datapath/actions.c

   1 /*
   2  * Copyright (c) 2007-2013 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, bool keep_skb);
  42
  43 static int make_writable(struct sk_buff *skb, int write_len)
  44 {
  45         if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
  46                 return 0;
  47
  48         return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
  49 }
  50
  51 /* remove VLAN header from packet and update csum accordingly. */
  52 static int __pop_vlan_tci(struct sk_buff *skb, __be16 *current_tci)
  53 {
  54         struct vlan_hdr *vhdr;
  55         int err;
  56
  57         err = make_writable(skb, VLAN_ETH_HLEN);
  58         if (unlikely(err))
  59                 return err;
  60
  61         if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
  62                 skb->csum = csum_sub(skb->csum, csum_partial(skb->data
  63                                         + (2 * ETH_ALEN), VLAN_HLEN, 0));
  64
  65         vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
  66         *current_tci = vhdr->h_vlan_TCI;
  67
  68         memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
  69         __skb_pull(skb, VLAN_HLEN);
  70
  71         vlan_set_encap_proto(skb, vhdr);
  72         skb->mac_header += VLAN_HLEN;
  73         skb_reset_mac_len(skb);
  74
  75         return 0;
  76 }
  77
  78 static int pop_vlan(struct sk_buff *skb)
  79 {
  80         __be16 tci;
  81         int err;
  82
  83         if (likely(vlan_tx_tag_present(skb))) {
  84                 vlan_set_tci(skb, 0);
  85         } else {
  86                 if (unlikely(skb->protocol != htons(ETH_P_8021Q) ||
  87                              skb->len < VLAN_ETH_HLEN))
  88                         return 0;
  89
  90                 err = __pop_vlan_tci(skb, &tci);
  91                 if (err)
  92                         return err;
  93         }
  94         /* move next vlan tag to hw accel tag */
  95         if (likely(skb->protocol != htons(ETH_P_8021Q) ||
  96                    skb->len < VLAN_ETH_HLEN))
  97                 return 0;
  98
  99         err = __pop_vlan_tci(skb, &tci);
 100         if (unlikely(err))
 101                 return err;
 102
 103         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(tci));
 104         return 0;
 105 }
 106
 107 static int push_vlan(struct sk_buff *skb, const struct ovs_action_push_vlan *vlan)
 108 {
 109         if (unlikely(vlan_tx_tag_present(skb))) {
 110                 u16 current_tag;
 111
 112                 /* push down current VLAN tag */
 113                 current_tag = vlan_tx_tag_get(skb);
 114
 115                 if (!__vlan_put_tag(skb, skb->vlan_proto, current_tag))
 116                         return -ENOMEM;
 117
 118                 if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
 119                         skb->csum = csum_add(skb->csum, csum_partial(skb->data
 120                                         + (2 * ETH_ALEN), VLAN_HLEN, 0));
 121
 122         }
 123         __vlan_hwaccel_put_tag(skb, vlan->vlan_tpid, ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
 124         return 0;
 125 }
 126
 127 static int set_eth_addr(struct sk_buff *skb,
 128                         const struct ovs_key_ethernet *eth_key)
 129 {
 130         int err;
 131         err = make_writable(skb, ETH_HLEN);
 132         if (unlikely(err))
 133                 return err;
 134
 135         if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
 136                 skb->csum = csum_sub(skb->csum, csum_partial(eth_hdr(skb),
 137                                                              ETH_ALEN * 2, 0));
 138
 139         memcpy(eth_hdr(skb)->h_source, eth_key->eth_src, ETH_ALEN);
 140         memcpy(eth_hdr(skb)->h_dest, eth_key->eth_dst, ETH_ALEN);
 141
 142         if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
 143                 skb->csum = csum_add(skb->csum, csum_partial(eth_hdr(skb),
 144                                                              ETH_ALEN * 2, 0));
 145
 146         return 0;
 147 }
 148
