Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0-or-later
0002 /*
0003  * vrf.c: device driver to encapsulate a VRF space
0004  *
0005  * Copyright (c) 2015 Cumulus Networks. All rights reserved.
0006  * Copyright (c) 2015 Shrijeet Mukherjee <shm@cumulusnetworks.com>
0007  * Copyright (c) 2015 David Ahern <dsa@cumulusnetworks.com>
0008  *
0009  * Based on dummy, team and ipvlan drivers
0010  */
0011 
0012 #include <linux/ethtool.h>
0013 #include <linux/module.h>
0014 #include <linux/kernel.h>
0015 #include <linux/netdevice.h>
0016 #include <linux/etherdevice.h>
0017 #include <linux/ip.h>
0018 #include <linux/init.h>
0019 #include <linux/moduleparam.h>
0020 #include <linux/netfilter.h>
0021 #include <linux/rtnetlink.h>
0022 #include <net/rtnetlink.h>
0023 #include <linux/u64_stats_sync.h>
0024 #include <linux/hashtable.h>
0025 #include <linux/spinlock_types.h>
0026 
0027 #include <linux/inetdevice.h>
0028 #include <net/arp.h>
0029 #include <net/ip.h>
0030 #include <net/ip_fib.h>
0031 #include <net/ip6_fib.h>
0032 #include <net/ip6_route.h>
0033 #include <net/route.h>
0034 #include <net/addrconf.h>
0035 #include <net/l3mdev.h>
0036 #include <net/fib_rules.h>
0037 #include <net/sch_generic.h>
0038 #include <net/netns/generic.h>
0039 #include <net/netfilter/nf_conntrack.h>
0040 
0041 #define DRV_NAME    "vrf"
0042 #define DRV_VERSION "1.1"
0043 
0044 #define FIB_RULE_PREF  1000       /* default preference for FIB rules */
0045 
0046 #define HT_MAP_BITS 4
0047 #define HASH_INITVAL    ((u32)0xcafef00d)
0048 
0049 struct  vrf_map {
0050     DECLARE_HASHTABLE(ht, HT_MAP_BITS);
0051     spinlock_t vmap_lock;
0052 
0053     /* shared_tables:
0054      * count how many distinct tables do not comply with the strict mode
0055      * requirement.
0056      * shared_tables value must be 0 in order to enable the strict mode.
0057      *
0058      * example of the evolution of shared_tables:
0059      *                                                        | time
0060      * add  vrf0 --> table 100        shared_tables = 0       | t0
0061      * add  vrf1 --> table 101        shared_tables = 0       | t1
0062      * add  vrf2 --> table 100        shared_tables = 1       | t2
0063      * add  vrf3 --> table 100        shared_tables = 1       | t3
0064      * add  vrf4 --> table 101        shared_tables = 2       v t4
0065      *
0066      * shared_tables is a "step function" (or "staircase function")
0067      * and it is increased by one when the second vrf is associated to a
0068      * table.
0069      *
0070      * at t2, vrf0 and vrf2 are bound to table 100: shared_tables = 1.
0071      *
0072      * at t3, another dev (vrf3) is bound to the same table 100 but the
0073      * value of shared_tables is still 1.
0074      * This means that no matter how many new vrfs will register on the
0075      * table 100, the shared_tables will not increase (considering only
0076      * table 100).
0077      *
0078      * at t4, vrf4 is bound to table 101, and shared_tables = 2.
0079      *
0080      * Looking at the value of shared_tables we can immediately know if
0081      * the strict_mode can or cannot be enforced. Indeed, strict_mode
0082      * can be enforced iff shared_tables = 0.
0083      *
0084      * Conversely, shared_tables is decreased when a vrf is de-associated
0085      * from a table with exactly two associated vrfs.
0086      */
0087     u32 shared_tables;
0088 
0089     bool strict_mode;
0090 };
0091 
0092 struct vrf_map_elem {
0093     struct hlist_node hnode;
0094     struct list_head vrf_list;  /* VRFs registered to this table */
0095 
0096     u32 table_id;
0097     int users;
0098     int ifindex;
0099 };
0100 
0101 static unsigned int vrf_net_id;
0102 
0103 /* per netns vrf data */
0104 struct netns_vrf {
0105     /* protected by rtnl lock */
0106     bool add_fib_rules;
0107 
0108     struct vrf_map vmap;
0109     struct ctl_table_header *ctl_hdr;
0110 };
0111 
0112 struct net_vrf {
0113     struct rtable __rcu *rth;
0114     struct rt6_info __rcu   *rt6;
0115 #if IS_ENABLED(CONFIG_IPV6)
0116     struct fib6_table   *fib6_table;
0117 #endif
0118     u32                     tb_id;
0119 
0120     struct list_head    me_list;   /* entry in vrf_map_elem */
0121     int         ifindex;
0122 };
0123 
0124 struct pcpu_dstats {
0125     u64         tx_pkts;
0126     u64         tx_bytes;
0127     u64         tx_drps;
0128     u64         rx_pkts;
0129     u64         rx_bytes;
0130     u64         rx_drps;
0131     struct u64_stats_sync   syncp;
0132 };
0133 
0134 static void vrf_rx_stats(struct net_device *dev, int len)
0135 {
0136     struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
0137 
0138     u64_stats_update_begin(&dstats->syncp);
0139     dstats->rx_pkts++;
0140     dstats->rx_bytes += len;
0141     u64_stats_update_end(&dstats->syncp);
0142 }
0143 
0144 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
0145 {
0146     vrf_dev->stats.tx_errors++;
0147     kfree_skb(skb);
0148 }
0149 
0150 static void vrf_get_stats64(struct net_device *dev,
0151                 struct rtnl_link_stats64 *stats)
0152 {
0153     int i;
0154 
0155     for_each_possible_cpu(i) {
0156         const struct pcpu_dstats *dstats;
0157         u64 tbytes, tpkts, tdrops, rbytes, rpkts;
0158         unsigned int start;
0159 
0160         dstats = per_cpu_ptr(dev->dstats, i);
0161         do {
0162             start = u64_stats_fetch_begin_irq(&dstats->syncp);
0163             tbytes = dstats->tx_bytes;
0164             tpkts = dstats->tx_pkts;
0165             tdrops = dstats->tx_drps;
0166             rbytes = dstats->rx_bytes;
0167             rpkts = dstats->rx_pkts;
0168         } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
0169         stats->tx_bytes += tbytes;
0170         stats->tx_packets += tpkts;
0171         stats->tx_dropped += tdrops;
0172         stats->rx_bytes += rbytes;
0173         stats->rx_packets += rpkts;
0174     }
0175 }
0176 
0177 static struct vrf_map *netns_vrf_map(struct net *net)
0178 {
0179     struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
0180 
0181     return &nn_vrf->vmap;
0182 }
0183 
0184 static struct vrf_map *netns_vrf_map_by_dev(struct net_device *dev)
0185 {
0186     return netns_vrf_map(dev_net(dev));
0187 }
0188 
0189 static int vrf_map_elem_get_vrf_ifindex(struct vrf_map_elem *me)
0190 {
0191     struct list_head *me_head = &me->vrf_list;
0192     struct net_vrf *vrf;
0193 
0194     if (list_empty(me_head))
0195         return -ENODEV;
0196 
0197     vrf = list_first_entry(me_head, struct net_vrf, me_list);
0198 
0199     return vrf->ifindex;
0200 }
0201 
0202 static struct vrf_map_elem *vrf_map_elem_alloc(gfp_t flags)
0203 {
0204     struct vrf_map_elem *me;
0205 
0206     me = kmalloc(sizeof(*me), flags);
0207     if (!me)
0208         return NULL;
0209 
0210     return me;
0211 }
0212 
0213 static void vrf_map_elem_free(struct vrf_map_elem *me)
0214 {
0215     kfree(me);
0216 }
0217 
0218 static void vrf_map_elem_init(struct vrf_map_elem *me, int table_id,
0219                   int ifindex, int users)
0220 {
0221     me->table_id = table_id;
0222     me->ifindex = ifindex;
0223     me->users = users;
0224     INIT_LIST_HEAD(&me->vrf_list);
0225 }
0226 
0227 static struct vrf_map_elem *vrf_map_lookup_elem(struct vrf_map *vmap,
0228                         u32 table_id)
0229 {
0230     struct vrf_map_elem *me;
0231     u32 key;
0232 
0233     key = jhash_1word(table_id, HASH_INITVAL);
0234     hash_for_each_possible(vmap->ht, me, hnode, key) {
0235         if (me->table_id == table_id)
0236             return me;
0237     }
0238 
0239     return NULL;
0240 }
0241 
0242 static void vrf_map_add_elem(struct vrf_map *vmap, struct vrf_map_elem *me)
0243 {
0244     u32 table_id = me->table_id;
0245     u32 key;
0246 
0247     key = jhash_1word(table_id, HASH_INITVAL);
0248     hash_add(vmap->ht, &me->hnode, key);
0249 }
0250 
0251 static void vrf_map_del_elem(struct vrf_map_elem *me)
0252 {
0253     hash_del(&me->hnode);
0254 }
0255 
0256 static void vrf_map_lock(struct vrf_map *vmap) __acquires(&vmap->vmap_lock)
0257 {
0258     spin_lock(&vmap->vmap_lock);
0259 }
0260 
0261 static void vrf_map_unlock(struct vrf_map *vmap) __releases(&vmap->vmap_lock)
0262 {
0263     spin_unlock(&vmap->vmap_lock);
0264 }
0265 
0266 /* called with rtnl lock held */
0267 static int
0268 vrf_map_register_dev(struct net_device *dev, struct netlink_ext_ack *extack)
0269 {
0270     struct vrf_map *vmap = netns_vrf_map_by_dev(dev);
0271     struct net_vrf *vrf = netdev_priv(dev);
0272     struct vrf_map_elem *new_me, *me;
0273     u32 table_id = vrf->tb_id;
0274     bool free_new_me = false;
0275     int users;
0276     int res;
0277 
0278     /* we pre-allocate elements used in the spin-locked section (so that we
0279      * keep the spinlock as short as possible).
