datapath/actions.c

   1 /*
   2  * Copyright (c) 2007-2015 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
  32 #include <net/ip.h>
  33 #include <net/ipv6.h>
  34 #include <net/checksum.h>
  35 #include <net/dsfield.h>
  36 #include <net/mpls.h>
  37 #include <net/sctp/checksum.h>
  38
  39 #include "datapath.h"
  40 #include "gso.h"
  41 #include "vlan.h"
  42 #include "vport.h"
  43
  44 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
  45                               struct sw_flow_key *key,
  46                               const struct nlattr *attr, int len);
  47
  48 struct deferred_action {
  49         struct sk_buff *skb;
  50         const struct nlattr *actions;
  51
  52         /* Store pkt_key clone when creating deferred action. */
  53         struct sw_flow_key pkt_key;
  54 };
  55
  56 #define DEFERRED_ACTION_FIFO_SIZE 10
  57 struct action_fifo {
  58         int head;
  59         int tail;
  60         /* Deferred action fifo queue storage. */
  61         struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
  62 };
  63
  64 static struct action_fifo __percpu *action_fifos;
  65 #define EXEC_ACTIONS_LEVEL_LIMIT 4   /* limit used to detect packet
  66                                       * looping by the network stack
  67                                       */
  68 static DEFINE_PER_CPU(int, exec_actions_level);
  69
  70 static void action_fifo_init(struct action_fifo *fifo)
  71 {
  72         fifo->head = 0;
  73         fifo->tail = 0;
  74 }
  75
  76 static bool action_fifo_is_empty(const struct action_fifo *fifo)
  77 {
  78         return (fifo->head == fifo->tail);
  79 }
  80
  81 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
  82 {
  83         if (action_fifo_is_empty(fifo))
  84                 return NULL;
  85
  86         return &fifo->fifo[fifo->tail++];
  87 }
  88
  89 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
  90 {
  91         if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
  92                 return NULL;
  93
  94         return &fifo->fifo[fifo->head++];
  95 }
  96
  97 /* Return queue entry if fifo is not full */
  98 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
  99                                                     const struct sw_flow_key *key,
 100                                                     const struct nlattr *attr)
 101 {
 102         struct action_fifo *fifo;
 103         struct deferred_action *da;
 104
 105         fifo = this_cpu_ptr(action_fifos);
 106         da = action_fifo_put(fifo);
 107         if (da) {
 108                 da->skb = skb;
 109                 da->actions = attr;
 110                 da->pkt_key = *key;
 111         }
 112
 113         return da;
 114 }
 115
 116 static void invalidate_flow_key(struct sw_flow_key *key)
 117 {
 118         key->eth.type = htons(0);
 119 }
 120
 121 static bool is_flow_key_valid(const struct sw_flow_key *key)
 122 {
 123         return !!key->eth.type;
 124 }
 125
 126 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 127                      const struct ovs_action_push_mpls *mpls)
 128 {
 129         __be32 *new_mpls_lse;
 130         struct ethhdr *hdr;
 131
 132         /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
 133         if (skb_encapsulation(skb))
 134                 return -ENOTSUPP;
 135
 136         if (skb_cow_head(skb, MPLS_HLEN) < 0)
 137                 return -ENOMEM;
 138
 139         skb_push(skb, MPLS_HLEN);
 140         memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
 141                 skb->mac_len);
 142         skb_reset_mac_header(skb);
 143
 144         new_mpls_lse = (__be32 *)skb_mpls_header(skb);
 145         *new_mpls_lse = mpls->mpls_lse;
 146
 147         if (skb->ip_summed == CHECKSUM_COMPLETE)
 148                 skb->csum = csum_add(skb->csum, csum_partial(new_mpls_lse,
 149                                                              MPLS_HLEN, 0));
 150
 151         hdr = eth_hdr(skb);
 152         hdr->h_proto = mpls->mpls_ethertype;
 153         if (!ovs_skb_get_inner_protocol(skb))
 154                 ovs_skb_set_inner_protocol(skb, skb->protocol);
 155         skb->protocol = mpls->mpls_ethertype;
 156
 157         invalidate_flow_key(key);
 158         return 0;
 159 }
 160
 161 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 162                     const __be16 ethertype)
 163 {
 164         struct ethhdr *hdr;
 165         int err;
 166
 167         err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
 168         if (unlikely(err))
 169                 return err;
 170
 171         skb_postpull_rcsum(skb, skb_mpls_header(skb), MPLS_HLEN);
 172
 173         memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
 174                 skb->mac_len);
 175
 176         __skb_pull(skb, MPLS_HLEN);
 177         skb_reset_mac_header(skb);
 178
 179         /* skb_mpls_header() is used to locate the ethertype
 180          * field correctly in the presence of VLAN tags.