 149 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 150                                 __be32 *addr, __be32 new_addr)
 151 {
 152         int transport_len = skb->len - skb_transport_offset(skb);
 153
 154         if (nh->protocol == IPPROTO_TCP) {
 155                 if (likely(transport_len >= sizeof(struct tcphdr)))
 156                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 157                                                  *addr, new_addr, 1);
 158         } else if (nh->protocol == IPPROTO_UDP) {
 159                 if (likely(transport_len >= sizeof(struct udphdr))) {
 160                         struct udphdr *uh = udp_hdr(skb);
 161
 162                         if (uh->check ||
 163                             get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
 164                                 inet_proto_csum_replace4(&uh->check, skb,
 165                                                          *addr, new_addr, 1);
 166                                 if (!uh->check)
 167                                         uh->check = CSUM_MANGLED_0;
 168                         }
 169                 }
 170         }
 171
 172         csum_replace4(&nh->check, *addr, new_addr);
 173         skb_clear_rxhash(skb);
 174         *addr = new_addr;
 175 }
 176
 177 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 178                                  __be32 addr[4], const __be32 new_addr[4])
 179 {
 180         int transport_len = skb->len - skb_transport_offset(skb);
 181
 182         if (l4_proto == IPPROTO_TCP) {
 183                 if (likely(transport_len >= sizeof(struct tcphdr)))
 184                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 185                                                   addr, new_addr, 1);
 186         } else if (l4_proto == IPPROTO_UDP) {
 187                 if (likely(transport_len >= sizeof(struct udphdr))) {
 188                         struct udphdr *uh = udp_hdr(skb);
 189
 190                         if (uh->check ||
 191                             get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
 192                                 inet_proto_csum_replace16(&uh->check, skb,
 193                                                           addr, new_addr, 1);
 194                                 if (!uh->check)
 195                                         uh->check = CSUM_MANGLED_0;
 196                         }
 197                 }
 198         }
 199 }
 200
 201 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 202                           __be32 addr[4], const __be32 new_addr[4],
 203                           bool recalculate_csum)
 204 {
 205         if (recalculate_csum)
 206                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 207
 208         skb_clear_rxhash(skb);
 209         memcpy(addr, new_addr, sizeof(__be32[4]));
 210 }
 211
 212 static void set_ipv6_tc(struct ipv6hdr *nh, u8 tc)
 213 {
 214         nh->priority = tc >> 4;
 215         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0x0F) | ((tc & 0x0F) << 4);
 216 }
 217
 218 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl)
 219 {
 220         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0xF0) | (fl & 0x000F0000) >> 16;
 221         nh->flow_lbl[1] = (fl & 0x0000FF00) >> 8;
 222         nh->flow_lbl[2] = fl & 0x000000FF;
 223 }
 224
 225 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl)
 226 {
 227         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 228         nh->ttl = new_ttl;
 229 }
 230
 231 static int set_ipv4(struct sk_buff *skb, const struct ovs_key_ipv4 *ipv4_key)
 232 {
 233         struct iphdr *nh;
 234         int err;
 235
 236         err = make_writable(skb, skb_network_offset(skb) +
 237                                  sizeof(struct iphdr));
 238         if (unlikely(err))
 239                 return err;
 240
 241         nh = ip_hdr(skb);
 242
 243         if (ipv4_key->ipv4_src != nh->saddr)
 244                 set_ip_addr(skb, nh, &nh->saddr, ipv4_key->ipv4_src);
 245
 246         if (ipv4_key->ipv4_dst != nh->daddr)
