vrf.c 45 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007
  1. // SPDX-License-Identifier: GPL-2.0-or-later
  2. /*
  3. * vrf.c: device driver to encapsulate a VRF space
  4. *
  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 drivers
  10. */
  11. #include <linux/module.h>
  12. #include <linux/kernel.h>
  13. #include <linux/netdevice.h>
  14. #include <linux/etherdevice.h>
  15. #include <linux/ip.h>
  16. #include <linux/init.h>
  17. #include <linux/moduleparam.h>
  18. #include <linux/netfilter.h>
  19. #include <linux/rtnetlink.h>
  20. #include <net/rtnetlink.h>
  21. #include <linux/u64_stats_sync.h>
  22. #include <linux/hashtable.h>
  23. #include <linux/spinlock_types.h>
  24. #include <linux/inetdevice.h>
  25. #include <net/arp.h>
  26. #include <net/ip.h>
  27. #include <net/ip_fib.h>
  28. #include <net/ip6_fib.h>
  29. #include <net/ip6_route.h>
  30. #include <net/route.h>
  31. #include <net/addrconf.h>
  32. #include <net/l3mdev.h>
  33. #include <net/fib_rules.h>
  34. #include <net/netns/generic.h>
  35. #include <net/netfilter/nf_conntrack.h>
  36. #define DRV_NAME "vrf"
  37. #define DRV_VERSION "1.1"
  38. #define FIB_RULE_PREF 1000 /* default preference for FIB rules */
  39. #define HT_MAP_BITS 4
  40. #define HASH_INITVAL ((u32)0xcafef00d)
  41. struct vrf_map {
  42. DECLARE_HASHTABLE(ht, HT_MAP_BITS);
  43. spinlock_t vmap_lock;
  44. /* shared_tables:
  45. * count how many distinct tables do not comply with the strict mode
  46. * requirement.
  47. * shared_tables value must be 0 in order to enable the strict mode.
  48. *
  49. * example of the evolution of shared_tables:
  50. * | time
  51. * add vrf0 --> table 100 shared_tables = 0 | t0
  52. * add vrf1 --> table 101 shared_tables = 0 | t1
  53. * add vrf2 --> table 100 shared_tables = 1 | t2
  54. * add vrf3 --> table 100 shared_tables = 1 | t3
  55. * add vrf4 --> table 101 shared_tables = 2 v t4
  56. *
  57. * shared_tables is a "step function" (or "staircase function")
  58. * and it is increased by one when the second vrf is associated to a
  59. * table.
  60. *
  61. * at t2, vrf0 and vrf2 are bound to table 100: shared_tables = 1.
  62. *
  63. * at t3, another dev (vrf3) is bound to the same table 100 but the
  64. * value of shared_tables is still 1.
  65. * This means that no matter how many new vrfs will register on the
  66. * table 100, the shared_tables will not increase (considering only
  67. * table 100).
  68. *
  69. * at t4, vrf4 is bound to table 101, and shared_tables = 2.
  70. *
  71. * Looking at the value of shared_tables we can immediately know if
  72. * the strict_mode can or cannot be enforced. Indeed, strict_mode
  73. * can be enforced iff shared_tables = 0.
  74. *
  75. * Conversely, shared_tables is decreased when a vrf is de-associated
  76. * from a table with exactly two associated vrfs.
  77. */
  78. u32 shared_tables;
  79. bool strict_mode;
  80. };
  81. struct vrf_map_elem {
  82. struct hlist_node hnode;
  83. struct list_head vrf_list; /* VRFs registered to this table */
  84. u32 table_id;
  85. int users;
  86. int ifindex;
  87. };
  88. static unsigned int vrf_net_id;
  89. /* per netns vrf data */
  90. struct netns_vrf {
  91. /* protected by rtnl lock */
  92. bool add_fib_rules;
  93. struct vrf_map vmap;
  94. struct ctl_table_header *ctl_hdr;
  95. };
  96. struct net_vrf {
  97. struct rtable __rcu *rth;
  98. struct rt6_info __rcu *rt6;
  99. #if IS_ENABLED(CONFIG_IPV6)
  100. struct fib6_table *fib6_table;
  101. #endif
  102. u32 tb_id;
  103. struct list_head me_list; /* entry in vrf_map_elem */
  104. int ifindex;
  105. };
  106. struct pcpu_dstats {
  107. u64 tx_pkts;
  108. u64 tx_bytes;
  109. u64 tx_drps;
  110. u64 rx_pkts;
  111. u64 rx_bytes;
  112. u64 rx_drps;
  113. struct u64_stats_sync syncp;
  114. };
  115. static void vrf_rx_stats(struct net_device *dev, int len)
  116. {
  117. struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
  118. u64_stats_update_begin(&dstats->syncp);
  119. dstats->rx_pkts++;
  120. dstats->rx_bytes += len;
  121. u64_stats_update_end(&dstats->syncp);
  122. }
  123. static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
  124. {
  125. vrf_dev->stats.tx_errors++;
  126. kfree_skb(skb);
  127. }
  128. static void vrf_get_stats64(struct net_device *dev,
  129. struct rtnl_link_stats64 *stats)
  130. {
  131. int i;
  132. for_each_possible_cpu(i) {
  133. const struct pcpu_dstats *dstats;
  134. u64 tbytes, tpkts, tdrops, rbytes, rpkts;
  135. unsigned int start;
  136. dstats = per_cpu_ptr(dev->dstats, i);
  137. do {
  138. start = u64_stats_fetch_begin_irq(&dstats->syncp);
  139. tbytes = dstats->tx_bytes;
  140. tpkts = dstats->tx_pkts;
  141. tdrops = dstats->tx_drps;
  142. rbytes = dstats->rx_bytes;
  143. rpkts = dstats->rx_pkts;
  144. } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
  145. stats->tx_bytes += tbytes;
  146. stats->tx_packets += tpkts;
  147. stats->tx_dropped += tdrops;
  148. stats->rx_bytes += rbytes;
  149. stats->rx_packets += rpkts;
  150. }
  151. }
  152. static struct vrf_map *netns_vrf_map(struct net *net)
  153. {
  154. struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
  155. return &nn_vrf->vmap;
  156. }
  157. static struct vrf_map *netns_vrf_map_by_dev(struct net_device *dev)
  158. {
  159. return netns_vrf_map(dev_net(dev));
  160. }
  161. static int vrf_map_elem_get_vrf_ifindex(struct vrf_map_elem *me)
  162. {
  163. struct list_head *me_head = &me->vrf_list;
  164. struct net_vrf *vrf;
  165. if (list_empty(me_head))
  166. return -ENODEV;
  167. vrf = list_first_entry(me_head, struct net_vrf, me_list);
  168. return vrf->ifindex;
  169. }
  170. static struct vrf_map_elem *vrf_map_elem_alloc(gfp_t flags)
  171. {
  172. struct vrf_map_elem *me;
  173. me = kmalloc(sizeof(*me), flags);
  174. if (!me)
  175. return NULL;
  176. return me;
  177. }
  178. static void vrf_map_elem_free(struct vrf_map_elem *me)
  179. {
  180. kfree(me);
  181. }
  182. static 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. static struct vrf_map_elem *vrf_map_lookup_elem(struct vrf_map *vmap,
  191. u32 table_id)
  192. {
  193. struct vrf_map_elem *me;
  194. u32 key;
  195. key = jhash_1word(table_id, HASH_INITVAL);
  196. hash_for_each_possible(vmap->ht, me, hnode, key) {
  197. if (me->table_id == table_id)
  198. return me;
  199. }
  200. return NULL;
  201. }
  202. static void vrf_map_add_elem(struct vrf_map *vmap, struct vrf_map_elem *me)
  203. {
  204. u32 table_id = me->table_id;
  205. u32 key;
  206. key = jhash_1word(table_id, HASH_INITVAL);
  207. hash_add(vmap->ht, &me->hnode, key);
  208. }
  209. static void vrf_map_del_elem(struct vrf_map_elem *me)
  210. {
  211. hash_del(&me->hnode);
  212. }
  213. static void vrf_map_lock(struct vrf_map *vmap) __acquires(&vmap->vmap_lock)
  214. {
  215. spin_lock(&vmap->vmap_lock);
  216. }
  217. static void vrf_map_unlock(struct vrf_map *vmap) __releases(&vmap->vmap_lock)
  218. {
  219. spin_unlock(&vmap->vmap_lock);
  220. }
  221. /* called with rtnl lock held */
  222. static int
  223. vrf_map_register_dev(struct net_device *dev, struct netlink_ext_ack *extack)
  224. {
  225. struct vrf_map *vmap = netns_vrf_map_by_dev(dev);
  226. struct net_vrf *vrf = netdev_priv(dev);
  227. struct vrf_map_elem *new_me, *me;
  228. u32 table_id = vrf->tb_id;
  229. bool free_new_me = false;
  230. int users;
  231. int res;
  232. /* we pre-allocate elements used in the spin-locked section (so that we
  233. * keep the spinlock as short as possibile).
