net/openvswitch/actions.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2007-2017 Nicira, Inc.
   4  */
   5
   6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   7
   8 #include <linux/skbuff.h>
   9 #include <linux/in.h>
  10 #include <linux/ip.h>
  11 #include <linux/openvswitch.h>
  12 #include <linux/sctp.h>
  13 #include <linux/tcp.h>
  14 #include <linux/udp.h>
  15 #include <linux/in6.h>
  16 #include <linux/if_arp.h>
  17 #include <linux/if_vlan.h>
  18
  19 #include <net/dst.h>
  20 #include <net/gso.h>
  21 #include <net/ip.h>
  22 #include <net/ipv6.h>
  23 #include <net/ip6_fib.h>
  24 #include <net/checksum.h>
  25 #include <net/dsfield.h>
  26 #include <net/mpls.h>
  27
  28 #if IS_ENABLED(CONFIG_PSAMPLE)
  29 #include <net/psample.h>
  30 #endif
  31
  32 #include <net/sctp/checksum.h>
  33
  34 #include "datapath.h"
  35 #include "drop.h"
  36 #include "flow.h"
  37 #include "conntrack.h"
  38 #include "vport.h"
  39 #include "flow_netlink.h"
  40 #include "openvswitch_trace.h"
  41
  42 struct deferred_action {
  43         struct sk_buff *skb;
  44         const struct nlattr *actions;
  45         int actions_len;
  46
  47         /* Store pkt_key clone when creating deferred action. */
  48         struct sw_flow_key pkt_key;
  49 };
  50
  51 #define MAX_L2_LEN      (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
  52 struct ovs_frag_data {
  53         unsigned long dst;
  54         struct vport *vport;
  55         struct ovs_skb_cb cb;
  56         __be16 inner_protocol;
  57         u16 network_offset;     /* valid only for MPLS */
  58         u16 vlan_tci;
  59         __be16 vlan_proto;
  60         unsigned int l2_len;
  61         u8 mac_proto;
  62         u8 l2_data[MAX_L2_LEN];
  63 };
  64
  65 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
  66
  67 #define DEFERRED_ACTION_FIFO_SIZE 10
  68 #define OVS_RECURSION_LIMIT 5
  69 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
  70 struct action_fifo {
  71         int head;
  72         int tail;
  73         /* Deferred action fifo queue storage. */
  74         struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
  75 };
  76
  77 struct action_flow_keys {
  78         struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
  79 };
  80
  81 static struct action_fifo __percpu *action_fifos;
  82 static struct action_flow_keys __percpu *flow_keys;
  83 static DEFINE_PER_CPU(int, exec_actions_level);
  84
  85 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
  86  * space. Return NULL if out of key spaces.
  87  */
  88 static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
  89 {
  90         struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
  91         int level = this_cpu_read(exec_actions_level);
  92         struct sw_flow_key *key = NULL;
  93
  94         if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
  95                 key = &keys->key[level - 1];
  96                 *key = *key_;
  97         }
  98
  99         return key;
 100 }
 101
 102 static void action_fifo_init(struct action_fifo *fifo)
 103 {
 104         fifo->head = 0;
 105         fifo->tail = 0;
 106 }
 107
 108 static bool action_fifo_is_empty(const struct action_fifo *fifo)
 109 {
 110         return (fifo->head == fifo->tail);
 111 }
 112
 113 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
 114 {
 115         if (action_fifo_is_empty(fifo))
 116                 return NULL;
 117
 118         return &fifo->fifo[fifo->tail++];
 119 }
 120
 121 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
 122 {
 123         if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
 124                 return NULL;
 125
 126         return &fifo->fifo[fifo->head++];
 127 }
 128
 129 /* Return true if fifo is not full */
 130 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
 131                                     const struct sw_flow_key *key,
 132                                     const struct nlattr *actions,
 133                                     const int actions_len)
 134 {
 135         struct action_fifo *fifo;
 136         struct deferred_action *da;
 137
 138         fifo = this_cpu_ptr(action_fifos);
 139         da = action_fifo_put(fifo);
 140         if (da) {
 141                 da->skb = skb;
 142                 da->actions = actions;
 143                 da->actions_len = actions_len;
 144                 da->pkt_key = *key;
 145         }
 146
 147         return da;
 148 }
 149
 150 static void invalidate_flow_key(struct sw_flow_key *key)
 151 {
 152         key->mac_proto |= SW_FLOW_KEY_INVALID;
 153 }
 154
 155 static bool is_flow_key_valid(const struct sw_flow_key *key)
 156 {
 157         return !(key->mac_proto & SW_FLOW_KEY_INVALID);
 158 }
 159
 160 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
 161                          struct sw_flow_key *key,
 162                          u32 recirc_id,
 163                          const struct nlattr *actions, int len,
 164                          bool last, bool clone_flow_key);
 165
 166 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 167                               struct sw_flow_key *key,
 168                               const struct nlattr *attr, int len);
 169
 170 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 171                      __be32 mpls_lse, __be16 mpls_ethertype, __u16 mac_len)
 172 {
 173         int err;
 174
 175         err = skb_mpls_push(skb, mpls_lse, mpls_ethertype, mac_len, !!mac_len);
 176         if (err)
 177                 return err;
 178
 179         if (!mac_len)
 180                 key->mac_proto = MAC_PROTO_NONE;
 181
 182         invalidate_flow_key(key);
 183         return 0;
 184 }
 185
 186 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 187                     const __be16 ethertype)
 188 {
 189         int err;
 190
 191         err = skb_mpls_pop(skb, ethertype, skb->mac_len,
 192                            ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET);
 193         if (err)
 194                 return err;
 195
 196         if (ethertype == htons(ETH_P_TEB))
 197                 key->mac_proto = MAC_PROTO_ETHERNET;
 198
 199         invalidate_flow_key(key);
 200         return 0;
 201 }
 202
 203 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
 204                     const __be32 *mpls_lse, const __be32 *mask)
 205 {
 206         struct mpls_shim_hdr *stack;
 207         __be32 lse;
 208         int err;
 209
 210         if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
 211                 return -ENOMEM;
 212
 213         stack = mpls_hdr(skb);
 214         lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
 215         err = skb_mpls_update_lse(skb, lse);
 216         if (err)
 217                 return err;
 218
 219         flow_key->mpls.lse[0] = lse;
 220         return 0;
 221 }
 222
 223 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 224 {
 225         int err;
 226
 227         err = skb_vlan_pop(skb);
 228         if (skb_vlan_tag_present(skb)) {
 229                 invalidate_flow_key(key);
 230         } else {
 231                 key->eth.vlan.tci = 0;
 232                 key->eth.vlan.tpid = 0;
 233         }
 234         return err;
 235 }
 236
 237 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
 238                      const struct ovs_action_push_vlan *vlan)
 239 {
 240         if (skb_vlan_tag_present(skb)) {
 241                 invalidate_flow_key(key);
 242         } else {
 243                 key->eth.vlan.tci = vlan->vlan_tci;
 244                 key->eth.vlan.tpid = vlan->vlan_tpid;
 245         }
 246         return skb_vlan_push(skb, vlan->vlan_tpid,
 247                              ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
 248 }
 249
 250 /* 'src' is already properly masked. */
 251 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
 252 {
 253         u16 *dst = (u16 *)dst_;
 254         const u16 *src = (const u16 *)src_;
 255         const u16 *mask = (const u16 *)mask_;
 256
 257         OVS_SET_MASKED(dst[0], src[0], mask[0]);
 258         OVS_SET_MASKED(dst[1], src[1], mask[1]);
 259         OVS_SET_MASKED(dst[2], src[2], mask[2]);
 260 }
 261
 262 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
 263                         const struct ovs_key_ethernet *key,
 264                         const struct ovs_key_ethernet *mask)
 265 {
 266         int err;
 267
 268         err = skb_ensure_writable(skb, ETH_HLEN);
 269         if (unlikely(err))
 270                 return err;
 271
 272         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 273
 274         ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
 275                                mask->eth_src);
 276         ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
 277                                mask->eth_dst);
 278
 279         skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 280
 281         ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
 282         ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
 283         return 0;
 284 }
 285
 286 /* pop_eth does not support VLAN packets as this action is never called
 287  * for them.
