fragmentation.c 16 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554
  1. // SPDX-License-Identifier: GPL-2.0
  2. /* Copyright (C) 2013-2020 B.A.T.M.A.N. contributors:
  3. *
  4. * Martin Hundebøll <martin@hundeboll.net>
  5. */
  6. #include "fragmentation.h"
  7. #include "main.h"
  8. #include <linux/atomic.h>
  9. #include <linux/byteorder/generic.h>
  10. #include <linux/errno.h>
  11. #include <linux/etherdevice.h>
  12. #include <linux/gfp.h>
  13. #include <linux/if_ether.h>
  14. #include <linux/jiffies.h>
  15. #include <linux/kernel.h>
  16. #include <linux/lockdep.h>
  17. #include <linux/netdevice.h>
  18. #include <linux/skbuff.h>
  19. #include <linux/slab.h>
  20. #include <linux/spinlock.h>
  21. #include <linux/string.h>
  22. #include <uapi/linux/batadv_packet.h>
  23. #include "hard-interface.h"
  24. #include "originator.h"
  25. #include "routing.h"
  26. #include "send.h"
  27. #include "soft-interface.h"
  28. /**
  29. * batadv_frag_clear_chain() - delete entries in the fragment buffer chain
  30. * @head: head of chain with entries.
  31. * @dropped: whether the chain is cleared because all fragments are dropped
  32. *
  33. * Free fragments in the passed hlist. Should be called with appropriate lock.
  34. */
  35. static void batadv_frag_clear_chain(struct hlist_head *head, bool dropped)
  36. {
  37. struct batadv_frag_list_entry *entry;
  38. struct hlist_node *node;
  39. hlist_for_each_entry_safe(entry, node, head, list) {
  40. hlist_del(&entry->list);
  41. if (dropped)
  42. kfree_skb(entry->skb);
  43. else
  44. consume_skb(entry->skb);
  45. kfree(entry);
  46. }
  47. }
  48. /**
  49. * batadv_frag_purge_orig() - free fragments associated to an orig
  50. * @orig_node: originator to free fragments from
  51. * @check_cb: optional function to tell if an entry should be purged
  52. */
  53. void batadv_frag_purge_orig(struct batadv_orig_node *orig_node,
  54. bool (*check_cb)(struct batadv_frag_table_entry *))
  55. {
  56. struct batadv_frag_table_entry *chain;
  57. u8 i;
  58. for (i = 0; i < BATADV_FRAG_BUFFER_COUNT; i++) {
  59. chain = &orig_node->fragments[i];
  60. spin_lock_bh(&chain->lock);
  61. if (!check_cb || check_cb(chain)) {
  62. batadv_frag_clear_chain(&chain->fragment_list, true);
  63. chain->size = 0;
  64. }
  65. spin_unlock_bh(&chain->lock);
  66. }
  67. }
  68. /**
  69. * batadv_frag_size_limit() - maximum possible size of packet to be fragmented
  70. *
  71. * Return: the maximum size of payload that can be fragmented.
  72. */
  73. static int batadv_frag_size_limit(void)
  74. {
  75. int limit = BATADV_FRAG_MAX_FRAG_SIZE;
  76. limit -= sizeof(struct batadv_frag_packet);
  77. limit *= BATADV_FRAG_MAX_FRAGMENTS;
  78. return limit;
  79. }
  80. /**
  81. * batadv_frag_init_chain() - check and prepare fragment chain for new fragment
  82. * @chain: chain in fragments table to init
  83. * @seqno: sequence number of the received fragment
  84. *
  85. * Make chain ready for a fragment with sequence number "seqno". Delete existing
  86. * entries if they have an "old" sequence number.
  87. *
  88. * Caller must hold chain->lock.
  89. *
  90. * Return: true if chain is empty and the caller can just insert the new
  91. * fragment without searching for the right position.
  92. */
  93. static bool batadv_frag_init_chain(struct batadv_frag_table_entry *chain,
  94. u16 seqno)
  95. {
  96. lockdep_assert_held(&chain->lock);
  97. if (chain->seqno == seqno)
  98. return false;
  99. if (!hlist_empty(&chain->fragment_list))
  100. batadv_frag_clear_chain(&chain->fragment_list, true);
  101. chain->size = 0;
  102. chain->seqno = seqno;
  103. return true;
  104. }
  105. /**
  106. * batadv_frag_insert_packet() - insert a fragment into a fragment chain
  107. * @orig_node: originator that the fragment was received from
  108. * @skb: skb to insert
  109. * @chain_out: list head to attach complete chains of fragments to
  110. *
  111. * Insert a new fragment into the reverse ordered chain in the right table
  112. * entry. The hash table entry is cleared if "old" fragments exist in it.