  234. */
  235. new_me = vrf_map_elem_alloc(GFP_KERNEL);
  236. if (!new_me)
  237. return -ENOMEM;
  238. vrf_map_elem_init(new_me, table_id, dev->ifindex, 0);
  239. vrf_map_lock(vmap);
  240. me = vrf_map_lookup_elem(vmap, table_id);
  241. if (!me) {
  242. me = new_me;
  243. vrf_map_add_elem(vmap, me);
  244. goto link_vrf;
  245. }
  246. /* we already have an entry in the vrf_map, so it means there is (at
  247. * least) a vrf registered on the specific table.
  248. */
  249. free_new_me = true;
  250. if (vmap->strict_mode) {
  251. /* vrfs cannot share the same table */
  252. NL_SET_ERR_MSG(extack, "Table is used by another VRF");
  253. res = -EBUSY;
  254. goto unlock;
  255. }
  256. link_vrf:
  257. users = ++me->users;
  258. if (users == 2)
  259. ++vmap->shared_tables;
  260. list_add(&vrf->me_list, &me->vrf_list);
  261. res = 0;
  262. unlock:
  263. vrf_map_unlock(vmap);
  264. /* clean-up, if needed */
  265. if (free_new_me)
  266. vrf_map_elem_free(new_me);
  267. return res;
  268. }
  269. /* called with rtnl lock held */
  270. static void vrf_map_unregister_dev(struct net_device *dev)
  271. {
  272. struct vrf_map *vmap = netns_vrf_map_by_dev(dev);
  273. struct net_vrf *vrf = netdev_priv(dev);
  274. u32 table_id = vrf->tb_id;
  275. struct vrf_map_elem *me;
  276. int users;
  277. vrf_map_lock(vmap);
  278. me = vrf_map_lookup_elem(vmap, table_id);
  279. if (!me)
  280. goto unlock;
  281. list_del(&vrf->me_list);
  282. users = --me->users;
  283. if (users == 1) {
  284. --vmap->shared_tables;
  285. } else if (users == 0) {
  286. vrf_map_del_elem(me);
  287. /* no one will refer to this element anymore */
  288. vrf_map_elem_free(me);
  289. }
  290. unlock:
  291. vrf_map_unlock(vmap);
  292. }
  293. /* return the vrf device index associated with the table_id */
  294. static int vrf_ifindex_lookup_by_table_id(struct net *net, u32 table_id)
  295. {
  296. struct vrf_map *vmap = netns_vrf_map(net);
  297. struct vrf_map_elem *me;
  298. int ifindex;
  299. vrf_map_lock(vmap);
  300. if (!vmap->strict_mode) {
  301. ifindex = -EPERM;
  302. goto unlock;
  303. }
  304. me = vrf_map_lookup_elem(vmap, table_id);
  305. if (!me) {
  306. ifindex = -ENODEV;
  307. goto unlock;
  308. }
  309. ifindex = vrf_map_elem_get_vrf_ifindex(me);
  310. unlock:
  311. vrf_map_unlock(vmap);
  312. return ifindex;
  313. }
  314. /* by default VRF devices do not have a qdisc and are expected
  315. * to be created with only a single queue.
  316. */
  317. static bool qdisc_tx_is_default(const struct net_device *dev)
  318. {
  319. struct netdev_queue *txq;
  320. struct Qdisc *qdisc;
  321. if (dev->num_tx_queues > 1)
  322. return false;
  323. txq = netdev_get_tx_queue(dev, 0);
  324. qdisc = rcu_access_pointer(txq->qdisc);
  325. return !qdisc->enqueue;
  326. }
  327. /* Local traffic destined to local address. Reinsert the packet to rx
  328. * path, similar to loopback handling.
  329. */
  330. static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
  331. struct dst_entry *dst)
  332. {
  333. int len = skb->len;
  334. skb_orphan(skb);
  335. skb_dst_set(skb, dst);
  336. /* set pkt_type to avoid skb hitting packet taps twice -
  337. * once on Tx and again in Rx processing
  338. */
  339. skb->pkt_type = PACKET_LOOPBACK;
  340. skb->protocol = eth_type_trans(skb, dev);
  341. if (likely(netif_rx(skb) == NET_RX_SUCCESS))
  342. vrf_rx_stats(dev, len);
  343. else
  344. this_cpu_inc(dev->dstats->rx_drps);
  345. return NETDEV_TX_OK;
  346. }
  347. static void vrf_nf_set_untracked(struct sk_buff *skb)
  348. {
  349. if (skb_get_nfct(skb) == 0)
  350. nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
  351. }
  352. static void vrf_nf_reset_ct(struct sk_buff *skb)
  353. {
  354. if (skb_get_nfct(skb) == IP_CT_UNTRACKED)
  355. nf_reset_ct(skb);
  356. }
  357. #if IS_ENABLED(CONFIG_IPV6)
  358. static int vrf_ip6_local_out(struct net *net, struct sock *sk,
  359. struct sk_buff *skb)
  360. {
  361. int err;
  362. vrf_nf_reset_ct(skb);
  363. err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
  364. sk, skb, NULL, skb_dst(skb)->dev, dst_output);
  365. if (likely(err == 1))
  366. err = dst_output(net, sk, skb);
  367. return err;
  368. }
  369. static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
  370. struct net_device *dev)
  371. {
  372. const struct ipv6hdr *iph;
  373. struct net *net = dev_net(skb->dev);
  374. struct flowi6 fl6;
  375. int ret = NET_XMIT_DROP;
  376. struct dst_entry *dst;
  377. struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
  378. if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr)))
  379. goto err;
  380. iph = ipv6_hdr(skb);
  381. memset(&fl6, 0, sizeof(fl6));
  382. /* needed to match OIF rule */
  383. fl6.flowi6_oif = dev->ifindex;
  384. fl6.flowi6_iif = LOOPBACK_IFINDEX;
  385. fl6.daddr = iph->daddr;
  386. fl6.saddr = iph->saddr;
  387. fl6.flowlabel = ip6_flowinfo(iph);
  388. fl6.flowi6_mark = skb->mark;
  389. fl6.flowi6_proto = iph->nexthdr;
  390. fl6.flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF;
  391. dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL);
  392. if (IS_ERR(dst) || dst == dst_null)
  393. goto err;
  394. skb_dst_drop(skb);
  395. /* if dst.dev is loopback or the VRF device again this is locally
  396. * originated traffic destined to a local address. Short circuit
  397. * to Rx path
  398. */
  399. if (dst->dev == dev)
  400. return vrf_local_xmit(skb, dev, dst);
  401. skb_dst_set(skb, dst);
  402. /* strip the ethernet header added for pass through VRF device */
  403. __skb_pull(skb, skb_network_offset(skb));
  404. memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
