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                 .daddr = ip4h->daddr,
 305         };
 306
 307         if (vrf_send_v4_prep(skb, &fl4, vrf_dev))
 308                 goto err;
 309
 310         if (!ip4h->saddr) {
 311                 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
 312                                                RT_SCOPE_LINK);
 313         }
 314
 315         ret = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
 316         if (unlikely(net_xmit_eval(ret)))
 317                 vrf_dev->stats.tx_errors++;
 318         else
 319                 ret = NET_XMIT_SUCCESS;
 320
 321 out:
 322         return ret;
 323 err:
 324         vrf_tx_error(vrf_dev, skb);
 325         goto out;
 326 }
 327
 328 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
 329 {
 330         /* strip the ethernet header added for pass through VRF device */
 331         __skb_pull(skb, skb_network_offset(skb));
 332
 333         switch (skb->protocol) {
 334         case htons(ETH_P_IP):
 335                 return vrf_process_v4_outbound(skb, dev);
 336         case htons(ETH_P_IPV6):
 337                 return vrf_process_v6_outbound(skb, dev);
 338         default:
 339                 vrf_tx_error(dev, skb);
 340                 return NET_XMIT_DROP;
 341         }
 342 }
 343
 344 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
 345 {
 346         netdev_tx_t ret = is_ip_tx_frame(skb, dev);
 347
 348         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 349                 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
 350
 351                 u64_stats_update_begin(&dstats->syncp);
 352                 dstats->tx_pkts++;
 353                 dstats->tx_bytes += skb->len;
 354                 u64_stats_update_end(&dstats->syncp);
 355         } else {
 356                 this_cpu_inc(dev->dstats->tx_drps);
 357         }
 358
 359         return ret;
 360 }
 361
 362 #if IS_ENABLED(CONFIG_IPV6)
 363 static struct dst_entry *vrf_ip6_check(struct dst_entry *dst, u32 cookie)
 364 {
 365         return dst;
 366 }
 367
 368 static struct dst_ops vrf_dst_ops6 = {
 369         .family         = AF_INET6,
 370         .local_out      = ip6_local_out,
 371         .check          = vrf_ip6_check,
 372         .mtu            = vrf_v4_mtu,
 373         .destroy        = vrf_dst_destroy,
 374         .default_advmss = vrf_default_advmss,
 375 };
 376
 377 static int init_dst_ops6_kmem_cachep(void)
 378 {
 379         vrf_dst_ops6.kmem_cachep = kmem_cache_create("vrf_ip6_dst_cache",
 380                                                      sizeof(struct rt6_info),
 381                                                      0,
 382                                                      SLAB_HWCACHE_ALIGN,
 383                                                      NULL);
 384
 385         if (!vrf_dst_ops6.kmem_cachep)
 386                 return -ENOMEM;
 387
 388         return 0;
 389 }
 390
 391 static void free_dst_ops6_kmem_cachep(void)
 392 {
 393         kmem_cache_destroy(vrf_dst_ops6.kmem_cachep);
 394 }
 395
 396 static int vrf_input6(struct sk_buff *skb)
 397 {
 398         skb->dev->stats.rx_errors++;
 399         kfree_skb(skb);
 400         return 0;
 401 }
 402
 403 /* modelled after ip6_finish_output2 */
 404 static int vrf_finish_output6(struct net *net, struct sock *sk,
 405                               struct sk_buff *skb)
 406 {
 407         struct dst_entry *dst = skb_dst(skb);
 408         struct net_device *dev = dst->dev;
 409         struct neighbour *neigh;
 410         struct in6_addr *nexthop;
 411         int ret;
 412
 413         skb->protocol = htons(ETH_P_IPV6);
 414         skb->dev = dev;
 415
 416         rcu_read_lock_bh();
 417         nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
 418         neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
 419         if (unlikely(!neigh))
 420                 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