  113. *
  114. * Return: true if skb is buffered, false on error. If the chain has all the
  115. * fragments needed to merge the packet, the chain is moved to the passed head
  116. * to avoid locking the chain in the table.
  117. */
  118. static bool batadv_frag_insert_packet(struct batadv_orig_node *orig_node,
  119. struct sk_buff *skb,
  120. struct hlist_head *chain_out)
  121. {
  122. struct batadv_frag_table_entry *chain;
  123. struct batadv_frag_list_entry *frag_entry_new = NULL, *frag_entry_curr;
  124. struct batadv_frag_list_entry *frag_entry_last = NULL;
  125. struct batadv_frag_packet *frag_packet;
  126. u8 bucket;
  127. u16 seqno, hdr_size = sizeof(struct batadv_frag_packet);
  128. bool ret = false;
  129. /* Linearize packet to avoid linearizing 16 packets in a row when doing
  130. * the later merge. Non-linear merge should be added to remove this
  131. * linearization.
  132. */
  133. if (skb_linearize(skb) < 0)
  134. goto err;
  135. frag_packet = (struct batadv_frag_packet *)skb->data;
  136. seqno = ntohs(frag_packet->seqno);
  137. bucket = seqno % BATADV_FRAG_BUFFER_COUNT;
  138. frag_entry_new = kmalloc(sizeof(*frag_entry_new), GFP_ATOMIC);
  139. if (!frag_entry_new)
  140. goto err;
  141. frag_entry_new->skb = skb;
  142. frag_entry_new->no = frag_packet->no;
  143. /* Select entry in the "chain table" and delete any prior fragments
  144. * with another sequence number. batadv_frag_init_chain() returns true,
  145. * if the list is empty at return.
  146. */
  147. chain = &orig_node->fragments[bucket];
  148. spin_lock_bh(&chain->lock);
  149. if (batadv_frag_init_chain(chain, seqno)) {
  150. hlist_add_head(&frag_entry_new->list, &chain->fragment_list);
  151. chain->size = skb->len - hdr_size;
  152. chain->timestamp = jiffies;
  153. chain->total_size = ntohs(frag_packet->total_size);
  154. ret = true;
  155. goto out;
  156. }
  157. /* Find the position for the new fragment. */
  158. hlist_for_each_entry(frag_entry_curr, &chain->fragment_list, list) {
  159. /* Drop packet if fragment already exists. */
  160. if (frag_entry_curr->no == frag_entry_new->no)
  161. goto err_unlock;
  162. /* Order fragments from highest to lowest. */
  163. if (frag_entry_curr->no < frag_entry_new->no) {
  164. hlist_add_before(&frag_entry_new->list,
  165. &frag_entry_curr->list);
  166. chain->size += skb->len - hdr_size;
  167. chain->timestamp = jiffies;
  168. ret = true;
  169. goto out;
  170. }
  171. /* store current entry because it could be the last in list */
  172. frag_entry_last = frag_entry_curr;
  173. }
  174. /* Reached the end of the list, so insert after 'frag_entry_last'. */
  175. if (likely(frag_entry_last)) {
  176. hlist_add_behind(&frag_entry_new->list, &frag_entry_last->list);
  177. chain->size += skb->len - hdr_size;
  178. chain->timestamp = jiffies;
  179. ret = true;
  180. }
  181. out:
  182. if (chain->size > batadv_frag_size_limit() ||
  183. chain->total_size != ntohs(frag_packet->total_size) ||
  184. chain->total_size > batadv_frag_size_limit()) {
  185. /* Clear chain if total size of either the list or the packet
  186. * exceeds the maximum size of one merged packet. Don't allow
  187. * packets to have different total_size.