 288  */
 289 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
 290 {
 291         int err;
 292
 293         err = skb_eth_pop(skb);
 294         if (err)
 295                 return err;
 296
 297         /* safe right before invalidate_flow_key */
 298         key->mac_proto = MAC_PROTO_NONE;
 299         invalidate_flow_key(key);
 300         return 0;
 301 }
 302
 303 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
 304                     const struct ovs_action_push_eth *ethh)
 305 {
 306         int err;
 307
 308         err = skb_eth_push(skb, ethh->addresses.eth_dst,
 309                            ethh->addresses.eth_src);
 310         if (err)
 311                 return err;
 312
 313         /* safe right before invalidate_flow_key */
 314         key->mac_proto = MAC_PROTO_ETHERNET;
 315         invalidate_flow_key(key);
 316         return 0;
 317 }
 318
 319 static noinline_for_stack int push_nsh(struct sk_buff *skb,
 320                                        struct sw_flow_key *key,
 321                                        const struct nlattr *a)
 322 {
 323         u8 buffer[NSH_HDR_MAX_LEN];
 324         struct nshhdr *nh = (struct nshhdr *)buffer;
 325         int err;
 326
 327         err = nsh_hdr_from_nlattr(a, nh, NSH_HDR_MAX_LEN);
 328         if (err)
 329                 return err;
 330
 331         err = nsh_push(skb, nh);
 332         if (err)
 333                 return err;
 334
 335         /* safe right before invalidate_flow_key */
 336         key->mac_proto = MAC_PROTO_NONE;
 337         invalidate_flow_key(key);
 338         return 0;
 339 }
 340
 341 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
 342 {
 343         int err;
 344
 345         err = nsh_pop(skb);
 346         if (err)
 347                 return err;
 348
 349         /* safe right before invalidate_flow_key */
 350         if (skb->protocol == htons(ETH_P_TEB))
 351                 key->mac_proto = MAC_PROTO_ETHERNET;
 352         else
 353                 key->mac_proto = MAC_PROTO_NONE;
 354         invalidate_flow_key(key);
 355         return 0;
 356 }
 357
 358 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
 359                                   __be32 addr, __be32 new_addr)
 360 {
 361         int transport_len = skb->len - skb_transport_offset(skb);
 362
 363         if (nh->frag_off & htons(IP_OFFSET))
 364                 return;
 365
 366         if (nh->protocol == IPPROTO_TCP) {
 367                 if (likely(transport_len >= sizeof(struct tcphdr)))
 368                         inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 369                                                  addr, new_addr, true);
 370         } else if (nh->protocol == IPPROTO_UDP) {
 371                 if (likely(transport_len >= sizeof(struct udphdr))) {
 372                         struct udphdr *uh = udp_hdr(skb);
 373
 374                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 375                                 inet_proto_csum_replace4(&uh->check, skb,
 376                                                          addr, new_addr, true);
 377                                 if (!uh->check)
 378                                         uh->check = CSUM_MANGLED_0;
 379                         }
 380                 }
 381         }
 382 }
 383
 384 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 385                         __be32 *addr, __be32 new_addr)
 386 {
 387         update_ip_l4_checksum(skb, nh, *addr, new_addr);
 388         csum_replace4(&nh->check, *addr, new_addr);
 389         skb_clear_hash(skb);
 390         ovs_ct_clear(skb, NULL);
 391         *addr = new_addr;
 392 }
 393
 394 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 395                                  __be32 addr[4], const __be32 new_addr[4])
 396 {
 397         int transport_len = skb->len - skb_transport_offset(skb);
 398
 399         if (l4_proto == NEXTHDR_TCP) {
 400                 if (likely(transport_len >= sizeof(struct tcphdr)))
 401                         inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 402                                                   addr, new_addr, true);
 403         } else if (l4_proto == NEXTHDR_UDP) {
 404                 if (likely(transport_len >= sizeof(struct udphdr))) {
 405                         struct udphdr *uh = udp_hdr(skb);
 406
 407                         if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 408                                 inet_proto_csum_replace16(&uh->check, skb,
 409                                                           addr, new_addr, true);
 410                                 if (!uh->check)
 411                                         uh->check = CSUM_MANGLED_0;
 412                         }
 413                 }
 414         } else if (l4_proto == NEXTHDR_ICMP) {
 415                 if (likely(transport_len >= sizeof(struct icmp6hdr)))
 416                         inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 417                                                   skb, addr, new_addr, true);
 418         }
 419 }
 420
 421 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
 422                            const __be32 mask[4], __be32 masked[4])
 423 {
 424         masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
 425         masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
 426         masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
 427         masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
 428 }
 429
 430 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 431                           __be32 addr[4], const __be32 new_addr[4],
 432                           bool recalculate_csum)
 433 {
 434         if (recalculate_csum)
 435                 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 436
 437         skb_clear_hash(skb);
 438         ovs_ct_clear(skb, NULL);
 439         memcpy(addr, new_addr, sizeof(__be32[4]));
 440 }
 441
 442 static void set_ipv6_dsfield(struct sk_buff *skb, struct ipv6hdr *nh, u8 ipv6_tclass, u8 mask)
 443 {
 444         u8 old_ipv6_tclass = ipv6_get_dsfield(nh);
 445
 446         ipv6_tclass = OVS_MASKED(old_ipv6_tclass, ipv6_tclass, mask);
 447
 448         if (skb->ip_summed == CHECKSUM_COMPLETE)
 449                 csum_replace(&skb->csum, (__force __wsum)(old_ipv6_tclass << 12),
 450                              (__force __wsum)(ipv6_tclass << 12));
 451
 452         ipv6_change_dsfield(nh, ~mask, ipv6_tclass);
 453 }
 454
 455 static void set_ipv6_fl(struct sk_buff *skb, struct ipv6hdr *nh, u32 fl, u32 mask)
 456 {
 457         u32 ofl;
 458
 459         ofl = nh->flow_lbl[0] << 16 |  nh->flow_lbl[1] << 8 |  nh->flow_lbl[2];
 460         fl = OVS_MASKED(ofl, fl, mask);
 461
 462         /* Bits 21-24 are always unmasked, so this retains their values. */
 463         nh->flow_lbl[0] = (u8)(fl >> 16);
 464         nh->flow_lbl[1] = (u8)(fl >> 8);
 465         nh->flow_lbl[2] = (u8)fl;
 466
 467         if (skb->ip_summed == CHECKSUM_COMPLETE)
 468                 csum_replace(&skb->csum, (__force __wsum)htonl(ofl), (__force __wsum)htonl(fl));
 469 }
 470
 471 static void set_ipv6_ttl(struct sk_buff *skb, struct ipv6hdr *nh, u8 new_ttl, u8 mask)
 472 {
 473         new_ttl = OVS_MASKED(nh->hop_limit, new_ttl, mask);
 474
 475         if (skb->ip_summed == CHECKSUM_COMPLETE)
 476                 csum_replace(&skb->csum, (__force __wsum)(nh->hop_limit << 8),
 477                              (__force __wsum)(new_ttl << 8));
 478         nh->hop_limit = new_ttl;
 479 }
 480
 481 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
 482                        u8 mask)
 483 {
 484         new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
 485
 486         csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 487         nh->ttl = new_ttl;
 488 }
 489
 490 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
 491                     const struct ovs_key_ipv4 *key,
 492                     const struct ovs_key_ipv4 *mask)
 493 {
 494         struct iphdr *nh;
 495         __be32 new_addr;
 496         int err;
 497
 498         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 499                                   sizeof(struct iphdr));
 500         if (unlikely(err))
 501                 return err;
 502
 503         nh = ip_hdr(skb);
 504
 505         /* Setting an IP addresses is typically only a side effect of
 506          * matching on them in the current userspace implementation, so it
 507          * makes sense to check if the value actually changed.
