drivers/net/vrf.c

   1 /*
   2  * vrf.c: device driver to encapsulate a VRF space
   3  *
   4  * Copyright (c) 2015 Cumulus Networks. All rights reserved.
   5  * Copyright (c) 2015 Shrijeet Mukherjee <shm@cumulusnetworks.com>
   6  * Copyright (c) 2015 David Ahern <dsa@cumulusnetworks.com>
   7  *
   8  * Based on dummy, team and ipvlan drivers
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License as published by
  12  * the Free Software Foundation; either version 2 of the License, or
  13  * (at your option) any later version.
  14  */
  15
  16 #include <linux/module.h>
  17 #include <linux/kernel.h>
  18 #include <linux/netdevice.h>
  19 #include <linux/etherdevice.h>
  20 #include <linux/ip.h>
  21 #include <linux/init.h>
  22 #include <linux/moduleparam.h>
  23 #include <linux/netfilter.h>
  24 #include <linux/rtnetlink.h>
  25 #include <net/rtnetlink.h>
  26 #include <linux/u64_stats_sync.h>
  27 #include <linux/hashtable.h>
  28
  29 #include <linux/inetdevice.h>
  30 #include <net/arp.h>
  31 #include <net/ip.h>
  32 #include <net/ip_fib.h>
  33 #include <net/ip6_fib.h>
  34 #include <net/ip6_route.h>
  35 #include <net/rtnetlink.h>
  36 #include <net/route.h>
  37 #include <net/addrconf.h>
  38 #include <net/l3mdev.h>
  39
  40 #define RT_FL_TOS(oldflp4) \
  41         ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
  42
  43 #define DRV_NAME        "vrf"
  44 #define DRV_VERSION     "1.0"
  45
  46 #define vrf_master_get_rcu(dev) \
  47         ((struct net_device *)rcu_dereference(dev->rx_handler_data))
  48
  49 struct slave {
  50         struct list_head        list;
  51         struct net_device       *dev;
  52 };
  53
  54 struct slave_queue {
  55         struct list_head        all_slaves;
  56 };
  57
  58 struct net_vrf {
  59         struct slave_queue      queue;
  60         struct rtable           *rth;
  61         struct rt6_info         *rt6;
  62         u32                     tb_id;
  63 };
  64
  65 struct pcpu_dstats {
  66         u64                     tx_pkts;
  67         u64                     tx_bytes;
  68         u64                     tx_drps;
  69         u64                     rx_pkts;
  70         u64                     rx_bytes;
  71         struct u64_stats_sync   syncp;
  72 };
  73
  74 static struct dst_entry *vrf_ip_check(struct dst_entry *dst, u32 cookie)
  75 {
  76         return dst;
  77 }
  78
  79 static int vrf_ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
  80 {
  81         return ip_local_out(net, sk, skb);
  82 }
  83
  84 static unsigned int vrf_v4_mtu(const struct dst_entry *dst)
  85 {
  86         /* TO-DO: return max ethernet size? */
  87         return dst->dev->mtu;
  88 }
  89
  90 static void vrf_dst_destroy(struct dst_entry *dst)
  91 {
  92         /* our dst lives forever - or until the device is closed */
  93 }
  94
  95 static unsigned int vrf_default_advmss(const struct dst_entry *dst)
  96 {
  97         return 65535 - 40;
  98 }
  99
 100 static struct dst_ops vrf_dst_ops = {
 101         .family         = AF_INET,
 102         .local_out      = vrf_ip_local_out,
 103         .check          = vrf_ip_check,
 104         .mtu            = vrf_v4_mtu,
 105         .destroy        = vrf_dst_destroy,
 106         .default_advmss = vrf_default_advmss,
 107 };
 108
 109 /* neighbor handling is done with actual device; do not want
 110  * to flip skb->dev for those ndisc packets. This really fails
 111  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
 112  * a start.
