datapath/actions.c

   1 /*
   2  * Copyright (c) 2007-2014 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 "datapath.h"
  38 #include "gso.h"
  39 #include "mpls.h"
  40 #include "vlan.h"
  41 #include "vport.h"
  42
  43 struct deferred_action {
  44         struct sk_buff *skb;
  45         const struct nlattr *actions;
  46
  47         /* Store pkt_key clone when creating deferred action. */
  48         struct sw_flow_key pkt_key;
  49 };
  50
  51 #define DEFERRED_ACTION_FIFO_SIZE 10
  52 struct action_fifo {
  53         int head;
  54         int tail;
  55         /* Deferred action fifo queue storage. */
  56         struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
  57 };
  58
  59 static struct action_fifo __percpu *action_fifos;
  60 #define EXEC_ACTIONS_LEVEL_LIMIT 4   /* limit used to detect packet
  61                                         looping by the network stack */
  62 static DEFINE_PER_CPU(int, exec_actions_level);
  63
  64 static void action_fifo_init(struct action_fifo *fifo)
  65 {
  66         fifo->head = 0;
  67         fifo->tail = 0;
  68 }
  69
  70 static bool action_fifo_is_empty(struct action_fifo *fifo)
  71 {
  72         return (fifo->head == fifo->tail);
  73 }
  74
  75 static struct deferred_action *
  76 action_fifo_get(struct action_fifo *fifo)
  77 {
  78         if (action_fifo_is_empty(fifo))
  79                 return NULL;
  80
  81         return &fifo->fifo[fifo->tail++];
  82 }
  83
  84 static struct deferred_action *
  85 action_fifo_put(struct action_fifo *fifo)
  86 {
  87         if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
  88                 return NULL;
  89
  90         return &fifo->fifo[fifo->head++];
  91 }
  92
  93 static void flow_key_clone(struct sk_buff *skb, struct sw_flow_key *new_key)
  94 {
  95         *new_key = *OVS_CB(skb)->pkt_key;
  96         OVS_CB(skb)->pkt_key = new_key;
  97 }
  98
  99 /* Return true if fifo is not full */
 100 static bool add_deferred_actions(struct sk_buff *skb,
 101                                  const struct nlattr *attr)
 102 {
 103         struct action_fifo *fifo;
 104         struct deferred_action *da;
 105
 106         fifo = this_cpu_ptr(action_fifos);
 107         da = action_fifo_put(fifo);
 108         if (da) {
 109                 da->skb = skb;
 110                 da->actions = attr;
 111                 flow_key_clone(skb, &da->pkt_key);
 112         }
 113
 114         return (da != NULL);
 115 }
 116
 117 static void flow_key_set_recirc_id(struct sk_buff *skb, u32 recirc_id)
 118 {
 119         OVS_CB(skb)->pkt_key->recirc_id = recirc_id;
 120 }
 121
 122 static void flow_key_set_priority(struct sk_buff *skb, u32 priority)
 123 {
 124         OVS_CB(skb)->pkt_key->phy.priority = priority;
 125 }
 126
 127 static void flow_key_set_skb_mark(struct sk_buff *skb, u32 skb_mark)
 128 {
 129         OVS_CB(skb)->pkt_key->phy.skb_mark = skb_mark;
 130 }
 131
 132 static void flow_key_set_eth_src(struct sk_buff *skb, const u8 addr[])
 133 {
 134         ether_addr_copy(OVS_CB(skb)->pkt_key->eth.src, addr);
 135 }
 136
 137 static void flow_key_set_eth_dst(struct sk_buff *skb, const u8 addr[])
 138 {
 139         ether_addr_copy(OVS_CB(skb)->pkt_key->eth.dst, addr);
 140 }
 141
 142 static void flow_key_set_vlan_tci(struct sk_buff *skb, __be16 tci)
 143 {
 144         OVS_CB(skb)->pkt_key->eth.tci = tci;
 145 }
 146
 147 static void flow_key_set_mpls_top_lse(struct sk_buff *skb, __be32 top_lse)
 148 {
 149         OVS_CB(skb)->pkt_key->mpls.top_lse = top_lse;
 150 }
 151
 152 static void flow_key_set_ipv4_src(struct sk_buff *skb, __be32 addr)
 153 {
 154         OVS_CB(skb)->pkt_key->ipv4.addr.src = addr;
 155 }
 156
 157 static void flow_key_set_ipv4_dst(struct sk_buff *skb, __be32 addr)
 158 {
 159         OVS_CB(skb)->pkt_key->ipv4.addr.src = addr;
 160 }
 161
 162 static void flow_key_set_ip_tos(struct sk_buff *skb, u8 tos)
 163 {
 164         OVS_CB(skb)->pkt_key->ip.tos = tos;
 165 }
 166
 167 static void flow_key_set_ip_ttl(struct sk_buff *skb, u8 ttl)
 168 {
 169         OVS_CB(skb)->pkt_key->ip.ttl = ttl;
 170 }
 171
 172 static void flow_key_set_ipv6_src(struct sk_buff *skb,
