brintos

brintos / linux-shallow public Read only

0
0
Text · 46.1 KiB · 4087f72 Raw
2034 lines · c
1// SPDX-License-Identifier: GPL-2.0-or-later2/*3 * vrf.c: device driver to encapsulate a VRF space4 *5 * Copyright (c) 2015 Cumulus Networks. All rights reserved.6 * Copyright (c) 2015 Shrijeet Mukherjee <shm@cumulusnetworks.com>7 * Copyright (c) 2015 David Ahern <dsa@cumulusnetworks.com>8 *9 * Based on dummy, team and ipvlan drivers10 */11 12#include <linux/ethtool.h>13#include <linux/module.h>14#include <linux/kernel.h>15#include <linux/netdevice.h>16#include <linux/etherdevice.h>17#include <linux/ip.h>18#include <linux/init.h>19#include <linux/moduleparam.h>20#include <linux/netfilter.h>21#include <linux/rtnetlink.h>22#include <net/rtnetlink.h>23#include <linux/u64_stats_sync.h>24#include <linux/hashtable.h>25#include <linux/spinlock_types.h>26 27#include <linux/inetdevice.h>28#include <net/arp.h>29#include <net/ip.h>30#include <net/ip_fib.h>31#include <net/ip6_fib.h>32#include <net/ip6_route.h>33#include <net/route.h>34#include <net/addrconf.h>35#include <net/l3mdev.h>36#include <net/fib_rules.h>37#include <net/sch_generic.h>38#include <net/netns/generic.h>39#include <net/netfilter/nf_conntrack.h>40#include <net/inet_dscp.h>41 42#define DRV_NAME	"vrf"43#define DRV_VERSION	"1.1"44 45#define FIB_RULE_PREF  1000       /* default preference for FIB rules */46 47#define HT_MAP_BITS	448#define HASH_INITVAL	((u32)0xcafef00d)49 50struct  vrf_map {51	DECLARE_HASHTABLE(ht, HT_MAP_BITS);52	spinlock_t vmap_lock;53 54	/* shared_tables:55	 * count how many distinct tables do not comply with the strict mode56	 * requirement.57	 * shared_tables value must be 0 in order to enable the strict mode.58	 *59	 * example of the evolution of shared_tables:60	 *                                                        | time61	 * add  vrf0 --> table 100        shared_tables = 0       | t062	 * add  vrf1 --> table 101        shared_tables = 0       | t163	 * add  vrf2 --> table 100        shared_tables = 1       | t264	 * add  vrf3 --> table 100        shared_tables = 1       | t365	 * add  vrf4 --> table 101        shared_tables = 2       v t466	 *67	 * shared_tables is a "step function" (or "staircase function")68	 * and it is increased by one when the second vrf is associated to a69	 * table.70	 *71	 * at t2, vrf0 and vrf2 are bound to table 100: shared_tables = 1.72	 *73	 * at t3, another dev (vrf3) is bound to the same table 100 but the74	 * value of shared_tables is still 1.75	 * This means that no matter how many new vrfs will register on the76	 * table 100, the shared_tables will not increase (considering only77	 * table 100).78	 *79	 * at t4, vrf4 is bound to table 101, and shared_tables = 2.80	 *81	 * Looking at the value of shared_tables we can immediately know if82	 * the strict_mode can or cannot be enforced. Indeed, strict_mode83	 * can be enforced iff shared_tables = 0.84	 *85	 * Conversely, shared_tables is decreased when a vrf is de-associated86	 * from a table with exactly two associated vrfs.87	 */88	u32 shared_tables;89 90	bool strict_mode;91};92 93struct vrf_map_elem {94	struct hlist_node hnode;95	struct list_head vrf_list;  /* VRFs registered to this table */96 97	u32 table_id;98	int users;99	int ifindex;100};101 102static unsigned int vrf_net_id;103 104/* per netns vrf data */105struct netns_vrf {106	/* protected by rtnl lock */107	bool add_fib_rules;108 109	struct vrf_map vmap;110	struct ctl_table_header	*ctl_hdr;111};112 113struct net_vrf {114	struct rtable __rcu	*rth;115	struct rt6_info	__rcu	*rt6;116#if IS_ENABLED(CONFIG_IPV6)117	struct fib6_table	*fib6_table;118#endif119	u32                     tb_id;120 121	struct list_head	me_list;   /* entry in vrf_map_elem */122	int			ifindex;123};124 125static void vrf_rx_stats(struct net_device *dev, int len)126{127	struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);128 129	u64_stats_update_begin(&dstats->syncp);130	u64_stats_inc(&dstats->rx_packets);131	u64_stats_add(&dstats->rx_bytes, len);132	u64_stats_update_end(&dstats->syncp);133}134 135static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)136{137	vrf_dev->stats.tx_errors++;138	kfree_skb(skb);139}140 141static struct vrf_map *netns_vrf_map(struct net *net)142{143	struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);144 145	return &nn_vrf->vmap;146}147 148static struct vrf_map *netns_vrf_map_by_dev(struct net_device *dev)149{150	return netns_vrf_map(dev_net(dev));151}152 153static int vrf_map_elem_get_vrf_ifindex(struct vrf_map_elem *me)154{155	struct list_head *me_head = &me->vrf_list;156	struct net_vrf *vrf;157 158	if (list_empty(me_head))159		return -ENODEV;160 161	vrf = list_first_entry(me_head, struct net_vrf, me_list);162 163	return vrf->ifindex;164}165 166static struct vrf_map_elem *vrf_map_elem_alloc(gfp_t flags)167{168	struct vrf_map_elem *me;169 170	me = kmalloc(sizeof(*me), flags);171	if (!me)172		return NULL;173 174	return me;175}176 177static void vrf_map_elem_free(struct vrf_map_elem *me)178{179	kfree(me);180}181 182static void vrf_map_elem_init(struct vrf_map_elem *me, int table_id,183			      int ifindex, int users)184{185	me->table_id = table_id;186	me->ifindex = ifindex;187	me->users = users;188	INIT_LIST_HEAD(&me->vrf_list);189}190 191static struct vrf_map_elem *vrf_map_lookup_elem(struct vrf_map *vmap,192						u32 table_id)193{194	struct vrf_map_elem *me;195	u32 key;196 197	key = jhash_1word(table_id, HASH_INITVAL);198	hash_for_each_possible(vmap->ht, me, hnode, key) {199		if (me->table_id == table_id)200			return me;201	}202 203	return NULL;204}205 206static void vrf_map_add_elem(struct vrf_map *vmap, struct vrf_map_elem *me)207{208	u32 table_id = me->table_id;209	u32 key;210 211	key = jhash_1word(table_id, HASH_INITVAL);212	hash_add(vmap->ht, &me->hnode, key);213}214 215static void vrf_map_del_elem(struct vrf_map_elem *me)216{217	hash_del(&me->hnode);218}219 220static void vrf_map_lock(struct vrf_map *vmap) __acquires(&vmap->vmap_lock)221{222	spin_lock(&vmap->vmap_lock);223}224 225static void vrf_map_unlock(struct vrf_map *vmap) __releases(&vmap->vmap_lock)226{227	spin_unlock(&vmap->vmap_lock);228}229 230/* called with rtnl lock held */231static int232vrf_map_register_dev(struct net_device *dev, struct netlink_ext_ack *extack)233{234	struct vrf_map *vmap = netns_vrf_map_by_dev(dev);235	struct net_vrf *vrf = netdev_priv(dev);236	struct vrf_map_elem *new_me, *me;237	u32 table_id = vrf->tb_id;238	bool free_new_me = false;239	int users;240	int res;241 242	/* we pre-allocate elements used in the spin-locked section (so that we243	 * keep the spinlock as short as possible).244	 */245	new_me = vrf_map_elem_alloc(GFP_KERNEL);246	if (!new_me)247		return -ENOMEM;248 249	vrf_map_elem_init(new_me, table_id, dev->ifindex, 0);250 251	vrf_map_lock(vmap);252 253	me = vrf_map_lookup_elem(vmap, table_id);254	if (!me) {255		me = new_me;256		vrf_map_add_elem(vmap, me);257		goto link_vrf;258	}259 260	/* we already have an entry in the vrf_map, so it means there is (at261	 * least) a vrf registered on the specific table.262	 */263	free_new_me = true;264	if (vmap->strict_mode) {265		/* vrfs cannot share the same table */266		NL_SET_ERR_MSG(extack, "Table is used by another VRF");267		res = -EBUSY;268		goto unlock;269	}270 271link_vrf:272	users = ++me->users;273	if (users == 2)274		++vmap->shared_tables;275 276	list_add(&vrf->me_list, &me->vrf_list);277 278	res = 0;279 280unlock:281	vrf_map_unlock(vmap);282 283	/* clean-up, if needed */284	if (free_new_me)285		vrf_map_elem_free(new_me);286 287	return res;288}289 290/* called with rtnl lock held */291static void vrf_map_unregister_dev(struct net_device *dev)292{293	struct vrf_map *vmap = netns_vrf_map_by_dev(dev);294	struct net_vrf *vrf = netdev_priv(dev);295	u32 table_id = vrf->tb_id;296	struct vrf_map_elem *me;297	int users;298 299	vrf_map_lock(vmap);300 301	me = vrf_map_lookup_elem(vmap, table_id);302	if (!me)303		goto unlock;304 305	list_del(&vrf->me_list);306 307	users = --me->users;308	if (users == 1) {309		--vmap->shared_tables;310	} else if (users == 0) {311		vrf_map_del_elem(me);312 313		/* no one will refer to this element anymore */314		vrf_map_elem_free(me);315	}316 317unlock:318	vrf_map_unlock(vmap);319}320 321/* return the vrf device index associated with the table_id */322static int vrf_ifindex_lookup_by_table_id(struct net *net, u32 table_id)323{324	struct vrf_map *vmap = netns_vrf_map(net);325	struct vrf_map_elem *me;326	int ifindex;327 328	vrf_map_lock(vmap);329 330	if (!vmap->strict_mode) {331		ifindex = -EPERM;332		goto unlock;333	}334 335	me = vrf_map_lookup_elem(vmap, table_id);336	if (!me) {337		ifindex = -ENODEV;338		goto unlock;339	}340 341	ifindex = vrf_map_elem_get_vrf_ifindex(me);342 343unlock:344	vrf_map_unlock(vmap);345 346	return ifindex;347}348 349/* by default VRF devices do not have a qdisc and are expected350 * to be created with only a single queue.351 */352static bool qdisc_tx_is_default(const struct net_device *dev)353{354	struct netdev_queue *txq;355	struct Qdisc *qdisc;356 357	if (dev->num_tx_queues > 1)358		return false;359 360	txq = netdev_get_tx_queue(dev, 0);361	qdisc = rcu_access_pointer(txq->qdisc);362 363	return !qdisc->enqueue;364}365 366/* Local traffic destined to local address. Reinsert the packet to rx367 * path, similar to loopback handling.368 */369static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,370			  struct dst_entry *dst)371{372	int len = skb->len;373 374	skb_orphan(skb);375 376	skb_dst_set(skb, dst);377 378	/* set pkt_type to avoid skb hitting packet taps twice -379	 * once on Tx and again in Rx processing380	 */381	skb->pkt_type = PACKET_LOOPBACK;382 383	skb->protocol = eth_type_trans(skb, dev);384 385	if (likely(__netif_rx(skb) == NET_RX_SUCCESS)) {386		vrf_rx_stats(dev, len);387	} else {388		struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);389 390		u64_stats_update_begin(&dstats->syncp);391		u64_stats_inc(&dstats->rx_drops);392		u64_stats_update_end(&dstats->syncp);393	}394 395	return NETDEV_TX_OK;396}397 398static void vrf_nf_set_untracked(struct sk_buff *skb)399{400	if (skb_get_nfct(skb) == 0)401		nf_ct_set(skb, NULL, IP_CT_UNTRACKED);402}403 404static void vrf_nf_reset_ct(struct sk_buff *skb)405{406	if (skb_get_nfct(skb) == IP_CT_UNTRACKED)407		nf_reset_ct(skb);408}409 410#if IS_ENABLED(CONFIG_IPV6)411static int vrf_ip6_local_out(struct net *net, struct sock *sk,412			     struct sk_buff *skb)413{414	int err;415 416	vrf_nf_reset_ct(skb);417 418	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,419		      sk, skb, NULL, skb_dst(skb)->dev, dst_output);420 421	if (likely(err == 1))422		err = dst_output(net, sk, skb);423 424	return err;425}426 427static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,428					   struct net_device *dev)429{430	const struct ipv6hdr *iph;431	struct net *net = dev_net(skb->dev);432	struct flowi6 fl6;433	int ret = NET_XMIT_DROP;434	struct dst_entry *dst;435	struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;436 437	if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr)))438		goto err;439 440	iph = ipv6_hdr(skb);441 442	memset(&fl6, 0, sizeof(fl6));443	/* needed to match OIF rule */444	fl6.flowi6_l3mdev = dev->ifindex;445	fl6.flowi6_iif = LOOPBACK_IFINDEX;446	fl6.daddr = iph->daddr;447	fl6.saddr = iph->saddr;448	fl6.flowlabel = ip6_flowinfo(iph);449	fl6.flowi6_mark = skb->mark;450	fl6.flowi6_proto = iph->nexthdr;451 452	dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL);453	if (IS_ERR(dst) || dst == dst_null)454		goto err;455 456	skb_dst_drop(skb);457 458	/* if dst.dev is the VRF device again this is locally originated traffic459	 * destined to a local address. Short circuit to Rx path.460	 */461	if (dst->dev == dev)462		return vrf_local_xmit(skb, dev, dst);463 464	skb_dst_set(skb, dst);465 466	/* strip the ethernet header added for pass through VRF device */467	__skb_pull(skb, skb_network_offset(skb));468 469	memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));470	ret = vrf_ip6_local_out(net, skb->sk, skb);471	if (unlikely(net_xmit_eval(ret)))472		dev->stats.tx_errors++;473	else474		ret = NET_XMIT_SUCCESS;475 476	return ret;477err:478	vrf_tx_error(dev, skb);479	return NET_XMIT_DROP;480}481#else482static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,483					   struct net_device *dev)484{485	vrf_tx_error(dev, skb);486	return NET_XMIT_DROP;487}488#endif489 490/* based on ip_local_out; can't use it b/c the dst is switched pointing to us */491static int vrf_ip_local_out(struct net *net, struct sock *sk,492			    struct sk_buff *skb)493{494	int err;495 496	vrf_nf_reset_ct(skb);497 498	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,499		      skb, NULL, skb_dst(skb)->dev, dst_output);500	if (likely(err == 1))501		err = dst_output(net, sk, skb);502 503	return err;504}505 506static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,507					   struct net_device *vrf_dev)508{509	struct iphdr *ip4h;510	int ret = NET_XMIT_DROP;511	struct flowi4 fl4;512	struct net *net = dev_net(vrf_dev);513	struct rtable *rt;514 515	if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr)))516		