lwt_bpf.c 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658
  1. // SPDX-License-Identifier: GPL-2.0-only
  2. /* Copyright (c) 2016 Thomas Graf <tgraf@tgraf.ch>
  3. */
  4. #include <linux/kernel.h>
  5. #include <linux/module.h>
  6. #include <linux/skbuff.h>
  7. #include <linux/types.h>
  8. #include <linux/bpf.h>
  9. #include <net/lwtunnel.h>
  10. #include <net/gre.h>
  11. #include <net/ip6_route.h>
  12. #include <net/ipv6_stubs.h>
  13. struct bpf_lwt_prog {
  14. struct bpf_prog *prog;
  15. char *name;
  16. };
  17. struct bpf_lwt {
  18. struct bpf_lwt_prog in;
  19. struct bpf_lwt_prog out;
  20. struct bpf_lwt_prog xmit;
  21. int family;
  22. };
  23. #define MAX_PROG_NAME 256
  24. static inline struct bpf_lwt *bpf_lwt_lwtunnel(struct lwtunnel_state *lwt)
  25. {
  26. return (struct bpf_lwt *)lwt->data;
  27. }
  28. #define NO_REDIRECT false
  29. #define CAN_REDIRECT true
  30. static int run_lwt_bpf(struct sk_buff *skb, struct bpf_lwt_prog *lwt,
  31. struct dst_entry *dst, bool can_redirect)
  32. {
  33. int ret;
  34. /* Migration disable and BH disable are needed to protect per-cpu
  35. * redirect_info between BPF prog and skb_do_redirect().
  36. */
  37. migrate_disable();
  38. local_bh_disable();
  39. bpf_compute_data_pointers(skb);
  40. ret = bpf_prog_run_save_cb(lwt->prog, skb);
  41. switch (ret) {
  42. case BPF_OK:
  43. case BPF_LWT_REROUTE:
  44. break;
  45. case BPF_REDIRECT:
  46. if (unlikely(!can_redirect)) {
  47. pr_warn_once("Illegal redirect return code in prog %s\n",
  48. lwt->name ? : "<unknown>");
  49. ret = BPF_OK;
  50. } else {
  51. skb_reset_mac_header(skb);
  52. ret = skb_do_redirect(skb);
  53. if (ret == 0)
  54. ret = BPF_REDIRECT;
  55. }
  56. break;
  57. case BPF_DROP:
  58. kfree_skb(skb);
  59. ret = -EPERM;
  60. break;
  61. default:
  62. pr_warn_once("bpf-lwt: Illegal return value %u, expect packet loss\n", ret);
  63. kfree_skb(skb);
  64. ret = -EINVAL;
  65. break;
  66. }
  67. local_bh_enable();
  68. migrate_enable();
  69. return ret;
  70. }
  71. static int bpf_lwt_input_reroute(struct sk_buff *skb)
  72. {
  73. int err = -EINVAL;
  74. if (skb->protocol == htons(ETH_P_IP)) {
  75. struct net_device *dev = skb_dst(skb)->dev;
  76. struct iphdr *iph = ip_hdr(skb);
  77. dev_hold(dev);
  78. skb_dst_drop(skb);
  79. err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
  80. iph->tos, dev);
  81. dev_put(dev);
  82. } else if (skb->protocol == htons(ETH_P_IPV6)) {
  83. skb_dst_drop(skb);
  84. err = ipv6_stub->ipv6_route_input(skb);
  85. } else {
  86. err = -EAFNOSUPPORT;
  87. }
  88. if (err)
  89. goto err;
  90. return dst_input(skb);
  91. err:
  92. kfree_skb(skb);
  93. return err;
  94. }
  95. static int bpf_input(struct sk_buff *skb)
  96. {
  97. struct dst_entry *dst = skb_dst(skb);
  98. struct bpf_lwt *bpf;
  99. int ret;
  100. bpf = bpf_lwt_lwtunnel(dst->lwtstate);
  101. if (bpf->in.prog) {
  102. ret = run_lwt_bpf(skb, &bpf->in, dst, NO_REDIRECT);
  103. if (ret < 0)
  104. return ret;
  105. if (ret == BPF_LWT_REROUTE)
  106. return bpf_lwt_input_reroute(skb);
  107. }
  108. if (unlikely(!dst->lwtstate->orig_input)) {
  109. kfree_skb(skb);