0280      */
0281     new_me = vrf_map_elem_alloc(GFP_KERNEL);
0282     if (!new_me)
0283         return -ENOMEM;
0284 
0285     vrf_map_elem_init(new_me, table_id, dev->ifindex, 0);
0286 
0287     vrf_map_lock(vmap);
0288 
0289     me = vrf_map_lookup_elem(vmap, table_id);
0290     if (!me) {
0291         me = new_me;
0292         vrf_map_add_elem(vmap, me);
0293         goto link_vrf;
0294     }
0295 
0296     /* we already have an entry in the vrf_map, so it means there is (at
0297      * least) a vrf registered on the specific table.
0298      */
0299     free_new_me = true;
0300     if (vmap->strict_mode) {
0301         /* vrfs cannot share the same table */
0302         NL_SET_ERR_MSG(extack, "Table is used by another VRF");
0303         res = -EBUSY;
0304         goto unlock;
0305     }
0306 
0307 link_vrf:
0308     users = ++me->users;
0309     if (users == 2)
0310         ++vmap->shared_tables;
0311 
0312     list_add(&vrf->me_list, &me->vrf_list);
0313 
0314     res = 0;
0315 
0316 unlock:
0317     vrf_map_unlock(vmap);
0318 
0319     /* clean-up, if needed */
0320     if (free_new_me)
0321         vrf_map_elem_free(new_me);
0322 
0323     return res;
0324 }
0325 
0326 /* called with rtnl lock held */
0327 static void vrf_map_unregister_dev(struct net_device *dev)
0328 {
0329     struct vrf_map *vmap = netns_vrf_map_by_dev(dev);
0330     struct net_vrf *vrf = netdev_priv(dev);
0331     u32 table_id = vrf->tb_id;
0332     struct vrf_map_elem *me;
0333     int users;
0334 
0335     vrf_map_lock(vmap);
0336 
0337     me = vrf_map_lookup_elem(vmap, table_id);
0338     if (!me)
0339         goto unlock;
0340 
0341     list_del(&vrf->me_list);
0342 
0343     users = --me->users;
0344     if (users == 1) {
0345         --vmap->shared_tables;
0346     } else if (users == 0) {
0347         vrf_map_del_elem(me);
0348 
0349         /* no one will refer to this element anymore */
0350         vrf_map_elem_free(me);
0351     }
0352 
0353 unlock:
0354     vrf_map_unlock(vmap);
0355 }
0356 
0357 /* return the vrf device index associated with the table_id */
0358 static int vrf_ifindex_lookup_by_table_id(struct net *net, u32 table_id)
0359 {
0360     struct vrf_map *vmap = netns_vrf_map(net);
0361     struct vrf_map_elem *me;
0362     int ifindex;
0363 
0364     vrf_map_lock(vmap);
0365 
0366     if (!vmap->strict_mode) {
0367         ifindex = -EPERM;
0368         goto unlock;
0369     }
0370 
0371     me = vrf_map_lookup_elem(vmap, table_id);
0372     if (!me) {
0373         ifindex = -ENODEV;
0374         goto unlock;
0375     }
0376 
0377     ifindex = vrf_map_elem_get_vrf_ifindex(me);
0378 
0379 unlock:
0380     vrf_map_unlock(vmap);
0381 
0382     return ifindex;
0383 }
0384 
0385 /* by default VRF devices do not have a qdisc and are expected
0386  * to be created with only a single queue.
0387  */
0388 static bool qdisc_tx_is_default(const struct net_device *dev)
0389 {
0390     struct netdev_queue *txq;
0391     struct Qdisc *qdisc;
0392 
0393     if (dev->num_tx_queues > 1)
0394         return false;
0395 
0396     txq = netdev_get_tx_queue(dev, 0);
0397     qdisc = rcu_access_pointer(txq->qdisc);
0398 
0399     return !qdisc->enqueue;
0400 }
0401 
0402 /* Local traffic destined to local address. Reinsert the packet to rx
0403  * path, similar to loopback handling.
0404  */
0405 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
0406               struct dst_entry *dst)
0407 {
0408     int len = skb->len;
0409 
0410     skb_orphan(skb);
0411 
0412     skb_dst_set(skb, dst);
0413 
0414     /* set pkt_type to avoid skb hitting packet taps twice -
0415      * once on Tx and again in Rx processing
0416      */
0417     skb->pkt_type = PACKET_LOOPBACK;
0418 
0419     skb->protocol = eth_type_trans(skb, dev);
0420 
0421     if (likely(__netif_rx(skb) == NET_RX_SUCCESS))
0422         vrf_rx_stats(dev, len);
0423     else
0424         this_cpu_inc(dev->dstats->rx_drps);
0425 
0426     return NETDEV_TX_OK;
0427 }
0428 
0429 static void vrf_nf_set_untracked(struct sk_buff *skb)
0430 {
0431     if (skb_get_nfct(skb) == 0)
0432         nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
0433 }
0434 
0435 static void vrf_nf_reset_ct(struct sk_buff *skb)
0436 {
0437     if (skb_get_nfct(skb) == IP_CT_UNTRACKED)
0438         nf_reset_ct(skb);
0439 }
0440 
0441 #if IS_ENABLED(CONFIG_IPV6)
0442 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
0443                  struct sk_buff *skb)
0444 {
0445     int err;
0446 
0447     vrf_nf_reset_ct(skb);
0448 
0449     err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
0450               sk, skb, NULL, skb_dst(skb)->dev, dst_output);
0451 
0452     if (likely(err == 1))
0453         err = dst_output(net, sk, skb);
0454 
0455     return err;
0456 }
0457 
0458 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
0459                        struct net_device *dev)
0460 {
0461     const struct ipv6hdr *iph;
0462     struct net *net = dev_net(skb->dev);
0463     struct flowi6 fl6;
0464     int ret = NET_XMIT_DROP;
0465     struct dst_entry *dst;
0466     struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
0467 
0468     if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr)))
0469         goto err;
0470 
0471     iph = ipv6_hdr(skb);
0472 
0473     memset(&fl6, 0, sizeof(fl6));
0474     /* needed to match OIF rule */
0475     fl6.flowi6_l3mdev = dev->ifindex;
0476     fl6.flowi6_iif = LOOPBACK_IFINDEX;
0477     fl6.daddr = iph->daddr;
0478     fl6.saddr = iph->saddr;
0479     fl6.flowlabel = ip6_flowinfo(iph);
0480     fl6.flowi6_mark = skb->mark;
0481     fl6.flowi6_proto = iph->nexthdr;
0482 
0483     dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL);
0484     if (IS_ERR(dst) || dst == dst_null)
0485         goto err;
0486 
0487     skb_dst_drop(skb);
0488 
0489     /* if dst.dev is the VRF device again this is locally originated traffic
0490      * destined to a local address. Short circuit to Rx path.
