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/sctp.h>
  26 #include <linux/tcp.h>
  27 #include <linux/udp.h>
  28 #include <linux/in6.h>
  29 #include <linux/if_arp.h>
  30 #include <linux/if_vlan.h>
  31 #include <net/ip.h>
  32 #include <net/ipv6.h>
  33 #include <net/checksum.h>
  34 #include <net/dsfield.h>
  35 #include <net/sctp/checksum.h>
  36
  37 #include "checksum.h"
  38 #include "datapath.h"
  39 #include "vlan.h"
  40 #include "vport.h"
  41
  42 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
  43                               const struct nlattr *attr, int len, bool keep_skb);
  44
  45 static int make_writable(struct sk_buff *skb, int write_len)
  46 {
  47         if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
  48                 return 0;
  49
  50         return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
  51 }
  52
  53 /* remove VLAN header from packet and update csum accordingly. */
  54 static int __pop_vlan_tci(struct sk_buff *skb, __be16 *current_tci)
  55 {
  56         struct vlan_hdr *vhdr;
  57         int err;
  58
  59         err = make_writable(skb, VLAN_ETH_HLEN);
  60         if (unlikely(err))
  61                 return err;
  62
  63         if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
  64                 skb->csum = csum_sub(skb->csum, csum_partial(skb->data
  65                                         + (2 * ETH_ALEN), VLAN_HLEN, 0));
  66
  67         vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
  68         *current_tci = vhdr->h_vlan_TCI;
  69
  70         memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
  71         __skb_pull(skb, VLAN_HLEN);
  72
  73         vlan_set_encap_proto(skb, vhdr);
  74         skb->mac_header += VLAN_HLEN;
  75         skb_reset_mac_len(skb);
  76
  77         return 0;
  78 }
  79
  80 static int pop_vlan(struct sk_buff *skb)
  81 {
  82         __be16 tci;
  83         int err;
  84
  85         if (likely(vlan_tx_tag_present(skb))) {
  86                 vlan_set_tci(skb, 0);
  87         } else {
  88                 if (unlikely(skb->protocol != htons(ETH_P_8021Q) ||
  89                              skb->len < VLAN_ETH_HLEN))
  90                         return 0;
  91
  92                 err = __pop_vlan_tci(skb, &tci);
  93                 if (err)
  94                         return err;
  95         }
  96         /* move next vlan tag to hw accel tag */
  97         if (likely(skb->protocol != htons(ETH_P_8021Q) ||
  98                    skb->len < VLAN_ETH_HLEN))
  99                 return 0;
 100
 101         err = __pop_vlan_tci(skb, &tci);
 102         if (unlikely(err))
 103                 return err;
 104
 105         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(tci));
 106         return 0;
 107 }
 108
 109 static int push_vlan(struct sk_buff *skb, const struct ovs_action_push_vlan *vlan)
 110 {
 111         if (unlikely(vlan_tx_tag_present(skb))) {
 112                 u16 current_tag;
 113
 114                 /* push down current VLAN tag */
 115                 current_tag = vlan_tx_tag_get(skb);
 116
 117                 if (!__vlan_put_tag(skb, skb->vlan_proto, current_tag))
 118                         return -ENOMEM;
 119
 120                 if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
 121                         skb->csum = csum_add(skb->csum, csum_partial(skb->data
 122                                         + (2 * ETH_ALEN), VLAN_HLEN, 0));
 123
 124         }
 125         __vlan_hwaccel_put_tag(skb, vlan->vlan_tpid, ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
 126         return 0;
 127 }
 128
 129 static int set_eth_addr(struct sk_buff *skb,
 130                         const struct ovs_key_ethernet *eth_key)
 131 {
 132         int err;
 133         err = make_writable(skb, ETH_HLEN);
 134         if (unlikely(err))
 135                 return err;
 136
 137         if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
 138                 skb->csum = csum_sub(skb->csum, csum_partial(eth_hdr(skb),
 139                                                              ETH_ALEN * 2, 0));
 140
 141         memcpy(eth_hdr(skb)->h_source, eth_key->eth_src, ETH_ALEN);
 142         memcpy(eth_hdr(skb)->h_dest, eth_key->eth_dst, ETH_ALEN);
 143
 144         if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
 145                 skb->csum = csum_add(skb->csum, csum_partial(eth_hdr(skb),