 113  */
 114 #if IS_ENABLED(CONFIG_IPV6)
 115 static bool check_ipv6_frame(const struct sk_buff *skb)
 116 {
 117         const struct ipv6hdr *ipv6h;
 118         struct ipv6hdr _ipv6h;
 119         bool rc = true;
 120
 121         ipv6h = skb_header_pointer(skb, 0, sizeof(_ipv6h), &_ipv6h);
 122         if (!ipv6h)
 123                 goto out;
 124
 125         if (ipv6h->nexthdr == NEXTHDR_ICMP) {
 126                 const struct icmp6hdr *icmph;
 127                 struct icmp6hdr _icmph;
 128
 129                 icmph = skb_header_pointer(skb, sizeof(_ipv6h),
 130                                            sizeof(_icmph), &_icmph);
 131                 if (!icmph)
 132                         goto out;
 133
 134                 switch (icmph->icmp6_type) {
 135                 case NDISC_ROUTER_SOLICITATION:
 136                 case NDISC_ROUTER_ADVERTISEMENT:
 137                 case NDISC_NEIGHBOUR_SOLICITATION:
 138                 case NDISC_NEIGHBOUR_ADVERTISEMENT:
 139                 case NDISC_REDIRECT:
 140                         rc = false;
 141                         break;
 142                 }
 143         }
 144
 145 out:
 146         return rc;
 147 }
 148 #else
 149 static bool check_ipv6_frame(const struct sk_buff *skb)
 150 {
 151         return false;
 152 }
 153 #endif
 154
 155 static bool is_ip_rx_frame(struct sk_buff *skb)
 156 {
 157         switch (skb->protocol) {
 158         case htons(ETH_P_IP):
 159                 return true;
 160         case htons(ETH_P_IPV6):
 161                 return check_ipv6_frame(skb);
 162         }
 163         return false;
 164 }
 165
 166 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
 167 {
 168         vrf_dev->stats.tx_errors++;
 169         kfree_skb(skb);
 170 }
 171
 172 /* note: already called with rcu_read_lock */
 173 static rx_handler_result_t vrf_handle_frame(struct sk_buff **pskb)
 174 {
 175         struct sk_buff *skb = *pskb;
 176
 177         if (is_ip_rx_frame(skb)) {
 178                 struct net_device *dev = vrf_master_get_rcu(skb->dev);
 179                 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
 180
 181                 u64_stats_update_begin(&dstats->syncp);
 182                 dstats->rx_pkts++;
 183                 dstats->rx_bytes += skb->len;
 184                 u64_stats_update_end(&dstats->syncp);
 185
 186                 skb->dev = dev;
 187
 188                 return RX_HANDLER_ANOTHER;
 189         }
 190         return RX_HANDLER_PASS;
 191 }
 192
 193 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev,
 194                                                  struct rtnl_link_stats64 *stats)
 195 {
 196         int i;
 197
 198         for_each_possible_cpu(i) {
 199                 const struct pcpu_dstats *dstats;
 200                 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
 201                 unsigned int start;
 202
 203                 dstats = per_cpu_ptr(dev->dstats, i);
 204                 do {
 205                         start = u64_stats_fetch_begin_irq(&dstats->syncp);
 206                         tbytes = dstats->tx_bytes;
 207                         tpkts = dstats->tx_pkts;
 208                         tdrops = dstats->tx_drps;
 209                         rbytes = dstats->rx_bytes;
 210                         rpkts = dstats->rx_pkts;
 211                 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
 212                 stats->tx_bytes += tbytes;
 213                 stats->tx_packets += tpkts;
 214                 stats->tx_dropped += tdrops;
 215                 stats->rx_bytes += rbytes;
 216                 stats->rx_packets += rpkts;
 217         }
 218         return stats;
 219 }
 220
 221 #if IS_ENABLED(CONFIG_IPV6)
 222 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
 223                                            struct net_device *dev)
 224 {
 225         const struct ipv6hdr *iph = ipv6_hdr(skb);
 226         struct net *net = dev_net(skb->dev);
 227         struct flowi6 fl6 = {
 228                 /* needed to match OIF rule */
 229                 .flowi6_oif = dev->ifindex,
 230                 .flowi6_iif = LOOPBACK_IFINDEX,
 231                 .daddr = iph->daddr,
 232                 .saddr = iph->saddr,
 233                 .flowlabel = ip6_flowinfo(iph),
 234                 .flowi6_mark = skb->mark,
 235                 .flowi6_proto = iph->nexthdr,