 421         if (!IS_ERR(neigh)) {
 422                 ret = dst_neigh_output(dst, neigh, skb);
 423                 rcu_read_unlock_bh();
 424                 return ret;
 425         }
 426         rcu_read_unlock_bh();
 427
 428         IP6_INC_STATS(dev_net(dst->dev),
 429                       ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
 430         kfree_skb(skb);
 431         return -EINVAL;
 432 }
 433
 434 /* modelled after ip6_output */
 435 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
 436 {
 437         return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
 438                             net, sk, skb, NULL, skb_dst(skb)->dev,
 439                             vrf_finish_output6,
 440                             !(IP6CB(skb)->flags & IP6SKB_REROUTED));
 441 }
 442
 443 static void vrf_rt6_destroy(struct net_vrf *vrf)
 444 {
 445         dst_destroy(&vrf->rt6->dst);
 446         free_percpu(vrf->rt6->rt6i_pcpu);
 447         vrf->rt6 = NULL;
 448 }
 449
 450 static int vrf_rt6_create(struct net_device *dev)
 451 {
 452         struct net_vrf *vrf = netdev_priv(dev);
 453         struct dst_entry *dst;
 454         struct rt6_info *rt6;
 455         int cpu;
 456         int rc = -ENOMEM;
 457
 458         rt6 = dst_alloc(&vrf_dst_ops6, dev, 0,
 459                         DST_OBSOLETE_NONE,
 460                         (DST_HOST | DST_NOPOLICY | DST_NOXFRM));
 461         if (!rt6)
 462                 goto out;
 463
 464         dst = &rt6->dst;
 465
 466         rt6->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, GFP_KERNEL);
 467         if (!rt6->rt6i_pcpu) {
 468                 dst_destroy(dst);
 469                 goto out;
 470         }
 471         for_each_possible_cpu(cpu) {
 472                 struct rt6_info **p = per_cpu_ptr(rt6->rt6i_pcpu, cpu);
 473                 *p =  NULL;
 474         }
 475
 476         memset(dst + 1, 0, sizeof(*rt6) - sizeof(*dst));
 477
 478         INIT_LIST_HEAD(&rt6->rt6i_siblings);
 479         INIT_LIST_HEAD(&rt6->rt6i_uncached);
 480
 481         rt6->dst.input  = vrf_input6;
 482         rt6->dst.output = vrf_output6;
 483
 484         rt6->rt6i_table = fib6_get_table(dev_net(dev), vrf->tb_id);
 485
 486         atomic_set(&rt6->dst.__refcnt, 2);
 487
 488         vrf->rt6 = rt6;
 489         rc = 0;
 490 out:
 491         return rc;
 492 }
 493 #else
 494 static int init_dst_ops6_kmem_cachep(void)
 495 {
 496         return 0;
 497 }
 498
 499 static void free_dst_ops6_kmem_cachep(void)
 500 {
 501 }
 502
 503 static void vrf_rt6_destroy(struct net_vrf *vrf)
 504 {
 505 }
 506
 507 static int vrf_rt6_create(struct net_device *dev)
 508 {
 509         return 0;
 510 }
 511 #endif
 512
 513 /* modelled after ip_finish_output2 */
 514 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 515 {
 516         struct dst_entry *dst = skb_dst(skb);
 517         struct rtable *rt = (struct rtable *)dst;
 518         struct net_device *dev = dst->dev;
 519         unsigned int hh_len = LL_RESERVED_SPACE(dev);
 520         struct neighbour *neigh;
 521         u32 nexthop;
 522         int ret = -EINVAL;
 523
 524         /* Be paranoid, rather than too clever. */
 525         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
 526                 struct sk_buff *skb2;
 527
 528                 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
 529                 if (!skb2) {
 530                         ret = -ENOMEM;
 531                         goto err;
 532                 }
 533                 if (skb->sk)
 534                         skb_set_owner_w(skb2, skb->sk);
 535
 536                 consume_skb(skb);
 537                 skb = skb2;
 538         }
 539
 540         rcu_read_lock_bh();
 541
 542         nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
 543         neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
 544         if (unlikely(!neigh))
 545                 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
 546         if (!IS_ERR(neigh))