  188. */
  189. batadv_frag_clear_chain(&chain->fragment_list, true);
  190. chain->size = 0;
  191. } else if (ntohs(frag_packet->total_size) == chain->size) {
  192. /* All fragments received. Hand over chain to caller. */
  193. hlist_move_list(&chain->fragment_list, chain_out);
  194. chain->size = 0;
  195. }
  196. err_unlock:
  197. spin_unlock_bh(&chain->lock);
  198. err:
  199. if (!ret) {
  200. kfree(frag_entry_new);
  201. kfree_skb(skb);
  202. }
  203. return ret;
  204. }
  205. /**
  206. * batadv_frag_merge_packets() - merge a chain of fragments
  207. * @chain: head of chain with fragments
  208. *
  209. * Expand the first skb in the chain and copy the content of the remaining
  210. * skb's into the expanded one. After doing so, clear the chain.
  211. *
  212. * Return: the merged skb or NULL on error.
  213. */
  214. static struct sk_buff *
  215. batadv_frag_merge_packets(struct hlist_head *chain)
  216. {
  217. struct batadv_frag_packet *packet;
  218. struct batadv_frag_list_entry *entry;
  219. struct sk_buff *skb_out;
  220. int size, hdr_size = sizeof(struct batadv_frag_packet);
  221. bool dropped = false;
  222. /* Remove first entry, as this is the destination for the rest of the
  223. * fragments.
  224. */
  225. entry = hlist_entry(chain->first, struct batadv_frag_list_entry, list);
  226. hlist_del(&entry->list);
  227. skb_out = entry->skb;
  228. kfree(entry);
  229. packet = (struct batadv_frag_packet *)skb_out->data;
  230. size = ntohs(packet->total_size) + hdr_size;
  231. /* Make room for the rest of the fragments. */
  232. if (pskb_expand_head(skb_out, 0, size - skb_out->len, GFP_ATOMIC) < 0) {
  233. kfree_skb(skb_out);
  234. skb_out = NULL;
  235. dropped = true;
  236. goto free;
  237. }
  238. /* Move the existing MAC header to just before the payload. (Override
  239. * the fragment header.)
  240. */
  241. skb_pull(skb_out, hdr_size);
  242. skb_out->ip_summed = CHECKSUM_NONE;
  243. memmove(skb_out->data - ETH_HLEN, skb_mac_header(skb_out), ETH_HLEN);
  244. skb_set_mac_header(skb_out, -ETH_HLEN);
  245. skb_reset_network_header(skb_out);
  246. skb_reset_transport_header(skb_out);
  247. /* Copy the payload of the each fragment into the last skb */
  248. hlist_for_each_entry(entry, chain, list) {
  249. size = entry->skb->len - hdr_size;
  250. skb_put_data(skb_out, entry->skb->data + hdr_size, size);
  251. }
  252. free:
  253. /* Locking is not needed, because 'chain' is not part of any orig. */
  254. batadv_frag_clear_chain(chain, dropped);
  255. return skb_out;
  256. }
  257. /**
  258. * batadv_frag_skb_buffer() - buffer fragment for later merge
  259. * @skb: skb to buffer
  260. * @orig_node_src: originator that the skb is received from
  261. *
  262. * Add fragment to buffer and merge fragments if possible.
  263. *
  264. * There are three possible outcomes: 1) Packet is merged: Return true and
  265. * set *skb to merged packet; 2) Packet is buffered: Return true and set *skb
  266. * to NULL; 3) Error: Return false and free skb.
  267. *
  268. * Return: true when the packet is merged or buffered, false when skb is not
  269. * used.
  270. */
  271. bool batadv_frag_skb_buffer(struct sk_buff **skb,
  272. struct batadv_orig_node *orig_node_src)
  273. {
  274. struct sk_buff *skb_out = NULL;
  275. struct hlist_head head = HLIST_HEAD_INIT;
  276. bool ret = false;
  277. /* Add packet to buffer and table entry if merge is possible. */
  278. if (!batadv_frag_insert_packet(orig_node_src, *skb, &head))
  279. goto out_err;
  280. /* Leave if more fragments are needed to merge. */
  281. if (hlist_empty(&head))
  282. goto out;
  283. skb_out = batadv_frag_merge_packets(&head);
  284. if (!skb_out)
  285. goto out_err;
  286. out:
  287. ret = true;
  288. out_err:
  289. *skb = skb_out;
  290. return ret;
  291. }
  292. /**
  293. * batadv_frag_skb_fwd() - forward fragments that would exceed MTU when merged
  294. * @skb: skb to forward
  295. * @recv_if: interface that the skb is received on
  296. * @orig_node_src: originator that the skb is received from
  297. *
  298. * Look up the next-hop of the fragments payload and check if the merged packet
  299. * will exceed the MTU towards the next-hop. If so, the fragment is forwarded
  300. * without merging it.