  110. return -EINVAL;
  111. }
  112. return dst->lwtstate->orig_input(skb);
  113. }
  114. static int bpf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
  115. {
  116. struct dst_entry *dst = skb_dst(skb);
  117. struct bpf_lwt *bpf;
  118. int ret;
  119. bpf = bpf_lwt_lwtunnel(dst->lwtstate);
  120. if (bpf->out.prog) {
  121. ret = run_lwt_bpf(skb, &bpf->out, dst, NO_REDIRECT);
  122. if (ret < 0)
  123. return ret;
  124. }
  125. if (unlikely(!dst->lwtstate->orig_output)) {
  126. pr_warn_once("orig_output not set on dst for prog %s\n",
  127. bpf->out.name);
  128. kfree_skb(skb);
  129. return -EINVAL;
  130. }
  131. return dst->lwtstate->orig_output(net, sk, skb);
  132. }
  133. static int xmit_check_hhlen(struct sk_buff *skb)
  134. {
  135. int hh_len = skb_dst(skb)->dev->hard_header_len;
  136. if (skb_headroom(skb) < hh_len) {
  137. int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
  138. if (pskb_expand_head(skb, nhead, 0, GFP_ATOMIC))
  139. return -ENOMEM;
  140. }
  141. return 0;
  142. }
  143. static int bpf_lwt_xmit_reroute(struct sk_buff *skb)
  144. {
  145. struct net_device *l3mdev = l3mdev_master_dev_rcu(skb_dst(skb)->dev);
  146. int oif = l3mdev ? l3mdev->ifindex : 0;
  147. struct dst_entry *dst = NULL;
  148. int err = -EAFNOSUPPORT;
  149. struct sock *sk;
  150. struct net *net;
  151. bool ipv4;
  152. if (skb->protocol == htons(ETH_P_IP))
  153. ipv4 = true;
  154. else if (skb->protocol == htons(ETH_P_IPV6))
  155. ipv4 = false;
  156. else
  157. goto err;
  158. sk = sk_to_full_sk(skb->sk);
  159. if (sk) {
  160. if (sk->sk_bound_dev_if)
  161. oif = sk->sk_bound_dev_if;
  162. net = sock_net(sk);
  163. } else {
  164. net = dev_net(skb_dst(skb)->dev);
  165. }
  166. if (ipv4) {
  167. struct iphdr *iph = ip_hdr(skb);
  168. struct flowi4 fl4 = {};
  169. struct rtable *rt;
  170. fl4.flowi4_oif = oif;
  171. fl4.flowi4_mark = skb->mark;
  172. fl4.flowi4_uid = sock_net_uid(net, sk);
  173. fl4.flowi4_tos = RT_TOS(iph->tos);
  174. fl4.flowi4_flags = FLOWI_FLAG_ANYSRC;
  175. fl4.flowi4_proto = iph->protocol;
  176. fl4.daddr = iph->daddr;
  177. fl4.saddr = iph->saddr;
  178. rt = ip_route_output_key(net, &fl4);
  179. if (IS_ERR(rt)) {
  180. err = PTR_ERR(rt);
  181. goto err;
  182. }
  183. dst = &rt->dst;
  184. } else {
  185. struct ipv6hdr *iph6 = ipv6_hdr(skb);
  186. struct flowi6 fl6 = {};
  187. fl6.flowi6_oif = oif;
  188. fl6.flowi6_mark = skb->mark;
  189. fl6.flowi6_uid = sock_net_uid(net, sk);
  190. fl6.flowlabel = ip6_flowinfo(iph6);
  191. fl6.flowi6_proto = iph6->nexthdr;
  192. fl6.daddr = iph6->daddr;
  193. fl6.saddr = iph6->saddr;
  194. dst = ipv6_stub->ipv6_dst_lookup_flow(net, skb->sk, &fl6, NULL);
  195. if (IS_ERR(dst)) {
  196. err = PTR_ERR(dst);
  197. goto err;
  198. }
  199. }
  200. if (unlikely(dst->error)) {
  201. err = dst->error;
  202. dst_release(dst);
  203. goto err;
  204. }
  205. /* Although skb header was reserved in bpf_lwt_push_ip_encap(), it
  206. * was done for the previous dst, so we are doing it here again, in
  207. * case the new dst needs much more space. The call below is a noop
  208. * if there is enough header space in skb.
