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