 173                                   const __be32 addr[4])
 174 {
 175         memcpy(&OVS_CB(skb)->pkt_key->ipv6.addr.src, addr, sizeof(__be32[4]));
 176 }
 177
 178 static void flow_key_set_ipv6_dst(struct sk_buff *skb,
 179                                   const __be32 addr[4])
 180 {
 181         memcpy(&OVS_CB(skb)->pkt_key->ipv6.addr.dst, addr, sizeof(__be32[4]));
 182 }
 183
 184 static void flow_key_set_ipv6_fl(struct sk_buff *skb,
 185                                  const struct ipv6hdr *nh)
 186 {
 187         OVS_CB(skb)->pkt_key->ipv6.label = *(__be32 *)nh &
 188                                            htonl(IPV6_FLOWINFO_FLOWLABEL);
 189 }
 190
 191 static void flow_key_set_tp_src(struct sk_buff *skb, __be16 port)
 192 {
 193         OVS_CB(skb)->pkt_key->tp.src = port;
 194 }
 195
 196 static void flow_key_set_tp_dst(struct sk_buff *skb, __be16 port)
 197 {
 198         OVS_CB(skb)->pkt_key->tp.dst = port;
 199 }
 200
 201 static void invalidate_skb_flow_key(struct sk_buff *skb)
 202 {
 203         OVS_CB(skb)->pkt_key->eth.type = htons(0);
 204 }
 205
 206 static bool is_skb_flow_key_valid(struct sk_buff *skb)
 207 {
 208         return !!OVS_CB(skb)->pkt_key->eth.type;
 209 }
 210
 211 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 212                               const struct nlattr *attr, int len);
 213
 214 static int make_writable(struct sk_buff *skb, int write_len)
 215 {
 216         if (!pskb_may_pull(skb, write_len))
 217                 return -ENOMEM;
 218
 219         if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
 220                 return 0;
 221
 222         return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
 223 }
 224
 225 /* The end of the mac header.
 226  *
 227  * For non-MPLS skbs this will correspond to the network header.
 228  * For MPLS skbs it will be before the network_header as the MPLS
 229  * label stack lies between the end of the mac header and the network
 230  * header. That is, for MPLS skbs the end of the mac header
 231  * is the top of the MPLS label stack.
 232  */
 233 static unsigned char *mac_header_end(const struct sk_buff *skb)
 234 {
 235         return skb_mac_header(skb) + skb->mac_len;
 236 }
 237
 238 static int push_mpls(struct sk_buff *skb,
 239                      const struct ovs_action_push_mpls *mpls)
 240 {
 241         __be32 *new_mpls_lse;
 242         struct ethhdr *hdr;
 243
 244         if (skb_cow_head(skb, MPLS_HLEN) < 0)
 245                 return -ENOMEM;
 246
 247         skb_push(skb, MPLS_HLEN);
 248         memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
 249                 skb->mac_len);
 250         skb_reset_mac_header(skb);
 251
 252         new_mpls_lse = (__be32 *)mac_header_end(skb);
 253         *new_mpls_lse = mpls->mpls_lse;
 254
 255         if (skb->ip_summed == CHECKSUM_COMPLETE)
 256                 skb->csum = csum_add(skb->csum, csum_partial(new_mpls_lse,
 257                                                              MPLS_HLEN, 0));
 258
 259         hdr = eth_hdr(skb);
 260         hdr->h_proto = mpls->mpls_ethertype;
 261         if (!ovs_skb_get_inner_protocol(skb))
 262                 ovs_skb_set_inner_protocol(skb, skb->protocol);
 263         skb->protocol = mpls->mpls_ethertype;
 264         invalidate_skb_flow_key(skb);
 265         return 0;
 266 }
 267
 268 static int pop_mpls(struct sk_buff *skb, const __be16 ethertype)
 269 {
 270         struct ethhdr *hdr;
 271         int err;
 272
 273         err = make_writable(skb, skb->mac_len + MPLS_HLEN);
 274         if (unlikely(err))
 275                 return err;
 276
 277         if (skb->ip_summed == CHECKSUM_COMPLETE)
 278                 skb->csum = csum_sub(skb->csum,
 279                                      csum_partial(mac_header_end(skb),
 280                                                   MPLS_HLEN, 0));
 281
 282         memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
 283                 skb->mac_len);
 284
 285         __skb_pull(skb, MPLS_HLEN);
 286         skb_reset_mac_header(skb);
 287
 288         /* mac_header_end() is used to locate the ethertype
 289          * field correctly in the presence of VLAN tags.