 247                 set_ip_addr(skb, nh, &nh->daddr, ipv4_key->ipv4_dst);
 248
 249         if (ipv4_key->ipv4_tos != nh->tos)
 250                 ipv4_change_dsfield(nh, 0, ipv4_key->ipv4_tos);
 251
 252         if (ipv4_key->ipv4_ttl != nh->ttl)
 253                 set_ip_ttl(skb, nh, ipv4_key->ipv4_ttl);
 254
 255         return 0;
 256 }
 257
 258 static int set_ipv6(struct sk_buff *skb, const struct ovs_key_ipv6 *ipv6_key)
 259 {
 260         struct ipv6hdr *nh;
 261         int err;
 262         __be32 *saddr;
 263         __be32 *daddr;
 264
 265         err = make_writable(skb, skb_network_offset(skb) +
 266                             sizeof(struct ipv6hdr));
 267         if (unlikely(err))
 268                 return err;
 269
 270         nh = ipv6_hdr(skb);
 271         saddr = (__be32 *)&nh->saddr;
 272         daddr = (__be32 *)&nh->daddr;
 273
 274         if (memcmp(ipv6_key->ipv6_src, saddr, sizeof(ipv6_key->ipv6_src)))
 275                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, saddr,
 276                               ipv6_key->ipv6_src, true);
 277
 278         if (memcmp(ipv6_key->ipv6_dst, daddr, sizeof(ipv6_key->ipv6_dst))) {
 279                 unsigned int offset = 0;
 280                 int flags = OVS_IP6T_FH_F_SKIP_RH;
 281                 bool recalc_csum = true;
 282
 283                 if (ipv6_ext_hdr(nh->nexthdr))
 284                         recalc_csum = ipv6_find_hdr(skb, &offset,
 285                                                     NEXTHDR_ROUTING, NULL,
 286                                                     &flags) != NEXTHDR_ROUTING;
 287
 288                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, daddr,
 289                               ipv6_key->ipv6_dst, recalc_csum);
 290         }
 291
 292         set_ipv6_tc(nh, ipv6_key->ipv6_tclass);
 293         set_ipv6_fl(nh, ntohl(ipv6_key->ipv6_label));
 294         nh->hop_limit = ipv6_key->ipv6_hlimit;
 295
 296         return 0;
 297 }
 298
 299 /* Must follow make_writable() since that can move the skb data. */
 300 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 301                          __be16 new_port, __sum16 *check)
 302 {
 303         inet_proto_csum_replace2(check, skb, *port, new_port, 0);
 304         *port = new_port;
 305         skb_clear_rxhash(skb);
 306 }
 307
 308 static void set_udp_port(struct sk_buff *skb, __be16 *port, __be16 new_port)
 309 {
 310         struct udphdr *uh = udp_hdr(skb);
 311
 312         if (uh->check && get_ip_summed(skb) != OVS_CSUM_PARTIAL) {
 313                 set_tp_port(skb, port, new_port, &uh->check);
 314
 315                 if (!uh->check)
 316                         uh->check = CSUM_MANGLED_0;
 317         } else {
 318                 *port = new_port;
 319                 skb_clear_rxhash(skb);
 320         }
 321 }
 322
 323 static int set_udp(struct sk_buff *skb, const struct ovs_key_udp *udp_port_key)
 324 {
 325         struct udphdr *uh;
 326         int err;
 327
 328         err = make_writable(skb, skb_transport_offset(skb) +
 329                                  sizeof(struct udphdr));
 330         if (unlikely(err))
 331                 return err;
 332
 333         uh = udp_hdr(skb);
 334         if (udp_port_key->udp_src != uh->source)
 335                 set_udp_port(skb, &uh->source, udp_port_key->udp_src);
 336
 337         if (udp_port_key->udp_dst != uh->dest)
 338                 set_udp_port(skb, &uh->dest, udp_port_key->udp_dst);
 339
 340         return 0;
 341 }
 342
 343 static int set_tcp(struct sk_buff *skb, const struct ovs_key_tcp *tcp_port_key)
 344 {
 345         struct tcphdr *th;
 346         int err;
 347
 348         err = make_writable(skb, skb_transport_offset(skb) +
 349                                  sizeof(struct tcphdr));
 350         if (unlikely(err))
 351                 return err;
 352
 353         th = tcp_hdr(skb);
 354         if (tcp_port_key->tcp_src != th->source)
 355                 set_tp_port(skb, &th->source, tcp_port_key->tcp_src, &th->check);
 356
 357         if (tcp_port_key->tcp_dst != th->dest)