  209. */
  210. err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
  211. if (unlikely(err))
  212. goto err;
  213. skb_dst_drop(skb);
  214. skb_dst_set(skb, dst);
  215. err = dst_output(dev_net(skb_dst(skb)->dev), skb->sk, skb);
  216. if (unlikely(err))
  217. return err;
  218. /* ip[6]_finish_output2 understand LWTUNNEL_XMIT_DONE */
  219. return LWTUNNEL_XMIT_DONE;
  220. err:
  221. kfree_skb(skb);
  222. return err;
  223. }
  224. static int bpf_xmit(struct sk_buff *skb)
  225. {
  226. struct dst_entry *dst = skb_dst(skb);
  227. struct bpf_lwt *bpf;
  228. bpf = bpf_lwt_lwtunnel(dst->lwtstate);
  229. if (bpf->xmit.prog) {
  230. __be16 proto = skb->protocol;
  231. int ret;
  232. ret = run_lwt_bpf(skb, &bpf->xmit, dst, CAN_REDIRECT);
  233. switch (ret) {
  234. case BPF_OK:
  235. /* If the header changed, e.g. via bpf_lwt_push_encap,
  236. * BPF_LWT_REROUTE below should have been used if the
  237. * protocol was also changed.
  238. */
  239. if (skb->protocol != proto) {
  240. kfree_skb(skb);
  241. return -EINVAL;
  242. }
  243. /* If the header was expanded, headroom might be too
  244. * small for L2 header to come, expand as needed.
  245. */
  246. ret = xmit_check_hhlen(skb);
  247. if (unlikely(ret))
  248. return ret;
  249. return LWTUNNEL_XMIT_CONTINUE;
  250. case BPF_REDIRECT:
  251. return LWTUNNEL_XMIT_DONE;
  252. case BPF_LWT_REROUTE:
  253. return bpf_lwt_xmit_reroute(skb);
  254. default:
  255. return ret;
  256. }
  257. }
  258. return LWTUNNEL_XMIT_CONTINUE;
  259. }
  260. static void bpf_lwt_prog_destroy(struct bpf_lwt_prog *prog)
  261. {
  262. if (prog->prog)
  263. bpf_prog_put(prog->prog);
  264. kfree(prog->name);
  265. }
  266. static void bpf_destroy_state(struct lwtunnel_state *lwt)
  267. {
  268. struct bpf_lwt *bpf = bpf_lwt_lwtunnel(lwt);
  269. bpf_lwt_prog_destroy(&bpf->in);
  270. bpf_lwt_prog_destroy(&bpf->out);
  271. bpf_lwt_prog_destroy(&bpf->xmit);
  272. }
  273. static const struct nla_policy bpf_prog_policy[LWT_BPF_PROG_MAX + 1] = {
  274. [LWT_BPF_PROG_FD] = { .type = NLA_U32, },
  275. [LWT_BPF_PROG_NAME] = { .type = NLA_NUL_STRING,
  276. .len = MAX_PROG_NAME },
  277. };
  278. static int bpf_parse_prog(struct nlattr *attr, struct bpf_lwt_prog *prog,
  279. enum bpf_prog_type type)
  280. {
  281. struct nlattr *tb[LWT_BPF_PROG_MAX + 1];
  282. struct bpf_prog *p;
  283. int ret;
  284. u32 fd;
  285. ret = nla_parse_nested_deprecated(tb, LWT_BPF_PROG_MAX, attr,
  286. bpf_prog_policy, NULL);
  287. if (ret < 0)
  288. return ret;
  289. if (!tb[LWT_BPF_PROG_FD] || !tb[LWT_BPF_PROG_NAME])
  290. return -EINVAL;
  291. prog->name = nla_memdup(tb[LWT_BPF_PROG_NAME], GFP_ATOMIC);