 146                                                              ETH_ALEN * 2, 0));
 147
 148         return 0;
 149 }
 150
 151 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 152                                 __be32 *addr, __be32 new_addr)
 153 {
 154         int transport_len = skb->len - skb_transport_offset(skb);
 155
 156         if (nh->protocol == IPPROTO_TCP) {
 157                 if (likely(transport_len >= sizeof(struct tcphdr)))
 158                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 159                                                  *addr, new_addr, 1);
 160         } else if (nh->protocol == IPPROTO_UDP) {
 161                 if (likely(transport_len >= sizeof(struct udphdr))) {
 162                         struct udphdr *uh = udp_hdr(skb);
 163
 164                         if (uh->check ||
 165                             get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
 166                                 inet_proto_csum_replace4(&uh->check, skb,
 167                                                          *addr, new_addr, 1);
 168                                 if (!uh->check)
 169                                         uh->check = CSUM_MANGLED_0;
 170                         }
 171                 }
 172         }
 173
 174         csum_replace4(&nh->check, *addr, new_addr);
 175         skb_clear_rxhash(skb);
 176         *addr = new_addr;
 177 }
 178
 179 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 180                                  __be32 addr[4], const __be32 new_addr[4])
 181 {
 182         int transport_len = skb->len - skb_transport_offset(skb);
 183
 184         if (l4_proto == IPPROTO_TCP) {
 185                 if (likely(transport_len >= sizeof(struct tcphdr)))
 186                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 187                                                   addr, new_addr, 1);
 188         } else if (l4_proto == IPPROTO_UDP) {
 189                 if (likely(transport_len >= sizeof(struct udphdr))) {
 190                         struct udphdr *uh = udp_hdr(skb);
 191
 192                         if (uh->check ||
 193                             get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
 194                                 inet_proto_csum_replace16(&uh->check, skb,
 195                                                           addr, new_addr, 1);
 196                                 if (!uh->check)
 197                                         uh->check = CSUM_MANGLED_0;
 198                         }
 199                 }
 200         }
 201 }
 202
 203 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 204                           __be32 addr[4], const __be32 new_addr[4],
 205                           bool recalculate_csum)
 206 {
 207         if (recalculate_csum)
 208                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 209
 210         skb_clear_rxhash(skb);
 211         memcpy(addr, new_addr, sizeof(__be32[4]));
 212 }
 213
 214 static void set_ipv6_tc(struct ipv6hdr *nh, u8 tc)
 215 {
 216         nh->priority = tc >> 4;
 217         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0x0F) | ((tc & 0x0F) << 4);
 218 }
 219
 220 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl)
 221 {
 222         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0xF0) | (fl & 0x000F0000) >> 16;
 223         nh->flow_lbl[1] = (fl & 0x0000FF00) >> 8;
 224         nh->flow_lbl[2] = fl & 0x000000FF;
 225 }
 226
 227 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl)
 228 {
 229         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 230         nh->ttl = new_ttl;
 231 }
 232
 233 static int set_ipv4(struct sk_buff *skb, const struct ovs_key_ipv4 *ipv4_key)
 234 {
 235         struct iphdr *nh;
 236         int err;
 237
 238         err = make_writable(skb, skb_network_offset(skb) +
 239                                  sizeof(struct iphdr));
 240         if (unlikely(err))
 241                 return err;
 242
 243         nh = ip_hdr(skb);
 244
 245         if (ipv4_key->ipv4_src != nh->saddr)
 246                 set_ip_addr(skb, nh, &nh->saddr, ipv4_key->ipv4_src);
 247
 248         if (ipv4_key->ipv4_dst != nh->daddr)
 249                 set_ip_addr(skb, nh, &nh->daddr, ipv4_key->ipv4_dst);
 250
 251         if (ipv4_key->ipv4_tos != nh->tos)
 252                 ipv4_change_dsfield(nh, 0, ipv4_key->ipv4_tos);
 253
 254         if (ipv4_key->ipv4_ttl != nh->ttl)