 236                 .flowi6_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF,
 237         };
 238         int ret = NET_XMIT_DROP;
 239         struct dst_entry *dst;
 240         struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
 241
 242         dst = ip6_route_output(net, NULL, &fl6);
 243         if (dst == dst_null)
 244                 goto err;
 245
 246         skb_dst_drop(skb);
 247         skb_dst_set(skb, dst);
 248
 249         ret = ip6_local_out(net, skb->sk, skb);
 250         if (unlikely(net_xmit_eval(ret)))
 251                 dev->stats.tx_errors++;
 252         else
 253                 ret = NET_XMIT_SUCCESS;
 254
 255         return ret;
 256 err:
 257         vrf_tx_error(dev, skb);
 258         return NET_XMIT_DROP;
 259 }
 260 #else
 261 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
 262                                            struct net_device *dev)
 263 {
 264         vrf_tx_error(dev, skb);
 265         return NET_XMIT_DROP;
 266 }
 267 #endif
 268
 269 static int vrf_send_v4_prep(struct sk_buff *skb, struct flowi4 *fl4,
 270                             struct net_device *vrf_dev)
 271 {
 272         struct rtable *rt;
 273         int err = 1;
 274
 275         rt = ip_route_output_flow(dev_net(vrf_dev), fl4, NULL);
 276         if (IS_ERR(rt))
 277                 goto out;
 278
 279         /* TO-DO: what about broadcast ? */
 280         if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
 281                 ip_rt_put(rt);
 282                 goto out;
 283         }
 284
 285         skb_dst_drop(skb);
 286         skb_dst_set(skb, &rt->dst);
 287         err = 0;
 288 out:
 289         return err;
 290 }
 291
 292 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
 293                                            struct net_device *vrf_dev)
 294 {
 295         struct iphdr *ip4h = ip_hdr(skb);
 296         int ret = NET_XMIT_DROP;
 297         struct flowi4 fl4 = {
 298                 /* needed to match OIF rule */
 299                 .flowi4_oif = vrf_dev->ifindex,
 300                 .flowi4_iif = LOOPBACK_IFINDEX,
 301                 .flowi4_tos = RT_TOS(ip4h->tos),
 302                 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC |
 303                                 FLOWI_FLAG_SKIP_NH_OIF,
 304                 .flowi4_proto = ip4h->protocol,
 305                 .daddr = ip4h->daddr,
 306                 .saddr = ip4h->saddr,
 307         };
 308
 309         if (vrf_send_v4_prep(skb, &fl4, vrf_dev))
 310                 goto err;
 311
 312         if (!ip4h->saddr) {
 313                 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
 314                                                RT_SCOPE_LINK);
 315         }
 316
 317         ret = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
 318         if (unlikely(net_xmit_eval(ret)))
 319                 vrf_dev->stats.tx_errors++;
 320         else
 321                 ret = NET_XMIT_SUCCESS;
 322
 323 out:
 324         return ret;
 325 err:
 326         vrf_tx_error(vrf_dev, skb);
 327         goto out;
 328 }
 329
 330 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
 331 {
 332         /* strip the ethernet header added for pass through VRF device */
 333         __skb_pull(skb, skb_network_offset(skb));
 334
 335         switch (skb->protocol) {
 336         case htons(ETH_P_IP):
 337                 return vrf_process_v4_outbound(skb, dev);
 338         case htons(ETH_P_IPV6):
 339                 return vrf_process_v6_outbound(skb, dev);
 340         default:
 341                 vrf_tx_error(dev, skb);
 342                 return NET_XMIT_DROP;
 343         }
 344 }
 345
 346 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
 347 {
 348         int len = skb->len;
 349         netdev_tx_t ret = is_ip_tx_frame(skb, dev);
 350
 351         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 352                 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
 353
 354                 u64_stats_update_begin(&dstats->syncp);
 355                 dstats->tx_pkts++;
 356                 dstats->tx_bytes += len;
 357                 u64_stats_update_end(&dstats->syncp);
 358         } else {
 359                 this_cpu_inc(dev->dstats->tx_drps);
 360         }
 361
 362         return ret;
 363 }
 364
 365 #if IS_ENABLED(CONFIG_IPV6)
 366 static struct dst_entry *vrf_ip6_check(struct dst_entry *dst, u32 cookie)