  292. if (!prog->name)
  293. return -ENOMEM;
  294. fd = nla_get_u32(tb[LWT_BPF_PROG_FD]);
  295. p = bpf_prog_get_type(fd, type);
  296. if (IS_ERR(p))
  297. return PTR_ERR(p);
  298. prog->prog = p;
  299. return 0;
  300. }
  301. static const struct nla_policy bpf_nl_policy[LWT_BPF_MAX + 1] = {
  302. [LWT_BPF_IN] = { .type = NLA_NESTED, },
  303. [LWT_BPF_OUT] = { .type = NLA_NESTED, },
  304. [LWT_BPF_XMIT] = { .type = NLA_NESTED, },
  305. [LWT_BPF_XMIT_HEADROOM] = { .type = NLA_U32 },
  306. };
  307. static int bpf_build_state(struct net *net, struct nlattr *nla,
  308. unsigned int family, const void *cfg,
  309. struct lwtunnel_state **ts,
  310. struct netlink_ext_ack *extack)
  311. {
  312. struct nlattr *tb[LWT_BPF_MAX + 1];
  313. struct lwtunnel_state *newts;
  314. struct bpf_lwt *bpf;
  315. int ret;
  316. if (family != AF_INET && family != AF_INET6)
  317. return -EAFNOSUPPORT;
  318. ret = nla_parse_nested_deprecated(tb, LWT_BPF_MAX, nla, bpf_nl_policy,
  319. extack);
  320. if (ret < 0)
  321. return ret;
  322. if (!tb[LWT_BPF_IN] && !tb[LWT_BPF_OUT] && !tb[LWT_BPF_XMIT])
  323. return -EINVAL;
  324. newts = lwtunnel_state_alloc(sizeof(*bpf));
  325. if (!newts)
  326. return -ENOMEM;
  327. newts->type = LWTUNNEL_ENCAP_BPF;
  328. bpf = bpf_lwt_lwtunnel(newts);
  329. if (tb[LWT_BPF_IN]) {
  330. newts->flags |= LWTUNNEL_STATE_INPUT_REDIRECT;
  331. ret = bpf_parse_prog(tb[LWT_BPF_IN], &bpf->in,
  332. BPF_PROG_TYPE_LWT_IN);
  333. if (ret < 0)
  334. goto errout;
  335. }
  336. if (tb[LWT_BPF_OUT]) {
  337. newts->flags |= LWTUNNEL_STATE_OUTPUT_REDIRECT;
  338. ret = bpf_parse_prog(tb[LWT_BPF_OUT], &bpf->out,
  339. BPF_PROG_TYPE_LWT_OUT);
  340. if (ret < 0)
  341. goto errout;
  342. }
  343. if (tb[LWT_BPF_XMIT]) {
  344. newts->flags |= LWTUNNEL_STATE_XMIT_REDIRECT;
  345. ret = bpf_parse_prog(tb[LWT_BPF_XMIT], &bpf->xmit,
  346. BPF_PROG_TYPE_LWT_XMIT);
  347. if (ret < 0)
  348. goto errout;
  349. }
  350. if (tb[LWT_BPF_XMIT_HEADROOM]) {
  351. u32 headroom = nla_get_u32(tb[LWT_BPF_XMIT_HEADROOM]);
  352. if (headroom > LWT_BPF_MAX_HEADROOM) {
  353. ret = -ERANGE;
  354. goto errout;
  355. }
  356. newts->headroom = headroom;
  357. }
  358. bpf->family = family;
  359. *ts = newts;
  360. return 0;
  361. errout:
  362. bpf_destroy_state(newts);
  363. kfree(newts);
  364. return ret;
  365. }
  366. static int bpf_fill_lwt_prog(struct sk_buff *skb, int attr,
  367. struct bpf_lwt_prog *prog)
  368. {
  369. struct nlattr *nest;
  370. if (!prog->prog)
  371. return 0;
  372. nest = nla_nest_start_noflag(skb, attr);
  373. if (!nest)
  374. return -EMSGSIZE;
  375. if (prog->name &&
  376. nla_put_string(skb, LWT_BPF_PROG_NAME, prog->name))