 181          */
 182         hdr = (struct ethhdr *)(skb_mpls_header(skb) - ETH_HLEN);
 183         hdr->h_proto = ethertype;
 184         if (eth_p_mpls(skb->protocol))
 185                 skb->protocol = ethertype;
 186
 187         invalidate_flow_key(key);
 188         return 0;
 189 }
 190
 191 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
 192                     const __be32 *mpls_lse, const __be32 *mask)
 193 {
 194         __be32 *stack;
 195         __be32 lse;
 196         int err;
 197
 198         err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
 199         if (unlikely(err))
 200                 return err;
 201
 202         stack = (__be32 *)skb_mpls_header(skb);
 203         lse = OVS_MASKED(*stack, *mpls_lse, *mask);
 204         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 205                 __be32 diff[] = { ~(*stack), lse };
 206
 207                 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
 208                                           ~skb->csum);
 209         }
 210
 211         *stack = lse;
 212         flow_key->mpls.top_lse = lse;
 213         return 0;
 214 }
 215
 216 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 217 {
 218         int err;
 219
 220         err = skb_vlan_pop(skb);
 221         if (skb_vlan_tag_present(skb))
 222                 invalidate_flow_key(key);
 223         else
 224                 key->eth.tci = 0;
 225         return err;
 226 }
 227
 228 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
 229                      const struct ovs_action_push_vlan *vlan)
 230 {
 231         if (skb_vlan_tag_present(skb))
 232                 invalidate_flow_key(key);
 233         else
 234                 key->eth.tci = vlan->vlan_tci;
 235         return skb_vlan_push(skb, vlan->vlan_tpid,
 236                              ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
 237 }
 238
 239 /* 'src' is already properly masked. */
 240 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
 241 {
 242         u16 *dst = (u16 *)dst_;
 243         const u16 *src = (const u16 *)src_;
 244         const u16 *mask = (const u16 *)mask_;
 245
 246         OVS_SET_MASKED(dst[0], src[0], mask[0]);
 247         OVS_SET_MASKED(dst[1], src[1], mask[1]);
 248         OVS_SET_MASKED(dst[2], src[2], mask[2]);
 249 }
 250
 251 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
 252                         const struct ovs_key_ethernet *key,
 253                         const struct ovs_key_ethernet *mask)
 254 {
 255         int err;
 256
 257         err = skb_ensure_writable(skb, ETH_HLEN);
 258         if (unlikely(err))
 259                 return err;
 260
 261         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 262
 263         ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
 264                                mask->eth_src);
 265         ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
 266                                mask->eth_dst);
 267
 268         ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 269
 270         ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
 271         ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
 272         return 0;
 273 }
 274
 275 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
 276                                   __be32 addr, __be32 new_addr)
 277 {
 278         int transport_len = skb->len - skb_transport_offset(skb);
 279
 280         if (nh->frag_off & htons(IP_OFFSET))
 281                 return;
 282
 283         if (nh->protocol == IPPROTO_TCP) {
 284                 if (likely(transport_len >= sizeof(struct tcphdr)))
 285                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 286                                                  addr, new_addr, 1);
 287         } else if (nh->protocol == IPPROTO_UDP) {
 288                 if (likely(transport_len >= sizeof(struct udphdr))) {
 289                         struct udphdr *uh = udp_hdr(skb);
 290
 291                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 292                                 inet_proto_csum_replace4(&uh->check, skb,
 293                                                          addr, new_addr, 1);
 294                                 if (!uh->check)
 295                                         uh->check = CSUM_MANGLED_0;
 296                         }
 297                 }
 298         }
 299
 300 }
 301
 302 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 303                         __be32 *addr, __be32 new_addr)
 304 {
 305         update_ip_l4_checksum(skb, nh, *addr, new_addr);
 306         csum_replace4(&nh->check, *addr, new_addr);
 307         skb_clear_hash(skb);
 308         *addr = new_addr;
 309 }
 310
 311 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 312                                  __be32 addr[4], const __be32 new_addr[4])
 313 {
 314         int transport_len = skb->len - skb_transport_offset(skb);
 315
 316         if (l4_proto == NEXTHDR_TCP) {
 317                 if (likely(transport_len >= sizeof(struct tcphdr)))