  405. ret = vrf_ip6_local_out(net, skb->sk, skb);
  406. if (unlikely(net_xmit_eval(ret)))
  407. dev->stats.tx_errors++;
  408. else
  409. ret = NET_XMIT_SUCCESS;
  410. return ret;
  411. err:
  412. vrf_tx_error(dev, skb);
  413. return NET_XMIT_DROP;
  414. }
  415. #else
  416. static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
  417. struct net_device *dev)
  418. {
  419. vrf_tx_error(dev, skb);
  420. return NET_XMIT_DROP;
  421. }
  422. #endif
  423. /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
  424. static int vrf_ip_local_out(struct net *net, struct sock *sk,
  425. struct sk_buff *skb)
  426. {
  427. int err;
  428. vrf_nf_reset_ct(skb);
  429. err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
  430. skb, NULL, skb_dst(skb)->dev, dst_output);
  431. if (likely(err == 1))
  432. err = dst_output(net, sk, skb);
  433. return err;
  434. }
  435. static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
  436. struct net_device *vrf_dev)
  437. {
  438. struct iphdr *ip4h;
  439. int ret = NET_XMIT_DROP;
  440. struct flowi4 fl4;
  441. struct net *net = dev_net(vrf_dev);
  442. struct rtable *rt;
  443. if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr)))
  444. goto err;
  445. ip4h = ip_hdr(skb);
  446. memset(&fl4, 0, sizeof(fl4));
  447. /* needed to match OIF rule */
  448. fl4.flowi4_oif = vrf_dev->ifindex;
  449. fl4.flowi4_iif = LOOPBACK_IFINDEX;
  450. fl4.flowi4_tos = RT_TOS(ip4h->tos);
  451. fl4.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF;
  452. fl4.flowi4_proto = ip4h->protocol;
  453. fl4.daddr = ip4h->daddr;
  454. fl4.saddr = ip4h->saddr;
  455. rt = ip_route_output_flow(net, &fl4, NULL);
  456. if (IS_ERR(rt))
  457. goto err;
  458. skb_dst_drop(skb);
  459. /* if dst.dev is loopback or the VRF device again this is locally
  460. * originated traffic destined to a local address. Short circuit
  461. * to Rx path
  462. */
  463. if (rt->dst.dev == vrf_dev)
  464. return vrf_local_xmit(skb, vrf_dev, &rt->dst);
  465. skb_dst_set(skb, &rt->dst);
  466. /* strip the ethernet header added for pass through VRF device */
  467. __skb_pull(skb, skb_network_offset(skb));
  468. if (!ip4h->saddr) {
  469. ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
  470. RT_SCOPE_LINK);
  471. }
  472. memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
  473. ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
  474. if (unlikely(net_xmit_eval(ret)))
  475. vrf_dev->stats.tx_errors++;
  476. else
  477. ret = NET_XMIT_SUCCESS;
  478. out:
  479. return ret;
  480. err:
  481. vrf_tx_error(vrf_dev, skb);
  482. goto out;
  483. }
  484. static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
  485. {
  486. switch (skb->protocol) {
  487. case htons(ETH_P_IP):
  488. return vrf_process_v4_outbound(skb, dev);
  489. case htons(ETH_P_IPV6):
  490. return vrf_process_v6_outbound(skb, dev);
  491. default:
  492. vrf_tx_error(dev, skb);
  493. return NET_XMIT_DROP;
  494. }
  495. }
  496. static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
  497. {
  498. int len = skb->len;
  499. netdev_tx_t ret = is_ip_tx_frame(skb, dev);
  500. if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
  501. struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
  502. u64_stats_update_begin(&dstats->syncp);
  503. dstats->tx_pkts++;
  504. dstats->tx_bytes += len;
  505. u64_stats_update_end(&dstats->syncp);
  506. } else {
  507. this_cpu_inc(dev->dstats->tx_drps);
  508. }
  509. return ret;
  510. }
  511. static void vrf_finish_direct(struct sk_buff *skb)
  512. {
  513. struct net_device *vrf_dev = skb->dev;
  514. if (!list_empty(&vrf_dev->ptype_all) &&
  515. likely(skb_headroom(skb) >= ETH_HLEN)) {
  516. struct ethhdr *eth = skb_push(skb, ETH_HLEN);
  517. ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
  518. eth_zero_addr(eth->h_dest);
  519. eth->h_proto = skb->protocol;
  520. rcu_read_lock_bh();
  521. dev_queue_xmit_nit(skb, vrf_dev);
  522. rcu_read_unlock_bh();
  523. skb_pull(skb, ETH_HLEN);
  524. }
  525. vrf_nf_reset_ct(skb);
  526. }
  527. #if IS_ENABLED(CONFIG_IPV6)
  528. /* modelled after ip6_finish_output2 */
  529. static int vrf_finish_output6(struct net *net, struct sock *sk,
  530. struct sk_buff *skb)
  531. {
  532. struct dst_entry *dst = skb_dst(skb);
  533. struct net_device *dev = dst->dev;
  534. const struct in6_addr *nexthop;
  535. struct neighbour *neigh;
  536. int ret;
  537. vrf_nf_reset_ct(skb);
  538. skb->protocol = htons(ETH_P_IPV6);
  539. skb->dev = dev;
  540. rcu_read_lock_bh();
  541. nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
  542. neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
  543. if (unlikely(!neigh))
  544. neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
  545. if (!IS_ERR(neigh)) {
  546. sock_confirm_neigh(skb, neigh);
  547. ret = neigh_output(neigh, skb, false);
  548. rcu_read_unlock_bh();
  549. return ret;
  550. }
  551. rcu_read_unlock_bh();
  552. IP6_INC_STATS(dev_net(dst->dev),
  553. ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
  554. kfree_skb(skb);
  555. return -EINVAL;
  556. }
  557. /* modelled after ip6_output */
  558. static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
  559. {
  560. return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
  561. net, sk, skb, NULL, skb_dst(skb)->dev,
  562. vrf_finish_output6,
  563. !(IP6CB(skb)->flags & IP6SKB_REROUTED));
  564. }
  565. /* set dst on skb to send packet to us via dev_xmit path. Allows
  566. * packet to go through device based features such as qdisc, netfilter
  567. * hooks and packet sockets with skb->dev set to vrf device.