 358                 set_tp_port(skb, &th->dest, tcp_port_key->tcp_dst, &th->check);
 359
 360         return 0;
 361 }
 362
 363 static int do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
 364 {
 365         struct vport *vport;
 366
 367         if (unlikely(!skb))
 368                 return -ENOMEM;
 369
 370         vport = ovs_vport_rcu(dp, out_port);
 371         if (unlikely(!vport)) {
 372                 kfree_skb(skb);
 373                 return -ENODEV;
 374         }
 375
 376         ovs_vport_send(vport, skb);
 377         return 0;
 378 }
 379
 380 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 381                             const struct nlattr *attr)
 382 {
 383         struct dp_upcall_info upcall;
 384         const struct nlattr *a;
 385         int rem;
 386
 387         BUG_ON(!OVS_CB(skb)->pkt_key);
 388
 389         upcall.cmd = OVS_PACKET_CMD_ACTION;
 390         upcall.key = OVS_CB(skb)->pkt_key;
 391         upcall.userdata = NULL;
 392         upcall.portid = 0;
 393
 394         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 395                  a = nla_next(a, &rem)) {
 396                 switch (nla_type(a)) {
 397                 case OVS_USERSPACE_ATTR_USERDATA:
 398                         upcall.userdata = a;
 399                         break;
 400
 401                 case OVS_USERSPACE_ATTR_PID:
 402                         upcall.portid = nla_get_u32(a);
 403                         break;
 404                 }
 405         }
 406
 407         return ovs_dp_upcall(dp, skb, &upcall);
 408 }
 409
 410 static int sample(struct datapath *dp, struct sk_buff *skb,
 411                   const struct nlattr *attr)
 412 {
 413         const struct nlattr *acts_list = NULL;
 414         const struct nlattr *a;
 415         int rem;
 416
 417         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 418                  a = nla_next(a, &rem)) {
 419                 switch (nla_type(a)) {
 420                 case OVS_SAMPLE_ATTR_PROBABILITY:
 421                         if (net_random() >= nla_get_u32(a))
 422                                 return 0;
 423                         break;
 424
 425                 case OVS_SAMPLE_ATTR_ACTIONS:
 426                         acts_list = a;
 427                         break;
 428                 }
 429         }
 430
 431         return do_execute_actions(dp, skb, nla_data(acts_list),
 432                                   nla_len(acts_list), true);
 433 }
 434
 435 static int execute_set_action(struct sk_buff *skb,
 436                                  const struct nlattr *nested_attr)
 437 {
 438         int err = 0;
 439
 440         switch (nla_type(nested_attr)) {
 441         case OVS_KEY_ATTR_PRIORITY:
 442                 skb->priority = nla_get_u32(nested_attr);
 443                 break;
 444
 445         case OVS_KEY_ATTR_SKB_MARK:
 446                 skb_set_mark(skb, nla_get_u32(nested_attr));
 447                 break;
 448
 449         case OVS_KEY_ATTR_IPV4_TUNNEL:
 450                 OVS_CB(skb)->tun_key = nla_data(nested_attr);
 451                 break;
 452
 453         case OVS_KEY_ATTR_ETHERNET:
 454                 err = set_eth_addr(skb, nla_data(nested_attr));
 455                 break;
 456
 457         case OVS_KEY_ATTR_IPV4:
 458                 err = set_ipv4(skb, nla_data(nested_attr));
 459                 break;
 460
 461         case OVS_KEY_ATTR_IPV6:
 462                 err = set_ipv6(skb, nla_data(nested_attr));
 463                 break;
 464
 465         case OVS_KEY_ATTR_TCP:
 466                 err = set_tcp(skb, nla_data(nested_attr));
 467                 break;
 468
 469         case OVS_KEY_ATTR_UDP:
 470                 err = set_udp(skb, nla_data(nested_attr));
 471                 break;
 472         }
 473
 474         return err;
 475 }
 476
 477 /* Execute a list of actions against 'skb'. */
 478 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 479                         const struct nlattr *attr, int len, bool keep_skb)
 480 {
 481         /* Every output action needs a separate clone of 'skb', but the common