 318                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 319                                                   addr, new_addr, 1);
 320         } else if (l4_proto == NEXTHDR_UDP) {
 321                 if (likely(transport_len >= sizeof(struct udphdr))) {
 322                         struct udphdr *uh = udp_hdr(skb);
 323
 324                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 325                                 inet_proto_csum_replace16(&uh->check, skb,
 326                                                           addr, new_addr, 1);
 327                                 if (!uh->check)
 328                                         uh->check = CSUM_MANGLED_0;
 329                         }
 330                 }
 331         } else if (l4_proto == NEXTHDR_ICMP) {
 332                 if (likely(transport_len >= sizeof(struct icmp6hdr)))
 333                         inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 334                                                   skb, addr, new_addr, 1);
 335         }
 336 }
 337
 338 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
 339                            const __be32 mask[4], __be32 masked[4])
 340 {
 341         masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
 342         masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
 343         masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
 344         masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
 345 }
 346
 347 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 348                           __be32 addr[4], const __be32 new_addr[4],
 349                           bool recalculate_csum)
 350 {
 351         if (likely(recalculate_csum))
 352                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 353
 354         skb_clear_hash(skb);
 355         memcpy(addr, new_addr, sizeof(__be32[4]));
 356 }
 357
 358 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
 359 {
 360         /* Bits 21-24 are always unmasked, so this retains their values. */
 361         OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
 362         OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
 363         OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
 364 }
 365
 366 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
 367                        u8 mask)
 368 {
 369         new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
 370
 371         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 372         nh->ttl = new_ttl;
 373 }
 374
 375 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
 376                     const struct ovs_key_ipv4 *key,
 377                     const struct ovs_key_ipv4 *mask)
 378 {
 379         struct iphdr *nh;
 380         __be32 new_addr;
 381         int err;
 382
 383         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 384                                   sizeof(struct iphdr));
 385         if (unlikely(err))
 386                 return err;
 387
 388         nh = ip_hdr(skb);
 389
 390         /* Setting an IP addresses is typically only a side effect of
 391          * matching on them in the current userspace implementation, so it
 392          * makes sense to check if the value actually changed.
 393          */
 394         if (mask->ipv4_src) {
 395                 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
 396
 397                 if (unlikely(new_addr != nh->saddr)) {
 398                         set_ip_addr(skb, nh, &nh->saddr, new_addr);
 399                         flow_key->ipv4.addr.src = new_addr;
 400                 }
 401         }
 402         if (mask->ipv4_dst) {
 403                 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
 404
 405                 if (unlikely(new_addr != nh->daddr)) {
 406                         set_ip_addr(skb, nh, &nh->daddr, new_addr);
 407                         flow_key->ipv4.addr.dst = new_addr;
 408                 }
 409         }
 410         if (mask->ipv4_tos) {
 411                 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
 412                 flow_key->ip.tos = nh->tos;
 413         }
 414         if (mask->ipv4_ttl) {
 415                 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
 416                 flow_key->ip.ttl = nh->ttl;
 417         }
 418
 419         return 0;
 420 }
 421
 422 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
 423 {
 424         return !!(addr[0] | addr[1] | addr[2] | addr[3]);
 425 }
 426
 427 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
 428                     const struct ovs_key_ipv6 *key,
 429                     const struct ovs_key_ipv6 *mask)
 430 {
 431         struct ipv6hdr *nh;
 432         int err;
 433
 434         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 435                                   sizeof(struct ipv6hdr));
 436         if (unlikely(err))
 437                 return err;
 438
 439         nh = ipv6_hdr(skb);
 440
 441         /* Setting an IP addresses is typically only a side effect of
 442          * matching on them in the current userspace implementation, so it
 443          * makes sense to check if the value actually changed.