 508          */
 509         if (mask->ipv4_src) {
 510                 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
 511
 512                 if (unlikely(new_addr != nh->saddr)) {
 513                         set_ip_addr(skb, nh, &nh->saddr, new_addr);
 514                         flow_key->ipv4.addr.src = new_addr;
 515                 }
 516         }
 517         if (mask->ipv4_dst) {
 518                 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
 519
 520                 if (unlikely(new_addr != nh->daddr)) {
 521                         set_ip_addr(skb, nh, &nh->daddr, new_addr);
 522                         flow_key->ipv4.addr.dst = new_addr;
 523                 }
 524         }
 525         if (mask->ipv4_tos) {
 526                 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
 527                 flow_key->ip.tos = nh->tos;
 528         }
 529         if (mask->ipv4_ttl) {
 530                 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
 531                 flow_key->ip.ttl = nh->ttl;
 532         }
 533
 534         return 0;
 535 }
 536
 537 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
 538 {
 539         return !!(addr[0] | addr[1] | addr[2] | addr[3]);
 540 }
 541
 542 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
 543                     const struct ovs_key_ipv6 *key,
 544                     const struct ovs_key_ipv6 *mask)
 545 {
 546         struct ipv6hdr *nh;
 547         int err;
 548
 549         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 550                                   sizeof(struct ipv6hdr));
 551         if (unlikely(err))
 552                 return err;
 553
 554         nh = ipv6_hdr(skb);
 555
 556         /* Setting an IP addresses is typically only a side effect of
 557          * matching on them in the current userspace implementation, so it
 558          * makes sense to check if the value actually changed.
 559          */
 560         if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
 561                 __be32 *saddr = (__be32 *)&nh->saddr;
 562                 __be32 masked[4];
 563
 564                 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
 565
 566                 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
 567                         set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
 568                                       true);
 569                         memcpy(&flow_key->ipv6.addr.src, masked,
 570                                sizeof(flow_key->ipv6.addr.src));
 571                 }
 572         }
 573         if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
 574                 unsigned int offset = 0;
 575                 int flags = IP6_FH_F_SKIP_RH;
 576                 bool recalc_csum = true;
 577                 __be32 *daddr = (__be32 *)&nh->daddr;
 578                 __be32 masked[4];
 579
 580                 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
 581
 582                 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
 583                         if (ipv6_ext_hdr(nh->nexthdr))
 584                                 recalc_csum = (ipv6_find_hdr(skb, &offset,
 585                                                              NEXTHDR_ROUTING,
 586                                                              NULL, &flags)
 587                                                != NEXTHDR_ROUTING);
 588
 589                         set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
 590                                       recalc_csum);
 591                         memcpy(&flow_key->ipv6.addr.dst, masked,
 592                                sizeof(flow_key->ipv6.addr.dst));
 593                 }
 594         }
 595         if (mask->ipv6_tclass) {
 596                 set_ipv6_dsfield(skb, nh, key->ipv6_tclass, mask->ipv6_tclass);
 597                 flow_key->ip.tos = ipv6_get_dsfield(nh);
 598         }
 599         if (mask->ipv6_label) {
 600                 set_ipv6_fl(skb, nh, ntohl(key->ipv6_label),
 601                             ntohl(mask->ipv6_label));
 602                 flow_key->ipv6.label =
 603                     *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 604         }
 605         if (mask->ipv6_hlimit) {
 606                 set_ipv6_ttl(skb, nh, key->ipv6_hlimit, mask->ipv6_hlimit);
 607                 flow_key->ip.ttl = nh->hop_limit;
 608         }
 609         return 0;
 610 }
 611
 612 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
 613                    const struct nlattr *a)
 614 {
 615         struct nshhdr *nh;
 616         size_t length;
 617         int err;
 618         u8 flags;
 619         u8 ttl;
 620         int i;
 621
 622         struct ovs_key_nsh key;
 623         struct ovs_key_nsh mask;
 624
 625         err = nsh_key_from_nlattr(a, &key, &mask);
 626         if (err)
 627                 return err;
 628
 629         /* Make sure the NSH base header is there */
 630         if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
 631                 return -ENOMEM;
 632
 633         nh = nsh_hdr(skb);
 634         length = nsh_hdr_len(nh);
 635
 636         /* Make sure the whole NSH header is there */
 637         err = skb_ensure_writable(skb, skb_network_offset(skb) +
 638                                        length);
 639         if (unlikely(err))
 640                 return err;
 641
 642         nh = nsh_hdr(skb);
 643         skb_postpull_rcsum(skb, nh, length);
 644         flags = nsh_get_flags(nh);
 645         flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
 646         flow_key->nsh.base.flags = flags;
 647         ttl = nsh_get_ttl(nh);
 648         ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
 649         flow_key->nsh.base.ttl = ttl;
 650         nsh_set_flags_and_ttl(nh, flags, ttl);
 651         nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
 652                                   mask.base.path_hdr);
 653         flow_key->nsh.base.path_hdr = nh->path_hdr;
 654         switch (nh->mdtype) {
 655         case NSH_M_TYPE1:
 656                 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
 657                         nh->md1.context[i] =
 658                             OVS_MASKED(nh->md1.context[i], key.context[i],
 659                                        mask.context[i]);
 660                 }
 661                 memcpy(flow_key->nsh.context, nh->md1.context,
 662                        sizeof(nh->md1.context));
 663                 break;
 664         case NSH_M_TYPE2:
 665                 memset(flow_key->nsh.context, 0,
 666                        sizeof(flow_key->nsh.context));
 667                 break;
 668         default:
 669                 return -EINVAL;
 670         }
 671         skb_postpush_rcsum(skb, nh, length);
 672         return 0;
 673 }
 674
 675 /* Must follow skb_ensure_writable() since that can move the skb data. */
 676 static void set_tp_port(struct sk_buff *skb, __be16 *port,
 677                         __be16 new_port, __sum16 *check)
 678 {
 679         ovs_ct_clear(skb, NULL);
 680         inet_proto_csum_replace2(check, skb, *port, new_port, false);
 681         *port = new_port;
 682 }
 683
 684 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 685                    const struct ovs_key_udp *key,
 686                    const struct ovs_key_udp *mask)
 687 {
 688         struct udphdr *uh;
 689         __be16 src, dst;
 690         int err;
 691
 692         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 693                                   sizeof(struct udphdr));
 694         if (unlikely(err))
 695                 return err;
 696
 697         uh = udp_hdr(skb);
 698         /* Either of the masks is non-zero, so do not bother checking them. */
 699         src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
 700         dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
 701
 702         if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 703                 if (likely(src != uh->source)) {
 704                         set_tp_port(skb, &uh->source, src, &uh->check);
 705                         flow_key->tp.src = src;
 706                 }
 707                 if (likely(dst != uh->dest)) {
 708                         set_tp_port(skb, &uh->dest, dst, &uh->check);
 709                         flow_key->tp.dst = dst;
 710                 }
 711
 712                 if (unlikely(!uh->check))
 713                         uh->check = CSUM_MANGLED_0;
 714         } else {
 715                 uh->source = src;
 716                 uh->dest = dst;
 717                 flow_key->tp.src = src;
 718                 flow_key->tp.dst = dst;
 719                 ovs_ct_clear(skb, NULL);