  377. return -EMSGSIZE;
  378. return nla_nest_end(skb, nest);
  379. }
  380. static int bpf_fill_encap_info(struct sk_buff *skb, struct lwtunnel_state *lwt)
  381. {
  382. struct bpf_lwt *bpf = bpf_lwt_lwtunnel(lwt);
  383. if (bpf_fill_lwt_prog(skb, LWT_BPF_IN, &bpf->in) < 0 ||
  384. bpf_fill_lwt_prog(skb, LWT_BPF_OUT, &bpf->out) < 0 ||
  385. bpf_fill_lwt_prog(skb, LWT_BPF_XMIT, &bpf->xmit) < 0)
  386. return -EMSGSIZE;
  387. return 0;
  388. }
  389. static int bpf_encap_nlsize(struct lwtunnel_state *lwtstate)
  390. {
  391. int nest_len = nla_total_size(sizeof(struct nlattr)) +
  392. nla_total_size(MAX_PROG_NAME) + /* LWT_BPF_PROG_NAME */
  393. 0;
  394. return nest_len + /* LWT_BPF_IN */
  395. nest_len + /* LWT_BPF_OUT */
  396. nest_len + /* LWT_BPF_XMIT */
  397. 0;
  398. }
  399. static int bpf_lwt_prog_cmp(struct bpf_lwt_prog *a, struct bpf_lwt_prog *b)
  400. {
  401. /* FIXME:
  402. * The LWT state is currently rebuilt for delete requests which
  403. * results in a new bpf_prog instance. Comparing names for now.
  404. */
  405. if (!a->name && !b->name)
  406. return 0;
  407. if (!a->name || !b->name)
  408. return 1;
  409. return strcmp(a->name, b->name);
  410. }
  411. static int bpf_encap_cmp(struct lwtunnel_state *a, struct lwtunnel_state *b)
  412. {
  413. struct bpf_lwt *a_bpf = bpf_lwt_lwtunnel(a);
  414. struct bpf_lwt *b_bpf = bpf_lwt_lwtunnel(b);
  415. return bpf_lwt_prog_cmp(&a_bpf->in, &b_bpf->in) ||
  416. bpf_lwt_prog_cmp(&a_bpf->out, &b_bpf->out) ||
  417. bpf_lwt_prog_cmp(&a_bpf->xmit, &b_bpf->xmit);
  418. }
  419. static const struct lwtunnel_encap_ops bpf_encap_ops = {
  420. .build_state = bpf_build_state,
  421. .destroy_state = bpf_destroy_state,
  422. .input = bpf_input,
  423. .output = bpf_output,
  424. .xmit = bpf_xmit,
  425. .fill_encap = bpf_fill_encap_info,
  426. .get_encap_size = bpf_encap_nlsize,
  427. .cmp_encap = bpf_encap_cmp,
  428. .owner = THIS_MODULE,
  429. };
  430. static int handle_gso_type(struct sk_buff *skb, unsigned int gso_type,
  431. int encap_len)
  432. {
  433. struct skb_shared_info *shinfo = skb_shinfo(skb);
  434. gso_type |= SKB_GSO_DODGY;
  435. shinfo->gso_type |= gso_type;
  436. skb_decrease_gso_size(shinfo, encap_len);
  437. shinfo->gso_segs = 0;
  438. return 0;
  439. }
  440. static int handle_gso_encap(struct sk_buff *skb, bool ipv4, int encap_len)
  441. {
  442. int next_hdr_offset;
  443. void *next_hdr;
  444. __u8 protocol;
  445. /* SCTP and UDP_L4 gso need more nuanced handling than what
  446. * handle_gso_type() does above: skb_decrease_gso_size() is not enough.
  447. * So at the moment only TCP GSO packets are let through.