 444          */
 445         if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
 446                 __be32 *saddr = (__be32 *)&nh->saddr;
 447                 __be32 masked[4];
 448
 449                 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
 450
 451                 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
 452                         set_ipv6_addr(skb, key->ipv6_proto, saddr, masked,
 453                                       true);
 454                         memcpy(&flow_key->ipv6.addr.src, masked,
 455                                sizeof(flow_key->ipv6.addr.src));
 456                 }
 457         }
 458         if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
 459                 unsigned int offset = 0;
 460                 int flags = IP6_FH_F_SKIP_RH;
 461                 bool recalc_csum = true;
 462                 __be32 *daddr = (__be32 *)&nh->daddr;
 463                 __be32 masked[4];
 464
 465                 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
 466
 467                 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
 468                         if (ipv6_ext_hdr(nh->nexthdr))
 469                                 recalc_csum = (ipv6_find_hdr(skb, &offset,
 470                                                              NEXTHDR_ROUTING,
 471                                                              NULL, &flags)
 472                                                != NEXTHDR_ROUTING);
 473
 474                         set_ipv6_addr(skb, key->ipv6_proto, daddr, masked,
 475                                       recalc_csum);
 476                         memcpy(&flow_key->ipv6.addr.dst, masked,
 477                                sizeof(flow_key->ipv6.addr.dst));
 478                 }
 479         }
 480         if (mask->ipv6_tclass) {
 481                 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
 482                 flow_key->ip.tos = ipv6_get_dsfield(nh);
 483         }
 484         if (mask->ipv6_label) {
 485                 set_ipv6_fl(nh, ntohl(key->ipv6_label),
 486                             ntohl(mask->ipv6_label));
 487                 flow_key->ipv6.label =
 488                     *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 489         }
 490         if (mask->ipv6_hlimit) {
 491                 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
 492                                mask->ipv6_hlimit);
 493                 flow_key->ip.ttl = nh->hop_limit;
 494         }
 495         return 0;
 496 }
 497
 498 /* Must follow skb_ensure_writable() since that can move the skb data. */
 499 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 500                         __be16 new_port, __sum16 *check)
 501 {
 502         inet_proto_csum_replace2(check, skb, *port, new_port, 0);
 503         *port = new_port;
 504 }
 505
 506 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 507                    const struct ovs_key_udp *key,
 508                    const struct ovs_key_udp *mask)
 509 {
 510         struct udphdr *uh;
 511         __be16 src, dst;
 512         int err;
 513
 514         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 515                                   sizeof(struct udphdr));
 516         if (unlikely(err))
 517                 return err;
 518
 519         uh = udp_hdr(skb);
 520         /* Either of the masks is non-zero, so do not bother checking them. */
 521         src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
 522         dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
 523
 524         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 525                 if (likely(src != uh->source)) {
 526                         set_tp_port(skb, &uh->source, src, &uh->check);
 527                         flow_key->tp.src = src;
 528                 }
 529                 if (likely(dst != uh->dest)) {
 530                         set_tp_port(skb, &uh->dest, dst, &uh->check);
 531                         flow_key->tp.dst = dst;
 532                 }
 533
 534                 if (unlikely(!uh->check))
 535                         uh->check = CSUM_MANGLED_0;
 536         } else {
 537                 uh->source = src;
 538                 uh->dest = dst;
 539                 flow_key->tp.src = src;
 540                 flow_key->tp.dst = dst;
 541         }
 542
 543         skb_clear_hash(skb);
 544
 545         return 0;
 546 }
 547
 548 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 549                    const struct ovs_key_tcp *key,
 550                    const struct ovs_key_tcp *mask)
 551 {
 552         struct tcphdr *th;
 553         __be16 src, dst;
 554         int err;
 555
 556         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 557                                   sizeof(struct tcphdr));
 558         if (unlikely(err))
 559                 return err;
 560
 561         th = tcp_hdr(skb);
 562         src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
 563         if (likely(src != th->source)) {
 564                 set_tp_port(skb, &th->source, src, &th->check);
 565                 flow_key->tp.src = src;
 566         }