 547                 ret = dst_neigh_output(dst, neigh, skb);
 548
 549         rcu_read_unlock_bh();
 550 err:
 551         if (unlikely(ret < 0))
 552                 vrf_tx_error(skb->dev, skb);
 553         return ret;
 554 }
 555
 556 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 557 {
 558         struct net_device *dev = skb_dst(skb)->dev;
 559
 560         IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
 561
 562         skb->dev = dev;
 563         skb->protocol = htons(ETH_P_IP);
 564
 565         return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
 566                             net, sk, skb, NULL, dev,
 567                             vrf_finish_output,
 568                             !(IPCB(skb)->flags & IPSKB_REROUTED));
 569 }
 570
 571 static void vrf_rtable_destroy(struct net_vrf *vrf)
 572 {
 573         struct dst_entry *dst = (struct dst_entry *)vrf->rth;
 574
 575         dst_destroy(dst);
 576         vrf->rth = NULL;
 577 }
 578
 579 static struct rtable *vrf_rtable_create(struct net_device *dev)
 580 {
 581         struct net_vrf *vrf = netdev_priv(dev);
 582         struct rtable *rth;
 583
 584         rth = dst_alloc(&vrf_dst_ops, dev, 2,
 585                         DST_OBSOLETE_NONE,
 586                         (DST_HOST | DST_NOPOLICY | DST_NOXFRM));
 587         if (rth) {
 588                 rth->dst.output = vrf_output;
 589                 rth->rt_genid   = rt_genid_ipv4(dev_net(dev));
 590                 rth->rt_flags   = 0;
 591                 rth->rt_type    = RTN_UNICAST;
 592                 rth->rt_is_input = 0;
 593                 rth->rt_iif     = 0;
 594                 rth->rt_pmtu    = 0;
 595                 rth->rt_gateway = 0;
 596                 rth->rt_uses_gateway = 0;
 597                 rth->rt_table_id = vrf->tb_id;
 598                 INIT_LIST_HEAD(&rth->rt_uncached);
 599                 rth->rt_uncached_list = NULL;
 600         }
 601
 602         return rth;
 603 }
 604
 605 /**************************** device handling ********************/
 606
 607 /* cycle interface to flush neighbor cache and move routes across tables */
 608 static void cycle_netdev(struct net_device *dev)
 609 {
 610         unsigned int flags = dev->flags;
 611         int ret;
 612
 613         if (!netif_running(dev))
 614                 return;
 615
 616         ret = dev_change_flags(dev, flags & ~IFF_UP);
 617         if (ret >= 0)
 618                 ret = dev_change_flags(dev, flags);
 619
 620         if (ret < 0) {
 621                 netdev_err(dev,
 622                            "Failed to cycle device %s; route tables might be wrong!\n",
 623                            dev->name);
 624         }
 625 }
 626
 627 static struct slave *__vrf_find_slave_dev(struct slave_queue *queue,
 628                                           struct net_device *dev)
 629 {
 630         struct list_head *head = &queue->all_slaves;
 631         struct slave *slave;
 632
 633         list_for_each_entry(slave, head, list) {
 634                 if (slave->dev == dev)
 635                         return slave;
 636         }
 637
 638         return NULL;
 639 }
 640
 641 /* inverse of __vrf_insert_slave */
 642 static void __vrf_remove_slave(struct slave_queue *queue, struct slave *slave)
 643 {
 644         list_del(&slave->list);
 645 }
 646
 647 static void __vrf_insert_slave(struct slave_queue *queue, struct slave *slave)
 648 {
 649         list_add(&slave->list, &queue->all_slaves);
 650 }
 651
 652 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
 653 {
 654         struct slave *slave = kzalloc(sizeof(*slave), GFP_KERNEL);
 655         struct net_vrf *vrf = netdev_priv(dev);
 656         struct slave_queue *queue = &vrf->queue;
 657         int ret = -ENOMEM;
 658
 659         if (!slave)
 660                 goto out_fail;
 661
 662         slave->dev = port_dev;
 663
 664         /* register the packet handler for slave ports */