 255                 set_ip_ttl(skb, nh, ipv4_key->ipv4_ttl);
 256
 257         return 0;
 258 }
 259
 260 static int set_ipv6(struct sk_buff *skb, const struct ovs_key_ipv6 *ipv6_key)
 261 {
 262         struct ipv6hdr *nh;
 263         int err;
 264         __be32 *saddr;
 265         __be32 *daddr;
 266
 267         err = make_writable(skb, skb_network_offset(skb) +
 268                             sizeof(struct ipv6hdr));
 269         if (unlikely(err))
 270                 return err;
 271
 272         nh = ipv6_hdr(skb);
 273         saddr = (__be32 *)&nh->saddr;
 274         daddr = (__be32 *)&nh->daddr;
 275
 276         if (memcmp(ipv6_key->ipv6_src, saddr, sizeof(ipv6_key->ipv6_src)))
 277                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, saddr,
 278                               ipv6_key->ipv6_src, true);
 279
 280         if (memcmp(ipv6_key->ipv6_dst, daddr, sizeof(ipv6_key->ipv6_dst))) {
 281                 unsigned int offset = 0;
 282                 int flags = OVS_IP6T_FH_F_SKIP_RH;
 283                 bool recalc_csum = true;
 284
 285                 if (ipv6_ext_hdr(nh->nexthdr))
 286                         recalc_csum = ipv6_find_hdr(skb, &offset,
 287                                                     NEXTHDR_ROUTING, NULL,
 288                                                     &flags) != NEXTHDR_ROUTING;
 289
 290                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, daddr,
 291                               ipv6_key->ipv6_dst, recalc_csum);
 292         }
 293
 294         set_ipv6_tc(nh, ipv6_key->ipv6_tclass);
 295         set_ipv6_fl(nh, ntohl(ipv6_key->ipv6_label));
 296         nh->hop_limit = ipv6_key->ipv6_hlimit;
 297
 298         return 0;
 299 }
 300
 301 /* Must follow make_writable() since that can move the skb data. */
 302 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 303                          __be16 new_port, __sum16 *check)
 304 {
 305         inet_proto_csum_replace2(check, skb, *port, new_port, 0);
 306         *port = new_port;
 307         skb_clear_rxhash(skb);
 308 }
 309
 310 static void set_udp_port(struct sk_buff *skb, __be16 *port, __be16 new_port)
 311 {
 312         struct udphdr *uh = udp_hdr(skb);
 313
 314         if (uh->check && get_ip_summed(skb) != OVS_CSUM_PARTIAL) {
 315                 set_tp_port(skb, port, new_port, &uh->check);
 316
 317                 if (!uh->check)
 318                         uh->check = CSUM_MANGLED_0;
 319         } else {
 320                 *port = new_port;
 321                 skb_clear_rxhash(skb);
 322         }
 323 }
 324
 325 static int set_udp(struct sk_buff *skb, const struct ovs_key_udp *udp_port_key)
 326 {
 327         struct udphdr *uh;
 328         int err;
 329
 330         err = make_writable(skb, skb_transport_offset(skb) +
 331                                  sizeof(struct udphdr));
 332         if (unlikely(err))
 333                 return err;
 334
 335         uh = udp_hdr(skb);
 336         if (udp_port_key->udp_src != uh->source)
 337                 set_udp_port(skb, &uh->source, udp_port_key->udp_src);
 338
 339         if (udp_port_key->udp_dst != uh->dest)
 340                 set_udp_port(skb, &uh->dest, udp_port_key->udp_dst);
 341
 342         return 0;
 343 }
 344
 345 static int set_tcp(struct sk_buff *skb, const struct ovs_key_tcp *tcp_port_key)
 346 {
 347         struct tcphdr *th;
 348         int err;
 349
 350         err = make_writable(skb, skb_transport_offset(skb) +
 351                                  sizeof(struct tcphdr));
 352         if (unlikely(err))
 353                 return err;
 354
 355         th = tcp_hdr(skb);
 356         if (tcp_port_key->tcp_src != th->source)
 357                 set_tp_port(skb, &th->source, tcp_port_key->tcp_src, &th->check);
 358
 359         if (tcp_port_key->tcp_dst != th->dest)
 360                 set_tp_port(skb, &th->dest, tcp_port_key->tcp_dst, &th->check);
 361
 362         return 0;
 363 }
 364
 365 static int set_sctp(struct sk_buff *skb,
 366                      const struct ovs_key_sctp *sctp_port_key)
 367 {
 368         struct sctphdr *sh;
 369         int err;
 370         unsigned int sctphoff = skb_transport_offset(skb);
 371
 372         err = make_writable(skb, sctphoff + sizeof(struct sctphdr));
 373         if (unlikely(err))
 374                 return err;
 375
 376         sh = sctp_hdr(skb);
 377         if (sctp_port_key->sctp_src != sh->source ||