  568. */
  569. static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
  570. struct sk_buff *skb)
  571. {
  572. struct net_vrf *vrf = netdev_priv(vrf_dev);
  573. struct dst_entry *dst = NULL;
  574. struct rt6_info *rt6;
  575. rcu_read_lock();
  576. rt6 = rcu_dereference(vrf->rt6);
  577. if (likely(rt6)) {
  578. dst = &rt6->dst;
  579. dst_hold(dst);
  580. }
  581. rcu_read_unlock();
  582. if (unlikely(!dst)) {
  583. vrf_tx_error(vrf_dev, skb);
  584. return NULL;
  585. }
  586. skb_dst_drop(skb);
  587. skb_dst_set(skb, dst);
  588. return skb;
  589. }
  590. static int vrf_output6_direct_finish(struct net *net, struct sock *sk,
  591. struct sk_buff *skb)
  592. {
  593. vrf_finish_direct(skb);
  594. return vrf_ip6_local_out(net, sk, skb);
  595. }
  596. static int vrf_output6_direct(struct net *net, struct sock *sk,
  597. struct sk_buff *skb)
  598. {
  599. int err = 1;
  600. skb->protocol = htons(ETH_P_IPV6);
  601. if (!(IPCB(skb)->flags & IPSKB_REROUTED))
  602. err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb,
  603. NULL, skb->dev, vrf_output6_direct_finish);
  604. if (likely(err == 1))
  605. vrf_finish_direct(skb);
  606. return err;
  607. }
  608. static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk,
  609. struct sk_buff *skb)
  610. {
  611. int err;
  612. err = vrf_output6_direct(net, sk, skb);
  613. if (likely(err == 1))
  614. err = vrf_ip6_local_out(net, sk, skb);
  615. return err;
  616. }
  617. static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
  618. struct sock *sk,
  619. struct sk_buff *skb)
  620. {
  621. struct net *net = dev_net(vrf_dev);
  622. int err;
  623. skb->dev = vrf_dev;
  624. err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
  625. skb, NULL, vrf_dev, vrf_ip6_out_direct_finish);
  626. if (likely(err == 1))
  627. err = vrf_output6_direct(net, sk, skb);
  628. if (likely(err == 1))
  629. return skb;
  630. return NULL;
  631. }
  632. static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
  633. struct sock *sk,
  634. struct sk_buff *skb)
  635. {
  636. /* don't divert link scope packets */
  637. if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
  638. return skb;
  639. vrf_nf_set_untracked(skb);
  640. if (qdisc_tx_is_default(vrf_dev) ||
  641. IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED)
  642. return vrf_ip6_out_direct(vrf_dev, sk, skb);
  643. return vrf_ip6_out_redirect(vrf_dev, skb);
  644. }
  645. /* holding rtnl */
  646. static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
  647. {
  648. struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
  649. struct net *net = dev_net(dev);
  650. struct dst_entry *dst;
  651. RCU_INIT_POINTER(vrf->rt6, NULL);
  652. synchronize_rcu();
  653. /* move dev in dst's to loopback so this VRF device can be deleted
  654. * - based on dst_ifdown
  655. */
  656. if (rt6) {
  657. dst = &rt6->dst;
  658. dev_put(dst->dev);
  659. dst->dev = net->loopback_dev;
  660. dev_hold(dst->dev);
  661. dst_release(dst);
  662. }
  663. }
  664. static int vrf_rt6_create(struct net_device *dev)
  665. {
  666. int flags = DST_NOPOLICY | DST_NOXFRM;
  667. struct net_vrf *vrf = netdev_priv(dev);
  668. struct net *net = dev_net(dev);
  669. struct rt6_info *rt6;
  670. int rc = -ENOMEM;
  671. /* IPv6 can be CONFIG enabled and then disabled runtime */
  672. if (!ipv6_mod_enabled())
  673. return 0;
  674. vrf->fib6_table = fib6_new_table(net, vrf->tb_id);
  675. if (!vrf->fib6_table)
  676. goto out;
  677. /* create a dst for routing packets out a VRF device */
  678. rt6 = ip6_dst_alloc(net, dev, flags);
  679. if (!rt6)
  680. goto out;
  681. rt6->dst.output = vrf_output6;
  682. rcu_assign_pointer(vrf->rt6, rt6);
  683. rc = 0;
  684. out:
  685. return rc;
  686. }
  687. #else
  688. static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
  689. struct sock *sk,
  690. struct sk_buff *skb)
  691. {
  692. return skb;
  693. }
  694. static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
  695. {
  696. }
  697. static int vrf_rt6_create(struct net_device *dev)
  698. {
  699. return 0;
  700. }
  701. #endif
  702. /* modelled after ip_finish_output2 */
  703. static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
  704. {
  705. struct dst_entry *dst = skb_dst(skb);
  706. struct rtable *rt = (struct rtable *)dst;
  707. struct net_device *dev = dst->dev;
  708. unsigned int hh_len = LL_RESERVED_SPACE(dev);
  709. struct neighbour *neigh;
  710. bool is_v6gw = false;
  711. int ret = -EINVAL;
  712. vrf_nf_reset_ct(skb);
  713. /* Be paranoid, rather than too clever. */
  714. if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
  715. struct sk_buff *skb2;
  716. skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
  717. if (!skb2) {
  718. ret = -ENOMEM;
  719. goto err;
  720. }
  721. if (skb->sk)
  722. skb_set_owner_w(skb2, skb->sk);
  723. consume_skb(skb);
  724. skb = skb2;
  725. }
  726. rcu_read_lock_bh();
  727. neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
  728. if (!IS_ERR(neigh)) {
  729. sock_confirm_neigh(skb, neigh);
  730. /* if crossing protocols, can not use the cached header */
  731. ret = neigh_output(neigh, skb, is_v6gw);
  732. rcu_read_unlock_bh();
  733. return ret;
  734. }
  735. rcu_read_unlock_bh();
  736. err:
  737. vrf_tx_error(skb->dev, skb);
  738. return ret;
  739. }
  740. static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
  741. {
  742. struct net_device *dev = skb_dst(skb)->dev;
  743. IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
  744. skb->dev = dev;
  745. skb->protocol = htons(ETH_P_IP);
  746. return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
  747. net, sk, skb, NULL, dev,
  748. vrf_finish_output,
  749. !(IPCB(skb)->flags & IPSKB_REROUTED));
  750. }
  751. /* set dst on skb to send packet to us via dev_xmit path. Allows
  752. * packet to go through device based features such as qdisc, netfilter
  753. * hooks and packet sockets with skb->dev set to vrf device.
