reassembly.c 19 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773
  1. /*
  2. * IPv6 fragment reassembly
  3. * Linux INET6 implementation
  4. *
  5. * Authors:
  6. * Pedro Roque <roque@di.fc.ul.pt>
  7. *
  8. * $Id: reassembly.c,v 1.26 2001/03/07 22:00:57 davem Exp $
  9. *
  10. * Based on: net/ipv4/ip_fragment.c
  11. *
  12. * This program is free software; you can redistribute it and/or
  13. * modify it under the terms of the GNU General Public License
  14. * as published by the Free Software Foundation; either version
  15. * 2 of the License, or (at your option) any later version.
  16. */
  17. /*
  18. * Fixes:
  19. * Andi Kleen Make it work with multiple hosts.
  20. * More RFC compliance.
  21. *
  22. * Horst von Brand Add missing #include <linux/string.h>
  23. * Alexey Kuznetsov SMP races, threading, cleanup.
  24. * Patrick McHardy LRU queue of frag heads for evictor.
  25. * Mitsuru KANDA @USAGI Register inet6_protocol{}.
  26. * David Stevens and
  27. * YOSHIFUJI,H. @USAGI Always remove fragment header to
  28. * calculate ICV correctly.
  29. */
  30. #include <linux/errno.h>
  31. #include <linux/types.h>
  32. #include <linux/string.h>
  33. #include <linux/socket.h>
  34. #include <linux/sockios.h>
  35. #include <linux/jiffies.h>
  36. #include <linux/net.h>
  37. #include <linux/list.h>
  38. #include <linux/netdevice.h>
  39. #include <linux/in6.h>
  40. #include <linux/ipv6.h>
  41. #include <linux/icmpv6.h>
  42. #include <linux/random.h>
  43. #include <linux/jhash.h>
  44. #include <linux/skbuff.h>
  45. #include <net/sock.h>
  46. #include <net/snmp.h>
  47. #include <net/ipv6.h>
  48. #include <net/ip6_route.h>
  49. #include <net/protocol.h>
  50. #include <net/transp_v6.h>
  51. #include <net/rawv6.h>
  52. #include <net/ndisc.h>
  53. #include <net/addrconf.h>
  54. #include <net/inet_frag.h>
  55. struct ip6frag_skb_cb
  56. {
  57. struct inet6_skb_parm h;
  58. int offset;
  59. };
  60. #define FRAG6_CB(skb) ((struct ip6frag_skb_cb*)((skb)->cb))
  61. /*
  62. * Equivalent of ipv4 struct ipq
  63. */
  64. struct frag_queue
  65. {
  66. struct inet_frag_queue q;
  67. __be32 id; /* fragment id */
  68. struct in6_addr saddr;
  69. struct in6_addr daddr;
  70. int iif;
  71. unsigned int csum;
  72. __u16 nhoffset;
  73. };
  74. static struct inet_frags ip6_frags;
  75. int ip6_frag_nqueues(struct net *net)
  76. {
  77. return net->ipv6.frags.nqueues;
  78. }
  79. int ip6_frag_mem(struct net *net)
  80. {
  81. return atomic_read(&net->ipv6.frags.mem);
  82. }
  83. static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
  84. struct net_device *dev);
  85. /*
  86. * callers should be careful not to use the hash value outside the ipfrag_lock
  87. * as doing so could race with ipfrag_hash_rnd being recalculated.