 290          */
 291         hdr = (struct ethhdr *)(mac_header_end(skb) - ETH_HLEN);
 292         hdr->h_proto = ethertype;
 293         if (eth_p_mpls(skb->protocol))
 294                 skb->protocol = ethertype;
 295         invalidate_skb_flow_key(skb);
 296         return 0;
 297 }
 298
 299 static int set_mpls(struct sk_buff *skb, const __be32 *mpls_lse)
 300 {
 301         __be32 *stack = (__be32 *)mac_header_end(skb);
 302         int err;
 303
 304         err = make_writable(skb, skb->mac_len + MPLS_HLEN);
 305         if (unlikely(err))
 306                 return err;
 307
 308         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 309                 __be32 diff[] = { ~(*stack), *mpls_lse };
 310                 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
 311                                           ~skb->csum);
 312         }
 313
 314         *stack = *mpls_lse;
 315         flow_key_set_mpls_top_lse(skb, *stack);
 316         return 0;
 317 }
 318
 319 /* remove VLAN header from packet and update csum accordingly. */
 320 static int __pop_vlan_tci(struct sk_buff *skb, __be16 *current_tci)
 321 {
 322         struct vlan_hdr *vhdr;
 323         int err;
 324
 325         err = make_writable(skb, VLAN_ETH_HLEN);
 326         if (unlikely(err))
 327                 return err;
 328
 329         if (skb->ip_summed == CHECKSUM_COMPLETE)
 330                 skb->csum = csum_sub(skb->csum, csum_partial(skb->data
 331                                         + (2 * ETH_ALEN), VLAN_HLEN, 0));
 332
 333         vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
 334         *current_tci = vhdr->h_vlan_TCI;
 335
 336         memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
 337         __skb_pull(skb, VLAN_HLEN);
 338
 339         vlan_set_encap_proto(skb, vhdr);
 340         skb->mac_header += VLAN_HLEN;
 341         /* Update mac_len for subsequent MPLS actions */
 342         skb->mac_len -= VLAN_HLEN;
 343
 344         return 0;
 345 }
 346
 347 static int pop_vlan(struct sk_buff *skb)
 348 {
 349         __be16 tci;
 350         int err;
 351
 352         if (likely(vlan_tx_tag_present(skb))) {
 353                 vlan_set_tci(skb, 0);
 354         } else {
 355                 if (unlikely(skb->protocol != htons(ETH_P_8021Q) ||
 356                              skb->len < VLAN_ETH_HLEN))
 357                         return 0;
 358
 359                 err = __pop_vlan_tci(skb, &tci);
 360                 if (err)
 361                         return err;
 362         }
 363         /* move next vlan tag to hw accel tag */
 364         if (likely(skb->protocol != htons(ETH_P_8021Q) ||
 365                    skb->len < VLAN_ETH_HLEN)) {
 366                 flow_key_set_vlan_tci(skb, 0);
 367                 return 0;
 368         }
 369
 370         invalidate_skb_flow_key(skb);
 371         err = __pop_vlan_tci(skb, &tci);
 372         if (unlikely(err))
 373                 return err;
 374
 375         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(tci));
 376         return 0;
 377 }
 378
 379 static int push_vlan(struct sk_buff *skb, const struct ovs_action_push_vlan *vlan)
 380 {
 381         if (unlikely(vlan_tx_tag_present(skb))) {
 382                 u16 current_tag;
 383
 384                 /* push down current VLAN tag */
 385                 current_tag = vlan_tx_tag_get(skb);
 386
 387                 if (!__vlan_put_tag(skb, skb->vlan_proto, current_tag))
 388                         return -ENOMEM;
 389
 390                 /* Update mac_len for subsequent MPLS actions */
 391                 skb->mac_len += VLAN_HLEN;
 392
 393                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 394                         skb->csum = csum_add(skb->csum, csum_partial(skb->data
 395                                         + (2 * ETH_ALEN), VLAN_HLEN, 0));
 396
 397                 invalidate_skb_flow_key(skb);
 398         } else {
 399                 flow_key_set_vlan_tci(skb,  vlan->vlan_tci);
 400         }
 401         __vlan_hwaccel_put_tag(skb, vlan->vlan_tpid, ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
 402         return 0;
 403 }
 404
 405 static int set_eth_addr(struct sk_buff *skb,
 406                         const struct ovs_key_ethernet *eth_key)
 407 {
 408         int err;
 409         err = make_writable(skb, ETH_HLEN);
 410         if (unlikely(err))
 411                 return err;
 412
 413         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 414
 415         ether_addr_copy(eth_hdr(skb)->h_source, eth_key->eth_src);
 416         ether_addr_copy(eth_hdr(skb)->h_dest, eth_key->eth_dst);
 417
 418         ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 419