  754. */
  755. static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
  756. struct sk_buff *skb)
  757. {
  758. struct net_vrf *vrf = netdev_priv(vrf_dev);
  759. struct dst_entry *dst = NULL;
  760. struct rtable *rth;
  761. rcu_read_lock();
  762. rth = rcu_dereference(vrf->rth);
  763. if (likely(rth)) {
  764. dst = &rth->dst;
  765. dst_hold(dst);
  766. }
  767. rcu_read_unlock();
  768. if (unlikely(!dst)) {
  769. vrf_tx_error(vrf_dev, skb);
  770. return NULL;
  771. }
  772. skb_dst_drop(skb);
  773. skb_dst_set(skb, dst);
  774. return skb;
  775. }
  776. static int vrf_output_direct_finish(struct net *net, struct sock *sk,
  777. struct sk_buff *skb)
  778. {
  779. vrf_finish_direct(skb);
  780. return vrf_ip_local_out(net, sk, skb);
  781. }
  782. static int vrf_output_direct(struct net *net, struct sock *sk,
  783. struct sk_buff *skb)
  784. {
  785. int err = 1;
  786. skb->protocol = htons(ETH_P_IP);
  787. if (!(IPCB(skb)->flags & IPSKB_REROUTED))
  788. err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb,
  789. NULL, skb->dev, vrf_output_direct_finish);
  790. if (likely(err == 1))
  791. vrf_finish_direct(skb);
  792. return err;
  793. }
  794. static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk,
  795. struct sk_buff *skb)
  796. {
  797. int err;
  798. err = vrf_output_direct(net, sk, skb);
  799. if (likely(err == 1))
  800. err = vrf_ip_local_out(net, sk, skb);
  801. return err;
  802. }
  803. static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
  804. struct sock *sk,
  805. struct sk_buff *skb)
  806. {
  807. struct net *net = dev_net(vrf_dev);
  808. int err;
  809. skb->dev = vrf_dev;
  810. err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
  811. skb, NULL, vrf_dev, vrf_ip_out_direct_finish);
  812. if (likely(err == 1))
  813. err = vrf_output_direct(net, sk, skb);
  814. if (likely(err == 1))
  815. return skb;
  816. return NULL;
  817. }
  818. static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
  819. struct sock *sk,
  820. struct sk_buff *skb)
  821. {
  822. /* don't divert multicast or local broadcast */
  823. if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
  824. ipv4_is_lbcast(ip_hdr(skb)->daddr))
  825. return skb;
  826. vrf_nf_set_untracked(skb);
  827. if (qdisc_tx_is_default(vrf_dev) ||
  828. IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED)
  829. return vrf_ip_out_direct(vrf_dev, sk, skb);
  830. return vrf_ip_out_redirect(vrf_dev, skb);
  831. }
  832. /* called with rcu lock held */
  833. static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
  834. struct sock *sk,
  835. struct sk_buff *skb,
  836. u16 proto)
  837. {
  838. switch (proto) {
  839. case AF_INET:
  840. return vrf_ip_out(vrf_dev, sk, skb);
  841. case AF_INET6:
  842. return vrf_ip6_out(vrf_dev, sk, skb);
  843. }
  844. return skb;
  845. }
  846. /* holding rtnl */
  847. static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
  848. {
  849. struct rtable *rth = rtnl_dereference(vrf->rth);
  850. struct net *net = dev_net(dev);
  851. struct dst_entry *dst;
  852. RCU_INIT_POINTER(vrf->rth, NULL);
  853. synchronize_rcu();
  854. /* move dev in dst's to loopback so this VRF device can be deleted
  855. * - based on dst_ifdown
  856. */
  857. if (rth) {
  858. dst = &rth->dst;
  859. dev_put(dst->dev);
  860. dst->dev = net->loopback_dev;
  861. dev_hold(dst->dev);
  862. dst_release(dst);
  863. }
  864. }
  865. static int vrf_rtable_create(struct net_device *dev)
  866. {
  867. struct net_vrf *vrf = netdev_priv(dev);
  868. struct rtable *rth;
  869. if (!fib_new_table(dev_net(dev), vrf->tb_id))
  870. return -ENOMEM;
  871. /* create a dst for routing packets out through a VRF device */
  872. rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1);
  873. if (!rth)
  874. return -ENOMEM;
  875. rth->dst.output = vrf_output;
  876. rcu_assign_pointer(vrf->rth, rth);
  877. return 0;
  878. }
  879. /**************************** device handling ********************/
  880. /* cycle interface to flush neighbor cache and move routes across tables */
  881. static void cycle_netdev(struct net_device *dev,
  882. struct netlink_ext_ack *extack)
  883. {
  884. unsigned int flags = dev->flags;
  885. int ret;
  886. if (!netif_running(dev))
  887. return;
  888. ret = dev_change_flags(dev, flags & ~IFF_UP, extack);
  889. if (ret >= 0)
  890. ret = dev_change_flags(dev, flags, extack);
  891. if (ret < 0) {
  892. netdev_err(dev,
  893. "Failed to cycle device %s; route tables might be wrong!\n",
  894. dev->name);
  895. }
  896. }
  897. static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
  898. struct netlink_ext_ack *extack)
  899. {
  900. int ret;
  901. /* do not allow loopback device to be enslaved to a VRF.
  902. * The vrf device acts as the loopback for the vrf.
  903. */
  904. if (port_dev == dev_net(dev)->loopback_dev) {
  905. NL_SET_ERR_MSG(extack,
  906. "Can not enslave loopback device to a VRF");
  907. return -EOPNOTSUPP;
  908. }
  909. port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
  910. ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
  911. if (ret < 0)
  912. goto err;
  913. cycle_netdev(port_dev, extack);
  914. return 0;
  915. err:
  916. port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
  917. return ret;
  918. }
  919. static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
  920. struct netlink_ext_ack *extack)
  921. {
  922. if (netif_is_l3_master(port_dev)) {
  923. NL_SET_ERR_MSG(extack,
  924. "Can not enslave an L3 master device to a VRF");
  925. return -EINVAL;
  926. }
  927. if (netif_is_l3_slave(port_dev))
  928. return -EINVAL;
  929. return do_vrf_add_slave(dev, port_dev, extack);
  930. }
  931. /* inverse of do_vrf_add_slave */
  932. static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
  933. {
  934. netdev_upper_dev_unlink(port_dev, dev);
  935. port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
  936. cycle_netdev(port_dev, NULL);
  937. return 0;
  938. }
  939. static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
  940. {
  941. return do_vrf_del_slave(dev, port_dev);
  942. }
  943. static void vrf_dev_uninit(struct net_device *dev)
  944. {
  945. struct net_vrf *vrf = netdev_priv(dev);
  946. vrf_rtable_release(dev, vrf);
  947. vrf_rt6_release(dev, vrf);
  948. free_percpu(dev->dstats);
  949. dev->dstats = NULL;
  950. }
  951. static int vrf_dev_init(struct net_device *dev)
  952. {
  953. struct net_vrf *vrf = netdev_priv(dev);
  954. dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
  955. if (!dev->dstats)
  956. goto out_nomem;
  957. /* create the default dst which points back to us */
  958. if (vrf_rtable_create(dev) != 0)
  959. goto out_stats;
  960. if (vrf_rt6_create(dev) != 0)
  961. goto out_rth;
  962. dev->flags = IFF_MASTER | IFF_NOARP;
  963. /* similarly, oper state is irrelevant; set to up to avoid confusion */
  964. dev->operstate = IF_OPER_UP;
  965. netdev_lockdep_set_classes(dev);
  966. return 0;
  967. out_rth:
  968. vrf_rtable_release(dev, vrf);
  969. out_stats:
  970. free_percpu(dev->dstats);
  971. dev->dstats = NULL;
  972. out_nomem:
  973. return -ENOMEM;
  974. }
  975. static const struct net_device_ops vrf_netdev_ops = {
  976. .ndo_init = vrf_dev_init,
  977. .ndo_uninit = vrf_dev_uninit,
  978. .ndo_start_xmit = vrf_xmit,
  979. .ndo_set_mac_address = eth_mac_addr,
  980. .ndo_get_stats64 = vrf_get_stats64,
  981. .ndo_add_slave = vrf_add_slave,
  982. .ndo_del_slave = vrf_del_slave,
  983. };
  984. static u32 vrf_fib_table(const struct net_device *dev)
  985. {
  986. struct net_vrf *vrf = netdev_priv(dev);
  987. return vrf->tb_id;
  988. }
  989. static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
  990. {
  991. kfree_skb(skb);
  992. return 0;
  993. }
  994. static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
  995. struct sk_buff *skb,
  996. struct net_device *dev)
  997. {
  998. struct net *net = dev_net(dev);
  999. if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
  1000. skb = NULL; /* kfree_skb(skb) handled by nf code */
  1001. return skb;
  1002. }
  1003. #if IS_ENABLED(CONFIG_IPV6)
  1004. /* neighbor handling is done with actual device; do not want
  1005. * to flip skb->dev for those ndisc packets. This really fails
  1006. * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
  1007. * a start.