 367 {
 368         return dst;
 369 }
 370
 371 static struct dst_ops vrf_dst_ops6 = {
 372         .family         = AF_INET6,
 373         .local_out      = ip6_local_out,
 374         .check          = vrf_ip6_check,
 375         .mtu            = vrf_v4_mtu,
 376         .destroy        = vrf_dst_destroy,
 377         .default_advmss = vrf_default_advmss,
 378 };
 379
 380 static int init_dst_ops6_kmem_cachep(void)
 381 {
 382         vrf_dst_ops6.kmem_cachep = kmem_cache_create("vrf_ip6_dst_cache",
 383                                                      sizeof(struct rt6_info),
 384                                                      0,
 385                                                      SLAB_HWCACHE_ALIGN,
 386                                                      NULL);
 387
 388         if (!vrf_dst_ops6.kmem_cachep)
 389                 return -ENOMEM;
 390
 391         return 0;
 392 }
 393
 394 static void free_dst_ops6_kmem_cachep(void)
 395 {
 396         kmem_cache_destroy(vrf_dst_ops6.kmem_cachep);
 397 }
 398
 399 static int vrf_input6(struct sk_buff *skb)
 400 {
 401         skb->dev->stats.rx_errors++;
 402         kfree_skb(skb);
 403         return 0;
 404 }
 405
 406 /* modelled after ip6_finish_output2 */
 407 static int vrf_finish_output6(struct net *net, struct sock *sk,
 408                               struct sk_buff *skb)
 409 {
 410         struct dst_entry *dst = skb_dst(skb);
 411         struct net_device *dev = dst->dev;
 412         struct neighbour *neigh;
 413         struct in6_addr *nexthop;
 414         int ret;
 415
 416         nf_reset(skb);
 417
 418         skb->protocol = htons(ETH_P_IPV6);
 419         skb->dev = dev;
 420
 421         rcu_read_lock_bh();
 422         nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
 423         neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
 424         if (unlikely(!neigh))
 425                 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
 426         if (!IS_ERR(neigh)) {
 427                 ret = dst_neigh_output(dst, neigh, skb);
 428                 rcu_read_unlock_bh();
 429                 return ret;
 430         }
 431         rcu_read_unlock_bh();
 432
 433         IP6_INC_STATS(dev_net(dst->dev),
 434                       ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
 435         kfree_skb(skb);
 436         return -EINVAL;
 437 }
 438
 439 /* modelled after ip6_output */
 440 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
 441 {
 442         return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
 443                             net, sk, skb, NULL, skb_dst(skb)->dev,
 444                             vrf_finish_output6,
 445                             !(IP6CB(skb)->flags & IP6SKB_REROUTED));
 446 }
 447
 448 static void vrf_rt6_destroy(struct net_vrf *vrf)
 449 {
 450         dst_destroy(&vrf->rt6->dst);
 451         free_percpu(vrf->rt6->rt6i_pcpu);
 452         vrf->rt6 = NULL;
 453 }
 454
 455 static int vrf_rt6_create(struct net_device *dev)
 456 {
 457         struct net_vrf *vrf = netdev_priv(dev);
 458         struct dst_entry *dst;
 459         struct rt6_info *rt6;
 460         int cpu;
 461         int rc = -ENOMEM;
 462
 463         rt6 = dst_alloc(&vrf_dst_ops6, dev, 0,
 464                         DST_OBSOLETE_NONE,
 465                         (DST_HOST | DST_NOPOLICY | DST_NOXFRM));
 466         if (!rt6)
 467                 goto out;
 468
 469         dst = &rt6->dst;
 470
 471         rt6->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, GFP_KERNEL);
 472         if (!rt6->rt6i_pcpu) {
 473                 dst_destroy(dst);
 474                 goto out;
 475         }
 476         for_each_possible_cpu(cpu) {
 477                 struct rt6_info **p = per_cpu_ptr(rt6->rt6i_pcpu, cpu);
 478                 *p =  NULL;
 479         }
 480
 481         memset(dst + 1, 0, sizeof(*rt6) - sizeof(*dst));
 482
 483         INIT_LIST_HEAD(&rt6->rt6i_siblings);
 484         INIT_LIST_HEAD(&rt6->rt6i_uncached);
 485
 486         rt6->dst.input  = vrf_input6;
 487         rt6->dst.output = vrf_output6;
 488
 489         rt6->rt6i_table = fib6_get_table(dev_net(dev), vrf->tb_id);
 490
 491         atomic_set(&rt6->dst.__refcnt, 2);
 492
 493         vrf->rt6 = rt6;
 494         rc = 0;
 495 out:
 496         return rc;
 497 }
 498 #else
 499 static int init_dst_ops6_kmem_cachep(void)
 500 {
 501         return 0;