  88. */
  89. static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
  90. struct in6_addr *daddr)
  91. {
  92. u32 a, b, c;
  93. a = (__force u32)saddr->s6_addr32[0];
  94. b = (__force u32)saddr->s6_addr32[1];
  95. c = (__force u32)saddr->s6_addr32[2];
  96. a += JHASH_GOLDEN_RATIO;
  97. b += JHASH_GOLDEN_RATIO;
  98. c += ip6_frags.rnd;
  99. __jhash_mix(a, b, c);
  100. a += (__force u32)saddr->s6_addr32[3];
  101. b += (__force u32)daddr->s6_addr32[0];
  102. c += (__force u32)daddr->s6_addr32[1];
  103. __jhash_mix(a, b, c);
  104. a += (__force u32)daddr->s6_addr32[2];
  105. b += (__force u32)daddr->s6_addr32[3];
  106. c += (__force u32)id;
  107. __jhash_mix(a, b, c);
  108. return c & (INETFRAGS_HASHSZ - 1);
  109. }
  110. static unsigned int ip6_hashfn(struct inet_frag_queue *q)
  111. {
  112. struct frag_queue *fq;
  113. fq = container_of(q, struct frag_queue, q);
  114. return ip6qhashfn(fq->id, &fq->saddr, &fq->daddr);
  115. }
  116. int ip6_frag_match(struct inet_frag_queue *q, void *a)
  117. {
  118. struct frag_queue *fq;
  119. struct ip6_create_arg *arg = a;
  120. fq = container_of(q, struct frag_queue, q);
  121. return (fq->id == arg->id &&
  122. ipv6_addr_equal(&fq->saddr, arg->src) &&
  123. ipv6_addr_equal(&fq->daddr, arg->dst));
  124. }
  125. EXPORT_SYMBOL(ip6_frag_match);
  126. /* Memory Tracking Functions. */
  127. static inline void frag_kfree_skb(struct netns_frags *nf,
  128. struct sk_buff *skb, int *work)
  129. {
  130. if (work)
  131. *work -= skb->truesize;
  132. atomic_sub(skb->truesize, &nf->mem);
  133. kfree_skb(skb);
  134. }
  135. void ip6_frag_init(struct inet_frag_queue *q, void *a)
  136. {
  137. struct frag_queue *fq = container_of(q, struct frag_queue, q);
  138. struct ip6_create_arg *arg = a;
  139. fq->id = arg->id;
  140. ipv6_addr_copy(&fq->saddr, arg->src);
  141. ipv6_addr_copy(&fq->daddr, arg->dst);
  142. }
  143. EXPORT_SYMBOL(ip6_frag_init);
  144. /* Destruction primitives. */
  145. static __inline__ void fq_put(struct frag_queue *fq)
  146. {
  147. inet_frag_put(&fq->q, &ip6_frags);
  148. }
  149. /* Kill fq entry. It is not destroyed immediately,
  150. * because caller (and someone more) holds reference count.
  151. */
  152. static __inline__ void fq_kill(struct frag_queue *fq)
  153. {
  154. inet_frag_kill(&fq->q, &ip6_frags);
  155. }
  156. static void ip6_evictor(struct net *net, struct inet6_dev *idev)
  157. {
  158. int evicted;
  159. evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
  160. if (evicted)
  161. IP6_ADD_STATS_BH(idev, IPSTATS_MIB_REASMFAILS, evicted);
  162. }
  163. static void ip6_frag_expire(unsigned long data)
  164. {
  165. struct frag_queue *fq;
  166. struct net_device *dev = NULL;
  167. struct net *net;
  168. fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
  169. spin_lock(&fq->q.lock);
  170. if (fq->q.last_in & INET_FRAG_COMPLETE)
  171. goto out;
  172. fq_kill(fq);
  173. net = container_of(fq->q.net, struct net, ipv6.frags);
  174. dev = dev_get_by_index(net, fq->iif);
  175. if (!dev)
  176. goto out;
  177. rcu_read_lock();
  178. IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
  179. IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
  180. rcu_read_unlock();
  181. /* Don't send error if the first segment did not arrive. */
  182. if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
  183. goto out;
  184. /*
  185. But use as source device on which LAST ARRIVED
  186. segment was received. And do not use fq->dev
  187. pointer directly, device might already disappeared.