 720         }
 721
 722         skb_clear_hash(skb);
 723
 724         return 0;
 725 }
 726
 727 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 728                    const struct ovs_key_tcp *key,
 729                    const struct ovs_key_tcp *mask)
 730 {
 731         struct tcphdr *th;
 732         __be16 src, dst;
 733         int err;
 734
 735         err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 736                                   sizeof(struct tcphdr));
 737         if (unlikely(err))
 738                 return err;
 739
 740         th = tcp_hdr(skb);
 741         src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
 742         if (likely(src != th->source)) {
 743                 set_tp_port(skb, &th->source, src, &th->check);
 744                 flow_key->tp.src = src;
 745         }
 746         dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
 747         if (likely(dst != th->dest)) {
 748                 set_tp_port(skb, &th->dest, dst, &th->check);
 749                 flow_key->tp.dst = dst;
 750         }
 751         skb_clear_hash(skb);
 752
 753         return 0;
 754 }
 755
 756 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 757                     const struct ovs_key_sctp *key,
 758                     const struct ovs_key_sctp *mask)
 759 {
 760         unsigned int sctphoff = skb_transport_offset(skb);
 761         struct sctphdr *sh;
 762         __le32 old_correct_csum, new_csum, old_csum;
 763         int err;
 764
 765         err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
 766         if (unlikely(err))
 767                 return err;
 768
 769         sh = sctp_hdr(skb);
 770         old_csum = sh->checksum;
 771         old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 772
 773         sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
 774         sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
 775
 776         new_csum = sctp_compute_cksum(skb, sctphoff);
 777
 778         /* Carry any checksum errors through. */
 779         sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 780
 781         skb_clear_hash(skb);
 782         ovs_ct_clear(skb, NULL);
 783
 784         flow_key->tp.src = sh->source;
 785         flow_key->tp.dst = sh->dest;
 786
 787         return 0;
 788 }
 789
 790 static int ovs_vport_output(struct net *net, struct sock *sk,
 791                             struct sk_buff *skb)
 792 {
 793         struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
 794         struct vport *vport = data->vport;
 795
 796         if (skb_cow_head(skb, data->l2_len) < 0) {
 797                 kfree_skb_reason(skb, SKB_DROP_REASON_NOMEM);
 798                 return -ENOMEM;
 799         }
 800
 801         __skb_dst_copy(skb, data->dst);
 802         *OVS_CB(skb) = data->cb;
 803         skb->inner_protocol = data->inner_protocol;
 804         if (data->vlan_tci & VLAN_CFI_MASK)
 805                 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
 806         else
 807                 __vlan_hwaccel_clear_tag(skb);
 808
 809         /* Reconstruct the MAC header.  */
 810         skb_push(skb, data->l2_len);
 811         memcpy(skb->data, &data->l2_data, data->l2_len);
 812         skb_postpush_rcsum(skb, skb->data, data->l2_len);
 813         skb_reset_mac_header(skb);
 814
 815         if (eth_p_mpls(skb->protocol)) {
 816                 skb->inner_network_header = skb->network_header;
 817                 skb_set_network_header(skb, data->network_offset);
 818                 skb_reset_mac_len(skb);
 819         }
 820
 821         ovs_vport_send(vport, skb, data->mac_proto);
 822         return 0;
 823 }
 824
 825 static unsigned int
 826 ovs_dst_get_mtu(const struct dst_entry *dst)
 827 {
 828         return dst->dev->mtu;
 829 }
 830
 831 static struct dst_ops ovs_dst_ops = {
 832         .family = AF_UNSPEC,
 833         .mtu = ovs_dst_get_mtu,
 834 };
 835
 836 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
 837  * ovs_vport_output(), which is called once per fragmented packet.
 838  */
 839 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
 840                          u16 orig_network_offset, u8 mac_proto)
 841 {
 842         unsigned int hlen = skb_network_offset(skb);
 843         struct ovs_frag_data *data;
 844
 845         data = this_cpu_ptr(&ovs_frag_data_storage);
 846         data->dst = skb->_skb_refdst;
 847         data->vport = vport;
 848         data->cb = *OVS_CB(skb);
 849         data->inner_protocol = skb->inner_protocol;
 850         data->network_offset = orig_network_offset;
 851         if (skb_vlan_tag_present(skb))
 852                 data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
 853         else
 854                 data->vlan_tci = 0;
 855         data->vlan_proto = skb->vlan_proto;
 856         data->mac_proto = mac_proto;
 857         data->l2_len = hlen;
 858         memcpy(&data->l2_data, skb->data, hlen);
 859
 860         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 861         skb_pull(skb, hlen);
 862 }
 863
 864 static void ovs_fragment(struct net *net, struct vport *vport,
 865                          struct sk_buff *skb, u16 mru,
 866                          struct sw_flow_key *key)
 867 {
 868         enum ovs_drop_reason reason;
 869         u16 orig_network_offset = 0;
 870
 871         if (eth_p_mpls(skb->protocol)) {
 872                 orig_network_offset = skb_network_offset(skb);
 873                 skb->network_header = skb->inner_network_header;
 874         }
 875
 876         if (skb_network_offset(skb) > MAX_L2_LEN) {
 877                 OVS_NLERR(1, "L2 header too long to fragment");
 878                 reason = OVS_DROP_FRAG_L2_TOO_LONG;
 879                 goto err;
 880         }
 881
 882         if (key->eth.type == htons(ETH_P_IP)) {
 883                 struct rtable ovs_rt = { 0 };
 884                 unsigned long orig_dst;
 885
 886                 prepare_frag(vport, skb, orig_network_offset,
 887                              ovs_key_mac_proto(key));
 888                 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL,
 889                          DST_OBSOLETE_NONE, DST_NOCOUNT);
 890                 ovs_rt.dst.dev = vport->dev;
 891
 892                 orig_dst = skb->_skb_refdst;
 893                 skb_dst_set_noref(skb, &ovs_rt.dst);
 894                 IPCB(skb)->frag_max_size = mru;
 895
 896                 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
 897                 refdst_drop(orig_dst);
 898         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 899                 unsigned long orig_dst;
 900                 struct rt6_info ovs_rt;
 901
 902                 prepare_frag(vport, skb, orig_network_offset,
 903                              ovs_key_mac_proto(key));
 904                 memset(&ovs_rt, 0, sizeof(ovs_rt));
 905                 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL,
 906                          DST_OBSOLETE_NONE, DST_NOCOUNT);
 907                 ovs_rt.dst.dev = vport->dev;
 908
 909                 orig_dst = skb->_skb_refdst;
 910                 skb_dst_set_noref(skb, &ovs_rt.dst);
 911                 IP6CB(skb)->frag_max_size = mru;
 912
 913                 ipv6_stub->ipv6_fragment(net, skb->sk, skb, ovs_vport_output);
 914                 refdst_drop(orig_dst);
 915         } else {
 916                 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
 917                           ovs_vport_name(vport), ntohs(key->eth.type), mru,
 918                           vport->dev->mtu);
 919                 reason = OVS_DROP_FRAG_INVALID_PROTO;
 920                 goto err;
 921         }
 922
 923         return;
 924 err:
 925         ovs_kfree_skb_reason(skb, reason);
 926 }
 927
 928 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
 929                       struct sw_flow_key *key)
 930 {
 931         struct vport *vport = ovs_vport_rcu(dp, out_port);
 932
 933         if (likely(vport && netif_carrier_ok(vport->dev))) {
 934                 u16 mru = OVS_CB(skb)->mru;
 935                 u32 cutlen = OVS_CB(skb)->cutlen;
 936
 937                 if (unlikely(cutlen > 0)) {
 938                         if (skb->len - cutlen > ovs_mac_header_len(key))
 939                                 pskb_trim(skb, skb->len - cutlen);
 940                         else
 941                                 pskb_trim(skb, ovs_mac_header_len(key));
 942                 }
 943
 944                 /* Need to set the pkt_type to involve the routing layer.  The
 945                  * packet movement through the OVS datapath doesn't generally
 946                  * use routing, but this is needed for tunnel cases.