0491      */
0492     if (dst->dev == dev)
0493         return vrf_local_xmit(skb, dev, dst);
0494 
0495     skb_dst_set(skb, dst);
0496 
0497     /* strip the ethernet header added for pass through VRF device */
0498     __skb_pull(skb, skb_network_offset(skb));
0499 
0500     memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
0501     ret = vrf_ip6_local_out(net, skb->sk, skb);
0502     if (unlikely(net_xmit_eval(ret)))
0503         dev->stats.tx_errors++;
0504     else
0505         ret = NET_XMIT_SUCCESS;
0506 
0507     return ret;
0508 err:
0509     vrf_tx_error(dev, skb);
0510     return NET_XMIT_DROP;
0511 }
0512 #else
0513 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
0514                        struct net_device *dev)
0515 {
0516     vrf_tx_error(dev, skb);
0517     return NET_XMIT_DROP;
0518 }
0519 #endif
0520 
0521 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
0522 static int vrf_ip_local_out(struct net *net, struct sock *sk,
0523                 struct sk_buff *skb)
0524 {
0525     int err;
0526 
0527     vrf_nf_reset_ct(skb);
0528 
0529     err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
0530               skb, NULL, skb_dst(skb)->dev, dst_output);
0531     if (likely(err == 1))
0532         err = dst_output(net, sk, skb);
0533 
0534     return err;
0535 }
0536 
0537 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
0538                        struct net_device *vrf_dev)
0539 {
0540     struct iphdr *ip4h;
0541     int ret = NET_XMIT_DROP;
0542     struct flowi4 fl4;
0543     struct net *net = dev_net(vrf_dev);
0544     struct rtable *rt;
0545 
0546     if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr)))
0547         goto err;
0548 
0549     ip4h = ip_hdr(skb);
0550 
0551     memset(&fl4, 0, sizeof(fl4));
0552     /* needed to match OIF rule */
0553     fl4.flowi4_l3mdev = vrf_dev->ifindex;
0554     fl4.flowi4_iif = LOOPBACK_IFINDEX;
0555     fl4.flowi4_tos = RT_TOS(ip4h->tos);
0556     fl4.flowi4_flags = FLOWI_FLAG_ANYSRC;
0557     fl4.flowi4_proto = ip4h->protocol;
0558     fl4.daddr = ip4h->daddr;
0559     fl4.saddr = ip4h->saddr;
0560 
0561     rt = ip_route_output_flow(net, &fl4, NULL);
0562     if (IS_ERR(rt))
0563         goto err;
0564 
0565     skb_dst_drop(skb);
0566 
0567     /* if dst.dev is the VRF device again this is locally originated traffic
0568      * destined to a local address. Short circuit to Rx path.
0569      */
0570     if (rt->dst.dev == vrf_dev)
0571         return vrf_local_xmit(skb, vrf_dev, &rt->dst);
0572 
0573     skb_dst_set(skb, &rt->dst);
0574 
0575     /* strip the ethernet header added for pass through VRF device */
0576     __skb_pull(skb, skb_network_offset(skb));
0577 
0578     if (!ip4h->saddr) {
0579         ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
0580                            RT_SCOPE_LINK);
0581     }
0582 
0583     memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
0584     ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
0585     if (unlikely(net_xmit_eval(ret)))
0586         vrf_dev->stats.tx_errors++;
0587     else
0588         ret = NET_XMIT_SUCCESS;
0589 
0590 out:
0591     return ret;
0592 err:
0593     vrf_tx_error(vrf_dev, skb);
0594     goto out;
0595 }
0596 
0597 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
0598 {
0599     switch (skb->protocol) {
0600     case htons(ETH_P_IP):
0601         return vrf_process_v4_outbound(skb, dev);
0602     case htons(ETH_P_IPV6):
0603         return vrf_process_v6_outbound(skb, dev);
0604     default:
0605         vrf_tx_error(dev, skb);
0606         return NET_XMIT_DROP;
0607     }
0608 }
0609 
0610 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
0611 {
0612     int len = skb->len;
0613     netdev_tx_t ret = is_ip_tx_frame(skb, dev);
0614 
0615     if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
0616         struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
0617 
0618         u64_stats_update_begin(&dstats->syncp);
0619         dstats->tx_pkts++;
0620         dstats->tx_bytes += len;
0621         u64_stats_update_end(&dstats->syncp);
0622     } else {
0623         this_cpu_inc(dev->dstats->tx_drps);
0624     }
0625 
0626     return ret;
0627 }
0628 
0629 static void vrf_finish_direct(struct sk_buff *skb)
0630 {
0631     struct net_device *vrf_dev = skb->dev;
0632 
0633     if (!list_empty(&vrf_dev->ptype_all) &&
0634         likely(skb_headroom(skb) >= ETH_HLEN)) {
0635         struct ethhdr *eth = skb_push(skb, ETH_HLEN);
0636 
0637         ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
0638         eth_zero_addr(eth->h_dest);
0639         eth->h_proto = skb->protocol;
0640 
0641         rcu_read_lock_bh();
0642         dev_queue_xmit_nit(skb, vrf_dev);
0643         rcu_read_unlock_bh();
0644 
0645         skb_pull(skb, ETH_HLEN);
0646     }
0647 
0648     vrf_nf_reset_ct(skb);
0649 }
0650 
0651 #if IS_ENABLED(CONFIG_IPV6)
0652 /* modelled after ip6_finish_output2 */
0653 static int vrf_finish_output6(struct net *net, struct sock *sk,
0654                   struct sk_buff *skb)
0655 {
0656     struct dst_entry *dst = skb_dst(skb);
0657     struct net_device *dev = dst->dev;
0658     const struct in6_addr *nexthop;
0659     struct neighbour *neigh;
0660     int ret;
0661 
0662     vrf_nf_reset_ct(skb);
0663 
0664     skb->protocol = htons(ETH_P_IPV6);
0665     skb->dev = dev;
0666 
0667     rcu_read_lock_bh();
0668     nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
0669     neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
0670     if (unlikely(!neigh))
0671         neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
0672     if (!IS_ERR(neigh)) {
0673         sock_confirm_neigh(skb, neigh);
0674         ret = neigh_output(neigh, skb, false);
0675         rcu_read_unlock_bh();
0676         return ret;
0677     }
0678     rcu_read_unlock_bh();
0679 
0680     IP6_INC_STATS(dev_net(dst->dev),
0681               ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
0682     kfree_skb(skb);
0683     return -EINVAL;
0684 }
0685 
0686 /* modelled after ip6_output */
0687 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
0688 {
0689     return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
0690                 net, sk, skb, NULL, skb_dst(skb)->dev,
0691                 vrf_finish_output6,
0692                 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
0693 }
0694 
0695 /* set dst on skb to send packet to us via dev_xmit path. Allows
0696  * packet to go through device based features such as qdisc, netfilter
0697  * hooks and packet sockets with skb->dev set to vrf device.