  188. */
  189. fq->q.fragments->dev = dev;
  190. icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
  191. out:
  192. if (dev)
  193. dev_put(dev);
  194. spin_unlock(&fq->q.lock);
  195. fq_put(fq);
  196. }
  197. static __inline__ struct frag_queue *
  198. fq_find(struct net *net, __be32 id, struct in6_addr *src, struct in6_addr *dst,
  199. struct inet6_dev *idev)
  200. {
  201. struct inet_frag_queue *q;
  202. struct ip6_create_arg arg;
  203. unsigned int hash;
  204. arg.id = id;
  205. arg.src = src;
  206. arg.dst = dst;
  207. hash = ip6qhashfn(id, src, dst);
  208. q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
  209. if (q == NULL)
  210. goto oom;
  211. return container_of(q, struct frag_queue, q);
  212. oom:
  213. IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
  214. return NULL;
  215. }
  216. static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
  217. struct frag_hdr *fhdr, int nhoff)
  218. {
  219. struct sk_buff *prev, *next;
  220. struct net_device *dev;
  221. int offset, end;
  222. if (fq->q.last_in & INET_FRAG_COMPLETE)
  223. goto err;
  224. offset = ntohs(fhdr->frag_off) & ~0x7;
  225. end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
  226. ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
  227. if ((unsigned int)end > IPV6_MAXPLEN) {
  228. IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
  229. IPSTATS_MIB_INHDRERRORS);
  230. icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
  231. ((u8 *)&fhdr->frag_off -
  232. skb_network_header(skb)));
  233. return -1;
  234. }
  235. if (skb->ip_summed == CHECKSUM_COMPLETE) {
  236. const unsigned char *nh = skb_network_header(skb);
  237. skb->csum = csum_sub(skb->csum,
  238. csum_partial(nh, (u8 *)(fhdr + 1) - nh,
  239. 0));
  240. }
  241. /* Is this the final fragment? */
  242. if (!(fhdr->frag_off & htons(IP6_MF))) {
  243. /* If we already have some bits beyond end
  244. * or have different end, the segment is corrupted.
  245. */
  246. if (end < fq->q.len ||
  247. ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
  248. goto err;
  249. fq->q.last_in |= INET_FRAG_LAST_IN;
  250. fq->q.len = end;
  251. } else {
  252. /* Check if the fragment is rounded to 8 bytes.
  253. * Required by the RFC.
  254. */
  255. if (end & 0x7) {
  256. /* RFC2460 says always send parameter problem in
  257. * this case. -DaveM
  258. */
  259. IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
  260. IPSTATS_MIB_INHDRERRORS);
  261. icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
  262. offsetof(struct ipv6hdr, payload_len));
  263. return -1;
  264. }
  265. if (end > fq->q.len) {
  266. /* Some bits beyond end -> corruption. */
  267. if (fq->q.last_in & INET_FRAG_LAST_IN)
  268. goto err;
  269. fq->q.len = end;
  270. }
  271. }
  272. if (end == offset)
  273. goto err;
  274. /* Point into the IP datagram 'data' part. */
  275. if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
  276. goto err;
  277. if (pskb_trim_rcsum(skb, end - offset))
  278. goto err;
  279. /* Find out which fragments are in front and at the back of us
  280. * in the chain of fragments so far. We must know where to put
  281. * this fragment, right?
  282. */
  283. prev = NULL;
  284. for(next = fq->q.fragments; next != NULL; next = next->next) {
  285. if (FRAG6_CB(next)->offset >= offset)
  286. break; /* bingo! */
  287. prev = next;
  288. }
  289. /* We found where to put this one. Check for overlap with
  290. * preceding fragment, and, if needed, align things so that
  291. * any overlaps are eliminated.
  292. */
  293. if (prev) {
  294. int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
  295. if (i > 0) {
  296. offset += i;
  297. if (end <= offset)
  298. goto err;
  299. if (!pskb_pull(skb, i))
  300. goto err;
  301. if (skb->ip_summed != CHECKSUM_UNNECESSARY)
  302. skb->ip_summed = CHECKSUM_NONE;
  303. }
  304. }
  305. /* Look for overlap with succeeding segments.
  306. * If we can merge fragments, do it.
  307. */
  308. while (next && FRAG6_CB(next)->offset < end) {
  309. int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
  310. if (i < next->len) {
  311. /* Eat head of the next overlapped fragment
  312. * and leave the loop. The next ones cannot overlap.
  313. */
  314. if (!pskb_pull(next, i))
  315. goto err;
  316. FRAG6_CB(next)->offset += i; /* next fragment */
  317. fq->q.meat -= i;
  318. if (next->ip_summed != CHECKSUM_UNNECESSARY)
  319. next->ip_summed = CHECKSUM_NONE;
  320. break;
  321. } else {
  322. struct sk_buff *free_it = next;
  323. /* Old fragment is completely overridden with
  324. * new one drop it.