 947                  */
 948                 skb->pkt_type = PACKET_OUTGOING;
 949
 950                 if (likely(!mru ||
 951                            (skb->len <= mru + vport->dev->hard_header_len))) {
 952                         ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
 953                 } else if (mru <= vport->dev->mtu) {
 954                         struct net *net = read_pnet(&dp->net);
 955
 956                         ovs_fragment(net, vport, skb, mru, key);
 957                 } else {
 958                         kfree_skb_reason(skb, SKB_DROP_REASON_PKT_TOO_BIG);
 959                 }
 960         } else {
 961                 kfree_skb_reason(skb, SKB_DROP_REASON_DEV_READY);
 962         }
 963 }
 964
 965 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 966                             struct sw_flow_key *key, const struct nlattr *attr,
 967                             const struct nlattr *actions, int actions_len,
 968                             uint32_t cutlen)
 969 {
 970         struct dp_upcall_info upcall;
 971         const struct nlattr *a;
 972         int rem;
 973
 974         memset(&upcall, 0, sizeof(upcall));
 975         upcall.cmd = OVS_PACKET_CMD_ACTION;
 976         upcall.mru = OVS_CB(skb)->mru;
 977
 978         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 979              a = nla_next(a, &rem)) {
 980                 switch (nla_type(a)) {
 981                 case OVS_USERSPACE_ATTR_USERDATA:
 982                         upcall.userdata = a;
 983                         break;
 984
 985                 case OVS_USERSPACE_ATTR_PID:
 986                         if (dp->user_features &
 987                             OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
 988                                 upcall.portid =
 989                                   ovs_dp_get_upcall_portid(dp,
 990                                                            smp_processor_id());
 991                         else
 992                                 upcall.portid = nla_get_u32(a);
 993                         break;
 994
 995                 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
 996                         /* Get out tunnel info. */
 997                         struct vport *vport;
 998
 999                         vport = ovs_vport_rcu(dp, nla_get_u32(a));
1000                         if (vport) {
1001                                 int err;
1002
1003                                 err = dev_fill_metadata_dst(vport->dev, skb);
1004                                 if (!err)
1005                                         upcall.egress_tun_info = skb_tunnel_info(skb);
1006                         }
1007
1008                         break;
1009                 }
1010
1011                 case OVS_USERSPACE_ATTR_ACTIONS: {
1012                         /* Include actions. */
1013                         upcall.actions = actions;
1014                         upcall.actions_len = actions_len;
1015                         break;
1016                 }
1017
1018                 } /* End of switch. */
1019         }
1020
1021         return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
1022 }
1023
1024 static int dec_ttl_exception_handler(struct datapath *dp, struct sk_buff *skb,
1025                                      struct sw_flow_key *key,
1026                                      const struct nlattr *attr)
1027 {
1028         /* The first attribute is always 'OVS_DEC_TTL_ATTR_ACTION'. */
1029         struct nlattr *actions = nla_data(attr);
1030
1031         if (nla_len(actions))
1032                 return clone_execute(dp, skb, key, 0, nla_data(actions),
1033                                      nla_len(actions), true, false);
1034
1035         ovs_kfree_skb_reason(skb, OVS_DROP_IP_TTL);
1036         return 0;
1037 }
1038
1039 /* When 'last' is true, sample() should always consume the 'skb'.
1040  * Otherwise, sample() should keep 'skb' intact regardless what
1041  * actions are executed within sample().
1042  */
1043 static int sample(struct datapath *dp, struct sk_buff *skb,
1044                   struct sw_flow_key *key, const struct nlattr *attr,
1045                   bool last)
1046 {
1047         struct nlattr *actions;
1048         struct nlattr *sample_arg;
1049         int rem = nla_len(attr);
1050         const struct sample_arg *arg;
1051         u32 init_probability;
1052         bool clone_flow_key;
1053         int err;
1054
1055         /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
1056         sample_arg = nla_data(attr);
1057         arg = nla_data(sample_arg);
1058         actions = nla_next(sample_arg, &rem);
1059         init_probability = OVS_CB(skb)->probability;
1060
1061         if ((arg->probability != U32_MAX) &&
1062             (!arg->probability || get_random_u32() > arg->probability)) {
1063                 if (last)
1064                         ovs_kfree_skb_reason(skb, OVS_DROP_LAST_ACTION);
1065                 return 0;
1066         }
1067
1068         OVS_CB(skb)->probability = arg->probability;
1069
1070         clone_flow_key = !arg->exec;
1071         err = clone_execute(dp, skb, key, 0, actions, rem, last,
1072                             clone_flow_key);
1073
1074         if (!last)
1075                 OVS_CB(skb)->probability = init_probability;
1076
1077         return err;
1078 }
1079
1080 /* When 'last' is true, clone() should always consume the 'skb'.
1081  * Otherwise, clone() should keep 'skb' intact regardless what
1082  * actions are executed within clone().
1083  */
1084 static int clone(struct datapath *dp, struct sk_buff *skb,
1085                  struct sw_flow_key *key, const struct nlattr *attr,
1086                  bool last)
1087 {
1088         struct nlattr *actions;
1089         struct nlattr *clone_arg;
1090         int rem = nla_len(attr);
1091         bool dont_clone_flow_key;
1092
1093         /* The first action is always 'OVS_CLONE_ATTR_EXEC'. */
1094         clone_arg = nla_data(attr);
1095         dont_clone_flow_key = nla_get_u32(clone_arg);
1096         actions = nla_next(clone_arg, &rem);
1097
1098         return clone_execute(dp, skb, key, 0, actions, rem, last,
1099                              !dont_clone_flow_key);
1100 }
1101
1102 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1103                          const struct nlattr *attr)
1104 {
1105         struct ovs_action_hash *hash_act = nla_data(attr);
1106         u32 hash = 0;
1107
1108         if (hash_act->hash_alg == OVS_HASH_ALG_L4) {
1109                 /* OVS_HASH_ALG_L4 hasing type. */
1110                 hash = skb_get_hash(skb);
1111         } else if (hash_act->hash_alg == OVS_HASH_ALG_SYM_L4) {
1112                 /* OVS_HASH_ALG_SYM_L4 hashing type.  NOTE: this doesn't
1113                  * extend past an encapsulated header.