goto err;517 518	ip4h = ip_hdr(skb);519 520	memset(&fl4, 0, sizeof(fl4));521	/* needed to match OIF rule */522	fl4.flowi4_l3mdev = vrf_dev->ifindex;523	fl4.flowi4_iif = LOOPBACK_IFINDEX;524	fl4.flowi4_tos = ip4h->tos & INET_DSCP_MASK;525	fl4.flowi4_flags = FLOWI_FLAG_ANYSRC;526	fl4.flowi4_proto = ip4h->protocol;527	fl4.daddr = ip4h->daddr;528	fl4.saddr = ip4h->saddr;529 530	rt = ip_route_output_flow(net, &fl4, NULL);531	if (IS_ERR(rt))532		goto err;533 534	skb_dst_drop(skb);535 536	/* if dst.dev is the VRF device again this is locally originated traffic537	 * destined to a local address. Short circuit to Rx path.538	 */539	if (rt->dst.dev == vrf_dev)540		return vrf_local_xmit(skb, vrf_dev, &rt->dst);541 542	skb_dst_set(skb, &rt->dst);543 544	/* strip the ethernet header added for pass through VRF device */545	__skb_pull(skb, skb_network_offset(skb));546 547	if (!ip4h->saddr) {548		ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,549					       RT_SCOPE_LINK);550	}551 552	memset(IPCB(skb), 0, sizeof(*IPCB(skb)));553	ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);554	if (unlikely(net_xmit_eval(ret)))555		vrf_dev->stats.tx_errors++;556	else557		ret = NET_XMIT_SUCCESS;558 559out:560	return ret;561err:562	vrf_tx_error(vrf_dev, skb);563	goto out;564}565 566static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)567{568	switch (skb->protocol) {569	case htons(ETH_P_IP):570		return vrf_process_v4_outbound(skb, dev);571	case htons(ETH_P_IPV6):572		return vrf_process_v6_outbound(skb, dev);573	default:574		vrf_tx_error(dev, skb);575		return NET_XMIT_DROP;576	}577}578 579static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)580{581	struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);582 583	int len = skb->len;584	netdev_tx_t ret = is_ip_tx_frame(skb, dev);585 586	u64_stats_update_begin(&dstats->syncp);587	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {588 589		u64_stats_inc(&dstats->tx_packets);590		u64_stats_add(&dstats->tx_bytes, len);591	} else {592		u64_stats_inc(&dstats->tx_drops);593	}594	u64_stats_update_end(&dstats->syncp);595 596	return ret;597}598 599static void vrf_finish_direct(struct sk_buff *skb)600{601	struct net_device *vrf_dev = skb->dev;602 603	if (!list_empty(&vrf_dev->ptype_all) &&604	    likely(skb_headroom(skb) >= ETH_HLEN)) {605		struct ethhdr *eth = skb_push(skb, ETH_HLEN);606 607		ether_addr_copy(eth->h_source, vrf_dev->dev_addr);608		eth_zero_addr(eth->h_dest);609		eth->h_proto = skb->protocol;610 611		rcu_read_lock_bh();612		dev_queue_xmit_nit(skb, vrf_dev);613		rcu_read_unlock_bh();614 615		skb_pull(skb, ETH_HLEN);616	}617 618	vrf_nf_reset_ct(skb);619}620 621#if IS_ENABLED(CONFIG_IPV6)622/* modelled after ip6_finish_output2 */623static int vrf_finish_output6(struct net *net, struct sock *sk,624			      struct sk_buff *skb)625{626	struct dst_entry *dst = skb_dst(skb);627	struct net_device *dev = dst->dev;628	const struct in6_addr *nexthop;629	struct neighbour *neigh;630	int ret;631 632	vrf_nf_reset_ct(skb);633 634	skb->protocol = htons(ETH_P_IPV6);635	skb->dev = dev;636 637	rcu_read_lock();638	nexthop = rt6_nexthop(dst_rt6_info(dst), &ipv6_hdr(skb)->daddr);639	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);640	if (unlikely(!neigh))641		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);642	if (!IS_ERR(neigh)) {643		sock_confirm_neigh(skb, neigh);644		ret = neigh_output(neigh, skb, false);645		rcu_read_unlock();646		return ret;647	}648	rcu_read_unlock();649 650	IP6_INC_STATS(dev_net(dst->dev),651		      ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);652	kfree_skb(skb);653	return -EINVAL;654}655 656/* modelled after ip6_output */657static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)658{659	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,660			    net, sk, skb, NULL, skb_dst(skb)->dev,661			    vrf_finish_output6,662			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));663}664 665/* set dst on skb to send packet to us via dev_xmit path. Allows666 * packet to go through device based features such as qdisc, netfilter667 * hooks and packet sockets with skb->dev set to vrf device.668 */669static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,670					    struct sk_buff *skb)671{672	struct net_vrf *vrf = netdev_priv(vrf_dev);673	struct dst_entry *dst = NULL;674	struct rt6_info *rt6;675 676	rcu_read_lock();677 678	rt6 = rcu_dereference(vrf->rt6);679	if (likely(rt6)) {680		dst = &rt6->dst;681		dst_hold(dst);682	}683 684	rcu_read_unlock();685 686	if (unlikely(!dst)) {687		vrf_tx_error(vrf_dev, skb);688		return NULL;689	}690 691	skb_dst_drop(skb);692	skb_dst_set(skb, dst);693 694	return skb;695}696 697static int vrf_output6_direct_finish(struct net *net, struct sock *sk,698				     struct sk_buff *skb)699{700	vrf_finish_direct(skb);701 702	return vrf_ip6_local_out(net, sk, skb);703}704 705static int vrf_output6_direct(struct net *net, struct sock *sk,706			      struct sk_buff *skb)707{708	int err = 1;709 710	skb->protocol = htons(ETH_P_IPV6);711 712	if (!(IPCB(skb)->flags & IPSKB_REROUTED))713		err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb,714			      NULL, skb->dev, vrf_output6_direct_finish);715 716	if (likely(err == 1))717		vrf_finish_direct(skb);718 719	return err;720}721 722static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk,723				     struct sk_buff *skb)724{725	int err;726 727	err = vrf_output6_direct(net, sk, skb);728	if (likely(err == 1))729		err = vrf_ip6_local_out(net, sk, skb);730 731	return err;732}733 734static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,735					  struct sock *sk,736					  struct sk_buff *skb)737{738	struct net *net = dev_net(vrf_dev);739	int err;740 741	skb->dev = vrf_dev;742 743	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,744		      skb, NULL, vrf_dev, vrf_ip6_out_direct_finish);745 746	if (likely(err == 1))747		err = vrf_output6_direct(net, sk, skb);748 749	if (likely(err == 1))750		return skb;751 752	return NULL;753}754 755static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,756				   struct sock *sk,757				   struct sk_buff *skb)758{759	/* don't divert link scope packets */760	if (rt6_need_strict(&ipv6_hdr(skb)->daddr))761		return skb;762 763	vrf_nf_set_untracked(skb);764 765	if (qdisc_tx_is_default(vrf_dev) ||766	    IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED)767		return vrf_ip6_out_direct(vrf_dev, sk, skb);768 769	return vrf_ip6_out_redirect(vrf_dev, skb);770}771 772/* holding rtnl */773static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)774{775	struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);776	struct net *net = dev_net(dev);777	struct dst_entry *dst;778 779	RCU_INIT_POINTER(vrf->rt6, NULL);780	synchronize_rcu();781 782	/* move dev in dst's to loopback so this VRF device can be deleted783	 * - based on dst_ifdown784	 */785	if (rt6) {786		dst = &rt6->dst;787		netdev_ref_replace(dst->dev, net->loopback_dev,788				   &dst->dev_tracker, GFP_KERNEL);789		dst->dev = net->loopback_dev;790		dst_release(dst);791	}792}793 794static int vrf_rt6_create(struct net_device *dev)795{796	int flags = DST_NOPOLICY | DST_NOXFRM;797	struct net_vrf *vrf = netdev_priv(dev);798	struct net *net = dev_net(dev);799	struct rt6_info *rt6;800	int rc = -ENOMEM;801 802	/* IPv6 can be CONFIG enabled and then disabled runtime */803	if (!ipv6_mod_enabled())804		return 0;805 806	vrf->fib6_table = fib6_new_table(net, vrf->tb_id);807	if (!vrf->fib6_table)808		goto out;809 810	/* create a dst for routing packets out a VRF device */811	rt6 = ip6_dst_alloc(net, dev, flags);812	if (!rt6)813		goto out;814 815	rt6->dst.output	= vrf_output6;816 817	rcu_assign_pointer(vrf->rt6, rt6);818 819	rc = 0;820out:821	return rc;822}823#else824static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,825				   struct sock *sk,826				   struct sk_buff *skb)827{828	return skb;829}830 831static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)832{833}834 835static int vrf_rt6_create(struct net_device *dev)836{837	return 0;838}839#endif840 841/* modelled after ip_finish_output2 */842static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)843{844	struct dst_entry *dst = skb_dst(skb);845	struct rtable *rt = dst_rtable(dst);846	struct net_device *dev = dst->dev;847	unsigned int hh_len = LL_RESERVED_SPACE(dev);848	struct neighbour *neigh;849	bool is_v6gw = false;850 851	vrf_nf_reset_ct(skb);852 853	/* Be paranoid, rather than too clever. */854	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {855		skb = skb_expand_head(skb, hh_len);856		if (!skb) {857			dev->stats.tx_errors++;858			return -ENOMEM;859		}860	}861 862	rcu_read_lock();863 864	neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);865	if (!IS_ERR(neigh)) {866		int ret;867 868		sock_confirm_neigh(skb, neigh);869		/* if crossing protocols, can not use the cached header */870		ret = neigh_output(neigh, skb, is_v6gw);871		rcu_read_unlock();872		return ret;873	}874 875	rcu_read_unlock();876	vrf_tx_error(skb->dev, skb);877	return -EINVAL;878}879 880static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)881{882	struct net_device *dev = skb_dst(skb)->dev;883 884	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);885 886	skb->dev = dev;887	skb->protocol = htons(ETH_P_IP);888 889	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,890			    net, sk, skb, NULL, dev,891			    vrf_finish_output,892			    !(IPCB(skb)->flags & IPSKB_REROUTED));893}894 895/* set dst on skb to send packet to us via dev_xmit path. Allows896 * packet to go through device based features such as qdisc, netfilter897 * hooks and packet sockets with skb->dev set to vrf device.898 */899static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,900					   struct sk_buff *skb)901{902	struct net_vrf *vrf = netdev_priv(vrf_dev);903	struct dst_entry *dst = NULL;904	struct rtable *rth;905 906	rcu_read_lock();907 908	rth = rcu_dereference(vrf->rth);909	if (likely(rth)) {910		dst = &rth->dst;911		dst_hold(dst);912	}913 914	rcu_read_unlock();915 916	if (unlikely(!dst)) {917		vrf_tx_error(vrf_dev, skb);918		return NULL;919	}920 921	skb_dst_drop(skb);922	skb_dst_set(skb, dst);923 924	return skb;925}926 927static int vrf_output_direct_finish(struct net *net, struct sock *sk,928				    struct sk_buff *skb)929{930	vrf_finish_direct(skb);931 932	return vrf_ip_local_out(net, sk, skb);933}934 935static int vrf_output_direct(struct net *net, struct sock *sk,936			     struct sk_buff *skb)937{938	int err = 1;939 940	skb->protocol = htons(ETH_P_IP);941 942	if (!(IPCB(skb)->flags & IPSKB_REROUTED))943		err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb,944			      NULL, skb->dev, vrf_output_direct_finish);945 946	if (likely(err == 1))947		vrf_finish_direct(skb);948 949	return err;950}951 952static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk,953				    struct sk_buff *skb)954{955	int err;956 957	err = vrf_output_direct(net, sk, skb);958	if (likely(err == 1))959		err = vrf_ip_local_out(net, sk, skb);960 961	return err;962}963 964static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,965					 struct sock *sk,966					 struct sk_buff *skb)967{968	struct net *net = dev_net(vrf_dev);969	int err;970 971	skb->dev = vrf_dev;972 973	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,974		      skb, NULL, vrf_dev, vrf_ip_out_direct_finish);975 976	if (likely(err == 1))977		err = vrf_output_direct(net, sk, skb);978 979	if (likely(err == 1))980		return skb;981 982	return NULL;983}984 985static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,986				  struct sock *sk,987				  struct sk_buff *skb)988{989	/* don't divert multicast or local broadcast */990	if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||991	    ipv4_is_lbcast(ip_hdr(skb)->daddr))992		return skb;993 994	vrf_nf_set_untracked(skb);995 996	if (qdisc_tx_is_default(vrf_dev) ||997	    IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED)998		return vrf_ip_out_direct(vrf_dev, sk, skb);999 1000	return vrf_ip_out_redirect(vrf_dev, skb);1001}1002 1003/* called with rcu lock held */1004static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,1005				  struct sock *sk,1006				  struct sk_buff *skb,1007				  u16 proto)1008{1009	switch (proto) {1010	case AF_INET:1011		return vrf_ip_out(vrf_dev, sk, skb);1012	case AF_INET6:1013		return vrf_ip6_out(vrf_dev, sk, skb);1014	}1015 1016	return skb;1017}1018 1019/* holding rtnl */1020static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)1021{1022	struct rtable *rth = rtnl_dereference(vrf->rth);1023	struct net *net = dev_net(dev);1024	struct dst_entry *dst;1025 1026	RCU_INIT_POINTER(vrf->rth, NULL);1027	synchronize_rcu();1028 1029	/* move dev in dst's to loopback so this VRF device can be deleted1030	 * - based on dst_ifdown1031	 */1032	if (rth) {1033		dst = &rth->dst;1034		netdev_ref_replace(dst->dev, net->loopback_dev,1035				   &dst->dev_tracker, GFP_KERNEL);1036		dst->dev = net->loopback_dev;1037		dst_release(dst);1038	}1039}1040 1041static int vrf_rtable_create(struct net_device *dev)1042{1043	struct net_vrf *vrf = netdev_priv(dev);1044	struct rtable *rth;1045 1046	if (!fib_new_table(dev_net(dev), vrf->tb_id))1047		return -ENOMEM;1048 1049	/* create a dst for routing packets out through a VRF device */1050	rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1);1051	if (!rth)1052		return -ENOMEM;1053 1054	rth->dst.output	= vrf_output;1055 1056	rcu_assign_pointer(vrf->rth, rth);1057 1058	return 0;1059}1060 1061/**************************** device handling ********************/1062 1063/* cycle interface to flush neighbor cache and move routes across tables */1064static void cycle_netdev(struct net_device *dev,1065			 struct netlink_ext_ack *extack)1066{1067	unsigned int flags = dev->flags;1068	int ret;1069 1070	if (!netif_running(dev))1071		return;1072 1073	ret = dev_change_flags(dev, flags & ~IFF_UP, extack);1074	if (ret >= 0)1075		ret = dev_change_flags(dev, flags, extack);1076 1077	if (ret < 0) {1078		netdev_err(dev,1079			   "Failed to cycle device %s; route tables might be wrong!