 378             sctp_port_key->sctp_dst != sh->dest) {
 379                 __le32 old_correct_csum, new_csum, old_csum;
 380
 381                 old_csum = sh->checksum;
 382                 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 383
 384                 sh->source = sctp_port_key->sctp_src;
 385                 sh->dest = sctp_port_key->sctp_dst;
 386
 387                 new_csum = sctp_compute_cksum(skb, sctphoff);
 388
 389                 /* Carry any checksum errors through. */
 390                 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 391
 392                 skb_clear_rxhash(skb);
 393         }
 394
 395         return 0;
 396 }
 397
 398 static int do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
 399 {
 400         struct vport *vport;
 401
 402         if (unlikely(!skb))
 403                 return -ENOMEM;
 404
 405         vport = ovs_vport_rcu(dp, out_port);
 406         if (unlikely(!vport)) {
 407                 kfree_skb(skb);
 408                 return -ENODEV;
 409         }
 410
 411         ovs_vport_send(vport, skb);
 412         return 0;
 413 }
 414
 415 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 416                             const struct nlattr *attr)
 417 {
 418         struct dp_upcall_info upcall;
 419         const struct nlattr *a;
 420         int rem;
 421
 422         BUG_ON(!OVS_CB(skb)->pkt_key);
 423
 424         upcall.cmd = OVS_PACKET_CMD_ACTION;
 425         upcall.key = OVS_CB(skb)->pkt_key;
 426         upcall.userdata = NULL;
 427         upcall.portid = 0;
 428
 429         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 430                  a = nla_next(a, &rem)) {
 431                 switch (nla_type(a)) {
 432                 case OVS_USERSPACE_ATTR_USERDATA:
 433                         upcall.userdata = a;
 434                         break;
 435
 436                 case OVS_USERSPACE_ATTR_PID:
 437                         upcall.portid = nla_get_u32(a);
 438                         break;
 439                 }
 440         }
 441
 442         return ovs_dp_upcall(dp, skb, &upcall);
 443 }
 444
 445 static int sample(struct datapath *dp, struct sk_buff *skb,
 446                   const struct nlattr *attr)
 447 {
 448         const struct nlattr *acts_list = NULL;
 449         const struct nlattr *a;
 450         int rem;
 451
 452         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 453                  a = nla_next(a, &rem)) {
 454                 switch (nla_type(a)) {
 455                 case OVS_SAMPLE_ATTR_PROBABILITY:
 456                         if (net_random() >= nla_get_u32(a))
 457                                 return 0;
 458                         break;
 459
 460                 case OVS_SAMPLE_ATTR_ACTIONS:
 461                         acts_list = a;
 462                         break;
 463                 }
 464         }
 465
 466         return do_execute_actions(dp, skb, nla_data(acts_list),
 467                                   nla_len(acts_list), true);
 468 }
 469
 470 static int execute_set_action(struct sk_buff *skb,
 471                                  const struct nlattr *nested_attr)
 472 {
 473         int err = 0;
 474
 475         switch (nla_type(nested_attr)) {
 476         case OVS_KEY_ATTR_PRIORITY:
 477                 skb->priority = nla_get_u32(nested_attr);
 478                 break;
 479
 480         case OVS_KEY_ATTR_SKB_MARK:
 481                 skb->mark = nla_get_u32(nested_attr);
 482                 break;
 483
 484         case OVS_KEY_ATTR_IPV4_TUNNEL:
 485                 OVS_CB(skb)->tun_key = nla_data(nested_attr);
 486                 break;
 487
 488         case OVS_KEY_ATTR_ETHERNET:
 489                 err = set_eth_addr(skb, nla_data(nested_attr));
 490                 break;
 491
 492         case OVS_KEY_ATTR_IPV4:
 493                 err = set_ipv4(skb, nla_data(nested_attr));
 494                 break;
 495
 496         case OVS_KEY_ATTR_IPV6:
 497                 err = set_ipv6(skb, nla_data(nested_attr));
 498                 break;
 499
 500         case OVS_KEY_ATTR_TCP:
 501                 err = set_tcp(skb, nla_data(nested_attr));
 502                 break;
 503
 504         case OVS_KEY_ATTR_UDP:
 505                 err = set_udp(skb, nla_data(nested_attr));
 506                 break;
 507
 508         case OVS_KEY_ATTR_SCTP:
 509                 err = set_sctp(skb, nla_data(nested_attr));
 510                 break;