1114                  */
1115                 hash = __skb_get_hash_symmetric(skb);
1116         }
1117
1118         hash = jhash_1word(hash, hash_act->hash_basis);
1119         if (!hash)
1120                 hash = 0x1;
1121
1122         key->ovs_flow_hash = hash;
1123 }
1124
1125 static int execute_set_action(struct sk_buff *skb,
1126                               struct sw_flow_key *flow_key,
1127                               const struct nlattr *a)
1128 {
1129         /* Only tunnel set execution is supported without a mask. */
1130         if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1131                 struct ovs_tunnel_info *tun = nla_data(a);
1132
1133                 skb_dst_drop(skb);
1134                 dst_hold((struct dst_entry *)tun->tun_dst);
1135                 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1136                 return 0;
1137         }
1138
1139         return -EINVAL;
1140 }
1141
1142 /* Mask is at the midpoint of the data. */
1143 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1144
1145 static int execute_masked_set_action(struct sk_buff *skb,
1146                                      struct sw_flow_key *flow_key,
1147                                      const struct nlattr *a)
1148 {
1149         int err = 0;
1150
1151         switch (nla_type(a)) {
1152         case OVS_KEY_ATTR_PRIORITY:
1153                 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1154                                *get_mask(a, u32 *));
1155                 flow_key->phy.priority = skb->priority;
1156                 break;
1157
1158         case OVS_KEY_ATTR_SKB_MARK:
1159                 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1160                 flow_key->phy.skb_mark = skb->mark;
1161                 break;
1162
1163         case OVS_KEY_ATTR_TUNNEL_INFO:
1164                 /* Masked data not supported for tunnel. */
1165                 err = -EINVAL;
1166                 break;
1167
1168         case OVS_KEY_ATTR_ETHERNET:
1169                 err = set_eth_addr(skb, flow_key, nla_data(a),
1170                                    get_mask(a, struct ovs_key_ethernet *));
1171                 break;
1172
1173         case OVS_KEY_ATTR_NSH:
1174                 err = set_nsh(skb, flow_key, a);
1175                 break;
1176
1177         case OVS_KEY_ATTR_IPV4:
1178                 err = set_ipv4(skb, flow_key, nla_data(a),
1179                                get_mask(a, struct ovs_key_ipv4 *));
1180                 break;
1181
1182         case OVS_KEY_ATTR_IPV6:
1183                 err = set_ipv6(skb, flow_key, nla_data(a),
1184                                get_mask(a, struct ovs_key_ipv6 *));
1185                 break;
1186
1187         case OVS_KEY_ATTR_TCP:
1188                 err = set_tcp(skb, flow_key, nla_data(a),
1189                               get_mask(a, struct ovs_key_tcp *));
1190                 break;
1191
1192         case OVS_KEY_ATTR_UDP:
1193                 err = set_udp(skb, flow_key, nla_data(a),
1194                               get_mask(a, struct ovs_key_udp *));
1195                 break;
1196
1197         case OVS_KEY_ATTR_SCTP:
1198                 err = set_sctp(skb, flow_key, nla_data(a),
1199                                get_mask(a, struct ovs_key_sctp *));
1200                 break;
1201
1202         case OVS_KEY_ATTR_MPLS:
1203                 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1204                                                                     __be32 *));
1205                 break;
1206
1207         case OVS_KEY_ATTR_CT_STATE:
1208         case OVS_KEY_ATTR_CT_ZONE:
1209         case OVS_KEY_ATTR_CT_MARK:
1210         case OVS_KEY_ATTR_CT_LABELS:
1211         case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1212         case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1213                 err = -EINVAL;
1214                 break;
1215         }
1216
1217         return err;
1218 }
1219
1220 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1221                           struct sw_flow_key *key,
1222                           const struct nlattr *a, bool last)
1223 {
1224         u32 recirc_id;
1225
1226         if (!is_flow_key_valid(key)) {
1227                 int err;
1228
1229                 err = ovs_flow_key_update(skb, key);
1230                 if (err)
1231                         return err;
1232         }
1233         BUG_ON(!is_flow_key_valid(key));
1234
1235         recirc_id = nla_get_u32(a);
1236         return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1237 }
1238
1239 static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1240                                  struct sw_flow_key *key,
1241                                  const struct nlattr *attr, bool last)
1242 {
1243         struct ovs_skb_cb *ovs_cb = OVS_CB(skb);
1244         const struct nlattr *actions, *cpl_arg;
1245         int len, max_len, rem = nla_len(attr);
1246         const struct check_pkt_len_arg *arg;
1247         bool clone_flow_key;
1248
1249         /* The first netlink attribute in 'attr' is always
1250          * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1251          */
1252         cpl_arg = nla_data(attr);
1253         arg = nla_data(cpl_arg);
1254
1255         len = ovs_cb->mru ? ovs_cb->mru + skb->mac_len : skb->len;
1256         max_len = arg->pkt_len;
1257
1258         if ((skb_is_gso(skb) && skb_gso_validate_mac_len(skb, max_len)) ||
1259             len <= max_len) {
1260                 /* Second netlink attribute in 'attr' is always
1261                  * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1262                  */
1263                 actions = nla_next(cpl_arg, &rem);
1264                 clone_flow_key = !arg->exec_for_lesser_equal;
1265         } else {
1266                 /* Third netlink attribute in 'attr' is always
1267                  * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1268                  */
1269                 actions = nla_next(cpl_arg, &rem);
1270                 actions = nla_next(actions, &rem);
1271                 clone_flow_key = !arg->exec_for_greater;
1272         }
1273
1274         return clone_execute(dp, skb, key, 0, nla_data(actions),
1275                              nla_len(actions), last, clone_flow_key);
1276 }
1277
1278 static int execute_dec_ttl(struct sk_buff *skb, struct sw_flow_key *key)
1279 {
1280         int err;
1281
1282         if (skb->protocol == htons(ETH_P_IPV6)) {
1283                 struct ipv6hdr *nh;
1284
1285                 err = skb_ensure_writable(skb, skb_network_offset(skb) +
1286                                           sizeof(*nh));
1287                 if (unlikely(err))
1288                         return err;
1289
1290                 nh = ipv6_hdr(skb);
1291
1292                 if (nh->hop_limit <= 1)