\n",1080			   dev->name);1081	}1082}1083 1084static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,1085			    struct netlink_ext_ack *extack)1086{1087	int ret;1088 1089	/* do not allow loopback device to be enslaved to a VRF.1090	 * The vrf device acts as the loopback for the vrf.1091	 */1092	if (port_dev == dev_net(dev)->loopback_dev) {1093		NL_SET_ERR_MSG(extack,1094			       "Can not enslave loopback device to a VRF");1095		return -EOPNOTSUPP;1096	}1097 1098	port_dev->priv_flags |= IFF_L3MDEV_SLAVE;1099	ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);1100	if (ret < 0)1101		goto err;1102 1103	cycle_netdev(port_dev, extack);1104 1105	return 0;1106 1107err:1108	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;1109	return ret;1110}1111 1112static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,1113			 struct netlink_ext_ack *extack)1114{1115	if (netif_is_l3_master(port_dev)) {1116		NL_SET_ERR_MSG(extack,1117			       "Can not enslave an L3 master device to a VRF");1118		return -EINVAL;1119	}1120 1121	if (netif_is_l3_slave(port_dev))1122		return -EINVAL;1123 1124	return do_vrf_add_slave(dev, port_dev, extack);1125}1126 1127/* inverse of do_vrf_add_slave */1128static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)1129{1130	netdev_upper_dev_unlink(port_dev, dev);1131	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;1132 1133	cycle_netdev(port_dev, NULL);1134 1135	return 0;1136}1137 1138static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)1139{1140	return do_vrf_del_slave(dev, port_dev);1141}1142 1143static void vrf_dev_uninit(struct net_device *dev)1144{1145	struct net_vrf *vrf = netdev_priv(dev);1146 1147	vrf_rtable_release(dev, vrf);1148	vrf_rt6_release(dev, vrf);1149}1150 1151static int vrf_dev_init(struct net_device *dev)1152{1153	struct net_vrf *vrf = netdev_priv(dev);1154 1155	/* create the default dst which points back to us */1156	if (vrf_rtable_create(dev) != 0)1157		goto out_nomem;1158 1159	if (vrf_rt6_create(dev) != 0)1160		goto out_rth;1161 1162	dev->flags = IFF_MASTER | IFF_NOARP;1163 1164	/* similarly, oper state is irrelevant; set to up to avoid confusion */1165	dev->operstate = IF_OPER_UP;1166	netdev_lockdep_set_classes(dev);1167	return 0;1168 1169out_rth:1170	vrf_rtable_release(dev, vrf);1171out_nomem:1172	return -ENOMEM;1173}1174 1175static const struct net_device_ops vrf_netdev_ops = {1176	.ndo_init		= vrf_dev_init,1177	.ndo_uninit		= vrf_dev_uninit,1178	.ndo_start_xmit		= vrf_xmit,1179	.ndo_set_mac_address	= eth_mac_addr,1180	.ndo_add_slave		= vrf_add_slave,1181	.ndo_del_slave		= vrf_del_slave,1182};1183 1184static u32 vrf_fib_table(const struct net_device *dev)1185{1186	struct net_vrf *vrf = netdev_priv(dev);1187 1188	return vrf->tb_id;1189}1190 1191static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)1192{1193	kfree_skb(skb);1194	return 0;1195}1196 1197static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,1198				      struct sk_buff *skb,1199				      struct net_device *dev)1200{1201	struct net *net = dev_net(dev);1202 1203	if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)1204		skb = NULL;    /* kfree_skb(skb) handled by nf code */1205 1206	return skb;1207}1208 1209static int vrf_prepare_mac_header(struct sk_buff *skb,1210				  struct net_device *vrf_dev, u16 proto)1211{1212	struct ethhdr *eth;1213	int err;1214 1215	/* in general, we do not know if there is enough space in the head of1216	 * the packet for hosting the mac header.1217	 */1218	err = skb_cow_head(skb, LL_RESERVED_SPACE(vrf_dev));1219	if (unlikely(err))1220		/* no space in the skb head */1221		return -ENOBUFS;1222 1223	__skb_push(skb, ETH_HLEN);1224	eth = (struct ethhdr *)skb->data;1225 1226	skb_reset_mac_header(skb);1227	skb_reset_mac_len(skb);1228 1229	/* we set the ethernet destination and the source addresses to the1230	 * address of the VRF device.1231	 */1232	ether_addr_copy(eth->h_dest, vrf_dev->dev_addr);1233	ether_addr_copy(eth->h_source, vrf_dev->dev_addr);1234	eth->h_proto = htons(proto);1235 1236	/* the destination address of the Ethernet frame corresponds to the1237	 * address set on the VRF interface; therefore, the packet is intended1238	 * to be processed locally.1239	 */1240	skb->protocol = eth->h_proto;1241	skb->pkt_type = PACKET_HOST;1242 1243	skb_postpush_rcsum(skb, skb->data, ETH_HLEN);1244 1245	skb_pull_inline(skb, ETH_HLEN);1246 1247	return 0;1248}1249 1250/* prepare and add the mac header to the packet if it was not set previously.1251 * In this way, packet sniffers such as tcpdump can parse the packet correctly.1252 * If the mac header was already set, the original mac header is left1253 * untouched and the function returns immediately.1254 */1255static int vrf_add_mac_header_if_unset(struct sk_buff *skb,1256				       struct net_device *vrf_dev,1257				       u16 proto, struct net_device *orig_dev)1258{1259	if (skb_mac_header_was_set(skb) && dev_has_header(orig_dev))1260		return 0;1261 1262	return vrf_prepare_mac_header(skb, vrf_dev, proto);1263}1264 1265#if IS_ENABLED(CONFIG_IPV6)1266/* neighbor handling is done with actual device; do not want1267 * to flip skb->dev for those ndisc packets. This really fails1268 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is1269 * a start.1270 */1271static bool ipv6_ndisc_frame(const struct sk_buff *skb)1272{1273	const struct ipv6hdr *iph = ipv6_hdr(skb);1274	bool rc = false;1275 1276	if (iph->nexthdr == NEXTHDR_ICMP) {1277		const struct icmp6hdr *icmph;1278		struct icmp6hdr _icmph;1279 1280		icmph = skb_header_pointer(skb, sizeof(*iph),1281					   sizeof(_icmph), &_icmph);1282		if (!icmph)1283			goto out;1284 1285		switch (icmph->icmp6_type) {1286		case NDISC_ROUTER_SOLICITATION:1287		case NDISC_ROUTER_ADVERTISEMENT:1288		case NDISC_NEIGHBOUR_SOLICITATION:1289		case NDISC_NEIGHBOUR_ADVERTISEMENT:1290		case NDISC_REDIRECT:1291			rc = true;1292			break;1293		}1294	}1295 1296out:1297	return rc;1298}1299 1300static struct rt6_info *vrf_ip6_route_lookup(struct net *net,1301					     const struct net_device *dev,1302					     struct flowi6 *fl6,1303					     int ifindex,1304					     const struct sk_buff *skb,1305					     int flags)1306{1307	struct net_vrf *vrf = netdev_priv(dev);1308 1309	return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags);1310}1311 1312static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,1313			      int ifindex)1314{1315	const struct ipv6hdr *iph = ipv6_hdr(skb);1316	struct flowi6 fl6 = {1317		.flowi6_iif     = ifindex,1318		.flowi6_mark    = skb->mark,1319		.flowi6_proto   = iph->nexthdr,1320		.daddr          = iph->daddr,1321		.saddr          = iph->saddr,1322		.flowlabel      = ip6_flowinfo(iph),1323	};1324	struct net *net = dev_net(vrf_dev);1325	struct rt6_info *rt6;1326 1327	rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,1328				   RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);1329	if (unlikely(!rt6))1330		return;1331 1332	if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))1333		return;1334 1335	skb_dst_set(skb, &rt6->dst);1336}1337 1338static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,1339				   struct sk_buff *skb)1340{1341	int orig_iif = skb->skb_iif;1342	bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);1343	bool is_ndisc = ipv6_ndisc_frame(skb);1344 1345	/* loopback, multicast & non-ND link-local traffic; do not push through1346	 * packet taps again. Reset pkt_type for upper layers to process skb.1347	 * For non-loopback strict packets, determine the dst using the original1348	 * ifindex.1349	 */1350	if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) {1351		skb->dev = vrf_dev;1352		skb->skb_iif = vrf_dev->ifindex;1353		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;1354 1355		if (skb->pkt_type == PACKET_LOOPBACK)1356			skb->pkt_type = PACKET_HOST;1357		else1358			vrf_ip6_input_dst(skb, vrf_dev, orig_iif);1359 1360		goto out;1361	}1362 1363	/* if packet is NDISC then keep the ingress interface */1364	if (!is_ndisc) {1365		struct net_device *orig_dev = skb->dev;1366 1367		vrf_rx_stats(vrf_dev, skb->len);1368		skb->dev = vrf_dev;1369		skb->skb_iif = vrf_dev->ifindex;1370 1371		if (!list_empty(&vrf_dev->ptype_all)) {1372			int err;1373 1374			err = vrf_add_mac_header_if_unset(skb, vrf_dev,1375							  ETH_P_IPV6,1376							  orig_dev);1377			if (likely(!err)) {1378				skb_push(skb, skb->mac_len);1379				dev_queue_xmit_nit(skb, vrf_dev);1380				skb_pull(skb, skb->mac_len);1381			}1382		}1383 1384		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;1385	}1386 1387	if (need_strict)1388		vrf_ip6_input_dst(skb, vrf_dev, orig_iif);1389 1390	skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);1391out:1392	return skb;1393}1394 1395#else1396static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,1397				   struct sk_buff *skb)1398{1399	return skb;1400}1401#endif1402 1403static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,1404				  struct sk_buff *skb)1405{1406	struct net_device *orig_dev = skb->dev;1407 1408	skb->dev = vrf_dev;1409	skb->skb_iif = vrf_dev->ifindex;1410	IPCB(skb)->flags |= IPSKB_L3SLAVE;1411 1412	if (ipv4_is_multicast(ip_hdr(skb)->daddr))1413		goto out;1414 1415	/* loopback traffic; do not push through packet taps again.1416	 * Reset pkt_type for upper layers to process skb1417	 */1418	if (skb->pkt_type == PACKET_LOOPBACK) {1419		skb->pkt_type = PACKET_HOST;1420		goto out;1421	}1422 1423	vrf_rx_stats(vrf_dev, skb->len);1424 1425	if (!list_empty(&vrf_dev->ptype_all)) {1426		int err;1427 1428		err = vrf_add_mac_header_if_unset(skb, vrf_dev, ETH_P_IP,1429						  orig_dev);1430		if (likely(!err)) {1431			skb_push(skb, skb->mac_len);1432			dev_queue_xmit_nit(skb, vrf_dev);1433			skb_pull(skb, skb->mac_len);1434		}1435	}1436 1437	skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);1438out:1439	return skb;1440}1441 1442/* called with rcu lock held */1443static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,1444				  struct sk_buff *skb,1445				  u16 proto)1446{1447	switch (proto) {1448	case AF_INET:1449		return vrf_ip_rcv(vrf_dev, skb);1450	case AF_INET6:1451		return vrf_ip6_rcv(vrf_dev, skb);1452	}1453 1454	return skb;1455}1456 1457#if IS_ENABLED(CONFIG_IPV6)1458/* send to link-local or multicast address via interface enslaved to1459 * VRF device. Force lookup to VRF table without changing flow struct1460 * Note: Caller to this function must hold rcu_read_lock() and no refcnt1461 * is taken on the dst by this function.1462 */1463static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,1464					      struct flowi6 *fl6)1465{1466	struct net *net = dev_net(dev);1467	int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF;1468	struct dst_entry *dst = NULL;1469	struct rt6_info *rt;1470 1471	/* VRF device does not have a link-local address and1472	 * sending packets to link-local or mcast addresses over1473	 * a VRF device does not make sense1474	 */1475	if (fl6->flowi6_oif == dev->ifindex) {1476		dst = &net->ipv6.ip6_null_entry->dst;1477		return dst;1478	}1479 1480	if (!ipv6_addr_any(&fl6->saddr))1481		flags |= RT6_LOOKUP_F_HAS_SADDR;1482 1483	rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);1484	if (rt)1485		dst = &rt->dst;1486 1487	return dst;1488}1489#endif1490 1491static const struct l3mdev_ops vrf_l3mdev_ops = {1492	.l3mdev_fib_table	= vrf_fib_table,1493	.l3mdev_l3_rcv		= vrf_l3_rcv,1494	.l3mdev_l3_out		= vrf_l3_out,1495#if IS_ENABLED(CONFIG_IPV6)1496	.l3mdev_link_scope_lookup = vrf_link_scope_lookup,1497#endif1498};1499 1500static void vrf_get_drvinfo(struct net_device *dev,1501			    struct ethtool_drvinfo *info)1502{1503	strscpy(info->driver, DRV_NAME, sizeof(info->driver));1504	strscpy(info->version, DRV_VERSION, sizeof(info->version));1505}1506 1507static const struct ethtool_ops vrf_ethtool_ops = {1508	.get_drvinfo	= vrf_get_drvinfo,1509};1510 1511static inline size_t vrf_fib_rule_nl_size(void)1512{1513	size_t sz;1514 1515	sz  = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));1516	sz += nla_total_size(sizeof(u8));	/* FRA_L3MDEV */1517	sz += nla_total_size(sizeof(u32));	/* FRA_PRIORITY */1518	sz += nla_total_size(sizeof(u8));       /* FRA_PROTOCOL */1519 1520	return sz;1521}1522 1523static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)1524{1525	