 502 }
 503
 504 static void free_dst_ops6_kmem_cachep(void)
 505 {
 506 }
 507
 508 static void vrf_rt6_destroy(struct net_vrf *vrf)
 509 {
 510 }
 511
 512 static int vrf_rt6_create(struct net_device *dev)
 513 {
 514         return 0;
 515 }
 516 #endif
 517
 518 /* modelled after ip_finish_output2 */
 519 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 520 {
 521         struct dst_entry *dst = skb_dst(skb);
 522         struct rtable *rt = (struct rtable *)dst;
 523         struct net_device *dev = dst->dev;
 524         unsigned int hh_len = LL_RESERVED_SPACE(dev);
 525         struct neighbour *neigh;
 526         u32 nexthop;
 527         int ret = -EINVAL;
 528
 529         nf_reset(skb);
 530
 531         /* Be paranoid, rather than too clever. */
 532         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
 533                 struct sk_buff *skb2;
 534
 535                 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
 536                 if (!skb2) {
 537                         ret = -ENOMEM;
 538                         goto err;
 539                 }
 540                 if (skb->sk)
 541                         skb_set_owner_w(skb2, skb->sk);
 542
 543                 consume_skb(skb);
 544                 skb = skb2;
 545         }
 546
 547         rcu_read_lock_bh();
 548
 549         nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
 550         neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
 551         if (unlikely(!neigh))
 552                 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
 553         if (!IS_ERR(neigh))
 554                 ret = dst_neigh_output(dst, neigh, skb);
 555
 556         rcu_read_unlock_bh();
 557 err:
 558         if (unlikely(ret < 0))
 559                 vrf_tx_error(skb->dev, skb);
 560         return ret;
 561 }
 562
 563 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 564 {
 565         struct net_device *dev = skb_dst(skb)->dev;
 566
 567         IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
 568
 569         skb->dev = dev;
 570         skb->protocol = htons(ETH_P_IP);
 571
 572         return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
 573                             net, sk, skb, NULL, dev,
 574                             vrf_finish_output,
 575                             !(IPCB(skb)->flags & IPSKB_REROUTED));
 576 }
 577
 578 static void vrf_rtable_destroy(struct net_vrf *vrf)
 579 {
 580         struct dst_entry *dst = (struct dst_entry *)vrf->rth;
 581
 582         dst_destroy(dst);
 583         vrf->rth = NULL;
 584 }
 585
 586 static struct rtable *vrf_rtable_create(struct net_device *dev)
 587 {
 588         struct net_vrf *vrf = netdev_priv(dev);
 589         struct rtable *rth;
 590
 591         rth = dst_alloc(&vrf_dst_ops, dev, 2,
 592                         DST_OBSOLETE_NONE,
 593                         (DST_HOST | DST_NOPOLICY | DST_NOXFRM));
 594         if (rth) {
 595                 rth->dst.output = vrf_output;
 596                 rth->rt_genid   = rt_genid_ipv4(dev_net(dev));
 597                 rth->rt_flags   = 0;
 598                 rth->rt_type    = RTN_UNICAST;
 599                 rth->rt_is_input = 0;
 600                 rth->rt_iif     = 0;
 601                 rth->rt_pmtu    = 0;
 602                 rth->rt_gateway = 0;
 603                 rth->rt_uses_gateway = 0;
 604                 rth->rt_table_id = vrf->tb_id;
 605                 INIT_LIST_HEAD(&rth->rt_uncached);
 606                 rth->rt_uncached_list = NULL;
 607         }
 608
 609         return rth;
 610 }
 611
 612 /**************************** device handling ********************/
 613
 614 /* cycle interface to flush neighbor cache and move routes across tables */
 615 static void cycle_netdev(struct net_device *dev)
 616 {
 617         unsigned int flags = dev->flags;
 618         int ret;
 619
 620         if (!netif_running(dev))
 621                 return;
 622
 623         ret = dev_change_flags(dev, flags & ~IFF_UP);
 624         if (ret >= 0)
 625                 ret = dev_change_flags(dev, flags);
 626
 627         if (ret < 0) {
 628                 netdev_err(dev,
 629                            "Failed to cycle device %s; route tables might be wrong!\n",
 630                            dev->name);
 631         }
 632 }
 633
 634 static struct slave *__vrf_find_slave_dev(struct slave_queue *queue,