 482          * case is just a single output action, so that doing a clone and
 483          * then freeing the original skbuff is wasteful.  So the following code
 484          * is slightly obscure just to avoid that. */
 485         int prev_port = -1;
 486         const struct nlattr *a;
 487         int rem;
 488
 489         for (a = attr, rem = len; rem > 0;
 490              a = nla_next(a, &rem)) {
 491                 int err = 0;
 492
 493                 if (prev_port != -1) {
 494                         do_output(dp, skb_clone(skb, GFP_ATOMIC), prev_port);
 495                         prev_port = -1;
 496                 }
 497
 498                 switch (nla_type(a)) {
 499                 case OVS_ACTION_ATTR_OUTPUT:
 500                         prev_port = nla_get_u32(a);
 501                         break;
 502
 503                 case OVS_ACTION_ATTR_USERSPACE:
 504                         output_userspace(dp, skb, a);
 505                         break;
 506
 507                 case OVS_ACTION_ATTR_PUSH_VLAN:
 508                         err = push_vlan(skb, nla_data(a));
 509                         if (unlikely(err)) /* skb already freed. */
 510                                 return err;
 511                         break;
 512
 513                 case OVS_ACTION_ATTR_POP_VLAN:
 514                         err = pop_vlan(skb);
 515                         break;
 516
 517                 case OVS_ACTION_ATTR_SET:
 518                         err = execute_set_action(skb, nla_data(a));
 519                         break;
 520
 521                 case OVS_ACTION_ATTR_SAMPLE:
 522                         err = sample(dp, skb, a);
 523                         break;
 524                 }
 525
 526                 if (unlikely(err)) {
 527                         kfree_skb(skb);
 528                         return err;
 529                 }
 530         }
 531
 532         if (prev_port != -1) {
 533                 if (keep_skb)
 534                         skb = skb_clone(skb, GFP_ATOMIC);
 535
 536                 do_output(dp, skb, prev_port);
 537         } else if (!keep_skb)
 538                 consume_skb(skb);
 539
 540         return 0;
 541 }
 542
 543 /* We limit the number of times that we pass into execute_actions()
 544  * to avoid blowing out the stack in the event that we have a loop. */
 545 #define MAX_LOOPS 4
 546
 547 struct loop_counter {
 548         u8 count;               /* Count. */
 549         bool looping;           /* Loop detected? */
 550 };
 551
 552 static DEFINE_PER_CPU(struct loop_counter, loop_counters);
 553
 554 static int loop_suppress(struct datapath *dp, struct sw_flow_actions *actions)
 555 {
 556         if (net_ratelimit())
 557                 pr_warn("%s: flow looped %d times, dropping\n",
 558                                 ovs_dp_name(dp), MAX_LOOPS);
 559         actions->actions_len = 0;
 560         return -ELOOP;
 561 }
 562
 563 /* Execute a list of actions against 'skb'. */
 564 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb)
 565 {
 566         struct sw_flow_actions *acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
 567         struct loop_counter *loop;
 568         int error;
 569
 570         /* Check whether we've looped too much. */
 571         loop = &__get_cpu_var(loop_counters);
 572         if (unlikely(++loop->count > MAX_LOOPS))
 573                 loop->looping = true;
 574         if (unlikely(loop->looping)) {
 575                 error = loop_suppress(dp, acts);
 576                 kfree_skb(skb);
 577                 goto out_loop;
 578         }
 579
 580         OVS_CB(skb)->tun_key = NULL;
 581         error = do_execute_actions(dp, skb, acts->actions,
 582                                          acts->actions_len, false);
 583
 584         /* Check whether sub-actions looped too much. */
 585         if (unlikely(loop->looping))
 586                 error = loop_suppress(dp, acts);
 587
 588 out_loop:
 589         /* Decrement loop counter. */
 590         if (!--loop->count)
 591                 loop->looping = false;
 592
 593         return error;
 594 }