0698  */
0699 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
0700                         struct sk_buff *skb)
0701 {
0702     struct net_vrf *vrf = netdev_priv(vrf_dev);
0703     struct dst_entry *dst = NULL;
0704     struct rt6_info *rt6;
0705 
0706     rcu_read_lock();
0707 
0708     rt6 = rcu_dereference(vrf->rt6);
0709     if (likely(rt6)) {
0710         dst = &rt6->dst;
0711         dst_hold(dst);
0712     }
0713 
0714     rcu_read_unlock();
0715 
0716     if (unlikely(!dst)) {
0717         vrf_tx_error(vrf_dev, skb);
0718         return NULL;
0719     }
0720 
0721     skb_dst_drop(skb);
0722     skb_dst_set(skb, dst);
0723 
0724     return skb;
0725 }
0726 
0727 static int vrf_output6_direct_finish(struct net *net, struct sock *sk,
0728                      struct sk_buff *skb)
0729 {
0730     vrf_finish_direct(skb);
0731 
0732     return vrf_ip6_local_out(net, sk, skb);
0733 }
0734 
0735 static int vrf_output6_direct(struct net *net, struct sock *sk,
0736                   struct sk_buff *skb)
0737 {
0738     int err = 1;
0739 
0740     skb->protocol = htons(ETH_P_IPV6);
0741 
0742     if (!(IPCB(skb)->flags & IPSKB_REROUTED))
0743         err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb,
0744                   NULL, skb->dev, vrf_output6_direct_finish);
0745 
0746     if (likely(err == 1))
0747         vrf_finish_direct(skb);
0748 
0749     return err;
0750 }
0751 
0752 static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk,
0753                      struct sk_buff *skb)
0754 {
0755     int err;
0756 
0757     err = vrf_output6_direct(net, sk, skb);
0758     if (likely(err == 1))
0759         err = vrf_ip6_local_out(net, sk, skb);
0760 
0761     return err;
0762 }
0763 
0764 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
0765                       struct sock *sk,
0766                       struct sk_buff *skb)
0767 {
0768     struct net *net = dev_net(vrf_dev);
0769     int err;
0770 
0771     skb->dev = vrf_dev;
0772 
0773     err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
0774               skb, NULL, vrf_dev, vrf_ip6_out_direct_finish);
0775 
0776     if (likely(err == 1))
0777         err = vrf_output6_direct(net, sk, skb);
0778 
0779     if (likely(err == 1))
0780         return skb;
0781 
0782     return NULL;
0783 }
0784 
0785 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
0786                    struct sock *sk,
0787                    struct sk_buff *skb)
0788 {
0789     /* don't divert link scope packets */
0790     if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
0791         return skb;
0792 
0793     vrf_nf_set_untracked(skb);
0794 
0795     if (qdisc_tx_is_default(vrf_dev) ||
0796         IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED)
0797         return vrf_ip6_out_direct(vrf_dev, sk, skb);
0798 
0799     return vrf_ip6_out_redirect(vrf_dev, skb);
0800 }
0801 
0802 /* holding rtnl */
0803 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
0804 {
0805     struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
0806     struct net *net = dev_net(dev);
0807     struct dst_entry *dst;
0808 
0809     RCU_INIT_POINTER(vrf->rt6, NULL);
0810     synchronize_rcu();
0811 
0812     /* move dev in dst's to loopback so this VRF device can be deleted
0813      * - based on dst_ifdown
0814      */
0815     if (rt6) {
0816         dst = &rt6->dst;
0817         netdev_ref_replace(dst->dev, net->loopback_dev,
0818                    &dst->dev_tracker, GFP_KERNEL);
0819         dst->dev = net->loopback_dev;
0820         dst_release(dst);
0821     }
0822 }
0823 
0824 static int vrf_rt6_create(struct net_device *dev)
0825 {
0826     int flags = DST_NOPOLICY | DST_NOXFRM;
0827     struct net_vrf *vrf = netdev_priv(dev);
0828     struct net *net = dev_net(dev);
0829     struct rt6_info *rt6;
0830     int rc = -ENOMEM;
0831 
0832     /* IPv6 can be CONFIG enabled and then disabled runtime */
0833     if (!ipv6_mod_enabled())
0834         return 0;
0835 
0836     vrf->fib6_table = fib6_new_table(net, vrf->tb_id);
0837     if (!vrf->fib6_table)
0838         goto out;
0839 
0840     /* create a dst for routing packets out a VRF device */
0841     rt6 = ip6_dst_alloc(net, dev, flags);
0842     if (!rt6)
0843         goto out;
0844 
0845     rt6->dst.output = vrf_output6;
0846 
0847     rcu_assign_pointer(vrf->rt6, rt6);
0848 
0849     rc = 0;
0850 out:
0851     return rc;
0852 }
0853 #else
0854 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
0855                    struct sock *sk,
0856                    struct sk_buff *skb)
0857 {
0858     return skb;
0859 }
0860 
0861 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
0862 {
0863 }
0864 
0865 static int vrf_rt6_create(struct net_device *dev)
0866 {
0867     return 0;
0868 }
0869 #endif
0870 
0871 /* modelled after ip_finish_output2 */
0872 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
0873 {
0874     struct dst_entry *dst = skb_dst(skb);
0875     struct rtable *rt = (struct rtable *)dst;
0876     struct net_device *dev = dst->dev;
0877     unsigned int hh_len = LL_RESERVED_SPACE(dev);
0878     struct neighbour *neigh;
0879     bool is_v6gw = false;
0880 
0881     vrf_nf_reset_ct(skb);
0882 
0883     /* Be paranoid, rather than too clever. */
0884     if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
0885         skb = skb_expand_head(skb, hh_len);
0886         if (!skb) {
0887             dev->stats.tx_errors++;
0888             return -ENOMEM;
0889         }
0890     }
0891 
0892     rcu_read_lock_bh();
0893 
0894     neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
0895     if (!IS_ERR(neigh)) {
0896         int ret;
0897 
0898         sock_confirm_neigh(skb, neigh);
0899         /* if crossing protocols, can not use the cached header */
0900         ret = neigh_output(neigh, skb, is_v6gw);
0901         rcu_read_unlock_bh();
0902         return ret;
0903     }
0904 
0905     rcu_read_unlock_bh();
0906     vrf_tx_error(skb->dev, skb);
0907     return -EINVAL;
0908 }
0909 
0910 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
0911 {
0912     struct net_device *dev = skb_dst(skb)->dev;
0913 
0914     IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
0915 
0916     skb->dev = dev;
0917     skb->protocol = htons(ETH_P_IP);
0918 
0919     return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
0920                 net, sk, skb, NULL, dev,
0921                 vrf_finish_output,
0922                 !(IPCB(skb)->flags & IPSKB_REROUTED));
0923 }
0924 
0925 /* set dst on skb to send packet to us via dev_xmit path. Allows
0926  * packet to go through device based features such as qdisc, netfilter
0927  * hooks and packet sockets with skb->dev set to vrf device.