 420         flow_key_set_eth_src(skb, eth_key->eth_src);
 421         flow_key_set_eth_dst(skb, eth_key->eth_dst);
 422         return 0;
 423 }
 424
 425 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 426                         __be32 *addr, __be32 new_addr)
 427 {
 428         int transport_len = skb->len - skb_transport_offset(skb);
 429
 430         if (nh->protocol == IPPROTO_TCP) {
 431                 if (likely(transport_len >= sizeof(struct tcphdr)))
 432                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 433                                                  *addr, new_addr, 1);
 434         } else if (nh->protocol == IPPROTO_UDP) {
 435                 if (likely(transport_len >= sizeof(struct udphdr))) {
 436                         struct udphdr *uh = udp_hdr(skb);
 437
 438                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 439                                 inet_proto_csum_replace4(&uh->check, skb,
 440                                                          *addr, new_addr, 1);
 441                                 if (!uh->check)
 442                                         uh->check = CSUM_MANGLED_0;
 443                         }
 444                 }
 445         }
 446
 447         csum_replace4(&nh->check, *addr, new_addr);
 448         skb_clear_hash(skb);
 449         *addr = new_addr;
 450 }
 451
 452 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 453                                  __be32 addr[4], const __be32 new_addr[4])
 454 {
 455         int transport_len = skb->len - skb_transport_offset(skb);
 456
 457         if (l4_proto == NEXTHDR_TCP) {
 458                 if (likely(transport_len >= sizeof(struct tcphdr)))
 459                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 460                                                   addr, new_addr, 1);
 461         } else if (l4_proto == NEXTHDR_UDP) {
 462                 if (likely(transport_len >= sizeof(struct udphdr))) {
 463                         struct udphdr *uh = udp_hdr(skb);
 464
 465                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 466                                 inet_proto_csum_replace16(&uh->check, skb,
 467                                                           addr, new_addr, 1);
 468                                 if (!uh->check)
 469                                         uh->check = CSUM_MANGLED_0;
 470                         }
 471                 }
 472         } else if (l4_proto == NEXTHDR_ICMP) {
 473                 if (likely(transport_len >= sizeof(struct icmp6hdr)))
 474                         inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 475                                                   skb, addr, new_addr, 1);
 476         }
 477 }
 478
 479 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 480                           __be32 addr[4], const __be32 new_addr[4],
 481                           bool recalculate_csum)
 482 {
 483         if (likely(recalculate_csum))
 484                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 485
 486         skb_clear_hash(skb);
 487         memcpy(addr, new_addr, sizeof(__be32[4]));
 488 }
 489
 490 static void set_ipv6_tc(struct ipv6hdr *nh, u8 tc)
 491 {
 492         nh->priority = tc >> 4;
 493         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0x0F) | ((tc & 0x0F) << 4);
 494 }
 495
 496 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl)
 497 {
 498         nh->flow_lbl[0] = (nh->flow_lbl[0] & 0xF0) | (fl & 0x000F0000) >> 16;
 499         nh->flow_lbl[1] = (fl & 0x0000FF00) >> 8;
 500         nh->flow_lbl[2] = fl & 0x000000FF;
 501 }
 502
 503 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl)
 504 {
 505         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 506         nh->ttl = new_ttl;
 507 }
 508
 509 static int set_ipv4(struct sk_buff *skb, const struct ovs_key_ipv4 *ipv4_key)
 510 {
 511         struct iphdr *nh;
 512         int err;
 513
 514         err = make_writable(skb, skb_network_offset(skb) +
 515                                  sizeof(struct iphdr));
 516         if (unlikely(err))
 517                 return err;
 518
 519         nh = ip_hdr(skb);
 520
 521         if (ipv4_key->ipv4_src != nh->saddr) {
 522                 set_ip_addr(skb, nh, &nh->saddr, ipv4_key->ipv4_src);
 523                 flow_key_set_ipv4_src(skb, ipv4_key->ipv4_src);
 524         }
 525
 526         if (ipv4_key->ipv4_dst != nh->daddr) {
 527                 set_ip_addr(skb, nh, &nh->daddr, ipv4_key->ipv4_dst);
 528                 flow_key_set_ipv4_dst(skb, ipv4_key->ipv4_dst);
 529         }
 530
 531         if (ipv4_key->ipv4_tos != nh->tos) {