struct fib_rule_hdr *frh;1526	struct nlmsghdr *nlh;1527	struct sk_buff *skb;1528	int err;1529 1530	if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) &&1531	    !ipv6_mod_enabled())1532		return 0;1533 1534	skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);1535	if (!skb)1536		return -ENOMEM;1537 1538	nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);1539	if (!nlh)1540		goto nla_put_failure;1541 1542	/* rule only needs to appear once */1543	nlh->nlmsg_flags |= NLM_F_EXCL;1544 1545	frh = nlmsg_data(nlh);1546	memset(frh, 0, sizeof(*frh));1547	frh->family = family;1548	frh->action = FR_ACT_TO_TBL;1549 1550	if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))1551		goto nla_put_failure;1552 1553	if (nla_put_u8(skb, FRA_L3MDEV, 1))1554		goto nla_put_failure;1555 1556	if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))1557		goto nla_put_failure;1558 1559	nlmsg_end(skb, nlh);1560 1561	/* fib_nl_{new,del}rule handling looks for net from skb->sk */1562	skb->sk = dev_net(dev)->rtnl;1563	if (add_it) {1564		err = fib_nl_newrule(skb, nlh, NULL);1565		if (err == -EEXIST)1566			err = 0;1567	} else {1568		err = fib_nl_delrule(skb, nlh, NULL);1569		if (err == -ENOENT)1570			err = 0;1571	}1572	nlmsg_free(skb);1573 1574	return err;1575 1576nla_put_failure:1577	nlmsg_free(skb);1578 1579	return -EMSGSIZE;1580}1581 1582static int vrf_add_fib_rules(const struct net_device *dev)1583{1584	int err;1585 1586	err = vrf_fib_rule(dev, AF_INET,  true);1587	if (err < 0)1588		goto out_err;1589 1590	err = vrf_fib_rule(dev, AF_INET6, true);1591	if (err < 0)1592		goto ipv6_err;1593 1594#if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)1595	err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);1596	if (err < 0)1597		goto ipmr_err;1598#endif1599 1600#if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)1601	err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true);1602	if (err < 0)1603		goto ip6mr_err;1604#endif1605 1606	return 0;1607 1608#if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)1609ip6mr_err:1610	vrf_fib_rule(dev, RTNL_FAMILY_IPMR,  false);1611#endif1612 1613#if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)1614ipmr_err:1615	vrf_fib_rule(dev, AF_INET6,  false);1616#endif1617 1618ipv6_err:1619	vrf_fib_rule(dev, AF_INET,  false);1620 1621out_err:1622	netdev_err(dev, "Failed to add FIB rules.\n");1623	return err;1624}1625 1626static void vrf_setup(struct net_device *dev)1627{1628	ether_setup(dev);1629 1630	/* Initialize the device structure. */1631	dev->netdev_ops = &vrf_netdev_ops;1632	dev->l3mdev_ops = &vrf_l3mdev_ops;1633	dev->ethtool_ops = &vrf_ethtool_ops;1634	dev->needs_free_netdev = true;1635 1636	/* Fill in device structure with ethernet-generic values. */1637	eth_hw_addr_random(dev);1638 1639	/* don't acquire vrf device's netif_tx_lock when transmitting */1640	dev->lltx = true;1641 1642	/* don't allow vrf devices to change network namespaces. */1643	dev->netns_local = true;1644 1645	/* does not make sense for a VLAN to be added to a vrf device */1646	dev->features   |= NETIF_F_VLAN_CHALLENGED;1647 1648	/* enable offload features */1649	dev->features   |= NETIF_F_GSO_SOFTWARE;1650	dev->features   |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC;1651	dev->features   |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;1652 1653	dev->hw_features = dev->features;1654	dev->hw_enc_features = dev->features;1655 1656	/* default to no qdisc; user can add if desired */1657	dev->priv_flags |= IFF_NO_QUEUE;1658	dev->priv_flags |= IFF_NO_RX_HANDLER;1659	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;1660 1661	/* VRF devices do not care about MTU, but if the MTU is set1662	 * too low then the ipv4 and ipv6 protocols are disabled1663	 * which breaks networking.1664	 */1665	dev->min_mtu = IPV6_MIN_MTU;1666	dev->max_mtu = IP6_MAX_MTU;1667	dev->mtu = dev->max_mtu;1668 1669	dev->pcpu_stat_type = NETDEV_PCPU_STAT_DSTATS;1670}1671 1672static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],1673			struct netlink_ext_ack *extack)1674{1675	if (tb[IFLA_ADDRESS]) {1676		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {1677			NL_SET_ERR_MSG(extack, "Invalid hardware address");1678			return -EINVAL;1679		}1680		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {1681			NL_SET_ERR_MSG(extack, "Invalid hardware address");1682			return -EADDRNOTAVAIL;1683		}1684	}1685	return 0;1686}1687 1688static void vrf_dellink(struct net_device *dev, struct list_head *head)1689{1690	struct net_device *port_dev;1691	struct list_head *iter;1692 1693	netdev_for_each_lower_dev(dev, port_dev, iter)1694		vrf_del_slave(dev, port_dev);1695 1696	vrf_map_unregister_dev(dev);1697 1698	unregister_netdevice_queue(dev, head);1699}1700 1701static int vrf_newlink(struct net *src_net, struct net_device *dev,1702		       struct nlattr *tb[], struct nlattr *data[],1703		       struct netlink_ext_ack *extack)1704{1705	struct net_vrf *vrf = netdev_priv(dev);1706	struct netns_vrf *nn_vrf;1707	bool *add_fib_rules;1708	struct net *net;1709	int err;1710 1711	if (!data || !data[IFLA_VRF_TABLE]) {1712		NL_SET_ERR_MSG(extack, "VRF table id is missing");1713		return -EINVAL;1714	}1715 1716	vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);1717	if (vrf->tb_id == RT_TABLE_UNSPEC) {1718		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],1719				    "Invalid VRF table id");1720		return -EINVAL;1721	}1722 1723	dev->priv_flags |= IFF_L3MDEV_MASTER;1724 1725	err = register_netdevice(dev);1726	if (err)1727		goto out;1728 1729	/* mapping between table_id and vrf;1730	 * note: such binding could not be done in the dev init function1731	 * because dev->ifindex id is not available yet.1732	 */1733	vrf->ifindex = dev->ifindex;1734 1735	err = vrf_map_register_dev(dev, extack);1736	if (err) {1737		unregister_netdevice(dev);1738		goto out;1739	}1740 1741	net = dev_net(dev);1742	nn_vrf = net_generic(net, vrf_net_id);1743 1744	add_fib_rules = &nn_vrf->add_fib_rules;1745	if (*add_fib_rules) {1746		err = vrf_add_fib_rules(dev);1747		if (err) {1748			vrf_map_unregister_dev(dev);1749			unregister_netdevice(dev);1750			goto out;1751		}1752		*add_fib_rules = false;1753	}1754 1755out:1756	return err;1757}1758 1759static size_t vrf_nl_getsize(const struct net_device *dev)1760{1761	return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */1762}1763 1764static int vrf_fillinfo(struct sk_buff *skb,1765			const struct net_device *dev)1766{1767	struct net_vrf *vrf = netdev_priv(dev);1768 1769	return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);1770}1771 