0928  */
0929 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
0930                        struct sk_buff *skb)
0931 {
0932     struct net_vrf *vrf = netdev_priv(vrf_dev);
0933     struct dst_entry *dst = NULL;
0934     struct rtable *rth;
0935 
0936     rcu_read_lock();
0937 
0938     rth = rcu_dereference(vrf->rth);
0939     if (likely(rth)) {
0940         dst = &rth->dst;
0941         dst_hold(dst);
0942     }
0943 
0944     rcu_read_unlock();
0945 
0946     if (unlikely(!dst)) {
0947         vrf_tx_error(vrf_dev, skb);
0948         return NULL;
0949     }
0950 
0951     skb_dst_drop(skb);
0952     skb_dst_set(skb, dst);
0953 
0954     return skb;
0955 }
0956 
0957 static int vrf_output_direct_finish(struct net *net, struct sock *sk,
0958                     struct sk_buff *skb)
0959 {
0960     vrf_finish_direct(skb);
0961 
0962     return vrf_ip_local_out(net, sk, skb);
0963 }
0964 
0965 static int vrf_output_direct(struct net *net, struct sock *sk,
0966                  struct sk_buff *skb)
0967 {
0968     int err = 1;
0969 
0970     skb->protocol = htons(ETH_P_IP);
0971 
0972     if (!(IPCB(skb)->flags & IPSKB_REROUTED))
0973         err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb,
0974                   NULL, skb->dev, vrf_output_direct_finish);
0975 
0976     if (likely(err == 1))
0977         vrf_finish_direct(skb);
0978 
0979     return err;
0980 }
0981 
0982 static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk,
0983                     struct sk_buff *skb)
0984 {
0985     int err;
0986 
0987     err = vrf_output_direct(net, sk, skb);
0988     if (likely(err == 1))
0989         err = vrf_ip_local_out(net, sk, skb);
0990 
0991     return err;
0992 }
0993 
0994 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
0995                      struct sock *sk,
0996                      struct sk_buff *skb)
0997 {
0998     struct net *net = dev_net(vrf_dev);
0999     int err;
1000 
1001     skb->dev = vrf_dev;
1002 
1003     err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
1004               skb, NULL, vrf_dev, vrf_ip_out_direct_finish);
1005 
1006     if (likely(err == 1))
1007         err = vrf_output_direct(net, sk, skb);
1008 
1009     if (likely(err == 1))
1010         return skb;
1011 
1012     return NULL;
1013 }
1014 
1015 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
1016                   struct sock *sk,
1017                   struct sk_buff *skb)
1018 {
1019     /* don't divert multicast or local broadcast */
1020     if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
1021         ipv4_is_lbcast(ip_hdr(skb)->daddr))
1022         return skb;
1023 
1024     vrf_nf_set_untracked(skb);
1025 
1026     if (qdisc_tx_is_default(vrf_dev) ||
1027         IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED)
1028         return vrf_ip_out_direct(vrf_dev, sk, skb);
1029 
1030     return vrf_ip_out_redirect(vrf_dev, skb);
1031 }
1032 
1033 /* called with rcu lock held */
1034 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
1035                   struct sock *sk,
1036                   struct sk_buff *skb,
1037                   u16 proto)
1038 {
1039     switch (proto) {
1040     case AF_INET:
1041         return vrf_ip_out(vrf_dev, sk, skb);
1042     case AF_INET6:
1043         return vrf_ip6_out(vrf_dev, sk, skb);
1044     }
1045 
1046     return skb;
1047 }
1048 
1049 /* holding rtnl */
1050 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
1051 {
1052     struct rtable *rth = rtnl_dereference(vrf->rth);
1053     struct net *net = dev_net(dev);
1054     struct dst_entry *dst;
1055 
1056     RCU_INIT_POINTER(vrf->rth, NULL);
1057     synchronize_rcu();
1058 
1059     /* move dev in dst's to loopback so this VRF device can be deleted
1060      * - based on dst_ifdown
1061      */
1062     if (rth) {
1063         dst = &rth->dst;
1064         netdev_ref_replace(dst->dev, net->loopback_dev,
1065                    &dst->dev_tracker, GFP_KERNEL);
1066         dst->dev = net->loopback_dev;
1067         dst_release(dst);
1068     }
1069 }
1070 
1071 static int vrf_rtable_create(struct net_device *dev)
1072 {
1073     struct net_vrf *vrf = netdev_priv(dev);
1074     struct rtable *rth;
1075 
1076     if (!fib_new_table(dev_net(dev), vrf->tb_id))
1077         return -ENOMEM;
1078 
1079     /* create a dst for routing packets out through a VRF device */
1080     rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1);
1081     if (!rth)
1082         return -ENOMEM;
1083 
1084     rth->dst.output = vrf_output;
1085 
1086     rcu_assign_pointer(vrf->rth, rth);
1087 
1088     return 0;
1089 }
1090 
1091 /**************************** device handling ********************/
1092 
1093 /* cycle interface to flush neighbor cache and move routes across tables */
1094 static void cycle_netdev(struct net_device *dev,
1095              struct netlink_ext_ack *extack)
1096 {
1097     unsigned int flags = dev->flags;
1098     int ret;
1099 
1100     if (!netif_running(dev))
1101         return;
1102 
1103     ret = dev_change_flags(dev, flags & ~IFF_UP, extack);
1104     if (ret >= 0)
1105         ret = dev_change_flags(dev, flags, extack);
1106 
1107     if (ret < 0) {
1108         netdev_err(dev,
1109                "Failed to cycle device %s; route tables might be wrong!\n",
1110                dev->name);
1111     }
1112 }
1113 
1114 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
1115                 struct netlink_ext_ack *extack)
1116 {
1117     int ret;
1118 
1119     /* do not allow loopback device to be enslaved to a VRF.
1120      * The vrf device acts as the loopback for the vrf.
1121      */
1122     if (port_dev == dev_net(dev)->loopback_dev) {
1123         NL_SET_ERR_MSG(extack,
1124                    "Can not enslave loopback device to a VRF");
1125         return -EOPNOTSUPP;
1126     }
1127 
1128     port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
1129     ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
1130     if (ret < 0)
1131         goto err;
1132 
1133     cycle_netdev(port_dev, extack);
1134 
1135     return 0;
1136 
1137 err:
1138     port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
1139     return ret;
1140 }
1141 
1142 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
1143              struct netlink_ext_ack *extack)
1144 {
1145     if (netif_is_l3_master(port_dev)) {
1146         NL_SET_ERR_MSG(extack,
1147                    "Can not enslave an L3 master device to a VRF");
1148         return -EINVAL;
1149     }
1150 
1151     if (netif_is_l3_slave(port_dev))
1152         return -EINVAL;
1153 
1154     return do_vrf_add_slave(dev, port_dev, extack);
1155 }
1156 
1157 /* inverse of do_vrf_add_slave */
1158 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
1159 {
1160     netdev_upper_dev_unlink(port_dev, dev);
1161     port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
1162 
1163     cycle_netdev(port_dev, NULL);
1164 
1165     return 0;
1166 }
1167 
1168 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
1169 {
1170     return do_vrf_del_slave(dev, port_dev);
1171 }
1172 
1173 static void vrf_dev_uninit(struct net_device *dev)
1174 {
1175     struct net_vrf *vrf = netdev_priv(dev);
1176 
1177     vrf_rtable_release(dev, vrf);
1178     vrf_rt6_release(dev, vrf);
1179 
1180     free_percpu(dev->dstats);
1181     dev->dstats = NULL;
1182 }
1183 
1184 static int vrf_dev_init(struct net_device *dev)
1185 {
1186     struct net_vrf *vrf = netdev_priv(dev);
1187 
1188     dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
1189     if (!dev->dstats)
1190         goto out_nomem;
1191 
1192     /* create the default dst which points back to us */
1193     if (vrf_rtable_create(dev) != 0)
1194         goto out_stats;
1195 
1196     if (vrf_rt6_create(dev) != 0)
1197         goto out_rth;
1198 
1199     dev->flags = IFF_MASTER | IFF_NOARP;
1200 
1201     /* similarly, oper state is irrelevant; set to up to avoid confusion */
1202     dev->operstate = IF_OPER_UP;
1203     netdev_lockdep_set_classes(dev);
1204     return 0;
1205 
1206 out_rth:
1207     vrf_rtable_release(dev, vrf);
1208 out_stats:
1209     free_percpu(dev->dstats);
1210     dev->dstats = NULL;
1211 out_nomem:
1212     return -ENOMEM;
1213 }
1214 
1215 static const struct net_device_ops vrf_netdev_ops = {
1216     .ndo_init       = vrf_dev_init,
1217     .ndo_uninit     = vrf_dev_uninit,
1218     .ndo_start_xmit     = vrf_xmit,
1219     .ndo_set_mac_address    = eth_mac_addr,
1220     .ndo_get_stats64    = vrf_get_stats64,
1221     .ndo_add_slave      = vrf_add_slave,
1222     .ndo_del_slave      = vrf_del_slave,
1223 };
1224 
1225 static u32 vrf_fib_table(const struct net_device *dev)
1226 {
1227     struct net_vrf *vrf = netdev_priv(dev);
1228 
1229     return vrf->tb_id;
1230 }
1231 
1232 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
1233 {
1234     kfree_skb(skb);
1235     return 0;
1236 }
1237 
1238 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
1239                       struct sk_buff *skb,
1240                       struct net_device *dev)
1241 {
1242     struct net *net = dev_net(dev);
1243 
1244     if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
1245         skb = NULL;    /* kfree_skb(skb) handled by nf code */
1246 
1247     return skb;
1248 }
1249 
1250 static int vrf_prepare_mac_header(struct sk_buff *skb,
1251                   struct net_device *vrf_dev, u16 proto)
1252 {
1253     struct ethhdr *eth;
1254     int err;
1255 
1256     /* in general, we do not know if there is enough space in the head of
1257      * the packet for hosting the mac header.
1258      */
1259     err = skb_cow_head(skb, LL_RESERVED_SPACE(vrf_dev));
1260     if (unlikely(err))
1261         /* no space in the skb head */
1262         return -ENOBUFS;
1263 
1264     __skb_push(skb, ETH_HLEN);
1265     eth = (struct ethhdr *)skb->data;
1266 
1267     skb_reset_mac_header(skb);
1268     skb_reset_mac_len(skb);
1269 
1270     /* we set the ethernet destination and the source addresses to the
1271      * address of the VRF device.
1272      */
1273     ether_addr_copy(eth->h_dest, vrf_dev->dev_addr);
1274     ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
1275     eth->h_proto = htons(proto);
1276 
1277     /* the destination address of the Ethernet frame corresponds to the
1278      * address set on the VRF interface; therefore, the packet is intended
1279      * to be processed locally.