  301. *
  302. * Return: true if the fragment is consumed/forwarded, false otherwise.
  303. */
  304. bool batadv_frag_skb_fwd(struct sk_buff *skb,
  305. struct batadv_hard_iface *recv_if,
  306. struct batadv_orig_node *orig_node_src)
  307. {
  308. struct batadv_priv *bat_priv = netdev_priv(recv_if->soft_iface);
  309. struct batadv_orig_node *orig_node_dst;
  310. struct batadv_neigh_node *neigh_node = NULL;
  311. struct batadv_frag_packet *packet;
  312. u16 total_size;
  313. bool ret = false;
  314. packet = (struct batadv_frag_packet *)skb->data;
  315. orig_node_dst = batadv_orig_hash_find(bat_priv, packet->dest);
  316. if (!orig_node_dst)
  317. goto out;
  318. neigh_node = batadv_find_router(bat_priv, orig_node_dst, recv_if);
  319. if (!neigh_node)
  320. goto out;
  321. /* Forward the fragment, if the merged packet would be too big to
  322. * be assembled.
  323. */
  324. total_size = ntohs(packet->total_size);
  325. if (total_size > neigh_node->if_incoming->net_dev->mtu) {
  326. batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_FWD);
  327. batadv_add_counter(bat_priv, BATADV_CNT_FRAG_FWD_BYTES,
  328. skb->len + ETH_HLEN);
  329. packet->ttl--;
  330. batadv_send_unicast_skb(skb, neigh_node);
  331. ret = true;
  332. }
  333. out:
  334. if (orig_node_dst)
  335. batadv_orig_node_put(orig_node_dst);
  336. if (neigh_node)
  337. batadv_neigh_node_put(neigh_node);
  338. return ret;
  339. }
  340. /**
  341. * batadv_frag_create() - create a fragment from skb
  342. * @net_dev: outgoing device for fragment
  343. * @skb: skb to create fragment from
  344. * @frag_head: header to use in new fragment
  345. * @fragment_size: size of new fragment
  346. *
  347. * Split the passed skb into two fragments: A new one with size matching the
  348. * passed mtu and the old one with the rest. The new skb contains data from the
  349. * tail of the old skb.
  350. *
  351. * Return: the new fragment, NULL on error.
  352. */
  353. static struct sk_buff *batadv_frag_create(struct net_device *net_dev,
  354. struct sk_buff *skb,
  355. struct batadv_frag_packet *frag_head,
  356. unsigned int fragment_size)
  357. {
  358. unsigned int ll_reserved = LL_RESERVED_SPACE(net_dev);
  359. unsigned int tailroom = net_dev->needed_tailroom;
  360. struct sk_buff *skb_fragment;
  361. unsigned int header_size = sizeof(*frag_head);
  362. unsigned int mtu = fragment_size + header_size;
  363. skb_fragment = dev_alloc_skb(ll_reserved + mtu + tailroom);
  364. if (!skb_fragment)
  365. goto err;
  366. skb_fragment->priority = skb->priority;
  367. /* Eat the last mtu-bytes of the skb */
  368. skb_reserve(skb_fragment, ll_reserved + header_size);
  369. skb_split(skb, skb_fragment, skb->len - fragment_size);
  370. /* Add the header */
  371. skb_push(skb_fragment, header_size);
  372. memcpy(skb_fragment->data, frag_head, header_size);
  373. err:
  374. return skb_fragment;
  375. }
  376. /**
  377. * batadv_frag_send_packet() - create up to 16 fragments from the passed skb
  378. * @skb: skb to create fragments from
  379. * @orig_node: final destination of the created fragments
  380. * @neigh_node: next-hop of the created fragments
  381. *
  382. * Return: the netdev tx status or a negative errno code on a failure
  383. */
  384. int batadv_frag_send_packet(struct sk_buff *skb,
  385. struct batadv_orig_node *orig_node,
  386. struct batadv_neigh_node *neigh_node)
  387. {
  388. struct net_device *net_dev = neigh_node->if_incoming->net_dev;
  389. struct batadv_priv *bat_priv;
  390. struct batadv_hard_iface *primary_if = NULL;
