr,
 403                   struct ovs_key_ipv4_tunnel *tun_key)
 404 {
 405         const struct nlattr *acts_list = NULL;
 406         const struct nlattr *a;
 407         int rem;
 408
 409         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 410                  a = nla_next(a, &rem)) {
 411                 switch (nla_type(a)) {
 412                 case OVS_SAMPLE_ATTR_PROBABILITY:
 413                         if (net_random() >= nla_get_u32(a))
 414                                 return 0;
 415                         break;
 416
 417                 case OVS_SAMPLE_ATTR_ACTIONS:
 418                         acts_list = a;
 419                         break;
 420                 }
 421         }
 422
 423         return do_execute_actions(dp, skb, nla_data(acts_list),
 424                                   nla_len(acts_list), tun_key, true);
 425 }
 426
 427 static int execute_set_action(struct sk_buff *skb,
 428                                  const struct nlattr *nested_attr,
 429                                  struct ovs_key_ipv4_tunnel *tun_key)
 430 {
 431         int err = 0;
 432
 433         switch (nla_type(nested_attr)) {
 434         case OVS_KEY_ATTR_PRIORITY:
 435                 skb->priority = nla_get_u32(nested_attr);
 436                 break;
 437
 438         case OVS_KEY_ATTR_SKB_MARK:
 439                 skb_set_mark(skb, nla_get_u32(nested_attr));
 440                 break;
 441
 442         case OVS_KEY_ATTR_TUN_ID:
 443                 /* If we're only using the TUN_ID action, store the value in a
 444                  * temporary instance of struct ovs_key_ipv4_tunnel on the stack.
 445                  * If both IPV4_TUNNEL and TUN_ID are being used together we
 446                  * can't write into the IPV4_TUNNEL action, so make a copy and
 447                  * write into that version.
 448                  */
 449                 if (!OVS_CB(skb)->tun_key)
 450                         memset(tun_key, 0, sizeof(*tun_key));
 451                 else if (OVS_CB(skb)->tun_key != tun_key)
 452                         memcpy(tun_key, OVS_CB(skb)->tun_key, sizeof(*tun_key));
 453                 OVS_CB(skb)->tun_key = tun_key;
 454
 455                 OVS_CB(skb)->tun_key->tun_id = nla_get_be64(nested_attr);
 456                 break;
 457
 458         case OVS_KEY_ATTR_IPV4_TUNNEL:
 459                 OVS_CB(skb)->tun_key = nla_data(nested_attr);
 460                 break;
 461
 462         case OVS_KEY_ATTR_ETHERNET:
 463                 err = set_eth_addr(skb, nla_data(nested_attr));
 464                 break;
 465
 466         case OVS_KEY_ATTR_IPV4:
 467                 err = set_ipv4(skb, nla_data(nested_attr));
 468                 break;
 469
 470         case OVS_KEY_ATTR_IPV6:
 471                 err = set_ipv6(skb, nla_data(nested_attr));
 472                 break;
 473
 474         case OVS_KEY_ATTR_TCP:
 475                 err = set_tcp(skb, nla_data(nested_attr));
 476                 break;
 477
 478         case OVS_KEY_ATTR_UDP:
 479                 err = set_udp(skb, nla_data(nested_attr));
 480                 break;
 481         }
 482
 483         return err;
 484 }
 485
 486 /* Execute a list of actions against 'skb'. */
 487 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 488                         const struct nlattr *attr, int len,
 489                         struct ovs_key_ipv4_tunnel *tun_key, bool keep_skb)
 490 {
 491         /* Every output action needs a separate clone of 'skb', but the common
 492          * case is just a single output action, so that doing a clone and
 493          * then freeing the original skbuff is wasteful.  So the following code
 494          * is slightly obscure just to avoid that. */
 495         int prev_port = -1;
 496         const struct nlattr *a;
 497         int rem;
 498
 499         for (a = attr, rem = len; rem > 0;
 500              a = nla_next(a, &rem)) {
 501                 int err = 0;
 502
 503                 if (prev_port != -1) {
 504                         do_output(dp, skb_clone(skb, GFP_ATOMIC), prev_port);
 505                         prev_port = -1;
 506                 }
 507
 508                 switch (nla_type(a)) {
 509                 case OVS_ACTION_ATTR_OUTPUT:
 510                         prev_port = nla_get_u32(a);
 511                         break;
 512
 513                 case OVS_ACTION_ATTR_USERSPACE:
 514                         output_userspace(dp, skb, a);
 515                         break;
 516
 517                 case OVS_ACTION_ATTR_PUSH_VLAN:
 518                         err = push_vlan(skb, nla_data(a));
 519                         if (unlikely(err)) /* skb already freed. */
 520                                 return err;
 521                         break;
 522
 523                 case OVS_ACTION_ATTR_POP_VLAN:
 524                         err = pop_vlan(skb);
 525                         break;
 526
 527                 case OVS_ACTION_ATTR_SET:
 528                         err = execute_set_action(skb, nla_data(a), tun_key);
 529                         break;
 530
 531                 case OVS_ACTION_ATTR_SAMPLE:
 532                         err = sample(dp, skb, a, tun_key);
 533                         break;
 534                 }
 535
 536                 if (unlikely(err)) {
 537                         kfree_skb(skb);
 538                         return err;
 539                 }
 540         }
 541
 542         if (prev_port != -1) {
 543                 if (keep_skb)
 544                         skb = skb_clone(skb, GFP_ATOMIC);
 545
 546                 do_output(dp, skb, prev_port);
 547         } else if (!keep_skb)
 548                 consume_skb(skb);
 549
 550         return 0;
 551 }
 552
 553 /* We limit the number of times that we pass into execute_actions()
 554  * to avoid blowing out the stack in the event that we have a loop. */
 555 #define MAX_LOOPS 5
 556
 557 struct loop_counter {
 558         u8 count;               /* Count. */
 559         bool looping;           /* Loop detected? */
 560 };
 561
 562 static DEFINE_PER_CPU(struct loop_counter, loop_counters);
 563
 564 static int loop_suppress(struct datapath *dp, struct sw_flow_actions *actions)
 565 {
 566         if (net_ratelimit())
 567                 pr_warn("%s: flow looped %d times, dropping\n",
 568                                 ovs_dp_name(dp), MAX_LOOPS);
 569         actions->actions_len = 0;
 570         return -ELOOP;
 571 }
 572
 573 /* Execute a list of actions against 'skb'. */
 574 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb)
 575 {
 576         struct sw_flow_actions *acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
 577         struct loop_counter *loop;
 578         int error;
 579         struct ovs_key_ipv4_tunnel tun_key;
 580
 581         /* Check whether we've looped too much. */
 582         loop = &__get_cpu_var(loop_counters);
 583         if (unlikely(++loop->count > MAX_LOOPS))
 584                 loop->looping = true;
 585         if (unlikely(loop->looping)) {
 586                 error = loop_suppress(dp, acts);
 587                 kfree_skb(skb);
 588                 goto out_loop;
 589         }
 590
 591         OVS_CB(skb)->tun_key = NULL;
 592         error = do_execute_actions(dp, skb, acts->actions,
 593                                          acts->actions_len, &tun_key, false);
 594
 595         /* Check whether sub-actions looped too much. */
 596         if (unlikely(loop->looping))
 597                 error = loop_suppress(dp, acts);
 598
 599 out_loop:
 600         /* Decrement loop counter. */
 601         if (!--loop->count)
 602                 loop->looping = false;
 603
 604         return error;
 605 }