 567         dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
 568         if (likely(dst != th->dest)) {
 569                 set_tp_port(skb, &th->dest, dst, &th->check);
 570                 flow_key->tp.dst = dst;
 571         }
 572         skb_clear_hash(skb);
 573
 574         return 0;
 575 }
 576
 577 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 578                     const struct ovs_key_sctp *key,
 579                     const struct ovs_key_sctp *mask)
 580 {
 581         unsigned int sctphoff = skb_transport_offset(skb);
 582         struct sctphdr *sh;
 583         __le32 old_correct_csum, new_csum, old_csum;
 584         int err;
 585
 586         err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
 587         if (unlikely(err))
 588                 return err;
 589
 590         sh = sctp_hdr(skb);
 591         old_csum = sh->checksum;
 592         old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 593
 594         sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
 595         sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
 596
 597         new_csum = sctp_compute_cksum(skb, sctphoff);
 598
 599         /* Carry any checksum errors through. */
 600         sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 601
 602         skb_clear_hash(skb);
 603         flow_key->tp.src = sh->source;
 604         flow_key->tp.dst = sh->dest;
 605
 606         return 0;
 607 }
 608
 609 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
 610 {
 611         struct vport *vport = ovs_vport_rcu(dp, out_port);
 612
 613         if (likely(vport))
 614                 ovs_vport_send(vport, skb);
 615         else
 616                 kfree_skb(skb);
 617 }
 618 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 619                             struct sw_flow_key *key, const struct nlattr *attr,
 620                             const struct nlattr *actions, int actions_len)
 621 {
 622         struct ip_tunnel_info info;
 623         struct dp_upcall_info upcall;
 624         const struct nlattr *a;
 625         int rem;
 626
 627         memset(&upcall, 0, sizeof(upcall));
 628         upcall.cmd = OVS_PACKET_CMD_ACTION;
 629
 630         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 631                  a = nla_next(a, &rem)) {
 632                 switch (nla_type(a)) {
 633                 case OVS_USERSPACE_ATTR_USERDATA:
 634                         upcall.userdata = a;
 635                         break;
 636
 637                 case OVS_USERSPACE_ATTR_PID:
 638                         upcall.portid = nla_get_u32(a);
 639                         break;
 640
 641                 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
 642                         /* Get out tunnel info. */
 643                         struct vport *vport;
 644
 645                         vport = ovs_vport_rcu(dp, nla_get_u32(a));
 646                         if (vport) {
 647                                 int err;
 648
 649                                 upcall.egress_tun_info = &info;
 650                                 err = ovs_vport_get_egress_tun_info(vport, skb,
 651                                                                     &upcall);
 652                                 if (err)
 653                                         upcall.egress_tun_info = NULL;
 654                         }
 655
 656                         break;
 657                 }
 658
 659                 case OVS_USERSPACE_ATTR_ACTIONS: {
 660                         /* Include actions. */
 661                         upcall.actions = actions;
 662                         upcall.actions_len = actions_len;
 663                         break;
 664                 }
 665
 666                 } /* End of switch. */
 667         }
 668
 669         return ovs_dp_upcall(dp, skb, key, &upcall);
 670 }
 671
 672 static int sample(struct datapath *dp, struct sk_buff *skb,
 673                   struct sw_flow_key *key, const struct nlattr *attr,
 674                   const struct nlattr *actions, int actions_len)
 675 {
 676         const struct nlattr *acts_list = NULL;
 677         const struct nlattr *a;
 678         int rem;
 679
 680         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 681                  a = nla_next(a, &rem)) {
 682                 u32 probability;
 683
 684                 switch (nla_type(a)) {
 685                 case OVS_SAMPLE_ATTR_PROBABILITY:
 686                         probability = nla_get_u32(a);
 687                         if (!probability || prandom_u32() > probability)
 688                                 return 0;
 689                         break;
 690
 691                 case OVS_SAMPLE_ATTR_ACTIONS:
 692                         acts_list = a;
 693                         break;
 694                 }
 695         }
 696
 697         rem = nla_len(acts_list);
 698         a = nla_data(acts_list);
 699
 700         /* Actions list is empty, do nothing */
 701         if (unlikely(!rem))
 702                 return 0;
 703
 704         /* The only known usage of sample action is having a single user-space
 705          * action. Treat this usage as a special case.