1293                         return -EHOSTUNREACH;
1294
1295                 key->ip.ttl = --nh->hop_limit;
1296         } else if (skb->protocol == htons(ETH_P_IP)) {
1297                 struct iphdr *nh;
1298                 u8 old_ttl;
1299
1300                 err = skb_ensure_writable(skb, skb_network_offset(skb) +
1301                                           sizeof(*nh));
1302                 if (unlikely(err))
1303                         return err;
1304
1305                 nh = ip_hdr(skb);
1306                 if (nh->ttl <= 1)
1307                         return -EHOSTUNREACH;
1308
1309                 old_ttl = nh->ttl--;
1310                 csum_replace2(&nh->check, htons(old_ttl << 8),
1311                               htons(nh->ttl << 8));
1312                 key->ip.ttl = nh->ttl;
1313         }
1314         return 0;
1315 }
1316
1317 #if IS_ENABLED(CONFIG_PSAMPLE)
1318 static void execute_psample(struct datapath *dp, struct sk_buff *skb,
1319                             const struct nlattr *attr)
1320 {
1321         struct psample_group psample_group = {};
1322         struct psample_metadata md = {};
1323         const struct nlattr *a;
1324         u32 rate;
1325         int rem;
1326
1327         nla_for_each_attr(a, nla_data(attr), nla_len(attr), rem) {
1328                 switch (nla_type(a)) {
1329                 case OVS_PSAMPLE_ATTR_GROUP:
1330                         psample_group.group_num = nla_get_u32(a);
1331                         break;
1332
1333                 case OVS_PSAMPLE_ATTR_COOKIE:
1334                         md.user_cookie = nla_data(a);
1335                         md.user_cookie_len = nla_len(a);
1336                         break;
1337                 }
1338         }
1339
1340         psample_group.net = ovs_dp_get_net(dp);
1341         md.in_ifindex = OVS_CB(skb)->input_vport->dev->ifindex;
1342         md.trunc_size = skb->len - OVS_CB(skb)->cutlen;
1343         md.rate_as_probability = 1;
1344
1345         rate = OVS_CB(skb)->probability ? OVS_CB(skb)->probability : U32_MAX;
1346
1347         psample_sample_packet(&psample_group, skb, rate, &md);
1348 }
1349 #else
1350 static void execute_psample(struct datapath *dp, struct sk_buff *skb,
1351                             const struct nlattr *attr)
1352 {}
1353 #endif
1354
1355 /* Execute a list of actions against 'skb'. */
1356 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1357                               struct sw_flow_key *key,
1358                               const struct nlattr *attr, int len)
1359 {
1360         const struct nlattr *a;
1361         int rem;
1362
1363         for (a = attr, rem = len; rem > 0;
1364              a = nla_next(a, &rem)) {
1365                 int err = 0;
1366
1367                 if (trace_ovs_do_execute_action_enabled())
1368                         trace_ovs_do_execute_action(dp, skb, key, a, rem);
1369
1370                 /* Actions that rightfully have to consume the skb should do it
1371                  * and return directly.
1372                  */
1373                 switch (nla_type(a)) {
1374                 case OVS_ACTION_ATTR_OUTPUT: {
1375                         int port = nla_get_u32(a);
1376                         struct sk_buff *clone;
1377
1378                         /* Every output action needs a separate clone
1379                          * of 'skb', In case the output action is the
1380                          * last action, cloning can be avoided.
1381                          */
1382                         if (nla_is_last(a, rem)) {
1383                                 do_output(dp, skb, port, key);
1384                                 /* 'skb' has been used for output.
1385                                  */
1386                                 return 0;
1387                         }
1388
1389                         clone = skb_clone(skb, GFP_ATOMIC);
1390                         if (clone)
1391                                 do_output(dp, clone, port, key);
1392                         OVS_CB(skb)->cutlen = 0;
1393                         break;
1394                 }
1395
1396                 case OVS_ACTION_ATTR_TRUNC: {
1397                         struct ovs_action_trunc *trunc = nla_data(a);
1398
1399                         if (skb->len > trunc->max_len)
1400                                 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1401                         break;
1402                 }
1403
1404                 case OVS_ACTION_ATTR_USERSPACE:
1405                         output_userspace(dp, skb, key, a, attr,
1406                                                      len, OVS_CB(skb)->cutlen);
1407                         OVS_CB(skb)->cutlen = 0;
1408                         if (nla_is_last(a, rem)) {
1409                                 consume_skb(skb);
1410                                 return 0;
1411                         }
1412                         break;
1413
1414                 case OVS_ACTION_ATTR_HASH:
1415                         execute_hash(skb, key, a);
1416                         break;
1417
1418                 case OVS_ACTION_ATTR_PUSH_MPLS: {
1419                         struct ovs_action_push_mpls *mpls = nla_data(a);
1420
1421                         err = push_mpls(skb, key, mpls->mpls_lse,
1422                                         mpls->mpls_ethertype, skb->mac_len);
1423                         break;
1424                 }
1425                 case OVS_ACTION_ATTR_ADD_MPLS: {
1426                         struct ovs_action_add_mpls *mpls = nla_data(a);
1427                         __u16 mac_len = 0;
1428
1429                         if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK)
1430                                 mac_len = skb->mac_len;
1431
1432                         err = push_mpls(skb, key, mpls->mpls_lse,
1433                                         mpls->mpls_ethertype, mac_len);
1434                         break;
1435                 }
1436                 case OVS_ACTION_ATTR_POP_MPLS:
1437                         err = pop_mpls(skb, key, nla_get_be16(a));
1438                         break;
1439
1440                 case OVS_ACTION_ATTR_PUSH_VLAN:
1441                         err = push_vlan(skb, key, nla_data(a));
1442                         break;
1443
1444                 case OVS_ACTION_ATTR_POP_VLAN:
1445                         err = pop_vlan(skb, key);
1446                         break;
1447
1448                 case OVS_ACTION_ATTR_RECIRC: {
1449                         bool last = nla_is_last(a, rem);
1450
1451                         err = execute_recirc(dp, skb, key, a, last);
1452                         if (last) {
1453                                 /* If this is the last action, the skb has
1454                                  * been consumed or freed.
1455                                  * Return immediately.