 532                 ipv4_change_dsfield(nh, 0, ipv4_key->ipv4_tos);
 533                 flow_key_set_ip_tos(skb, nh->tos);
 534         }
 535
 536         if (ipv4_key->ipv4_ttl != nh->ttl) {
 537                 set_ip_ttl(skb, nh, ipv4_key->ipv4_ttl);
 538                 flow_key_set_ip_ttl(skb, ipv4_key->ipv4_ttl);
 539         }
 540
 541         return 0;
 542 }
 543
 544 static int set_ipv6(struct sk_buff *skb, const struct ovs_key_ipv6 *ipv6_key)
 545 {
 546         struct ipv6hdr *nh;
 547         int err;
 548         __be32 *saddr;
 549         __be32 *daddr;
 550
 551         err = make_writable(skb, skb_network_offset(skb) +
 552                             sizeof(struct ipv6hdr));
 553         if (unlikely(err))
 554                 return err;
 555
 556         nh = ipv6_hdr(skb);
 557         saddr = (__be32 *)&nh->saddr;
 558         daddr = (__be32 *)&nh->daddr;
 559
 560         if (memcmp(ipv6_key->ipv6_src, saddr, sizeof(ipv6_key->ipv6_src))) {
 561                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, saddr,
 562                               ipv6_key->ipv6_src, true);
 563                 flow_key_set_ipv6_src(skb, ipv6_key->ipv6_src);
 564         }
 565
 566         if (memcmp(ipv6_key->ipv6_dst, daddr, sizeof(ipv6_key->ipv6_dst))) {
 567                 unsigned int offset = 0;
 568                 int flags = OVS_IP6T_FH_F_SKIP_RH;
 569                 bool recalc_csum = true;
 570
 571                 if (ipv6_ext_hdr(nh->nexthdr))
 572                         recalc_csum = ipv6_find_hdr(skb, &offset,
 573                                                     NEXTHDR_ROUTING, NULL,
 574                                                     &flags) != NEXTHDR_ROUTING;
 575
 576                 set_ipv6_addr(skb, ipv6_key->ipv6_proto, daddr,
 577                               ipv6_key->ipv6_dst, recalc_csum);
 578                 flow_key_set_ipv6_dst(skb, ipv6_key->ipv6_dst);
 579         }
 580
 581         set_ipv6_tc(nh, ipv6_key->ipv6_tclass);
 582         flow_key_set_ip_tos(skb, ipv6_get_dsfield(nh));
 583
 584         set_ipv6_fl(nh, ntohl(ipv6_key->ipv6_label));
 585         flow_key_set_ipv6_fl(skb, nh);
 586
 587         nh->hop_limit = ipv6_key->ipv6_hlimit;
 588         flow_key_set_ip_ttl(skb, ipv6_key->ipv6_hlimit);
 589         return 0;
 590 }
 591
 592 /* Must follow make_writable() since that can move the skb data. */
 593 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 594                          __be16 new_port, __sum16 *check)
 595 {
 596         inet_proto_csum_replace2(check, skb, *port, new_port, 0);
 597         *port = new_port;
 598         skb_clear_hash(skb);
 599 }
 600
 601 static void set_udp_port(struct sk_buff *skb, __be16 *port, __be16 new_port)
 602 {
 603         struct udphdr *uh = udp_hdr(skb);
 604
 605         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 606                 set_tp_port(skb, port, new_port, &uh->check);
 607
 608                 if (!uh->check)
 609                         uh->check = CSUM_MANGLED_0;
 610         } else {
 611                 *port = new_port;
 612                 skb_clear_hash(skb);
 613         }
 614 }
 615
 616 static int set_udp(struct sk_buff *skb, const struct ovs_key_udp *udp_port_key)
 617 {
 618         struct udphdr *uh;
 619         int err;
 620
 621         err = make_writable(skb, skb_transport_offset(skb) +
 622                                  sizeof(struct udphdr));
 623         if (unlikely(err))
 624                 return err;
 625
 626         uh = udp_hdr(skb);
 627         if (udp_port_key->udp_src != uh->source) {
 628                 set_udp_port(skb, &uh->source, udp_port_key->udp_src);
 629                 flow_key_set_tp_src(skb, udp_port_key->udp_src);
 630         }
 631
 632         if (udp_port_key->udp_dst != uh->dest) {
 633                 set_udp_port(skb, &uh->dest, udp_port_key->udp_dst);
 634                 flow_key_set_tp_dst(skb, udp_port_key->udp_dst);
 635         }
 636
 637         return 0;
 638 }
 639
 640 static int set_tcp(struct sk_buff *skb, const struct ovs_key_tcp *tcp_port_key)
 641 {
 642         struct tcphdr *th;
 643         int err;
 644
 645         err = make_writable(skb, skb_transport_offset(skb) +
 646                                  sizeof(struct tcphdr));
 647         if (unlikely(err))
 648                 return err;
 649
 650         th = tcp_hdr(skb);
 651         if (tcp_port_key->tcp_src != th->source) {
 652                 set_tp_port(skb, &th->source, tcp_port_key->tcp_src, &th->check);
 653                 flow_key_set_tp_src(skb, tcp_port_key->tcp_src);
 654         }
 655
 656         if (tcp_port_key->tcp_dst != th->dest) {
 657                 set_tp_port(skb, &th->dest, tcp_port_key->tcp_dst, &th->check);