 635                                           struct net_device *dev)
 636 {
 637         struct list_head *head = &queue->all_slaves;
 638         struct slave *slave;
 639
 640         list_for_each_entry(slave, head, list) {
 641                 if (slave->dev == dev)
 642                         return slave;
 643         }
 644
 645         return NULL;
 646 }
 647
 648 /* inverse of __vrf_insert_slave */
 649 static void __vrf_remove_slave(struct slave_queue *queue, struct slave *slave)
 650 {
 651         list_del(&slave->list);
 652 }
 653
 654 static void __vrf_insert_slave(struct slave_queue *queue, struct slave *slave)
 655 {
 656         list_add(&slave->list, &queue->all_slaves);
 657 }
 658
 659 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
 660 {
 661         struct slave *slave = kzalloc(sizeof(*slave), GFP_KERNEL);
 662         struct net_vrf *vrf = netdev_priv(dev);
 663         struct slave_queue *queue = &vrf->queue;
 664         int ret = -ENOMEM;
 665
 666         if (!slave)
 667                 goto out_fail;
 668
 669         slave->dev = port_dev;
 670
 671         /* register the packet handler for slave ports */
 672         ret = netdev_rx_handler_register(port_dev, vrf_handle_frame, dev);
 673         if (ret) {
 674                 netdev_err(port_dev,
 675                            "Device %s failed to register rx_handler\n",
 676                            port_dev->name);
 677                 goto out_fail;
 678         }
 679
 680         ret = netdev_master_upper_dev_link(port_dev, dev);
 681         if (ret < 0)
 682                 goto out_unregister;
 683
 684         port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
 685         __vrf_insert_slave(queue, slave);
 686         cycle_netdev(port_dev);
 687
 688         return 0;
 689
 690 out_unregister:
 691         netdev_rx_handler_unregister(port_dev);
 692 out_fail:
 693         kfree(slave);
 694         return ret;
 695 }
 696
 697 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
 698 {
 699         if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
 700                 return -EINVAL;
 701
 702         return do_vrf_add_slave(dev, port_dev);
 703 }
 704
 705 /* inverse of do_vrf_add_slave */
 706 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
 707 {
 708         struct net_vrf *vrf = netdev_priv(dev);
 709         struct slave_queue *queue = &vrf->queue;
 710         struct slave *slave;
 711
 712         netdev_upper_dev_unlink(port_dev, dev);
 713         port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
 714
 715         netdev_rx_handler_unregister(port_dev);
 716
 717         cycle_netdev(port_dev);
 718
 719         slave = __vrf_find_slave_dev(queue, port_dev);
 720         if (slave)
 721                 __vrf_remove_slave(queue, slave);
 722
 723         kfree(slave);
 724
 725         return 0;
 726 }
 727
 728 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
 729 {
 730         return do_vrf_del_slave(dev, port_dev);
 731 }
 732
 733 static void vrf_dev_uninit(struct net_device *dev)
 734 {
 735         struct net_vrf *vrf = netdev_priv(dev);
 736         struct slave_queue *queue = &vrf->queue;
 737         struct list_head *head = &queue->all_slaves;
 738         struct slave *slave, *next;
 739
 740         vrf_rtable_destroy(vrf);
 741         vrf_rt6_destroy(vrf);
 742
 743         list_for_each_entry_safe(slave, next, head, list)
 744                 vrf_del_slave(dev, slave->dev);
 745
 746         free_percpu(dev->dstats);
 747         dev->dstats = NULL;
 748 }
 749
 750 static int vrf_dev_init(struct net_device *dev)
 751 {
 752         struct net_vrf *vrf = netdev_priv(dev);
 753
 754         INIT_LIST_HEAD(&vrf->queue.all_slaves);
 755
 756         dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
 757         if (!dev->dstats)
 758                 goto out_nomem;
 759
 760         /* create the default dst which points back to us */
 761         vrf->rth = vrf_rtable_create(dev);
 762         if (!vrf->rth)
 763                 goto out_stats;
 764
 765         if (vrf_rt6_create(dev) != 0)
 766                 goto out_rth;
 767
 768         dev->flags = IFF_MASTER | IFF_NOARP;
 769
 770         return 0;
 771
 772 out_rth:
 773         vrf_rtable_destroy(vrf);
 774 out_stats:
 775         free_percpu(dev->dstats);
 776         dev->dstats = NULL;
 777 out_nomem:
 778         return -ENOMEM;
 779 }
 780