1280      */
1281     skb->protocol = eth->h_proto;
1282     skb->pkt_type = PACKET_HOST;
1283 
1284     skb_postpush_rcsum(skb, skb->data, ETH_HLEN);
1285 
1286     skb_pull_inline(skb, ETH_HLEN);
1287 
1288     return 0;
1289 }
1290 
1291 /* prepare and add the mac header to the packet if it was not set previously.
1292  * In this way, packet sniffers such as tcpdump can parse the packet correctly.
1293  * If the mac header was already set, the original mac header is left
1294  * untouched and the function returns immediately.
1295  */
1296 static int vrf_add_mac_header_if_unset(struct sk_buff *skb,
1297                        struct net_device *vrf_dev,
1298                        u16 proto, struct net_device *orig_dev)
1299 {
1300     if (skb_mac_header_was_set(skb) && dev_has_header(orig_dev))
1301         return 0;
1302 
1303     return vrf_prepare_mac_header(skb, vrf_dev, proto);
1304 }
1305 
1306 #if IS_ENABLED(CONFIG_IPV6)
1307 /* neighbor handling is done with actual device; do not want
1308  * to flip skb->dev for those ndisc packets. This really fails
1309  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
1310  * a start.
1311  */
1312 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
1313 {
1314     const struct ipv6hdr *iph = ipv6_hdr(skb);
1315     bool rc = false;
1316 
1317     if (iph->nexthdr == NEXTHDR_ICMP) {
1318         const struct icmp6hdr *icmph;
1319         struct icmp6hdr _icmph;
1320 
1321         icmph = skb_header_pointer(skb, sizeof(*iph),
1322                        sizeof(_icmph), &_icmph);
1323         if (!icmph)
1324             goto out;
1325 
1326         switch (icmph->icmp6_type) {
1327         case NDISC_ROUTER_SOLICITATION:
1328         case NDISC_ROUTER_ADVERTISEMENT:
1329         case NDISC_NEIGHBOUR_SOLICITATION:
1330         case NDISC_NEIGHBOUR_ADVERTISEMENT:
1331         case NDISC_REDIRECT:
1332             rc = true;
1333             break;
1334         }
1335     }
1336 
1337 out:
1338     return rc;
1339 }
1340 
1341 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
1342                          const struct net_device *dev,
1343                          struct flowi6 *fl6,
1344                          int ifindex,
1345                          const struct sk_buff *skb,
1346                          int flags)
1347 {
1348     struct net_vrf *vrf = netdev_priv(dev);
1349 
1350     return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags);
1351 }
1352 
1353 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
1354                   int ifindex)
1355 {
1356     const struct ipv6hdr *iph = ipv6_hdr(skb);
1357     struct flowi6 fl6 = {
1358         .flowi6_iif     = ifindex,
1359         .flowi6_mark    = skb->mark,
1360         .flowi6_proto   = iph->nexthdr,
1361         .daddr          = iph->daddr,
1362         .saddr          = iph->saddr,
1363         .flowlabel      = ip6_flowinfo(iph),
1364     };
1365     struct net *net = dev_net(vrf_dev);
1366     struct rt6_info *rt6;
1367 
1368     rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
1369                    RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
1370     if (unlikely(!rt6))
1371         return;
1372 
1373     if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
1374         return;
1375 
1376     skb_dst_set(skb, &rt6->dst);
1377 }
1378 
1379 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1380                    struct sk_buff *skb)
1381 {
1382     int orig_iif = skb->skb_iif;
1383     bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1384     bool is_ndisc = ipv6_ndisc_frame(skb);
1385 
1386     /* loopback, multicast & non-ND link-local traffic; do not push through
1387      * packet taps again. Reset pkt_type for upper layers to process skb.
1388      * For strict packets with a source LLA, determine the dst using the
1389      * original ifindex.
1390      */
1391     if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) {
1392         skb->dev = vrf_dev;
1393         skb->skb_iif = vrf_dev->ifindex;
1394         IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1395 
1396         if (skb->pkt_type == PACKET_LOOPBACK)
1397             skb->pkt_type = PACKET_HOST;
1398         else if (ipv6_addr_type(&ipv6_hdr(skb)->saddr) & IPV6_ADDR_LINKLOCAL)
1399             vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1400 
1401         goto out;
1402     }
1403 
1404     /* if packet is NDISC then keep the ingress interface */
1405     if (!is_ndisc) {
1406         struct net_device *orig_dev = skb->dev;
1407 
1408         vrf_rx_stats(vrf_dev, skb->len);
1409         skb->dev = vrf_dev;
1410         skb->skb_iif = vrf_dev->ifindex;
1411 
1412         if (!list_empty(&vrf_dev->ptype_all)) {
1413             int err;
1414 
1415             err = vrf_add_mac_header_if_unset(skb, vrf_dev,
1416                               ETH_P_IPV6,
1417                               orig_dev);
1418             if (likely(!err)) {
1419                 skb_push(skb, skb->mac_len);
1420                 dev_queue_xmit_nit(skb, vrf_dev);
1421                 skb_pull(skb, skb->mac_len);
1422             }
1423         }
1424 
1425         IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1426     }
1427 
1428     if (need_strict)
1429         vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1430 
1431     skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1432 out:
1433     return skb;
1434 }
1435 
1436 #else
1437 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1438                    struct sk_buff *skb)
1439 {
1440     return skb;
1441 }
1442 #endif
1443 
1444 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1445                   struct sk_buff *skb)
1446 {
1447     struct net_device *orig_dev = skb->dev;
1448 
1449     skb->dev = vrf_dev;
1450     skb->skb_iif = vrf_dev->ifindex;
1451     IPCB(skb)->flags |= IPSKB_L3SLAVE;
1452 
1453     if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1454         goto out;
1455 
1456     /* loopback traffic; do not push through packet taps again.
1457      * Reset pkt_type for upper layers to process skb
1458      */
1459     if (skb->pkt_type == PACKET_LOOPBACK) {
1460         skb->pkt_type = PACKET_HOST;
1461         goto out;
1462     }
1463 
1464     vrf_rx_stats(vrf_dev, skb->len);
1465 
1466     if (!list_empty(&vrf_dev->ptype_all)) {
1467         int err;
1468 
1469         err = vrf_add_mac_header_if_unset(skb, vrf_dev, ETH_P_IP,
1470                           orig_dev);
1471         if (likely(!err)) {
1472             skb_push(skb, skb->mac_len);
1473             dev_queue_xmit_nit(skb, vrf_dev);
1474             skb_pull(skb, skb->mac_len);
1475         }
1476     }
1477 
1478     skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1479 out:
1480     return skb;
1481 }
1482 
1483 /* called with rcu lock held */
1484 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1485                   struct sk_buff *skb,
1486                   u16 proto)
1487 {
1488     switch (proto) {
1489     case AF_INET:
1490         return vrf_ip_rcv(vrf_dev, skb);
1491     case AF_INET6:
1492         return vrf_ip6_rcv(vrf_dev, skb);
1493     }
1494 
1495     return skb;
1496 }
1497 
1498 #if IS_ENABLED(CONFIG_IPV6)
1499 /* send to link-local or multicast address via interface enslaved to
1500  * VRF device. Force lookup to VRF table without changing flow struct
1501  * Note: Caller to this function must hold rcu_read_lock() and no refcnt
1502  * is taken on the dst by this function.