  325. */
  326. next = next->next;
  327. if (prev)
  328. prev->next = next;
  329. else
  330. fq->q.fragments = next;
  331. fq->q.meat -= free_it->len;
  332. frag_kfree_skb(fq->q.net, free_it, NULL);
  333. }
  334. }
  335. FRAG6_CB(skb)->offset = offset;
  336. /* Insert this fragment in the chain of fragments. */
  337. skb->next = next;
  338. if (prev)
  339. prev->next = skb;
  340. else
  341. fq->q.fragments = skb;
  342. dev = skb->dev;
  343. if (dev) {
  344. fq->iif = dev->ifindex;
  345. skb->dev = NULL;
  346. }
  347. fq->q.stamp = skb->tstamp;
  348. fq->q.meat += skb->len;
  349. atomic_add(skb->truesize, &fq->q.net->mem);
  350. /* The first fragment.
  351. * nhoffset is obtained from the first fragment, of course.
  352. */
  353. if (offset == 0) {
  354. fq->nhoffset = nhoff;
  355. fq->q.last_in |= INET_FRAG_FIRST_IN;
  356. }
  357. if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
  358. fq->q.meat == fq->q.len)
  359. return ip6_frag_reasm(fq, prev, dev);
  360. write_lock(&ip6_frags.lock);
  361. list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
  362. write_unlock(&ip6_frags.lock);
  363. return -1;
  364. err:
  365. IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
  366. kfree_skb(skb);
  367. return -1;
  368. }
  369. /*
  370. * Check if this packet is complete.
  371. * Returns NULL on failure by any reason, and pointer
  372. * to current nexthdr field in reassembled frame.
  373. *
  374. * It is called with locked fq, and caller must check that
  375. * queue is eligible for reassembly i.e. it is not COMPLETE,
  376. * the last and the first frames arrived and all the bits are here.
  377. */
  378. static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
  379. struct net_device *dev)
  380. {
  381. struct sk_buff *fp, *head = fq->q.fragments;
  382. int payload_len;
  383. unsigned int nhoff;
  384. fq_kill(fq);
  385. /* Make the one we just received the head. */
  386. if (prev) {
  387. head = prev->next;
  388. fp = skb_clone(head, GFP_ATOMIC);
  389. if (!fp)
  390. goto out_oom;
  391. fp->next = head->next;
  392. prev->next = fp;
  393. skb_morph(head, fq->q.fragments);
  394. head->next = fq->q.fragments->next;
  395. kfree_skb(fq->q.fragments);
  396. fq->q.fragments = head;
  397. }
  398. BUG_TRAP(head != NULL);
  399. BUG_TRAP(FRAG6_CB(head)->offset == 0);
  400. /* Unfragmented part is taken from the first segment. */
  401. payload_len = ((head->data - skb_network_header(head)) -
  402. sizeof(struct ipv6hdr) + fq->q.len -
  403. sizeof(struct frag_hdr));
  404. if (payload_len > IPV6_MAXPLEN)
  405. goto out_oversize;
  406. /* Head of list must not be cloned. */
  407. if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
  408. goto out_oom;
  409. /* If the first fragment is fragmented itself, we split
  410. * it to two chunks: the first with data and paged part
  411. * and the second, holding only fragments. */
  412. if (skb_shinfo(head)->frag_list) {
  413. struct sk_buff *clone;
  414. int i, plen = 0;
  415. if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
  416. goto out_oom;
  417. clone->next = head->next;
  418. head->next = clone;
  419. skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
  420. skb_shinfo(head)->frag_list = NULL;
  421. for (i=0; i<skb_shinfo(head)->nr_frags; i++)
  422. plen += skb_shinfo(head)->frags[i].size;
  423. clone->len = clone->data_len = head->data_len - plen;
  424. head->data_len -= clone->len;
  425. head->len -= clone->len;
  426. clone->csum = 0;
  427. clone->ip_summed = head->ip_summed;
  428. atomic_add(clone->truesize, &fq->q.net->mem);
  429. }
  430. /* We have to remove fragment header from datagram and to relocate
  431. * header in order to calculate ICV correctly. */
  432. nhoff = fq->nhoffset;