 781 static const struct net_device_ops vrf_netdev_ops = {
 782         .ndo_init               = vrf_dev_init,
 783         .ndo_uninit             = vrf_dev_uninit,
 784         .ndo_start_xmit         = vrf_xmit,
 785         .ndo_get_stats64        = vrf_get_stats64,
 786         .ndo_add_slave          = vrf_add_slave,
 787         .ndo_del_slave          = vrf_del_slave,
 788 };
 789
 790 static u32 vrf_fib_table(const struct net_device *dev)
 791 {
 792         struct net_vrf *vrf = netdev_priv(dev);
 793
 794         return vrf->tb_id;
 795 }
 796
 797 static struct rtable *vrf_get_rtable(const struct net_device *dev,
 798                                      const struct flowi4 *fl4)
 799 {
 800         struct rtable *rth = NULL;
 801
 802         if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) {
 803                 struct net_vrf *vrf = netdev_priv(dev);
 804
 805                 rth = vrf->rth;
 806                 atomic_inc(&rth->dst.__refcnt);
 807         }
 808
 809         return rth;
 810 }
 811
 812 /* called under rcu_read_lock */
 813 static int vrf_get_saddr(struct net_device *dev, struct flowi4 *fl4)
 814 {
 815         struct fib_result res = { .tclassid = 0 };
 816         struct net *net = dev_net(dev);
 817         u32 orig_tos = fl4->flowi4_tos;
 818         u8 flags = fl4->flowi4_flags;
 819         u8 scope = fl4->flowi4_scope;
 820         u8 tos = RT_FL_TOS(fl4);
 821         int rc;
 822
 823         if (unlikely(!fl4->daddr))
 824                 return 0;
 825
 826         fl4->flowi4_flags |= FLOWI_FLAG_SKIP_NH_OIF;
 827         fl4->flowi4_iif = LOOPBACK_IFINDEX;
 828         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
 829         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
 830                              RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
 831
 832         rc = fib_lookup(net, fl4, &res, 0);
 833         if (!rc) {
 834                 if (res.type == RTN_LOCAL)
 835                         fl4->saddr = res.fi->fib_prefsrc ? : fl4->daddr;
 836                 else
 837                         fib_select_path(net, &res, fl4, -1);
 838         }
 839
 840         fl4->flowi4_flags = flags;
 841         fl4->flowi4_tos = orig_tos;
 842         fl4->flowi4_scope = scope;
 843
 844         return rc;
 845 }
 846
 847 #if IS_ENABLED(CONFIG_IPV6)
 848 static struct dst_entry *vrf_get_rt6_dst(const struct net_device *dev,
 849                                          const struct flowi6 *fl6)
 850 {
 851         struct rt6_info *rt = NULL;
 852
 853         if (!(fl6->flowi6_flags & FLOWI_FLAG_L3MDEV_SRC)) {
 854                 struct net_vrf *vrf = netdev_priv(dev);
 855
 856                 rt = vrf->rt6;
 857                 atomic_inc(&rt->dst.__refcnt);
 858         }
 859
 860         return (struct dst_entry *)rt;
 861 }
 862 #endif
 863
 864 static const struct l3mdev_ops vrf_l3mdev_ops = {
 865         .l3mdev_fib_table       = vrf_fib_table,
 866         .l3mdev_get_rtable      = vrf_get_rtable,
 867         .l3mdev_get_saddr       = vrf_get_saddr,
 868 #if IS_ENABLED(CONFIG_IPV6)
 869         .l3mdev_get_rt6_dst     = vrf_get_rt6_dst,
 870 #endif
 871 };
 872
 873 static void vrf_get_drvinfo(struct net_device *dev,
 874                             struct ethtool_drvinfo *info)
 875 {
 876         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
 877         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
 878 }
 879
 880 static const struct ethtool_ops vrf_ethtool_ops = {
 881         .get_drvinfo    = vrf_get_drvinfo,
 882 };
 883
 884 static void vrf_setup(struct net_device *dev)
 885 {
 886         ether_setup(dev);
 887
 888         /* Initialize the device structure. */
 889         dev->netdev_ops = &vrf_netdev_ops;
 890         dev->l3mdev_ops = &vrf_l3mdev_ops;
 891         dev->ethtool_ops = &vrf_ethtool_ops;
 892         dev->destructor = free_netdev;
 893
 894         /* Fill in device structure with ethernet-generic values. */
 895         eth_hw_addr_random(dev);
 896
 897         /* don't acquire vrf device's netif_tx_lock when transmitting */
 898         dev->features |= NETIF_F_LLTX;
 899
 900         /* don't allow vrf devices to change network namespaces. */
 901         dev->features |= NETIF_F_NETNS_LOCAL;
 902 }
 903
 904 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
 905 {
 906         if (tb[IFLA_ADDRESS]) {