  1008. */
  1009. static bool ipv6_ndisc_frame(const struct sk_buff *skb)
  1010. {
  1011. const struct ipv6hdr *iph = ipv6_hdr(skb);
  1012. bool rc = false;
  1013. if (iph->nexthdr == NEXTHDR_ICMP) {
  1014. const struct icmp6hdr *icmph;
  1015. struct icmp6hdr _icmph;
  1016. icmph = skb_header_pointer(skb, sizeof(*iph),
  1017. sizeof(_icmph), &_icmph);
  1018. if (!icmph)
  1019. goto out;
  1020. switch (icmph->icmp6_type) {
  1021. case NDISC_ROUTER_SOLICITATION:
  1022. case NDISC_ROUTER_ADVERTISEMENT:
  1023. case NDISC_NEIGHBOUR_SOLICITATION:
  1024. case NDISC_NEIGHBOUR_ADVERTISEMENT:
  1025. case NDISC_REDIRECT:
  1026. rc = true;
  1027. break;
  1028. }
  1029. }
  1030. out:
  1031. return rc;
  1032. }
  1033. static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
  1034. const struct net_device *dev,
  1035. struct flowi6 *fl6,
  1036. int ifindex,
  1037. const struct sk_buff *skb,
  1038. int flags)
  1039. {
  1040. struct net_vrf *vrf = netdev_priv(dev);
  1041. return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags);
  1042. }
  1043. static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
  1044. int ifindex)
  1045. {
  1046. const struct ipv6hdr *iph = ipv6_hdr(skb);
  1047. struct flowi6 fl6 = {
  1048. .flowi6_iif = ifindex,
  1049. .flowi6_mark = skb->mark,
  1050. .flowi6_proto = iph->nexthdr,
  1051. .daddr = iph->daddr,
  1052. .saddr = iph->saddr,
  1053. .flowlabel = ip6_flowinfo(iph),
  1054. };
  1055. struct net *net = dev_net(vrf_dev);
  1056. struct rt6_info *rt6;
  1057. rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
  1058. RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
  1059. if (unlikely(!rt6))
  1060. return;
  1061. if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
  1062. return;
  1063. skb_dst_set(skb, &rt6->dst);
  1064. }
  1065. static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
  1066. struct sk_buff *skb)
  1067. {
  1068. int orig_iif = skb->skb_iif;
  1069. bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
  1070. bool is_ndisc = ipv6_ndisc_frame(skb);
  1071. /* loopback, multicast & non-ND link-local traffic; do not push through
  1072. * packet taps again. Reset pkt_type for upper layers to process skb.
  1073. * For strict packets with a source LLA, determine the dst using the
  1074. * original ifindex.
  1075. */
  1076. if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) {
  1077. skb->dev = vrf_dev;
  1078. skb->skb_iif = vrf_dev->ifindex;
  1079. IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
  1080. if (skb->pkt_type == PACKET_LOOPBACK)
  1081. skb->pkt_type = PACKET_HOST;
  1082. else if (ipv6_addr_type(&ipv6_hdr(skb)->saddr) & IPV6_ADDR_LINKLOCAL)
  1083. vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
  1084. goto out;
  1085. }
  1086. /* if packet is NDISC then keep the ingress interface */
  1087. if (!is_ndisc) {
  1088. vrf_rx_stats(vrf_dev, skb->len);
  1089. skb->dev = vrf_dev;
  1090. skb->skb_iif = vrf_dev->ifindex;
  1091. if (!list_empty(&vrf_dev->ptype_all)) {
  1092. skb_push(skb, skb->mac_len);
  1093. dev_queue_xmit_nit(skb, vrf_dev);
  1094. skb_pull(skb, skb->mac_len);
  1095. }
  1096. IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
  1097. }
  1098. if (need_strict)
  1099. vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
  1100. skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
  1101. out:
  1102. return skb;
  1103. }
  1104. #else
  1105. static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
  1106. struct sk_buff *skb)
  1107. {
  1108. return skb;
  1109. }
  1110. #endif
  1111. static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
  1112. struct sk_buff *skb)
  1113. {
  1114. skb->dev = vrf_dev;
  1115. skb->skb_iif = vrf_dev->ifindex;
  1116. IPCB(skb)->flags |= IPSKB_L3SLAVE;
  1117. if (ipv4_is_multicast(ip_hdr(skb)->daddr))
  1118. goto out;
  1119. /* loopback traffic; do not push through packet taps again.
  1120. * Reset pkt_type for upper layers to process skb
  1121. */
  1122. if (skb->pkt_type == PACKET_LOOPBACK) {
  1123. skb->pkt_type = PACKET_HOST;
  1124. goto out;
  1125. }
  1126. vrf_rx_stats(vrf_dev, skb->len);
  1127. if (!list_empty(&vrf_dev->ptype_all)) {
  1128. skb_push(skb, skb->mac_len);
  1129. dev_queue_xmit_nit(skb, vrf_dev);
  1130. skb_pull(skb, skb->mac_len);
  1131. }
  1132. skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
  1133. out:
  1134. return skb;
  1135. }
  1136. /* called with rcu lock held */
  1137. static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
  1138. struct sk_buff *skb,
  1139. u16 proto)
  1140. {
  1141. switch (proto) {
  1142. case AF_INET:
  1143. return vrf_ip_rcv(vrf_dev, skb);
  1144. case AF_INET6:
  1145. return vrf_ip6_rcv(vrf_dev, skb);
  1146. }
  1147. return skb;
  1148. }
  1149. #if IS_ENABLED(CONFIG_IPV6)
  1150. /* send to link-local or multicast address via interface enslaved to
  1151. * VRF device. Force lookup to VRF table without changing flow struct
  1152. * Note: Caller to this function must hold rcu_read_lock() and no refcnt
  1153. * is taken on the dst by this function.
  1154. */
  1155. static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
  1156. struct flowi6 *fl6)
  1157. {
  1158. struct net *net = dev_net(dev);
  1159. int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF;
  1160. struct dst_entry *dst = NULL;
  1161. struct rt6_info *rt;
  1162. /* VRF device does not have a link-local address and
  1163. * sending packets to link-local or mcast addresses over
  1164. * a VRF device does not make sense
  1165. */
  1166. if (fl6->flowi6_oif == dev->ifindex) {
  1167. dst = &net->ipv6.ip6_null_entry->dst;
  1168. return dst;
  1169. }
  1170. if (!ipv6_addr_any(&fl6->saddr))
  1171. flags |= RT6_LOOKUP_F_HAS_SADDR;
  1172. rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
  1173. if (rt)
  1174. dst = &rt->dst;
  1175. return dst;
  1176. }
  1177. #endif
  1178. static const struct l3mdev_ops vrf_l3mdev_ops = {
  1179. .l3mdev_fib_table = vrf_fib_table,
  1180. .l3mdev_l3_rcv = vrf_l3_rcv,
  1181. .l3mdev_l3_out = vrf_l3_out,
  1182. #if IS_ENABLED(CONFIG_IPV6)
  1183. .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
  1184. #endif
  1185. };
  1186. static void vrf_get_drvinfo(struct net_device *dev,
  1187. struct ethtool_drvinfo *info)
  1188. {
  1189. strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
  1190. strlcpy(info->version, DRV_VERSION, sizeof(info->version));
  1191. }