 665         ret = netdev_rx_handler_register(port_dev, vrf_handle_frame, dev);
 666         if (ret) {
 667                 netdev_err(port_dev,
 668                            "Device %s failed to register rx_handler\n",
 669                            port_dev->name);
 670                 goto out_fail;
 671         }
 672
 673         ret = netdev_master_upper_dev_link(port_dev, dev);
 674         if (ret < 0)
 675                 goto out_unregister;
 676
 677         port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
 678         __vrf_insert_slave(queue, slave);
 679         cycle_netdev(port_dev);
 680
 681         return 0;
 682
 683 out_unregister:
 684         netdev_rx_handler_unregister(port_dev);
 685 out_fail:
 686         kfree(slave);
 687         return ret;
 688 }
 689
 690 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
 691 {
 692         if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
 693                 return -EINVAL;
 694
 695         return do_vrf_add_slave(dev, port_dev);
 696 }
 697
 698 /* inverse of do_vrf_add_slave */
 699 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
 700 {
 701         struct net_vrf *vrf = netdev_priv(dev);
 702         struct slave_queue *queue = &vrf->queue;
 703         struct slave *slave;
 704
 705         netdev_upper_dev_unlink(port_dev, dev);
 706         port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
 707
 708         netdev_rx_handler_unregister(port_dev);
 709
 710         cycle_netdev(port_dev);
 711
 712         slave = __vrf_find_slave_dev(queue, port_dev);
 713         if (slave)
 714                 __vrf_remove_slave(queue, slave);
 715
 716         kfree(slave);
 717
 718         return 0;
 719 }
 720
 721 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
 722 {
 723         return do_vrf_del_slave(dev, port_dev);
 724 }
 725
 726 static void vrf_dev_uninit(struct net_device *dev)
 727 {
 728         struct net_vrf *vrf = netdev_priv(dev);
 729         struct slave_queue *queue = &vrf->queue;
 730         struct list_head *head = &queue->all_slaves;
 731         struct slave *slave, *next;
 732
 733         vrf_rtable_destroy(vrf);
 734         vrf_rt6_destroy(vrf);
 735
 736         list_for_each_entry_safe(slave, next, head, list)
 737                 vrf_del_slave(dev, slave->dev);
 738
 739         free_percpu(dev->dstats);
 740         dev->dstats = NULL;
 741 }
 742
 743 static int vrf_dev_init(struct net_device *dev)
 744 {
 745         struct net_vrf *vrf = netdev_priv(dev);
 746
 747         INIT_LIST_HEAD(&vrf->queue.all_slaves);
 748
 749         dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
 750         if (!dev->dstats)
 751                 goto out_nomem;
 752
 753         /* create the default dst which points back to us */
 754         vrf->rth = vrf_rtable_create(dev);
 755         if (!vrf->rth)
 756                 goto out_stats;
 757
 758         if (vrf_rt6_create(dev) != 0)
 759                 goto out_rth;
 760
 761         dev->flags = IFF_MASTER | IFF_NOARP;
 762
 763         return 0;
 764
 765 out_rth:
 766         vrf_rtable_destroy(vrf);
 767 out_stats:
 768         free_percpu(dev->dstats);
 769         dev->dstats = NULL;
 770 out_nomem:
 771         return -ENOMEM;
 772 }
 773
 774 static const struct net_device_ops vrf_netdev_ops = {
 775         .ndo_init               = vrf_dev_init,
 776         .ndo_uninit             = vrf_dev_uninit,
 777         .ndo_start_xmit         = vrf_xmit,
 778         .ndo_get_stats64        = vrf_get_stats64,
 779         .ndo_add_slave          = vrf_add_slave,
 780         .ndo_del_slave          = vrf_del_slave,
 781 };
 782
 783 static u32 vrf_fib_table(const struct net_device *dev)
 784 {
 785         struct net_vrf *vrf = netdev_priv(dev);
 786
 787         return vrf->tb_id;
 788 }
 789
 790 static struct rtable *vrf_get_rtable(const struct net_device *dev,
 791                                      const struct flowi4 *fl4)
 792 {
 793         struct rtable *rth = NULL;
 794