  448. */
  449. if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
  450. return -ENOTSUPP;
  451. if (ipv4) {
  452. protocol = ip_hdr(skb)->protocol;
  453. next_hdr_offset = sizeof(struct iphdr);
  454. next_hdr = skb_network_header(skb) + next_hdr_offset;
  455. } else {
  456. protocol = ipv6_hdr(skb)->nexthdr;
  457. next_hdr_offset = sizeof(struct ipv6hdr);
  458. next_hdr = skb_network_header(skb) + next_hdr_offset;
  459. }
  460. switch (protocol) {
  461. case IPPROTO_GRE:
  462. next_hdr_offset += sizeof(struct gre_base_hdr);
  463. if (next_hdr_offset > encap_len)
  464. return -EINVAL;
  465. if (((struct gre_base_hdr *)next_hdr)->flags & GRE_CSUM)
  466. return handle_gso_type(skb, SKB_GSO_GRE_CSUM,
  467. encap_len);
  468. return handle_gso_type(skb, SKB_GSO_GRE, encap_len);
  469. case IPPROTO_UDP:
  470. next_hdr_offset += sizeof(struct udphdr);
  471. if (next_hdr_offset > encap_len)
  472. return -EINVAL;
  473. if (((struct udphdr *)next_hdr)->check)
  474. return handle_gso_type(skb, SKB_GSO_UDP_TUNNEL_CSUM,
  475. encap_len);
  476. return handle_gso_type(skb, SKB_GSO_UDP_TUNNEL, encap_len);
  477. case IPPROTO_IP:
  478. case IPPROTO_IPV6:
  479. if (ipv4)
  480. return handle_gso_type(skb, SKB_GSO_IPXIP4, encap_len);
  481. else
  482. return handle_gso_type(skb, SKB_GSO_IPXIP6, encap_len);
  483. default:
  484. return -EPROTONOSUPPORT;
  485. }
  486. }
  487. int bpf_lwt_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len, bool ingress)
  488. {
  489. struct iphdr *iph;
  490. bool ipv4;
  491. int err;
  492. if (unlikely(len < sizeof(struct iphdr) || len > LWT_BPF_MAX_HEADROOM))
  493. return -EINVAL;
  494. /* validate protocol and length */
  495. iph = (struct iphdr *)hdr;
  496. if (iph->version == 4) {
  497. ipv4 = true;
  498. if (unlikely(len < iph->ihl * 4))
  499. return -EINVAL;
  500. } else if (iph->version == 6) {
  501. ipv4 = false;
  502. if (unlikely(len < sizeof(struct ipv6hdr)))
  503. return -EINVAL;
  504. } else {
  505. return -EINVAL;
  506. }
  507. if (ingress)
  508. err = skb_cow_head(skb, len + skb->mac_len);
  509. else
  510. err = skb_cow_head(skb,
  511. len + LL_RESERVED_SPACE(skb_dst(skb)->dev));
  512. if (unlikely(err))
  513. return err;
  514. /* push the encap headers and fix pointers */
  515. skb_reset_inner_headers(skb);
  516. skb_reset_inner_mac_header(skb); /* mac header is not yet set */
  517. skb_set_inner_protocol(skb, skb->protocol);
  518. skb->encapsulation = 1;
  519. skb_push(skb, len);
  520. if (ingress)
  521. skb_postpush_rcsum(skb, iph, len);
  522. skb_reset_network_header(skb);
  523. memcpy(skb_network_header(skb), hdr, len);
  524. bpf_compute_data_pointers(skb);
  525. skb_clear_hash(skb);
  526. if (ipv4) {
  527. skb->protocol = htons(ETH_P_IP);
  528. iph = ip_hdr(skb);
  529. if (!iph->check)
  530. iph->check = ip_fast_csum((unsigned char *)iph,
  531. iph->ihl);
  532. } else {
  533. skb->protocol = htons(ETH_P_IPV6);
  534. }
  535. if (skb_is_gso(skb))
  536. return handle_gso_encap(skb, ipv4, len);
  537. return 0;
  538. }
  539. static int __init bpf_lwt_init(void)
  540. {
  541. return lwtunnel_encap_add_ops(&bpf_encap_ops, LWTUNNEL_ENCAP_BPF);
  542. }
  543. subsys_initcall(bpf_lwt_init)