 706          * The output_userspace() should clone the skb to be sent to the
 707          * user space. This skb will be consumed by its caller.
 708          */
 709         if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
 710                    nla_is_last(a, rem)))
 711                 return output_userspace(dp, skb, key, a, actions, actions_len);
 712
 713         skb = skb_clone(skb, GFP_ATOMIC);
 714         if (!skb)
 715                 /* Skip the sample action when out of memory. */
 716                 return 0;
 717
 718         if (!add_deferred_actions(skb, key, a)) {
 719                 if (net_ratelimit())
 720                         pr_warn("%s: deferred actions limit reached, dropping sample action\n",
 721                                 ovs_dp_name(dp));
 722
 723                 kfree_skb(skb);
 724         }
 725         return 0;
 726 }
 727
 728 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
 729                          const struct nlattr *attr)
 730 {
 731         struct ovs_action_hash *hash_act = nla_data(attr);
 732         u32 hash = 0;
 733
 734         /* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
 735         hash = skb_get_hash(skb);
 736         hash = jhash_1word(hash, hash_act->hash_basis);
 737         if (!hash)
 738                 hash = 0x1;
 739
 740         key->ovs_flow_hash = hash;
 741 }
 742
 743 static int execute_set_action(struct sk_buff *skb,
 744                               struct sw_flow_key *flow_key,
 745                               const struct nlattr *a)
 746 {
 747         /* Only tunnel set execution is supported without a mask. */
 748         if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
 749                 struct ovs_tunnel_info *tun = nla_data(a);
 750
 751                 ovs_skb_dst_drop(skb);
 752                 ovs_dst_hold((struct dst_entry *)tun->tun_dst);
 753                 ovs_skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
 754                 return 0;
 755         }
 756
 757         return -EINVAL;
 758 }
 759
 760 /* Mask is at the midpoint of the data. */
 761 #define get_mask(a, type) ((const type)nla_data(a) + 1)
 762
 763 static int execute_masked_set_action(struct sk_buff *skb,
 764                                      struct sw_flow_key *flow_key,
 765                                      const struct nlattr *a)
 766 {
 767         int err = 0;
 768
 769         switch (nla_type(a)) {
 770         case OVS_KEY_ATTR_PRIORITY:
 771                 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
 772                                *get_mask(a, u32 *));
 773                 flow_key->phy.priority = skb->priority;
 774                 break;
 775
 776         case OVS_KEY_ATTR_SKB_MARK:
 777                 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
 778                 flow_key->phy.skb_mark = skb->mark;
 779                 break;
 780
 781         case OVS_KEY_ATTR_TUNNEL_INFO:
 782                 /* Masked data not supported for tunnel. */
 783                 err = -EINVAL;
 784                 break;
 785
 786         case OVS_KEY_ATTR_ETHERNET:
 787                 err = set_eth_addr(skb, flow_key, nla_data(a),
 788                                    get_mask(a, struct ovs_key_ethernet *));
 789                 break;
 790
 791         case OVS_KEY_ATTR_IPV4:
 792                 err = set_ipv4(skb, flow_key, nla_data(a),
 793                                get_mask(a, struct ovs_key_ipv4 *));
 794                 break;
 795
 796         case OVS_KEY_ATTR_IPV6:
 797                 err = set_ipv6(skb, flow_key, nla_data(a),
 798                                get_mask(a, struct ovs_key_ipv6 *));
 799                 break;
 800
 801         case OVS_KEY_ATTR_TCP:
 802                 err = set_tcp(skb, flow_key, nla_data(a),
 803                               get_mask(a, struct ovs_key_tcp *));
 804                 break;
 805
 806         case OVS_KEY_ATTR_UDP:
 807                 err = set_udp(skb, flow_key, nla_data(a),
 808                               get_mask(a, struct ovs_key_udp *));
 809                 break;
 810
 811         case OVS_KEY_ATTR_SCTP:
 812                 err = set_sctp(skb, flow_key, nla_data(a),
 813                                get_mask(a, struct ovs_key_sctp *));
 814                 break;