 795         if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) {
 796                 struct net_vrf *vrf = netdev_priv(dev);
 797
 798                 rth = vrf->rth;
 799                 atomic_inc(&rth->dst.__refcnt);
 800         }
 801
 802         return rth;
 803 }
 804
 805 /* called under rcu_read_lock */
 806 static int vrf_get_saddr(struct net_device *dev, struct flowi4 *fl4)
 807 {
 808         struct fib_result res = { .tclassid = 0 };
 809         struct net *net = dev_net(dev);
 810         u32 orig_tos = fl4->flowi4_tos;
 811         u8 flags = fl4->flowi4_flags;
 812         u8 scope = fl4->flowi4_scope;
 813         u8 tos = RT_FL_TOS(fl4);
 814         int rc;
 815
 816         if (unlikely(!fl4->daddr))
 817                 return 0;
 818
 819         fl4->flowi4_flags |= FLOWI_FLAG_SKIP_NH_OIF;
 820         fl4->flowi4_iif = LOOPBACK_IFINDEX;
 821         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
 822         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
 823                              RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
 824
 825         rc = fib_lookup(net, fl4, &res, 0);
 826         if (!rc) {
 827                 if (res.type == RTN_LOCAL)
 828                         fl4->saddr = res.fi->fib_prefsrc ? : fl4->daddr;
 829                 else
 830                         fib_select_path(net, &res, fl4, -1);
 831         }
 832
 833         fl4->flowi4_flags = flags;
 834         fl4->flowi4_tos = orig_tos;
 835         fl4->flowi4_scope = scope;
 836
 837         return rc;
 838 }
 839
 840 #if IS_ENABLED(CONFIG_IPV6)
 841 static struct dst_entry *vrf_get_rt6_dst(const struct net_device *dev,
 842                                          const struct flowi6 *fl6)
 843 {
 844         struct rt6_info *rt = NULL;
 845
 846         if (!(fl6->flowi6_flags & FLOWI_FLAG_L3MDEV_SRC)) {
 847                 struct net_vrf *vrf = netdev_priv(dev);
 848
 849                 rt = vrf->rt6;
 850                 atomic_inc(&rt->dst.__refcnt);
 851         }
 852
 853         return (struct dst_entry *)rt;
 854 }
 855 #endif
 856
 857 static const struct l3mdev_ops vrf_l3mdev_ops = {
 858         .l3mdev_fib_table       = vrf_fib_table,
 859         .l3mdev_get_rtable      = vrf_get_rtable,
 860         .l3mdev_get_saddr       = vrf_get_saddr,
 861 #if IS_ENABLED(CONFIG_IPV6)
 862         .l3mdev_get_rt6_dst     = vrf_get_rt6_dst,
 863 #endif
 864 };
 865
 866 static void vrf_get_drvinfo(struct net_device *dev,
 867                             struct ethtool_drvinfo *info)
 868 {
 869         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
 870         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
 871 }
 872
 873 static const struct ethtool_ops vrf_ethtool_ops = {
 874         .get_drvinfo    = vrf_get_drvinfo,
 875 };
 876
 877 static void vrf_setup(struct net_device *dev)
 878 {
 879         ether_setup(dev);
 880
 881         /* Initialize the device structure. */
 882         dev->netdev_ops = &vrf_netdev_ops;
 883         dev->l3mdev_ops = &vrf_l3mdev_ops;
 884         dev->ethtool_ops = &vrf_ethtool_ops;
 885         dev->destructor = free_netdev;
 886
 887         /* Fill in device structure with ethernet-generic values. */
 888         eth_hw_addr_random(dev);
 889
 890         /* don't acquire vrf device's netif_tx_lock when transmitting */
 891         dev->features |= NETIF_F_LLTX;
 892
 893         /* don't allow vrf devices to change network namespaces. */
 894         dev->features |= NETIF_F_NETNS_LOCAL;
 895 }
 896
 897 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
 898 {
 899         if (tb[IFLA_ADDRESS]) {
 900                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 901                         return -EINVAL;
 902                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 903                         return -EADDRNOTAVAIL;
 904         }
 905         return 0;
 906 }
 907
 908 static void vrf_dellink(struct net_device *dev, struct list_head *head)