  1192. static const struct ethtool_ops vrf_ethtool_ops = {
  1193. .get_drvinfo = vrf_get_drvinfo,
  1194. };
  1195. static inline size_t vrf_fib_rule_nl_size(void)
  1196. {
  1197. size_t sz;
  1198. sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
  1199. sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
  1200. sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
  1201. sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */
  1202. return sz;
  1203. }
  1204. static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
  1205. {
  1206. struct fib_rule_hdr *frh;
  1207. struct nlmsghdr *nlh;
  1208. struct sk_buff *skb;
  1209. int err;
  1210. if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) &&
  1211. !ipv6_mod_enabled())
  1212. return 0;
  1213. skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
  1214. if (!skb)
  1215. return -ENOMEM;
  1216. nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
  1217. if (!nlh)
  1218. goto nla_put_failure;
  1219. /* rule only needs to appear once */
  1220. nlh->nlmsg_flags |= NLM_F_EXCL;
  1221. frh = nlmsg_data(nlh);
  1222. memset(frh, 0, sizeof(*frh));
  1223. frh->family = family;
  1224. frh->action = FR_ACT_TO_TBL;
  1225. if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
  1226. goto nla_put_failure;
  1227. if (nla_put_u8(skb, FRA_L3MDEV, 1))
  1228. goto nla_put_failure;
  1229. if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
  1230. goto nla_put_failure;
  1231. nlmsg_end(skb, nlh);
  1232. /* fib_nl_{new,del}rule handling looks for net from skb->sk */
  1233. skb->sk = dev_net(dev)->rtnl;
  1234. if (add_it) {
  1235. err = fib_nl_newrule(skb, nlh, NULL);
  1236. if (err == -EEXIST)
  1237. err = 0;
  1238. } else {
  1239. err = fib_nl_delrule(skb, nlh, NULL);
  1240. if (err == -ENOENT)
  1241. err = 0;
  1242. }
  1243. nlmsg_free(skb);
  1244. return err;
  1245. nla_put_failure:
  1246. nlmsg_free(skb);
  1247. return -EMSGSIZE;
  1248. }
  1249. static int vrf_add_fib_rules(const struct net_device *dev)
  1250. {
  1251. int err;
  1252. err = vrf_fib_rule(dev, AF_INET, true);
  1253. if (err < 0)
  1254. goto out_err;
  1255. err = vrf_fib_rule(dev, AF_INET6, true);
  1256. if (err < 0)
  1257. goto ipv6_err;
  1258. #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
  1259. err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
  1260. if (err < 0)
  1261. goto ipmr_err;
  1262. #endif
  1263. #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
  1264. err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true);
  1265. if (err < 0)
  1266. goto ip6mr_err;
  1267. #endif
  1268. return 0;
  1269. #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
  1270. ip6mr_err:
  1271. vrf_fib_rule(dev, RTNL_FAMILY_IPMR, false);
  1272. #endif
  1273. #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
  1274. ipmr_err:
  1275. vrf_fib_rule(dev, AF_INET6, false);
  1276. #endif
  1277. ipv6_err:
  1278. vrf_fib_rule(dev, AF_INET, false);
  1279. out_err:
  1280. netdev_err(dev, "Failed to add FIB rules.\n");
  1281. return err;
  1282. }
  1283. static void vrf_setup(struct net_device *dev)
  1284. {
  1285. ether_setup(dev);
  1286. /* Initialize the device structure. */
  1287. dev->netdev_ops = &vrf_netdev_ops;
  1288. dev->l3mdev_ops = &vrf_l3mdev_ops;
  1289. dev->ethtool_ops = &vrf_ethtool_ops;
  1290. dev->needs_free_netdev = true;
  1291. /* Fill in device structure with ethernet-generic values. */
  1292. eth_hw_addr_random(dev);
  1293. /* don't acquire vrf device's netif_tx_lock when transmitting */
  1294. dev->features |= NETIF_F_LLTX;
  1295. /* don't allow vrf devices to change network namespaces. */
  1296. dev->features |= NETIF_F_NETNS_LOCAL;
  1297. /* does not make sense for a VLAN to be added to a vrf device */
  1298. dev->features |= NETIF_F_VLAN_CHALLENGED;
  1299. /* enable offload features */
  1300. dev->features |= NETIF_F_GSO_SOFTWARE;
  1301. dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC;
  1302. dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
  1303. dev->hw_features = dev->features;
  1304. dev->hw_enc_features = dev->features;
  1305. /* default to no qdisc; user can add if desired */
  1306. dev->priv_flags |= IFF_NO_QUEUE;
  1307. dev->priv_flags |= IFF_NO_RX_HANDLER;
  1308. dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
  1309. /* VRF devices do not care about MTU, but if the MTU is set
  1310. * too low then the ipv4 and ipv6 protocols are disabled
  1311. * which breaks networking.
  1312. */
  1313. dev->min_mtu = IPV6_MIN_MTU;
  1314. dev->max_mtu = IP6_MAX_MTU;
  1315. dev->mtu = dev->max_mtu;
  1316. }
  1317. static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
  1318. struct netlink_ext_ack *extack)
  1319. {
  1320. if (tb[IFLA_ADDRESS]) {
  1321. if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
  1322. NL_SET_ERR_MSG(extack, "Invalid hardware address");
  1323. return -EINVAL;
  1324. }
  1325. if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
  1326. NL_SET_ERR_MSG(extack, "Invalid hardware address");
  1327. return -EADDRNOTAVAIL;
  1328. }
  1329. }
  1330. return 0;
  1331. }
  1332. static void vrf_dellink(struct net_device *dev, struct list_head *head)
  1333. {
  1334. struct net_device *port_dev;
  1335. struct list_head *iter;
  1336. netdev_for_each_lower_dev(dev, port_dev, iter)
  1337. vrf_del_slave(dev, port_dev);
  1338. vrf_map_unregister_dev(dev);
  1339. unregister_netdevice_queue(dev, head);
  1340. }
  1341. static int vrf_newlink(struct net *src_net, struct net_device *dev,
  1342. struct nlattr *tb[], struct nlattr *data[],
  1343. struct netlink_ext_ack *extack)
  1344. {
  1345. struct net_vrf *vrf = netdev_priv(dev);
  1346. struct netns_vrf *nn_vrf;
  1347. bool *add_fib_rules;
  1348. struct net *net;
  1349. int err;
  1350. if (!data || !data[IFLA_VRF_TABLE]) {
  1351. NL_SET_ERR_MSG(extack, "VRF table id is missing");
  1352. return -EINVAL;
  1353. }
  1354. vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
  1355. if (vrf->tb_id == RT_TABLE_UNSPEC) {
  1356. NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
  1357. "Invalid VRF table id");
  1358. return -EINVAL;
  1359. }
  1360. dev->priv_flags |= IFF_L3MDEV_MASTER;
  1361. err = register_netdevice(dev);
  1362. if (err)
  1363. goto out;
  1364. /* mapping between table_id and vrf;
  1365. * note: such binding could not be done in the dev init function
  1366. * because dev->ifindex id is not available yet.