 658                 flow_key_set_tp_dst(skb, tcp_port_key->tcp_dst);
 659         }
 660
 661         return 0;
 662 }
 663
 664 static int set_sctp(struct sk_buff *skb,
 665                     const struct ovs_key_sctp *sctp_port_key)
 666 {
 667         struct sctphdr *sh;
 668         int err;
 669         unsigned int sctphoff = skb_transport_offset(skb);
 670
 671         err = make_writable(skb, sctphoff + sizeof(struct sctphdr));
 672         if (unlikely(err))
 673                 return err;
 674
 675         sh = sctp_hdr(skb);
 676         if (sctp_port_key->sctp_src != sh->source ||
 677             sctp_port_key->sctp_dst != sh->dest) {
 678                 __le32 old_correct_csum, new_csum, old_csum;
 679
 680                 old_csum = sh->checksum;
 681                 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 682
 683                 sh->source = sctp_port_key->sctp_src;
 684                 sh->dest = sctp_port_key->sctp_dst;
 685
 686                 new_csum = sctp_compute_cksum(skb, sctphoff);
 687
 688                 /* Carry any checksum errors through. */
 689                 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 690
 691                 skb_clear_hash(skb);
 692                 flow_key_set_tp_src(skb, sctp_port_key->sctp_src);
 693                 flow_key_set_tp_dst(skb, sctp_port_key->sctp_dst);
 694         }
 695
 696         return 0;
 697 }
 698
 699 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
 700 {
 701         struct vport *vport = ovs_vport_rcu(dp, out_port);
 702
 703         if (likely(vport))
 704                 ovs_vport_send(vport, skb);
 705         else
 706                 kfree_skb(skb);
 707 }
 708
 709 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 710                             const struct nlattr *attr)
 711 {
 712         struct dp_upcall_info upcall;
 713         const struct nlattr *a;
 714         int rem;
 715         struct ovs_tunnel_info info;
 716
 717         upcall.cmd = OVS_PACKET_CMD_ACTION;
 718         upcall.userdata = NULL;
 719         upcall.portid = 0;
 720         upcall.egress_tun_info = NULL;
 721
 722         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 723                  a = nla_next(a, &rem)) {
 724                 switch (nla_type(a)) {
 725                 case OVS_USERSPACE_ATTR_USERDATA:
 726                         upcall.userdata = a;
 727                         break;
 728
 729                 case OVS_USERSPACE_ATTR_PID:
 730                         upcall.portid = nla_get_u32(a);
 731                         break;
 732
 733                 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
 734                         /* Get out tunnel info. */
 735                         struct vport *vport;
 736
 737                         vport = ovs_vport_rcu(dp, nla_get_u32(a));
 738                         if (vport) {
 739                                 int err;
 740
 741                                 err = ovs_vport_get_egress_tun_info(vport, skb,
 742                                                                     &info);
 743                                 if (!err)
 744                                         upcall.egress_tun_info = &info;
 745                         }
 746                         break;
 747                 }
 748
 749                 } /* End of switch. */
 750         }
 751
 752         return ovs_dp_upcall(dp, skb, &upcall);
 753 }
 754
 755 static bool last_action(const struct nlattr *a, int rem)
 756 {
 757         return a->nla_len == rem;
 758 }
 759
 760 static int sample(struct datapath *dp, struct sk_buff *skb,
 761                   const struct nlattr *attr)
 762 {
 763         const struct nlattr *acts_list = NULL;
 764         const struct nlattr *a;
 765         int rem;
 766
 767         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 768                  a = nla_next(a, &rem)) {
 769                 switch (nla_type(a)) {
 770                 case OVS_SAMPLE_ATTR_PROBABILITY:
 771                         if (prandom_u32() >= nla_get_u32(a))
 772                                 return 0;
 773                         break;
 774
 775                 case OVS_SAMPLE_ATTR_ACTIONS:
 776                         acts_list = a;
 777                         break;
 778                 }
 779         }
 780
 781         rem = nla_len(acts_list);
 782         a = nla_data(acts_list);
 783
 784         /* Actions list is empty, do nothing */
 785         if (unlikely(!rem))
 786                 return 0;
 787
 788         /* The only known usage of sample action is having a single user-space
 789          * action. Treat this usage as a special case.