  433. skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
  434. memmove(head->head + sizeof(struct frag_hdr), head->head,
  435. (head->data - head->head) - sizeof(struct frag_hdr));
  436. head->mac_header += sizeof(struct frag_hdr);
  437. head->network_header += sizeof(struct frag_hdr);
  438. skb_shinfo(head)->frag_list = head->next;
  439. skb_reset_transport_header(head);
  440. skb_push(head, head->data - skb_network_header(head));
  441. atomic_sub(head->truesize, &fq->q.net->mem);
  442. for (fp=head->next; fp; fp = fp->next) {
  443. head->data_len += fp->len;
  444. head->len += fp->len;
  445. if (head->ip_summed != fp->ip_summed)
  446. head->ip_summed = CHECKSUM_NONE;
  447. else if (head->ip_summed == CHECKSUM_COMPLETE)
  448. head->csum = csum_add(head->csum, fp->csum);
  449. head->truesize += fp->truesize;
  450. atomic_sub(fp->truesize, &fq->q.net->mem);
  451. }
  452. head->next = NULL;
  453. head->dev = dev;
  454. head->tstamp = fq->q.stamp;
  455. ipv6_hdr(head)->payload_len = htons(payload_len);
  456. IP6CB(head)->nhoff = nhoff;
  457. /* Yes, and fold redundant checksum back. 8) */
  458. if (head->ip_summed == CHECKSUM_COMPLETE)
  459. head->csum = csum_partial(skb_network_header(head),
  460. skb_network_header_len(head),
  461. head->csum);
  462. rcu_read_lock();
  463. IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
  464. rcu_read_unlock();
  465. fq->q.fragments = NULL;
  466. return 1;
  467. out_oversize:
  468. if (net_ratelimit())
  469. printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
  470. goto out_fail;
  471. out_oom:
  472. if (net_ratelimit())
  473. printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
  474. out_fail:
  475. rcu_read_lock();
  476. IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
  477. rcu_read_unlock();
  478. return -1;
  479. }
  480. static int ipv6_frag_rcv(struct sk_buff *skb)
  481. {
  482. struct frag_hdr *fhdr;
  483. struct frag_queue *fq;
  484. struct ipv6hdr *hdr = ipv6_hdr(skb);
  485. struct net *net;
  486. IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
  487. /* Jumbo payload inhibits frag. header */
  488. if (hdr->payload_len==0) {
  489. IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
  490. icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
  491. skb_network_header_len(skb));
  492. return -1;
  493. }
  494. if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
  495. sizeof(struct frag_hdr)))) {
  496. IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
  497. icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
  498. skb_network_header_len(skb));
  499. return -1;
  500. }
  501. hdr = ipv6_hdr(skb);
  502. fhdr = (struct frag_hdr *)skb_transport_header(skb);
  503. if (!(fhdr->frag_off & htons(0xFFF9))) {
  504. /* It is not a fragmented frame */
  505. skb->transport_header += sizeof(struct frag_hdr);
  506. IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
  507. IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
  508. return 1;
  509. }
  510. net = dev_net(skb->dev);
  511. if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
  512. ip6_evictor(net, ip6_dst_idev(skb->dst));
  513. if ((fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
  514. ip6_dst_idev(skb->dst))) != NULL) {
  515. int ret;
  516. spin_lock(&fq->q.lock);
  517. ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
  518. spin_unlock(&fq->q.lock);
  519. fq_put(fq);
  520. return ret;
  521. }
  522. IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
  523. kfree_skb(skb);
  524. return -1;
  525. }
  526. static struct inet6_protocol frag_protocol =
  527. {
  528. .handler = ipv6_frag_rcv,
  529. .flags = INET6_PROTO_NOPOLICY,
  530. };
  531. #ifdef CONFIG_SYSCTL
  532. static struct ctl_table ip6_frags_ctl_table[] = {