  1367. */
  1368. vrf->ifindex = dev->ifindex;
  1369. err = vrf_map_register_dev(dev, extack);
  1370. if (err) {
  1371. unregister_netdevice(dev);
  1372. goto out;
  1373. }
  1374. net = dev_net(dev);
  1375. nn_vrf = net_generic(net, vrf_net_id);
  1376. add_fib_rules = &nn_vrf->add_fib_rules;
  1377. if (*add_fib_rules) {
  1378. err = vrf_add_fib_rules(dev);
  1379. if (err) {
  1380. vrf_map_unregister_dev(dev);
  1381. unregister_netdevice(dev);
  1382. goto out;
  1383. }
  1384. *add_fib_rules = false;
  1385. }
  1386. out:
  1387. return err;
  1388. }
  1389. static size_t vrf_nl_getsize(const struct net_device *dev)
  1390. {
  1391. return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
  1392. }
  1393. static int vrf_fillinfo(struct sk_buff *skb,
  1394. const struct net_device *dev)
  1395. {
  1396. struct net_vrf *vrf = netdev_priv(dev);
  1397. return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
  1398. }
  1399. static size_t vrf_get_slave_size(const struct net_device *bond_dev,
  1400. const struct net_device *slave_dev)
  1401. {
  1402. return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
  1403. }
  1404. static int vrf_fill_slave_info(struct sk_buff *skb,
  1405. const struct net_device *vrf_dev,
  1406. const struct net_device *slave_dev)
  1407. {
  1408. struct net_vrf *vrf = netdev_priv(vrf_dev);
  1409. if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
  1410. return -EMSGSIZE;
  1411. return 0;
  1412. }
  1413. static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
  1414. [IFLA_VRF_TABLE] = { .type = NLA_U32 },
  1415. };
  1416. static struct rtnl_link_ops vrf_link_ops __read_mostly = {
  1417. .kind = DRV_NAME,
  1418. .priv_size = sizeof(struct net_vrf),
  1419. .get_size = vrf_nl_getsize,
  1420. .policy = vrf_nl_policy,
  1421. .validate = vrf_validate,
  1422. .fill_info = vrf_fillinfo,
  1423. .get_slave_size = vrf_get_slave_size,
  1424. .fill_slave_info = vrf_fill_slave_info,
  1425. .newlink = vrf_newlink,
  1426. .dellink = vrf_dellink,
  1427. .setup = vrf_setup,
  1428. .maxtype = IFLA_VRF_MAX,
  1429. };
  1430. static int vrf_device_event(struct notifier_block *unused,
  1431. unsigned long event, void *ptr)
  1432. {
  1433. struct net_device *dev = netdev_notifier_info_to_dev(ptr);
  1434. /* only care about unregister events to drop slave references */
  1435. if (event == NETDEV_UNREGISTER) {
  1436. struct net_device *vrf_dev;
  1437. if (!netif_is_l3_slave(dev))
  1438. goto out;
  1439. vrf_dev = netdev_master_upper_dev_get(dev);
  1440. vrf_del_slave(vrf_dev, dev);
  1441. }
  1442. out:
  1443. return NOTIFY_DONE;
  1444. }
  1445. static struct notifier_block vrf_notifier_block __read_mostly = {
  1446. .notifier_call = vrf_device_event,
  1447. };
  1448. static int vrf_map_init(struct vrf_map *vmap)
  1449. {
  1450. spin_lock_init(&vmap->vmap_lock);
  1451. hash_init(vmap->ht);
  1452. vmap->strict_mode = false;
  1453. return 0;
  1454. }
  1455. #ifdef CONFIG_SYSCTL
  1456. static bool vrf_strict_mode(struct vrf_map *vmap)
  1457. {
  1458. bool strict_mode;
  1459. vrf_map_lock(vmap);
  1460. strict_mode = vmap->strict_mode;
  1461. vrf_map_unlock(vmap);
  1462. return strict_mode;
  1463. }
  1464. static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode)
  1465. {
  1466. bool *cur_mode;
  1467. int res = 0;
  1468. vrf_map_lock(vmap);
  1469. cur_mode = &vmap->strict_mode;
  1470. if (*cur_mode == new_mode)
  1471. goto unlock;
  1472. if (*cur_mode) {
  1473. /* disable strict mode */
  1474. *cur_mode = false;
  1475. } else {
  1476. if (vmap->shared_tables) {
  1477. /* we cannot allow strict_mode because there are some
  1478. * vrfs that share one or more tables.
  1479. */
  1480. res = -EBUSY;
  1481. goto unlock;
  1482. }
  1483. /* no tables are shared among vrfs, so we can go back
  1484. * to 1:1 association between a vrf with its table.
  1485. */
  1486. *cur_mode = true;
  1487. }
  1488. unlock:
  1489. vrf_map_unlock(vmap);
  1490. return res;
  1491. }
  1492. static int vrf_shared_table_handler(struct ctl_table *table, int write,
  1493. void *buffer, size_t *lenp, loff_t *ppos)
  1494. {
  1495. struct net *net = (struct net *)table->extra1;
  1496. struct vrf_map *vmap = netns_vrf_map(net);
  1497. int proc_strict_mode = 0;
  1498. struct ctl_table tmp = {
  1499. .procname = table->procname,
  1500. .data = &proc_strict_mode,
  1501. .maxlen = sizeof(int),
  1502. .mode = table->mode,
  1503. .extra1 = SYSCTL_ZERO,
  1504. .extra2 = SYSCTL_ONE,
  1505. };
  1506. int ret;
  1507. if (!write)
  1508. proc_strict_mode = vrf_strict_mode(vmap);
  1509. ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
  1510. if (write && ret == 0)
  1511. ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode);
  1512. return ret;
  1513. }
  1514. static const struct ctl_table vrf_table[] = {
  1515. {
  1516. .procname = "strict_mode",
  1517. .data = NULL,
  1518. .maxlen = sizeof(int),
  1519. .mode = 0644,
  1520. .proc_handler = vrf_shared_table_handler,
  1521. /* set by the vrf_netns_init */
  1522. .extra1 = NULL,
  1523. },
  1524. { },
  1525. };
  1526. static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)
  1527. {
  1528. struct ctl_table *table;
  1529. table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL);
  1530. if (!table)
  1531. return -ENOMEM;
  1532. /* init the extra1 parameter with the reference to current netns */
  1533. table[0].extra1 = net;
  1534. nn_vrf->ctl_hdr = register_net_sysctl(net, "net/vrf", table);
  1535. if (!nn_vrf->ctl_hdr) {
  1536. kfree(table);
  1537. return -ENOMEM;
  1538. }
  1539. return 0;
  1540. }
  1541. static void vrf_netns_exit_sysctl(struct net *net)
  1542. {
  1543. struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
  1544. struct ctl_table *table;
  1545. table = nn_vrf->ctl_hdr->ctl_table_arg;
  1546. unregister_net_sysctl_table(nn_vrf->ctl_hdr);
  1547. kfree(table);
  1548. }
  1549. #else
  1550. static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)
  1551. {
  1552. return 0;
  1553. }
  1554. static void vrf_netns_exit_sysctl(struct net *net)
  1555. {
  1556. }
  1557. #endif
  1558. /* Initialize per network namespace state */
  1559. static int __net_init vrf_netns_init(struct net *net)
  1560. {
  1561. struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
  1562. nn_vrf->add_fib_rules = true;
  1563. vrf_map_init(&nn_vrf->vmap);
  1564. return vrf_netns_init_sysctl(net, nn_vrf);
  1565. }
  1566. static void __net_exit vrf_netns_exit(struct net *net)
  1567. {
  1568. vrf_netns_exit_sysctl(net);
  1569. }
  1570. static struct pernet_operations vrf_net_ops __net_initdata = {
  1571. .init = vrf_netns_init,
  1572. .exit = vrf_netns_exit,
  1573. .id = &vrf_net_id,
  1574. .size = sizeof(struct netns_vrf),
  1575. };
  1576. static int __init vrf_init_module(void)
  1577. {
  1578. int rc;
  1579. register_netdevice_notifier(&vrf_notifier_block);
  1580. rc = register_pernet_subsys(&vrf_net_ops);
  1581. if (rc < 0)
  1582. goto error;
  1583. rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF,
  1584. vrf_ifindex_lookup_by_table_id);
  1585. if (rc < 0)
  1586. goto unreg_pernet;
  1587. rc = rtnl_link_register(&vrf_link_ops);
  1588. if (rc < 0)
  1589. goto table_lookup_unreg;
  1590. return 0;
  1591. table_lookup_unreg:
  1592. l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF,
  1593. vrf_ifindex_lookup_by_table_id);
  1594. unreg_pernet:
  1595. unregister_pernet_subsys(&vrf_net_ops);
  1596. error:
  1597. unregister_netdevice_notifier(&vrf_notifier_block);
  1598. return rc;
  1599. }
  1600. module_init(vrf_init_module);
  1601. MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
  1602. MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
  1603. MODULE_LICENSE("GPL");
  1604. MODULE_ALIAS_RTNL_LINK(DRV_NAME);
  1605. MODULE_VERSION(DRV_VERSION);