1456                                  */
1457                                 return err;
1458                         }
1459                         break;
1460                 }
1461
1462                 case OVS_ACTION_ATTR_SET:
1463                         err = execute_set_action(skb, key, nla_data(a));
1464                         break;
1465
1466                 case OVS_ACTION_ATTR_SET_MASKED:
1467                 case OVS_ACTION_ATTR_SET_TO_MASKED:
1468                         err = execute_masked_set_action(skb, key, nla_data(a));
1469                         break;
1470
1471                 case OVS_ACTION_ATTR_SAMPLE: {
1472                         bool last = nla_is_last(a, rem);
1473
1474                         err = sample(dp, skb, key, a, last);
1475                         if (last)
1476                                 return err;
1477
1478                         break;
1479                 }
1480
1481                 case OVS_ACTION_ATTR_CT:
1482                         if (!is_flow_key_valid(key)) {
1483                                 err = ovs_flow_key_update(skb, key);
1484                                 if (err)
1485                                         return err;
1486                         }
1487
1488                         err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1489                                              nla_data(a));
1490
1491                         /* Hide stolen IP fragments from user space. */
1492                         if (err)
1493                                 return err == -EINPROGRESS ? 0 : err;
1494                         break;
1495
1496                 case OVS_ACTION_ATTR_CT_CLEAR:
1497                         err = ovs_ct_clear(skb, key);
1498                         break;
1499
1500                 case OVS_ACTION_ATTR_PUSH_ETH:
1501                         err = push_eth(skb, key, nla_data(a));
1502                         break;
1503
1504                 case OVS_ACTION_ATTR_POP_ETH:
1505                         err = pop_eth(skb, key);
1506                         break;
1507
1508                 case OVS_ACTION_ATTR_PUSH_NSH:
1509                         err = push_nsh(skb, key, nla_data(a));
1510                         break;
1511
1512                 case OVS_ACTION_ATTR_POP_NSH:
1513                         err = pop_nsh(skb, key);
1514                         break;
1515
1516                 case OVS_ACTION_ATTR_METER:
1517                         if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1518                                 ovs_kfree_skb_reason(skb, OVS_DROP_METER);
1519                                 return 0;
1520                         }
1521                         break;
1522
1523                 case OVS_ACTION_ATTR_CLONE: {
1524                         bool last = nla_is_last(a, rem);
1525
1526                         err = clone(dp, skb, key, a, last);
1527                         if (last)
1528                                 return err;
1529
1530                         break;
1531                 }
1532
1533                 case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1534                         bool last = nla_is_last(a, rem);
1535
1536                         err = execute_check_pkt_len(dp, skb, key, a, last);
1537                         if (last)
1538                                 return err;
1539
1540                         break;
1541                 }
1542
1543                 case OVS_ACTION_ATTR_DEC_TTL:
1544                         err = execute_dec_ttl(skb, key);
1545                         if (err == -EHOSTUNREACH)
1546                                 return dec_ttl_exception_handler(dp, skb,
1547                                                                  key, a);
1548                         break;
1549
1550                 case OVS_ACTION_ATTR_DROP: {
1551                         enum ovs_drop_reason reason = nla_get_u32(a)
1552                                 ? OVS_DROP_EXPLICIT_WITH_ERROR
1553                                 : OVS_DROP_EXPLICIT;
1554
1555                         ovs_kfree_skb_reason(skb, reason);
1556                         return 0;
1557                 }
1558
1559                 case OVS_ACTION_ATTR_PSAMPLE:
1560                         execute_psample(dp, skb, a);
1561                         OVS_CB(skb)->cutlen = 0;
1562                         if (nla_is_last(a, rem)) {
1563                                 consume_skb(skb);
1564                                 return 0;
1565                         }
1566                         break;
1567                 }
1568
1569                 if (unlikely(err)) {
1570                         ovs_kfree_skb_reason(skb, OVS_DROP_ACTION_ERROR);
1571                         return err;
1572                 }
1573         }
1574
1575         ovs_kfree_skb_reason(skb, OVS_DROP_LAST_ACTION);
1576         return 0;
1577 }
1578
1579 /* Execute the actions on the clone of the packet. The effect of the
1580  * execution does not affect the original 'skb' nor the original 'key'.
1581  *
1582  * The execution may be deferred in case the actions can not be executed
1583  * immediately.
1584  */
1585 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1586                          struct sw_flow_key *key, u32 recirc_id,
1587                          const struct nlattr *actions, int len,
1588                          bool last, bool clone_flow_key)
1589 {
1590         struct deferred_action *da;
1591         struct sw_flow_key *clone;
1592
1593         skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1594         if (!skb) {
1595                 /* Out of memory, skip this action.
1596                  */
1597                 return 0;
1598         }
1599
1600         /* When clone_flow_key is false, the 'key' will not be change
1601          * by the actions, then the 'key' can be used directly.
1602          * Otherwise, try to clone key from the next recursion level of
1603          * 'flow_keys'. If clone is successful, execute the actions
1604          * without deferring.
1605          */
1606         clone = clone_flow_key ? clone_key(key) : key;
1607         if (clone) {
1608                 int err = 0;
1609
1610                 if (actions) { /* Sample action */
1611                         if (clone_flow_key)
1612                                 __this_cpu_inc(exec_actions_level);
1613
1614                         err = do_execute_actions(dp, skb, clone,
1615                                                  actions, len);
1616
1617                         if (clone_flow_key)
1618                                 __this_cpu_dec(exec_actions_level);
1619                 } else { /* Recirc action */
1620                         clone->recirc_id = recirc_id;
1621                         ovs_dp_process_packet(skb, clone);
1622                 }
1623                 return err;
1624         }
1625
1626         /* Out of 'flow_keys' space. Defer actions */
1627         da = add_deferred_actions(skb, key, actions, len);
1628         if (da) {
1629                 if (!actions) { /* Recirc action */
1630                         key = &da->pkt_key;
1631                         key->recirc_id = recirc_id;
1632                 }
1633         } else {
1634                 /* Out of per CPU action FIFO space. Drop the 'skb' and
1635                  * log an error.
1636                  */
1637                 ovs_kfree_skb_reason(skb, OVS_DROP_DEFERRED_LIMIT);
1638
1639                 if (net_ratelimit()) {
1640                         if (actions) { /* Sample action */
1641                                 pr_warn("%s: deferred action limit reached, drop sample action\n",
1642                                         ovs_dp_name(dp));
1643                         } else {  /* Recirc action */
1644                                 pr_warn("%s: deferred action limit reached, drop recirc action (recirc_id=%#x)\n",
1645                                         ovs_dp_name(dp), recirc_id);
1646                         }
1647                 }
1648         }
1649         return 0;
1650 }
1651
1652 static void process_deferred_actions(struct datapath *dp)
1653 {
1654         struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1655
1656         /* Do not touch the FIFO in case there is no deferred actions. */
1657         if (action_fifo_is_empty(fifo))
1658                 return;
1659
1660         /* Finishing executing all deferred actions. */
1661         do {
1662                 struct deferred_action *da = action_fifo_get(fifo);
1663                 struct sk_buff *skb = da->skb;
1664                 struct sw_flow_key *key = &da->pkt_key;
1665                 const struct nlattr *actions = da->actions;
1666                 int actions_len = da->actions_len;
1667
1668                 if (actions)
1669                         do_execute_actions(dp, skb, key, actions, actions_len);
1670                 else
1671                         ovs_dp_process_packet(skb, key);
1672         } while (!action_fifo_is_empty(fifo));
1673
1674         /* Reset FIFO for the next packet.  */
1675         action_fifo_init(fifo);
1676 }
1677
1678 /* Execute a list of actions against 'skb'. */
1679 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1680                         const struct sw_flow_actions *acts,
1681                         struct sw_flow_key *key)
1682 {
1683         int err, level;
1684
1685         level = __this_cpu_inc_return(exec_actions_level);
1686         if (unlikely(level > OVS_RECURSION_LIMIT)) {
1687                 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1688                                      ovs_dp_name(dp));
1689                 ovs_kfree_skb_reason(skb, OVS_DROP_RECURSION_LIMIT);
1690                 err = -ENETDOWN;
1691                 goto out;
1692         }
1693
1694         OVS_CB(skb)->acts_origlen = acts->orig_len;
1695         err = do_execute_actions(dp, skb, key,
1696                                  acts->actions, acts->actions_len);
1697
1698         if (level == 1)
1699                 process_deferred_actions(dp);
1700
1701 out:
1702         __this_cpu_dec(exec_actions_level);
1703         return err;
1704 }
1705
1706 int action_fifos_init(void)
1707 {
1708         action_fifos = alloc_percpu(struct action_fifo);
1709         if (!action_fifos)
1710                 return -ENOMEM;
1711
1712         flow_keys = alloc_percpu(struct action_flow_keys);
1713         if (!flow_keys) {
1714                 free_percpu(action_fifos);
1715                 return -ENOMEM;
1716         }
1717
1718         return 0;
1719 }
1720
1721 void action_fifos_exit(void)
1722 {
1723         free_percpu(action_fifos);
1724         free_percpu(flow_keys);
1725 }