 907                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 908                         return -EINVAL;
 909                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 910                         return -EADDRNOTAVAIL;
 911         }
 912         return 0;
 913 }
 914
 915 static void vrf_dellink(struct net_device *dev, struct list_head *head)
 916 {
 917         unregister_netdevice_queue(dev, head);
 918 }
 919
 920 static int vrf_newlink(struct net *src_net, struct net_device *dev,
 921                        struct nlattr *tb[], struct nlattr *data[])
 922 {
 923         struct net_vrf *vrf = netdev_priv(dev);
 924
 925         if (!data || !data[IFLA_VRF_TABLE])
 926                 return -EINVAL;
 927
 928         vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
 929         if (vrf->tb_id == RT_TABLE_UNSPEC)
 930                 return -EINVAL;
 931
 932         dev->priv_flags |= IFF_L3MDEV_MASTER;
 933
 934         return register_netdevice(dev);
 935 }
 936
 937 static size_t vrf_nl_getsize(const struct net_device *dev)
 938 {
 939         return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
 940 }
 941
 942 static int vrf_fillinfo(struct sk_buff *skb,
 943                         const struct net_device *dev)
 944 {
 945         struct net_vrf *vrf = netdev_priv(dev);
 946
 947         return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
 948 }
 949
 950 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
 951         [IFLA_VRF_TABLE] = { .type = NLA_U32 },
 952 };
 953
 954 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
 955         .kind           = DRV_NAME,
 956         .priv_size      = sizeof(struct net_vrf),
 957
 958         .get_size       = vrf_nl_getsize,
 959         .policy         = vrf_nl_policy,
 960         .validate       = vrf_validate,
 961         .fill_info      = vrf_fillinfo,
 962
 963         .newlink        = vrf_newlink,
 964         .dellink        = vrf_dellink,
 965         .setup          = vrf_setup,
 966         .maxtype        = IFLA_VRF_MAX,
 967 };
 968
 969 static int vrf_device_event(struct notifier_block *unused,
 970                             unsigned long event, void *ptr)
 971 {
 972         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 973
 974         /* only care about unregister events to drop slave references */
 975         if (event == NETDEV_UNREGISTER) {
 976                 struct net_device *vrf_dev;
 977
 978                 if (!netif_is_l3_slave(dev))
 979                         goto out;
 980
 981                 vrf_dev = netdev_master_upper_dev_get(dev);
 982                 vrf_del_slave(vrf_dev, dev);
 983         }
 984 out:
 985         return NOTIFY_DONE;
 986 }
 987
 988 static struct notifier_block vrf_notifier_block __read_mostly = {
 989         .notifier_call = vrf_device_event,
 990 };
 991
 992 static int __init vrf_init_module(void)
 993 {
 994         int rc;
 995
 996         vrf_dst_ops.kmem_cachep =
 997                 kmem_cache_create("vrf_ip_dst_cache",
 998                                   sizeof(struct rtable), 0,
 999                                   SLAB_HWCACHE_ALIGN,
1000                                   NULL);
1001
1002         if (!vrf_dst_ops.kmem_cachep)
1003                 return -ENOMEM;
1004
1005         rc = init_dst_ops6_kmem_cachep();
1006         if (rc != 0)
1007                 goto error2;
1008
1009         register_netdevice_notifier(&vrf_notifier_block);
1010
1011         rc = rtnl_link_register(&vrf_link_ops);
1012         if (rc < 0)
1013                 goto error;
1014
1015         return 0;
1016
1017 error:
1018         unregister_netdevice_notifier(&vrf_notifier_block);
1019         free_dst_ops6_kmem_cachep();
1020 error2:
1021         kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
1022         return rc;
1023 }
1024
1025 static void __exit vrf_cleanup_module(void)
1026 {
1027         rtnl_link_unregister(&vrf_link_ops);
1028         unregister_netdevice_notifier(&vrf_notifier_block);
1029         kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
1030         free_dst_ops6_kmem_cachep();
1031 }
1032
1033 module_init(vrf_init_module);
1034 module_exit(vrf_cleanup_module);
1035 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1036 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1037 MODULE_LICENSE("GPL");
1038 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1039 MODULE_VERSION(DRV_VERSION);