 511         }
 512
 513         return err;
 514 }
 515
 516 /* Execute a list of actions against 'skb'. */
 517 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 518                         const struct nlattr *attr, int len, bool keep_skb)
 519 {
 520         /* Every output action needs a separate clone of 'skb', but the common
 521          * case is just a single output action, so that doing a clone and
 522          * then freeing the original skbuff is wasteful.  So the following code
 523          * is slightly obscure just to avoid that. */
 524         int prev_port = -1;
 525         const struct nlattr *a;
 526         int rem;
 527
 528         for (a = attr, rem = len; rem > 0;
 529              a = nla_next(a, &rem)) {
 530                 int err = 0;
 531
 532                 if (prev_port != -1) {
 533                         do_output(dp, skb_clone(skb, GFP_ATOMIC), prev_port);
 534                         prev_port = -1;
 535                 }
 536
 537                 switch (nla_type(a)) {
 538                 case OVS_ACTION_ATTR_OUTPUT:
 539                         prev_port = nla_get_u32(a);
 540                         break;
 541
 542                 case OVS_ACTION_ATTR_USERSPACE:
 543                         output_userspace(dp, skb, a);
 544                         break;
 545
 546                 case OVS_ACTION_ATTR_PUSH_VLAN:
 547                         err = push_vlan(skb, nla_data(a));
 548                         if (unlikely(err)) /* skb already freed. */
 549                                 return err;
 550                         break;
 551
 552                 case OVS_ACTION_ATTR_POP_VLAN:
 553                         err = pop_vlan(skb);
 554                         break;
 555
 556                 case OVS_ACTION_ATTR_SET:
 557                         err = execute_set_action(skb, nla_data(a));
 558                         break;
 559
 560                 case OVS_ACTION_ATTR_SAMPLE:
 561                         err = sample(dp, skb, a);
 562                         break;
 563                 }
 564
 565                 if (unlikely(err)) {
 566                         kfree_skb(skb);
 567                         return err;
 568                 }
 569         }
 570
 571         if (prev_port != -1) {
 572                 if (keep_skb)
 573                         skb = skb_clone(skb, GFP_ATOMIC);
 574
 575                 do_output(dp, skb, prev_port);
 576         } else if (!keep_skb)
 577                 consume_skb(skb);
 578
 579         return 0;
 580 }
 581
 582 /* We limit the number of times that we pass into execute_actions()
 583  * to avoid blowing out the stack in the event that we have a loop. */
 584 #define MAX_LOOPS 4
 585
 586 struct loop_counter {
 587         u8 count;               /* Count. */
 588         bool looping;           /* Loop detected? */
 589 };
 590
 591 static DEFINE_PER_CPU(struct loop_counter, loop_counters);
 592
 593 static int loop_suppress(struct datapath *dp, struct sw_flow_actions *actions)
 594 {
 595         if (net_ratelimit())
 596                 pr_warn("%s: flow looped %d times, dropping\n",
 597                                 ovs_dp_name(dp), MAX_LOOPS);
 598         actions->actions_len = 0;
 599         return -ELOOP;
 600 }
 601
 602 /* Execute a list of actions against 'skb'. */
 603 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb)
 604 {
 605         struct sw_flow_actions *acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
 606         struct loop_counter *loop;
 607         int error;
 608
 609         /* Check whether we've looped too much. */
 610         loop = &__get_cpu_var(loop_counters);
 611         if (unlikely(++loop->count > MAX_LOOPS))
 612                 loop->looping = true;
 613         if (unlikely(loop->looping)) {
 614                 error = loop_suppress(dp, acts);
 615                 kfree_skb(skb);
 616                 goto out_loop;
 617         }
 618
 619         OVS_CB(skb)->tun_key = NULL;
 620         error = do_execute_actions(dp, skb, acts->actions,
 621                                          acts->actions_len, false);
 622
 623         /* Check whether sub-actions looped too much. */
 624         if (unlikely(loop->looping))
 625                 error = loop_suppress(dp, acts);
 626
 627 out_loop:
 628         /* Decrement loop counter. */
 629         if (!--loop->count)
 630                 loop->looping = false;
 631
 632         return error;
 633 }