 790          * The output_userspace() should clone the skb to be sent to the
 791          * user space. This skb will be consumed by its caller. */
 792         if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
 793                    last_action(a, rem)))
 794                 return output_userspace(dp, skb, a);
 795
 796         skb = skb_clone(skb, GFP_ATOMIC);
 797         if (!skb)
 798                 /* Skip the sample action when out of memory. */
 799                 return 0;
 800
 801         if (!add_deferred_actions(skb, a)) {
 802                 if (net_ratelimit())
 803                         pr_warn("%s: deferred actions limit reached, dropping sample action\n",
 804                                 ovs_dp_name(dp));
 805
 806                 kfree_skb(skb);
 807         }
 808
 809         return 0;
 810 }
 811
 812 static void execute_hash(struct sk_buff *skb, const struct nlattr *attr)
 813 {
 814         struct sw_flow_key *key = OVS_CB(skb)->pkt_key;
 815         struct ovs_action_hash *hash_act = nla_data(attr);
 816         u32 hash = 0;
 817
 818         /* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
 819         hash = skb_get_hash(skb);
 820         hash = jhash_1word(hash, hash_act->hash_basis);
 821         if (!hash)
 822                 hash = 0x1;
 823
 824         key->ovs_flow_hash = hash;
 825 }
 826
 827 static int execute_set_action(struct sk_buff *skb,
 828                               const struct nlattr *nested_attr)
 829 {
 830         int err = 0;
 831
 832         switch (nla_type(nested_attr)) {
 833         case OVS_KEY_ATTR_PRIORITY:
 834                 skb->priority = nla_get_u32(nested_attr);
 835                 flow_key_set_priority(skb, skb->priority);
 836                 break;
 837
 838         case OVS_KEY_ATTR_SKB_MARK:
 839                 skb->mark = nla_get_u32(nested_attr);
 840                 flow_key_set_skb_mark(skb, skb->mark);
 841                 break;
 842
 843         case OVS_KEY_ATTR_TUNNEL_INFO:
 844                 OVS_CB(skb)->egress_tun_info = nla_data(nested_attr);
 845                 break;
 846
 847         case OVS_KEY_ATTR_ETHERNET:
 848                 err = set_eth_addr(skb, nla_data(nested_attr));
 849                 break;
 850
 851         case OVS_KEY_ATTR_IPV4:
 852                 err = set_ipv4(skb, nla_data(nested_attr));
 853                 break;
 854
 855         case OVS_KEY_ATTR_IPV6:
 856                 err = set_ipv6(skb, nla_data(nested_attr));
 857                 break;
 858
 859         case OVS_KEY_ATTR_TCP:
 860                 err = set_tcp(skb, nla_data(nested_attr));
 861                 break;
 862
 863         case OVS_KEY_ATTR_UDP:
 864                 err = set_udp(skb, nla_data(nested_attr));
 865                 break;
 866
 867         case OVS_KEY_ATTR_SCTP:
 868                 err = set_sctp(skb, nla_data(nested_attr));
 869                 break;
 870
 871         case OVS_KEY_ATTR_MPLS:
 872                 err = set_mpls(skb, nla_data(nested_attr));
 873                 break;
 874         }
 875
 876         return err;
 877 }
 878
 879 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
 880                           const struct nlattr *a, int rem)
 881 {
 882         if (!is_skb_flow_key_valid(skb)) {
 883                 int err;
 884
 885                 err = ovs_flow_key_update(skb, OVS_CB(skb)->pkt_key);
 886                 if (err)
 887                         return err;
 888
 889         }
 890         BUG_ON(!is_skb_flow_key_valid(skb));
 891
 892         if (!last_action(a, rem)) {
 893                 /* Recirc action is the not the last action
 894                  * of the action list, need to clone the skb. */
 895                 skb = skb_clone(skb, GFP_ATOMIC);
 896
 897                 /* Skip the recirc action when out of memory, but
 898                  * continue on with the rest of the action list. */
 899                 if (!skb)
 900                         return 0;
 901         }
 902
 903         if (add_deferred_actions(skb, NULL)) {
 904                 flow_key_set_recirc_id(skb, nla_get_u32(a));
 905         } else {
 906                 kfree_skb(skb);
 907
 908                 if (net_ratelimit())
 909                         pr_warn("%s: deferred action limit reached, drop recirc action\n",
 910                                 ovs_dp_name(dp));
 911         }
 912
 913         return 0;
 914 }
 915
 916 /* Execute a list of actions against 'skb'. */
 917 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 918                                const struct nlattr *attr, int len)
 919 {
 920         /* Every output action needs a separate clone of 'skb', but the common
 921          * case is just a single output action, so that doing a clone and
 922          * then freeing the original skbuff is wasteful.  So the following code
 923          * is slightly obscure just to avoid that. */
 924         int prev_port = -1;
 925         const struct nlattr *a;
 926         int rem;
 927
 928         for (a = attr, rem = len; rem > 0;
 929              a = nla_next(a, &rem)) {