1503  */
1504 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1505                           struct flowi6 *fl6)
1506 {
1507     struct net *net = dev_net(dev);
1508     int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF;
1509     struct dst_entry *dst = NULL;
1510     struct rt6_info *rt;
1511 
1512     /* VRF device does not have a link-local address and
1513      * sending packets to link-local or mcast addresses over
1514      * a VRF device does not make sense
1515      */
1516     if (fl6->flowi6_oif == dev->ifindex) {
1517         dst = &net->ipv6.ip6_null_entry->dst;
1518         return dst;
1519     }
1520 
1521     if (!ipv6_addr_any(&fl6->saddr))
1522         flags |= RT6_LOOKUP_F_HAS_SADDR;
1523 
1524     rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
1525     if (rt)
1526         dst = &rt->dst;
1527 
1528     return dst;
1529 }
1530 #endif
1531 
1532 static const struct l3mdev_ops vrf_l3mdev_ops = {
1533     .l3mdev_fib_table   = vrf_fib_table,
1534     .l3mdev_l3_rcv      = vrf_l3_rcv,
1535     .l3mdev_l3_out      = vrf_l3_out,
1536 #if IS_ENABLED(CONFIG_IPV6)
1537     .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1538 #endif
1539 };
1540 
1541 static void vrf_get_drvinfo(struct net_device *dev,
1542                 struct ethtool_drvinfo *info)
1543 {
1544     strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1545     strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1546 }
1547 
1548 static const struct ethtool_ops vrf_ethtool_ops = {
1549     .get_drvinfo    = vrf_get_drvinfo,
1550 };
1551 
1552 static inline size_t vrf_fib_rule_nl_size(void)
1553 {
1554     size_t sz;
1555 
1556     sz  = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1557     sz += nla_total_size(sizeof(u8));   /* FRA_L3MDEV */
1558     sz += nla_total_size(sizeof(u32));  /* FRA_PRIORITY */
1559     sz += nla_total_size(sizeof(u8));       /* FRA_PROTOCOL */
1560 
1561     return sz;
1562 }
1563 
1564 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1565 {
1566     struct fib_rule_hdr *frh;
1567     struct nlmsghdr *nlh;
1568     struct sk_buff *skb;
1569     int err;
1570 
1571     if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) &&
1572         !ipv6_mod_enabled())
1573         return 0;
1574 
1575     skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1576     if (!skb)
1577         return -ENOMEM;
1578 
1579     nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1580     if (!nlh)
1581         goto nla_put_failure;
1582 
1583     /* rule only needs to appear once */
1584     nlh->nlmsg_flags |= NLM_F_EXCL;
1585 
1586     frh = nlmsg_data(nlh);
1587     memset(frh, 0, sizeof(*frh));
1588     frh->family = family;
1589     frh->action = FR_ACT_TO_TBL;
1590 
1591     if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
1592         goto nla_put_failure;
1593 
1594     if (nla_put_u8(skb, FRA_L3MDEV, 1))
1595         goto nla_put_failure;
1596 
1597     if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1598         goto nla_put_failure;
1599 
1600     nlmsg_end(skb, nlh);
1601 
1602     /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1603     skb->sk = dev_net(dev)->rtnl;
1604     if (add_it) {
1605         err = fib_nl_newrule(skb, nlh, NULL);
1606         if (err == -EEXIST)
1607             err = 0;
1608     } else {
1609         err = fib_nl_delrule(skb, nlh, NULL);
1610         if (err == -ENOENT)
1611             err = 0;
1612     }
1613     nlmsg_free(skb);
1614 
1615     return err;
1616 
1617 nla_put_failure:
1618     nlmsg_free(skb);
1619 
1620     return -EMSGSIZE;
1621 }
1622 
1623 static int vrf_add_fib_rules(const struct net_device *dev)
1624 {
1625     int err;
1626 
1627     err = vrf_fib_rule(dev, AF_INET,  true);
1628     if (err < 0)
1629         goto out_err;
1630 
1631     err = vrf_fib_rule(dev, AF_INET6, true);
1632     if (err < 0)
1633         goto ipv6_err;
1634 
1635 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1636     err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1637     if (err < 0)
1638         goto ipmr_err;
1639 #endif
1640 
1641 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1642     err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true);
1643     if (err < 0)
1644         goto ip6mr_err;
1645 #endif
1646 
1647     return 0;
1648 
1649 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1650 ip6mr_err:
1651     vrf_fib_rule(dev, RTNL_FAMILY_IPMR,  false);
1652 #endif
1653 
1654 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1655 ipmr_err:
1656     vrf_fib_rule(dev, AF_INET6,  false);
1657 #endif
1658 
1659 ipv6_err:
1660     vrf_fib_rule(dev, AF_INET,  false);
1661 
1662 out_err:
1663     netdev_err(dev, "Failed to add FIB rules.\n");
1664     return err;
1665 }
1666 
1667 static void vrf_setup(struct net_device *dev)
1668 {
1669     ether_setup(dev);
1670 
1671     /* Initialize the device structure. */
1672     dev->netdev_ops = &vrf_netdev_ops;
1673     dev->l3mdev_ops = &vrf_l3mdev_ops;
1674     dev->ethtool_ops = &vrf_ethtool_ops;
1675     dev->needs_free_netdev = true;
1676 
1677     /* Fill in device structure with ethernet-generic values. */
1678     eth_hw_addr_random(dev);
1679 
1680     /* don't acquire vrf device's netif_tx_lock when transmitting */
1681     dev->features |= NETIF_F_LLTX;
1682 
1683     /* don't allow vrf devices to change network namespaces. */
1684     dev->features |= NETIF_F_NETNS_LOCAL;
1685 
1686     /* does not make sense for a VLAN to be added to a vrf device */
1687     dev->features   |= NETIF_F_VLAN_CHALLENGED;
1688 
1689     /* enable offload features */
1690     dev->features   |= NETIF_F_GSO_SOFTWARE;
1691     dev->features   |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC;
1692     dev->features   |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1693 
1694     dev->hw_features = dev->features;
1695     dev->hw_enc_features = dev->features;
1696 
1697     /* default to no qdisc; user can add if desired */
1698     dev->priv_flags |= IFF_NO_QUEUE;
1699     dev->priv_flags |= IFF_NO_RX_HANDLER;
1700     dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1701 
1702     /* VRF devices do not care about MTU, but if the MTU is set
1703      * too low then the ipv4 and ipv6 protocols are disabled
1704      * which breaks networking.
1705      */
1706     dev->min_mtu = IPV6_MIN_MTU;
1707     dev->max_mtu = IP6_MAX_MTU;
1708     dev->mtu = dev->max_mtu;
1709 }
1710 
1711 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1712             struct netlink_ext_ack *extack)
1713 {
1714     if (tb[IFLA_ADDRESS]) {
1715         if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1716             NL_SET_ERR_MSG(extack, "Invalid hardware address");
1717             return -EINVAL;
1718         }
1719         if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1720             NL_SET_ERR_MSG(extack, "Invalid hardware address");
1721             return -EADDRNOTAVAIL;
1722         }
1723     }
1724     return 0;
1725 }
1726 
1727 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1728 {
1729     struct net_device *port_dev;
1730     struct list_head *iter;
1731 
1732     netdev_for_each_lower_dev(dev, port_dev, iter)
1733         vrf_del_slave(dev, port_dev);
1734 
1735     vrf_map_unregister_dev(dev);
1736 
1737     unregister_netdevice_queue(dev, head);
1738 }
1739 
1740 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1741                struct nlattr *tb[], struct nlattr *data[],
1742                struct netlink_ext_ack *extack)
1743 {
1744     struct net_vrf *vrf = netdev_priv(dev);
1745     struct netns_vrf *nn_vrf;
1746     bool *add_fib_rules;
1747     struct net *net;
1748     int err;
1749 
1750     if (!data || !data[IFLA_VRF_TABLE]) {
1751         NL_SET_ERR_MSG(extack, "VRF table id is missing");
1752         return -EINVAL;
1753     }
1754 
1755     vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1756     if (vrf->tb_id == RT_TABLE_UNSPEC) {
1757         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1758                     "Invalid VRF table id");
1759         return -EINVAL;
1760     }
1761 
1762     dev->priv_flags |= IFF_L3MDEV_MASTER;
1763 
1764     err = register_netdevice(dev);
1765     if (err)
1766         goto out;
1767 
1768     /* mapping between table_id and vrf;
1769      * note: such binding could not be done in the dev init function
1770      * because dev->ifindex id is not available yet.