  533. {
  534. .ctl_name = NET_IPV6_IP6FRAG_HIGH_THRESH,
  535. .procname = "ip6frag_high_thresh",
  536. .data = &init_net.ipv6.frags.high_thresh,
  537. .maxlen = sizeof(int),
  538. .mode = 0644,
  539. .proc_handler = &proc_dointvec
  540. },
  541. {
  542. .ctl_name = NET_IPV6_IP6FRAG_LOW_THRESH,
  543. .procname = "ip6frag_low_thresh",
  544. .data = &init_net.ipv6.frags.low_thresh,
  545. .maxlen = sizeof(int),
  546. .mode = 0644,
  547. .proc_handler = &proc_dointvec
  548. },
  549. {
  550. .ctl_name = NET_IPV6_IP6FRAG_TIME,
  551. .procname = "ip6frag_time",
  552. .data = &init_net.ipv6.frags.timeout,
  553. .maxlen = sizeof(int),
  554. .mode = 0644,
  555. .proc_handler = &proc_dointvec_jiffies,
  556. .strategy = &sysctl_jiffies,
  557. },
  558. {
  559. .ctl_name = NET_IPV6_IP6FRAG_SECRET_INTERVAL,
  560. .procname = "ip6frag_secret_interval",
  561. .data = &ip6_frags.secret_interval,
  562. .maxlen = sizeof(int),
  563. .mode = 0644,
  564. .proc_handler = &proc_dointvec_jiffies,
  565. .strategy = &sysctl_jiffies
  566. },
  567. { }
  568. };
  569. static int ip6_frags_sysctl_register(struct net *net)
  570. {
  571. struct ctl_table *table;
  572. struct ctl_table_header *hdr;
  573. table = ip6_frags_ctl_table;
  574. if (net != &init_net) {
  575. table = kmemdup(table, sizeof(ip6_frags_ctl_table), GFP_KERNEL);
  576. if (table == NULL)
  577. goto err_alloc;
  578. table[0].data = &net->ipv6.frags.high_thresh;
  579. table[1].data = &net->ipv6.frags.low_thresh;
  580. table[2].data = &net->ipv6.frags.timeout;
  581. table[3].mode &= ~0222;
  582. }
  583. hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
  584. if (hdr == NULL)
  585. goto err_reg;
  586. net->ipv6.sysctl.frags_hdr = hdr;
  587. return 0;
  588. err_reg:
  589. if (net != &init_net)
  590. kfree(table);
  591. err_alloc:
  592. return -ENOMEM;
  593. }
  594. static void ip6_frags_sysctl_unregister(struct net *net)
  595. {
  596. struct ctl_table *table;
  597. table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
  598. unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
  599. kfree(table);
  600. }
  601. #else
  602. static inline int ip6_frags_sysctl_register(struct net *net)
  603. {
  604. return 0;
  605. }
  606. static inline void ip6_frags_sysctl_unregister(struct net *net)
  607. {
  608. }
  609. #endif
  610. static int ipv6_frags_init_net(struct net *net)
  611. {
  612. net->ipv6.frags.high_thresh = 256 * 1024;
  613. net->ipv6.frags.low_thresh = 192 * 1024;
  614. net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
  615. inet_frags_init_net(&net->ipv6.frags);
  616. return ip6_frags_sysctl_register(net);
  617. }
  618. static void ipv6_frags_exit_net(struct net *net)
  619. {
  620. ip6_frags_sysctl_unregister(net);
  621. inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
  622. }
  623. static struct pernet_operations ip6_frags_ops = {
  624. .init = ipv6_frags_init_net,
  625. .exit = ipv6_frags_exit_net,
  626. };
  627. int __init ipv6_frag_init(void)
  628. {
  629. int ret;
  630. ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
  631. if (ret)
  632. goto out;
  633. register_pernet_subsys(&ip6_frags_ops);
  634. ip6_frags.hashfn = ip6_hashfn;
  635. ip6_frags.constructor = ip6_frag_init;
  636. ip6_frags.destructor = NULL;
  637. ip6_frags.skb_free = NULL;
  638. ip6_frags.qsize = sizeof(struct frag_queue);
  639. ip6_frags.match = ip6_frag_match;
  640. ip6_frags.frag_expire = ip6_frag_expire;
  641. ip6_frags.secret_interval = 10 * 60 * HZ;
  642. inet_frags_init(&ip6_frags);
  643. out:
  644. return ret;
  645. }
  646. void ipv6_frag_exit(void)
  647. {
  648. inet_frags_fini(&ip6_frags);
  649. unregister_pernet_subsys(&ip6_frags_ops);
  650. inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
  651. }