 930                 int err = 0;
 931
 932                 if (unlikely(prev_port != -1)) {
 933                         struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
 934
 935                         if (out_skb)
 936                                 do_output(dp, out_skb, prev_port);
 937
 938                         prev_port = -1;
 939                 }
 940
 941                 switch (nla_type(a)) {
 942                 case OVS_ACTION_ATTR_OUTPUT:
 943                         prev_port = nla_get_u32(a);
 944                         break;
 945
 946                 case OVS_ACTION_ATTR_USERSPACE:
 947                         output_userspace(dp, skb, a);
 948                         break;
 949
 950                 case OVS_ACTION_ATTR_HASH:
 951                         execute_hash(skb, a);
 952                         break;
 953
 954                 case OVS_ACTION_ATTR_PUSH_MPLS:
 955                         err = push_mpls(skb, nla_data(a));
 956                         break;
 957
 958                 case OVS_ACTION_ATTR_POP_MPLS:
 959                         err = pop_mpls(skb, nla_get_be16(a));
 960                         break;
 961
 962                 case OVS_ACTION_ATTR_PUSH_VLAN:
 963                         err = push_vlan(skb, nla_data(a));
 964                         if (unlikely(err)) /* skb already freed. */
 965                                 return err;
 966                         break;
 967
 968                 case OVS_ACTION_ATTR_POP_VLAN:
 969                         err = pop_vlan(skb);
 970                         break;
 971
 972                 case OVS_ACTION_ATTR_RECIRC:
 973                         err = execute_recirc(dp, skb, a, rem);
 974                         if (last_action(a, rem)) {
 975                                 /* If this is the last action, the skb has
 976                                  * been consumed or freed.
 977                                  * Return immediately. */
 978                                 return err;
 979                         }
 980                         break;
 981
 982                 case OVS_ACTION_ATTR_SET:
 983                         err = execute_set_action(skb, nla_data(a));
 984                         break;
 985
 986                 case OVS_ACTION_ATTR_SAMPLE:
 987                         err = sample(dp, skb, a);
 988                         break;
 989                 }
 990
 991                 if (unlikely(err)) {
 992                         kfree_skb(skb);
 993                         return err;
 994                 }
 995         }
 996
 997         if (prev_port != -1)
 998                 do_output(dp, skb, prev_port);
 999         else
1000                 consume_skb(skb);
1001
1002         return 0;
1003 }
1004
1005 static void process_deferred_actions(struct datapath *dp)
1006 {
1007         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1008
1009         /* Do not touch the FIFO in case there is no deferred actions. */
1010         if (action_fifo_is_empty(fifo))
1011                 return;
1012
1013         /* Finishing executing all deferred actions. */
1014         do {
1015                 struct deferred_action *da = action_fifo_get(fifo);
1016                 struct sk_buff *skb = da->skb;
1017                 const struct nlattr *actions = da->actions;
1018
1019                 if (actions)
1020                         do_execute_actions(dp, skb, actions,
1021                                            nla_len(actions));
1022                 else
1023                         ovs_dp_process_packet(skb);
1024         } while (!action_fifo_is_empty(fifo));
1025
1026         /* Reset FIFO for the next packet.  */
1027         action_fifo_init(fifo);
1028 }
1029
1030 /* Execute a list of actions against 'skb'. */
1031 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1032                         struct sw_flow_actions *acts)
1033 {
1034         int level = this_cpu_read(exec_actions_level);
1035         int err;
1036
1037         if (unlikely(level >= EXEC_ACTIONS_LEVEL_LIMIT)) {
1038                 if (net_ratelimit())
1039                         pr_warn("%s: packet loop detected, dropping.\n",
1040                                 ovs_dp_name(dp));
1041
1042                 kfree_skb(skb);
1043                 return -ELOOP;
1044         }
1045
1046         this_cpu_inc(exec_actions_level);
1047
1048         err = do_execute_actions(dp, skb, acts->actions, acts->actions_len);
1049
1050         if (!level)
1051                 process_deferred_actions(dp);
1052
1053         this_cpu_dec(exec_actions_level);
1054
1055         /* This return status currently does not reflect the errors
1056          * encounted during deferred actions execution. Probably needs to
1057          * be fixed in the future. */
1058         return err;
1059 }
1060
1061 int action_fifos_init(void)
1062 {
1063         action_fifos = alloc_percpu(struct action_fifo);
1064         if (!action_fifos)
1065                 return -ENOMEM;
1066
1067         return 0;
1068 }
1069
1070 void action_fifos_exit(void)
1071 {
1072         free_percpu(action_fifos);
1073 }