1771      */
1772     vrf->ifindex = dev->ifindex;
1773 
1774     err = vrf_map_register_dev(dev, extack);
1775     if (err) {
1776         unregister_netdevice(dev);
1777         goto out;
1778     }
1779 
1780     net = dev_net(dev);
1781     nn_vrf = net_generic(net, vrf_net_id);
1782 
1783     add_fib_rules = &nn_vrf->add_fib_rules;
1784     if (*add_fib_rules) {
1785         err = vrf_add_fib_rules(dev);
1786         if (err) {
1787             vrf_map_unregister_dev(dev);
1788             unregister_netdevice(dev);
1789             goto out;
1790         }
1791         *add_fib_rules = false;
1792     }
1793 
1794 out:
1795     return err;
1796 }
1797 
1798 static size_t vrf_nl_getsize(const struct net_device *dev)
1799 {
1800     return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
1801 }
1802 
1803 static int vrf_fillinfo(struct sk_buff *skb,
1804             const struct net_device *dev)
1805 {
1806     struct net_vrf *vrf = netdev_priv(dev);
1807 
1808     return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1809 }
1810 
1811 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1812                  const struct net_device *slave_dev)
1813 {
1814     return nla_total_size(sizeof(u32));  /* IFLA_VRF_PORT_TABLE */
1815 }
1816 
1817 static int vrf_fill_slave_info(struct sk_buff *skb,
1818                    const struct net_device *vrf_dev,
1819                    const struct net_device *slave_dev)
1820 {
1821     struct net_vrf *vrf = netdev_priv(vrf_dev);
1822 
1823     if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1824         return -EMSGSIZE;
1825 
1826     return 0;
1827 }
1828 
1829 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1830     [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1831 };
1832 
1833 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1834     .kind       = DRV_NAME,
1835     .priv_size  = sizeof(struct net_vrf),
1836 
1837     .get_size   = vrf_nl_getsize,
1838     .policy     = vrf_nl_policy,
1839     .validate   = vrf_validate,
1840     .fill_info  = vrf_fillinfo,
1841 
1842     .get_slave_size  = vrf_get_slave_size,
1843     .fill_slave_info = vrf_fill_slave_info,
1844 
1845     .newlink    = vrf_newlink,
1846     .dellink    = vrf_dellink,
1847     .setup      = vrf_setup,
1848     .maxtype    = IFLA_VRF_MAX,
1849 };
1850 
1851 static int vrf_device_event(struct notifier_block *unused,
1852                 unsigned long event, void *ptr)
1853 {
1854     struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1855 
1856     /* only care about unregister events to drop slave references */
1857     if (event == NETDEV_UNREGISTER) {
1858         struct net_device *vrf_dev;
1859 
1860         if (!netif_is_l3_slave(dev))
1861             goto out;
1862 
1863         vrf_dev = netdev_master_upper_dev_get(dev);
1864         vrf_del_slave(vrf_dev, dev);
1865     }
1866 out:
1867     return NOTIFY_DONE;
1868 }
1869 
1870 static struct notifier_block vrf_notifier_block __read_mostly = {
1871     .notifier_call = vrf_device_event,
1872 };
1873 
1874 static int vrf_map_init(struct vrf_map *vmap)
1875 {
1876     spin_lock_init(&vmap->vmap_lock);
1877     hash_init(vmap->ht);
1878 
1879     vmap->strict_mode = false;
1880 
1881     return 0;
1882 }
1883 
1884 #ifdef CONFIG_SYSCTL
1885 static bool vrf_strict_mode(struct vrf_map *vmap)
1886 {
1887     bool strict_mode;
1888 
1889     vrf_map_lock(vmap);
1890     strict_mode = vmap->strict_mode;
1891     vrf_map_unlock(vmap);
1892 
1893     return strict_mode;
1894 }
1895 
1896 static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode)
1897 {
1898     bool *cur_mode;
1899     int res = 0;
1900 
1901     vrf_map_lock(vmap);
1902 
1903     cur_mode = &vmap->strict_mode;
1904     if (*cur_mode == new_mode)
1905         goto unlock;
1906 
1907     if (*cur_mode) {
1908         /* disable strict mode */
1909         *cur_mode = false;
1910     } else {
1911         if (vmap->shared_tables) {
1912             /* we cannot allow strict_mode because there are some
1913              * vrfs that share one or more tables.
1914              */
1915             res = -EBUSY;
1916             goto unlock;
1917         }
1918 
1919         /* no tables are shared among vrfs, so we can go back
1920          * to 1:1 association between a vrf with its table.
1921          */
1922         *cur_mode = true;
1923     }
1924 
1925 unlock:
1926     vrf_map_unlock(vmap);
1927 
1928     return res;
1929 }
1930 
1931 static int vrf_shared_table_handler(struct ctl_table *table, int write,
1932                     void *buffer, size_t *lenp, loff_t *ppos)
1933 {
1934     struct net *net = (struct net *)table->extra1;
1935     struct vrf_map *vmap = netns_vrf_map(net);
1936     int proc_strict_mode = 0;
1937     struct ctl_table tmp = {
1938         .procname   = table->procname,
1939         .data       = &proc_strict_mode,
1940         .maxlen     = sizeof(int),
1941         .mode       = table->mode,
1942         .extra1     = SYSCTL_ZERO,
1943         .extra2     = SYSCTL_ONE,
1944     };
1945     int ret;
1946 
1947     if (!write)
1948         proc_strict_mode = vrf_strict_mode(vmap);
1949 
1950     ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
1951 
1952     if (write && ret == 0)
1953         ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode);
1954 
1955     return ret;
1956 }
1957 
1958 static const struct ctl_table vrf_table[] = {
1959     {
1960         .procname   = "strict_mode",
1961         .data       = NULL,
1962         .maxlen     = sizeof(int),
1963         .mode       = 0644,
1964         .proc_handler   = vrf_shared_table_handler,
1965         /* set by the vrf_netns_init */
1966         .extra1     = NULL,
1967     },
1968     { },
1969 };
1970 
1971 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)
1972 {
1973     struct ctl_table *table;
1974 
1975     table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL);
1976     if (!table)
1977         return -ENOMEM;
1978 
1979     /* init the extra1 parameter with the reference to current netns */
1980     table[0].extra1 = net;
1981 
1982     nn_vrf->ctl_hdr = register_net_sysctl(net, "net/vrf", table);
1983     if (!nn_vrf->ctl_hdr) {
1984         kfree(table);
1985         return -ENOMEM;
1986     }
1987 
1988     return 0;
1989 }
1990 
1991 static void vrf_netns_exit_sysctl(struct net *net)
1992 {
1993     struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
1994     struct ctl_table *table;
1995 
1996     table = nn_vrf->ctl_hdr->ctl_table_arg;
1997     unregister_net_sysctl_table(nn_vrf->ctl_hdr);
1998     kfree(table);
1999 }
2000 #else
2001 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)
2002 {
2003     return 0;
2004 }
2005 
2006 static void vrf_netns_exit_sysctl(struct net *net)
2007 {
2008 }
2009 #endif
2010 
2011 /* Initialize per network namespace state */
2012 static int __net_init vrf_netns_init(struct net *net)
2013 {
2014     struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
2015 
2016     nn_vrf->add_fib_rules = true;
2017     vrf_map_init(&nn_vrf->vmap);
2018 
2019     return vrf_netns_init_sysctl(net, nn_vrf);
2020 }
2021 
2022 static void __net_exit vrf_netns_exit(struct net *net)
2023 {
2024     vrf_netns_exit_sysctl(net);
2025 }
2026 
2027 static struct pernet_operations vrf_net_ops __net_initdata = {
2028     .init = vrf_netns_init,
2029     .exit = vrf_netns_exit,
2030     .id   = &vrf_net_id,
2031     .size = sizeof(struct netns_vrf),
2032 };
2033 
2034 static int __init vrf_init_module(void)
2035 {
2036     int rc;
2037 
2038     register_netdevice_notifier(&vrf_notifier_block);
2039 
2040     rc = register_pernet_subsys(&vrf_net_ops);
2041     if (rc < 0)
2042         goto error;
2043 
2044     rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF,
2045                       vrf_ifindex_lookup_by_table_id);
2046     if (rc < 0)
2047         goto unreg_pernet;
2048 
2049     rc = rtnl_link_register(&vrf_link_ops);
2050     if (rc < 0)
2051         goto table_lookup_unreg;
2052 
2053     return 0;
2054 
2055 table_lookup_unreg:
2056     l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF,
2057                        vrf_ifindex_lookup_by_table_id);
2058 
2059 unreg_pernet:
2060     unregister_pernet_subsys(&vrf_net_ops);
2061 
2062 error:
2063     unregister_netdevice_notifier(&vrf_notifier_block);
2064     return rc;
2065 }
2066 
2067 module_init(vrf_init_module);
2068 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
2069 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
2070 MODULE_LICENSE("GPL");
2071 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
2072 MODULE_VERSION(DRV_VERSION);
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!