 815
 816         case OVS_KEY_ATTR_MPLS:
 817                 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
 818                                                                     __be32 *));
 819                 break;
 820         }
 821
 822         return err;
 823 }
 824
 825 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
 826                           struct sw_flow_key *key,
 827                           const struct nlattr *a, int rem)
 828 {
 829         struct deferred_action *da;
 830
 831         if (!is_flow_key_valid(key)) {
 832                 int err;
 833
 834                 err = ovs_flow_key_update(skb, key);
 835                 if (err)
 836                         return err;
 837         }
 838         BUG_ON(!is_flow_key_valid(key));
 839
 840         if (!nla_is_last(a, rem)) {
 841                 /* Recirc action is the not the last action
 842                  * of the action list, need to clone the skb.
 843                  */
 844                 skb = skb_clone(skb, GFP_ATOMIC);
 845
 846                 /* Skip the recirc action when out of memory, but
 847                  * continue on with the rest of the action list.
 848                  */
 849                 if (!skb)
 850                         return 0;
 851         }
 852
 853         da = add_deferred_actions(skb, key, NULL);
 854         if (da) {
 855                 da->pkt_key.recirc_id = nla_get_u32(a);
 856         } else {
 857                 kfree_skb(skb);
 858
 859                 if (net_ratelimit())
 860                         pr_warn("%s: deferred action limit reached, drop recirc action\n",
 861                                 ovs_dp_name(dp));
 862         }
 863
 864         return 0;
 865 }
 866
 867 /* Execute a list of actions against 'skb'. */
 868 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 869                               struct sw_flow_key *key,
 870                               const struct nlattr *attr, int len)
 871 {
 872         /* Every output action needs a separate clone of 'skb', but the common
 873          * case is just a single output action, so that doing a clone and
 874          * then freeing the original skbuff is wasteful.  So the following code
 875          * is slightly obscure just to avoid that.
 876          */
 877         int prev_port = -1;
 878         const struct nlattr *a;
 879         int rem;
 880
 881         for (a = attr, rem = len; rem > 0;
 882              a = nla_next(a, &rem)) {
 883                 int err = 0;
 884
 885                 if (unlikely(prev_port != -1)) {
 886                         struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
 887
 888                         if (out_skb)
 889                                 do_output(dp, out_skb, prev_port);
 890
 891                         prev_port = -1;
 892                 }
 893
 894                 switch (nla_type(a)) {
 895                 case OVS_ACTION_ATTR_OUTPUT:
 896                         prev_port = nla_get_u32(a);
 897                         break;
 898
 899                 case OVS_ACTION_ATTR_USERSPACE:
 900                         output_userspace(dp, skb, key, a, attr, len);
 901                         break;
 902
 903                 case OVS_ACTION_ATTR_HASH:
 904                         execute_hash(skb, key, a);
 905                         break;
 906
 907                 case OVS_ACTION_ATTR_PUSH_MPLS:
 908                         err = push_mpls(skb, key, nla_data(a));
 909                         break;
 910
 911                 case OVS_ACTION_ATTR_POP_MPLS:
 912                         err = pop_mpls(skb, key, nla_get_be16(a));
 913                         break;
 914
 915                 case OVS_ACTION_ATTR_PUSH_VLAN:
 916                         err = push_vlan(skb, key, nla_data(a));
 917                         break;
 918
 919                 case OVS_ACTION_ATTR_POP_VLAN:
 920                         err = pop_vlan(skb, key);
 921                         break;
 922
 923                 case OVS_ACTION_ATTR_RECIRC:
 924                         err = execute_recirc(dp, skb, key, a, rem);
 925                         if (nla_is_last(a, rem)) {
 926                                 /* If this is the last action, the skb has
 927                                  * been consumed or freed.