1772static size_t vrf_get_slave_size(const struct net_device *bond_dev,1773				 const struct net_device *slave_dev)1774{1775	return nla_total_size(sizeof(u32));  /* IFLA_VRF_PORT_TABLE */1776}1777 1778static int vrf_fill_slave_info(struct sk_buff *skb,1779			       const struct net_device *vrf_dev,1780			       const struct net_device *slave_dev)1781{1782	struct net_vrf *vrf = netdev_priv(vrf_dev);1783 1784	if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))1785		return -EMSGSIZE;1786 1787	return 0;1788}1789 1790static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {1791	[IFLA_VRF_TABLE] = { .type = NLA_U32 },1792};1793 1794static struct rtnl_link_ops vrf_link_ops __read_mostly = {1795	.kind		= DRV_NAME,1796	.priv_size	= sizeof(struct net_vrf),1797 1798	.get_size	= vrf_nl_getsize,1799	.policy		= vrf_nl_policy,1800	.validate	= vrf_validate,1801	.fill_info	= vrf_fillinfo,1802 1803	.get_slave_size  = vrf_get_slave_size,1804	.fill_slave_info = vrf_fill_slave_info,1805 1806	.newlink	= vrf_newlink,1807	.dellink	= vrf_dellink,1808	.setup		= vrf_setup,1809	.maxtype	= IFLA_VRF_MAX,1810};1811 1812static int vrf_device_event(struct notifier_block *unused,1813			    unsigned long event, void *ptr)1814{1815	struct net_device *dev = netdev_notifier_info_to_dev(ptr);1816 1817	/* only care about unregister events to drop slave references */1818	if (event == NETDEV_UNREGISTER) {1819		struct net_device *vrf_dev;1820 1821		if (!netif_is_l3_slave(dev))1822			goto out;1823 1824		vrf_dev = netdev_master_upper_dev_get(dev);1825		vrf_del_slave(vrf_dev, dev);1826	}1827out:1828	return NOTIFY_DONE;1829}1830 1831static struct notifier_block vrf_notifier_block __read_mostly = {1832	.notifier_call = vrf_device_event,1833};1834 1835static int vrf_map_init(struct vrf_map *vmap)1836{1837	spin_lock_init(&vmap->vmap_lock);1838	hash_init(vmap->ht);1839 1840	vmap->strict_mode = false;1841 1842	return 0;1843}1844 1845#ifdef CONFIG_SYSCTL1846static bool vrf_strict_mode(struct vrf_map *vmap)1847{1848	bool strict_mode;1849 1850	vrf_map_lock(vmap);1851	strict_mode = vmap->strict_mode;1852	vrf_map_unlock(vmap);1853 1854	return strict_mode;1855}1856 1857static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode)1858{1859	bool *cur_mode;1860	int res = 0;1861 1862	vrf_map_lock(vmap);1863 1864	cur_mode = &vmap->strict_mode;1865	if (*cur_mode == new_mode)1866		goto unlock;1867 1868	if (*cur_mode) {1869		/* disable strict mode */1870		*cur_mode = false;1871	} else {1872		if (vmap->shared_tables) {1873			/* we cannot allow strict_mode because there are some1874			 * vrfs that share one or more tables.1875			 */1876			res = -EBUSY;1877			goto unlock;1878		}1879 1880		/* no tables are shared among vrfs, so we can go back1881		 * to 1:1 association between a vrf with its table.1882		 */1883		*cur_mode = true;1884	}1885 1886unlock:1887	vrf_map_unlock(vmap);1888 1889	return res;1890}1891 1892static int vrf_shared_table_handler(const struct ctl_table *table, int write,1893				    void *buffer, size_t *lenp, loff_t *ppos)1894{1895	struct net *net = (struct net *)table->extra1;1896	struct vrf_map *vmap = netns_vrf_map(net);1897	int proc_strict_mode = 0;1898	struct ctl_table tmp = {1899		.procname	= table->procname,1900		.data		= &proc_strict_mode,1901		.maxlen		= sizeof(int),1902		.mode		= table->mode,1903		.extra1		= SYSCTL_ZERO,1904		.extra2		= SYSCTL_ONE,1905	};1906	int ret;1907 1908	if (!write)1909		proc_strict_mode = vrf_strict_mode(vmap);1910 1911	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);1912 1913	if (write && ret == 0)1914		ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode);1915 1916	return ret;1917}1918 1919static const struct ctl_table vrf_table[] = {1920	{1921		.procname	= "strict_mode",1922		.data		= NULL,1923		.maxlen		= sizeof(int),1924		.mode		= 0644,1925		.proc_handler	= vrf_shared_table_handler,1926		/* set by the vrf_netns_init */1927		.extra1		= NULL,1928	},1929};1930 1931static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)1932{1933	struct ctl_table *table;1934 1935	table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL);1936	if (!table)1937		return -ENOMEM;1938 1939	/* init the extra1 parameter with the reference to current netns */1940	table[0].extra1 = net;1941 1942	nn_vrf->ctl_hdr = register_net_sysctl_sz(net, "net/vrf", table,1943						 ARRAY_SIZE(vrf_table));1944	if (!nn_vrf->ctl_hdr) {1945		kfree(table);1946		return -ENOMEM;1947	}1948 1949	return 0;1950}1951 1952static void vrf_netns_exit_sysctl(struct net *net)1953{1954	struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);1955	const struct ctl_table *table;1956 1957	table = nn_vrf->ctl_hdr->ctl_table_arg;1958	unregister_net_sysctl_table(nn_vrf->ctl_hdr);1959	kfree(table);1960}1961#else1962static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)1963{1964	return 0;1965}1966 1967static void vrf_netns_exit_sysctl(struct net *net)1968{1969}1970#endif1971 1972/* Initialize per network namespace state */1973static int __net_init vrf_netns_init(struct net *net)1974{1975	struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);1976 1977	nn_vrf->add_fib_rules = true;1978	vrf_map_init(&nn_vrf->vmap);1979 1980	return vrf_netns_init_sysctl(net, nn_vrf);1981}1982 1983static void __net_exit vrf_netns_exit(struct net *net)1984{1985	vrf_netns_exit_sysctl(net);1986}1987 1988static struct pernet_operations vrf_net_ops __net_initdata = {1989	.init = vrf_netns_init,1990	.exit = vrf_netns_exit,1991	.id   = &vrf_net_id,1992	.size = sizeof(struct netns_vrf),1993};1994 1995static int __init vrf_init_module(void)1996{1997	int rc;1998 1999	register_netdevice_notifier(&vrf_notifier_block);2000 2001	rc = register_pernet_subsys(&vrf_net_ops);2002	if (rc < 0)2003		goto error;2004 2005	rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF,2006					  vrf_ifindex_lookup_by_table_id);2007	if (rc < 0)2008		goto unreg_pernet;2009 2010	rc = rtnl_link_register(&vrf_link_ops);2011	if (rc < 0)2012		goto table_lookup_unreg;2013 2014	return 0;2015 2016table_lookup_unreg:2017	l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF,2018				       vrf_ifindex_lookup_by_table_id);2019 2020unreg_pernet:2021	unregister_pernet_subsys(&vrf_net_ops);2022 2023error:2024	unregister_netdevice_notifier(&vrf_notifier_block);2025	return rc;2026}2027 2028module_init(vrf_init_module);2029MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");2030MODULE_DESCRIPTION("Device driver to instantiate VRF domains");2031MODULE_LICENSE("GPL");2032MODULE_ALIAS_RTNL_LINK(DRV_NAME);2033MODULE_VERSION(DRV_VERSION);2034