  391. struct batadv_frag_packet frag_header;
  392. struct sk_buff *skb_fragment;
  393. unsigned int mtu = net_dev->mtu;
  394. unsigned int header_size = sizeof(frag_header);
  395. unsigned int max_fragment_size, num_fragments;
  396. int ret;
  397. /* To avoid merge and refragmentation at next-hops we never send
  398. * fragments larger than BATADV_FRAG_MAX_FRAG_SIZE
  399. */
  400. mtu = min_t(unsigned int, mtu, BATADV_FRAG_MAX_FRAG_SIZE);
  401. max_fragment_size = mtu - header_size;
  402. if (skb->len == 0 || max_fragment_size == 0)
  403. return -EINVAL;
  404. num_fragments = (skb->len - 1) / max_fragment_size + 1;
  405. max_fragment_size = (skb->len - 1) / num_fragments + 1;
  406. /* Don't even try to fragment, if we need more than 16 fragments */
  407. if (num_fragments > BATADV_FRAG_MAX_FRAGMENTS) {
  408. ret = -EAGAIN;
  409. goto free_skb;
  410. }
  411. bat_priv = orig_node->bat_priv;
  412. primary_if = batadv_primary_if_get_selected(bat_priv);
  413. if (!primary_if) {
  414. ret = -EINVAL;
  415. goto free_skb;
  416. }
  417. /* Create one header to be copied to all fragments */
  418. frag_header.packet_type = BATADV_UNICAST_FRAG;
  419. frag_header.version = BATADV_COMPAT_VERSION;
  420. frag_header.ttl = BATADV_TTL;
  421. frag_header.seqno = htons(atomic_inc_return(&bat_priv->frag_seqno));
  422. frag_header.reserved = 0;
  423. frag_header.no = 0;
  424. frag_header.total_size = htons(skb->len);
  425. /* skb->priority values from 256->263 are magic values to
  426. * directly indicate a specific 802.1d priority. This is used
  427. * to allow 802.1d priority to be passed directly in from VLAN
  428. * tags, etc.
  429. */
  430. if (skb->priority >= 256 && skb->priority <= 263)
  431. frag_header.priority = skb->priority - 256;
  432. else
  433. frag_header.priority = 0;
  434. ether_addr_copy(frag_header.orig, primary_if->net_dev->dev_addr);
  435. ether_addr_copy(frag_header.dest, orig_node->orig);
  436. /* Eat and send fragments from the tail of skb */
  437. while (skb->len > max_fragment_size) {
  438. /* The initial check in this function should cover this case */
  439. if (unlikely(frag_header.no == BATADV_FRAG_MAX_FRAGMENTS - 1)) {
  440. ret = -EINVAL;
  441. goto put_primary_if;
  442. }
  443. skb_fragment = batadv_frag_create(net_dev, skb, &frag_header,
  444. max_fragment_size);
  445. if (!skb_fragment) {
  446. ret = -ENOMEM;
  447. goto put_primary_if;
  448. }
  449. batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_TX);
  450. batadv_add_counter(bat_priv, BATADV_CNT_FRAG_TX_BYTES,
  451. skb_fragment->len + ETH_HLEN);
  452. ret = batadv_send_unicast_skb(skb_fragment, neigh_node);
  453. if (ret != NET_XMIT_SUCCESS) {
  454. ret = NET_XMIT_DROP;
  455. goto put_primary_if;
  456. }
  457. frag_header.no++;
  458. }
  459. /* make sure that there is at least enough head for the fragmentation
  460. * and ethernet headers
  461. */
  462. ret = skb_cow_head(skb, ETH_HLEN + header_size);
  463. if (ret < 0)
  464. goto put_primary_if;
  465. skb_push(skb, header_size);
  466. memcpy(skb->data, &frag_header, header_size);
  467. /* Send the last fragment */
  468. batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_TX);
  469. batadv_add_counter(bat_priv, BATADV_CNT_FRAG_TX_BYTES,
  470. skb->len + ETH_HLEN);
  471. ret = batadv_send_unicast_skb(skb, neigh_node);
  472. /* skb was consumed */
  473. skb = NULL;
  474. put_primary_if:
  475. batadv_hardif_put(primary_if);
  476. free_skb:
  477. kfree_skb(skb);
  478. return ret;
  479. }