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