 928                                  * Return immediately.
 929                                  */
 930                                 return err;
 931                         }
 932                         break;
 933
 934                 case OVS_ACTION_ATTR_SET:
 935                         err = execute_set_action(skb, key, nla_data(a));
 936                         break;
 937
 938                 case OVS_ACTION_ATTR_SET_MASKED:
 939                 case OVS_ACTION_ATTR_SET_TO_MASKED:
 940                         err = execute_masked_set_action(skb, key, nla_data(a));
 941                         break;
 942
 943                 case OVS_ACTION_ATTR_SAMPLE:
 944                         err = sample(dp, skb, key, a, attr, len);
 945                         break;
 946                 }
 947
 948                 if (unlikely(err)) {
 949                         kfree_skb(skb);
 950                         return err;
 951                 }
 952         }
 953
 954         if (prev_port != -1)
 955                 do_output(dp, skb, prev_port);
 956         else
 957                 consume_skb(skb);
 958
 959         return 0;
 960 }
 961
 962 static void process_deferred_actions(struct datapath *dp)
 963 {
 964         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
 965
 966         /* Do not touch the FIFO in case there is no deferred actions. */
 967         if (action_fifo_is_empty(fifo))
 968                 return;
 969
 970         /* Finishing executing all deferred actions. */
 971         do {
 972                 struct deferred_action *da = action_fifo_get(fifo);
 973                 struct sk_buff *skb = da->skb;
 974                 struct sw_flow_key *key = &da->pkt_key;
 975                 const struct nlattr *actions = da->actions;
 976
 977                 if (actions)
 978                         do_execute_actions(dp, skb, key, actions,
 979                                            nla_len(actions));
 980                 else
 981                         ovs_dp_process_packet(skb, key);
 982         } while (!action_fifo_is_empty(fifo));
 983
 984         /* Reset FIFO for the next packet.  */
 985         action_fifo_init(fifo);
 986 }
 987
 988 /* Execute a list of actions against 'skb'. */
 989 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
 990                         const struct sw_flow_actions *acts,
 991                         struct sw_flow_key *key)
 992 {
 993         int level = this_cpu_read(exec_actions_level);
 994         int err;
 995
 996         if (unlikely(level >= EXEC_ACTIONS_LEVEL_LIMIT)) {
 997                 if (net_ratelimit())
 998                         pr_warn("%s: packet loop detected, dropping.\n",
 999                                 ovs_dp_name(dp));
1000
1001                 kfree_skb(skb);
1002                 return -ELOOP;
1003         }
1004
1005         this_cpu_inc(exec_actions_level);
1006         err = do_execute_actions(dp, skb, key,
1007                                  acts->actions, acts->actions_len);
1008
1009         if (!level)
1010                 process_deferred_actions(dp);
1011
1012         this_cpu_dec(exec_actions_level);
1013
1014         /* This return status currently does not reflect the errors
1015          * encounted during deferred actions execution. Probably needs to
1016          * be fixed in the future.
1017          */
1018         return err;
1019 }
1020
1021 int action_fifos_init(void)
1022 {
1023         action_fifos = alloc_percpu(struct action_fifo);
1024         if (!action_fifos)
1025                 return -ENOMEM;
1026
1027         return 0;
1028 }
1029
1030 void action_fifos_exit(void)
1031 {
1032         free_percpu(action_fifos);
1033 }