 909 {
 910         unregister_netdevice_queue(dev, head);
 911 }
 912
 913 static int vrf_newlink(struct net *src_net, struct net_device *dev,
 914                        struct nlattr *tb[], struct nlattr *data[])
 915 {
 916         struct net_vrf *vrf = netdev_priv(dev);
 917
 918         if (!data || !data[IFLA_VRF_TABLE])
 919                 return -EINVAL;
 920
 921         vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
 922
 923         dev->priv_flags |= IFF_L3MDEV_MASTER;
 924
 925         return register_netdevice(dev);
 926 }
 927
 928 static size_t vrf_nl_getsize(const struct net_device *dev)
 929 {
 930         return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
 931 }
 932
 933 static int vrf_fillinfo(struct sk_buff *skb,
 934                         const struct net_device *dev)
 935 {
 936         struct net_vrf *vrf = netdev_priv(dev);
 937
 938         return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
 939 }
 940
 941 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
 942         [IFLA_VRF_TABLE] = { .type = NLA_U32 },
 943 };
 944
 945 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
 946         .kind           = DRV_NAME,
 947         .priv_size      = sizeof(struct net_vrf),
 948
 949         .get_size       = vrf_nl_getsize,
 950         .policy         = vrf_nl_policy,
 951         .validate       = vrf_validate,
 952         .fill_info      = vrf_fillinfo,
 953
 954         .newlink        = vrf_newlink,
 955         .dellink        = vrf_dellink,
 956         .setup          = vrf_setup,
 957         .maxtype        = IFLA_VRF_MAX,
 958 };
 959
 960 static int vrf_device_event(struct notifier_block *unused,
 961                             unsigned long event, void *ptr)
 962 {
 963         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 964
 965         /* only care about unregister events to drop slave references */
 966         if (event == NETDEV_UNREGISTER) {
 967                 struct net_device *vrf_dev;
 968
 969                 if (!netif_is_l3_slave(dev))
 970                         goto out;
 971
 972                 vrf_dev = netdev_master_upper_dev_get(dev);
 973                 vrf_del_slave(vrf_dev, dev);
 974         }
 975 out:
 976         return NOTIFY_DONE;
 977 }
 978
 979 static struct notifier_block vrf_notifier_block __read_mostly = {
 980         .notifier_call = vrf_device_event,
 981 };
 982
 983 static int __init vrf_init_module(void)
 984 {
 985         int rc;
 986
 987         vrf_dst_ops.kmem_cachep =
 988                 kmem_cache_create("vrf_ip_dst_cache",
 989                                   sizeof(struct rtable), 0,
 990                                   SLAB_HWCACHE_ALIGN,
 991                                   NULL);
 992
 993         if (!vrf_dst_ops.kmem_cachep)
 994                 return -ENOMEM;
 995
 996         rc = init_dst_ops6_kmem_cachep();
 997         if (rc != 0)
 998                 goto error2;
 999
1000         register_netdevice_notifier(&vrf_notifier_block);
1001
1002         rc = rtnl_link_register(&vrf_link_ops);
1003         if (rc < 0)
1004                 goto error;
1005
1006         return 0;
1007
1008 error:
1009         unregister_netdevice_notifier(&vrf_notifier_block);
1010         free_dst_ops6_kmem_cachep();
1011 error2:
1012         kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
1013         return rc;
1014 }
1015
1016 static void __exit vrf_cleanup_module(void)
1017 {
1018         rtnl_link_unregister(&vrf_link_ops);
1019         unregister_netdevice_notifier(&vrf_notifier_block);
1020         kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
1021         free_dst_ops6_kmem_cachep();
1022 }
1023
1024 module_init(vrf_init_module);
1025 module_exit(vrf_cleanup_module);
1026 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1027 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1028 MODULE_LICENSE("GPL");
1029 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1030 MODULE_VERSION(DRV_VERSION);