ematch.c 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528
  1. /*
  2. * net/sched/ematch.c Extended Match API
  3. *
  4. * This program is free software; you can redistribute it and/or
  5. * modify it under the terms of the GNU General Public License
  6. * as published by the Free Software Foundation; either version
  7. * 2 of the License, or (at your option) any later version.
  8. *
  9. * Authors: Thomas Graf <tgraf@suug.ch>
  10. *
  11. * ==========================================================================
  12. *
  13. * An extended match (ematch) is a small classification tool not worth
  14. * writing a full classifier for. Ematches can be interconnected to form
  15. * a logic expression and get attached to classifiers to extend their
  16. * functionatlity.
  17. *
  18. * The userspace part transforms the logic expressions into an array
  19. * consisting of multiple sequences of interconnected ematches separated
  20. * by markers. Precedence is implemented by a special ematch kind
  21. * referencing a sequence beyond the marker of the current sequence
  22. * causing the current position in the sequence to be pushed onto a stack
  23. * to allow the current position to be overwritten by the position referenced
  24. * in the special ematch. Matching continues in the new sequence until a
  25. * marker is reached causing the position to be restored from the stack.
  26. *
  27. * Example:
  28. * A AND (B1 OR B2) AND C AND D
  29. *
  30. * ------->-PUSH-------
  31. * -->-- / -->-- \ -->--
  32. * / \ / / \ \ / \
  33. * +-------+-------+-------+-------+-------+--------+
  34. * | A AND | B AND | C AND | D END | B1 OR | B2 END |
  35. * +-------+-------+-------+-------+-------+--------+
  36. * \ /
  37. * --------<-POP---------
  38. *
  39. * where B is a virtual ematch referencing to sequence starting with B1.
  40. *
  41. * ==========================================================================
  42. *
  43. * How to write an ematch in 60 seconds
  44. * ------------------------------------
  45. *
  46. * 1) Provide a matcher function:
  47. * static int my_match(struct sk_buff *skb, struct tcf_ematch *m,
  48. * struct tcf_pkt_info *info)
  49. * {
  50. * struct mydata *d = (struct mydata *) m->data;
  51. *
  52. * if (...matching goes here...)
  53. * return 1;
  54. * else
  55. * return 0;
  56. * }
  57. *
  58. * 2) Fill out a struct tcf_ematch_ops:
  59. * static struct tcf_ematch_ops my_ops = {
  60. * .kind = unique id,
  61. * .datalen = sizeof(struct mydata),
  62. * .match = my_match,
  63. * .owner = THIS_MODULE,
  64. * };
  65. *
  66. * 3) Register/Unregister your ematch:
  67. * static int __init init_my_ematch(void)
  68. * {
  69. * return tcf_em_register(&my_ops);
  70. * }
  71. *
  72. * static void __exit exit_my_ematch(void)
  73. * {
  74. * return tcf_em_unregister(&my_ops);
  75. * }
  76. *
  77. * module_init(init_my_ematch);
  78. * module_exit(exit_my_ematch);
  79. *
  80. * 4) By now you should have two more seconds left, barely enough to
  81. * open up a beer to watch the compilation going.
  82. */
  83. #include <linux/config.h>
  84. #include <linux/module.h>
  85. #include <linux/types.h>
  86. #include <linux/kernel.h>
  87. #include <linux/sched.h>
  88. #include <linux/mm.h>
  89. #include <linux/errno.h>
  90. #include <linux/interrupt.h>
  91. #include <linux/rtnetlink.h>
  92. #include <linux/skbuff.h>
  93. #include <net/pkt_cls.h>
  94. static LIST_HEAD(ematch_ops);
  95. static DEFINE_RWLOCK(ematch_mod_lock);
  96. static inline struct tcf_ematch_ops * tcf_em_lookup(u16 kind)
  97. {
  98. struct tcf_ematch_ops *e = NULL;
  99. read_lock(&ematch_mod_lock);
  100. list_for_each_entry(e, &ematch_ops, link) {
  101. if (kind == e->kind) {
  102. if (!try_module_get(e->owner))
  103. e = NULL;
  104. read_unlock(&ematch_mod_lock);
  105. return e;
  106. }
  107. }
  108. read_unlock(&ematch_mod_lock);
  109. return NULL;
  110. }
  111. /**
  112. * tcf_em_register - register an extended match
  113. *
  114. * @ops: ematch operations lookup table
  115. *
  116. * This function must be called by ematches to announce their presence.
  117. * The given @ops must have kind set to a unique identifier and the
  118. * callback match() must be implemented. All other callbacks are optional
  119. * and a fallback implementation is used instead.
  120. *
  121. * Returns -EEXISTS if an ematch of the same kind has already registered.
  122. */
  123. int tcf_em_register(struct tcf_ematch_ops *ops)
  124. {
  125. int err = -EEXIST;
  126. struct tcf_ematch_ops *e;
  127. if (ops->match == NULL)
  128. return -EINVAL;
  129. write_lock(&ematch_mod_lock);
  130. list_for_each_entry(e, &ematch_ops, link)
  131. if (ops->kind == e->kind)
  132. goto errout;
  133. list_add_tail(&ops->link, &ematch_ops);
  134. err = 0;
  135. errout:
  136. write_unlock(&ematch_mod_lock);
  137. return err;
  138. }
  139. /**
  140. * tcf_em_unregister - unregster and extended match
  141. *
  142. * @ops: ematch operations lookup table
  143. *
  144. * This function must be called by ematches to announce their disappearance
  145. * for examples when the module gets unloaded. The @ops parameter must be
  146. * the same as the one used for registration.
  147. *
  148. * Returns -ENOENT if no matching ematch was found.
  149. */
  150. int tcf_em_unregister(struct tcf_ematch_ops *ops)
  151. {
  152. int err = 0;
  153. struct tcf_ematch_ops *e;
  154. write_lock(&ematch_mod_lock);
  155. list_for_each_entry(e, &ematch_ops, link) {
  156. if (e == ops) {
  157. list_del(&e->link);
  158. goto out;
  159. }
  160. }
  161. err = -ENOENT;
  162. out:
  163. write_unlock(&ematch_mod_lock);
  164. return err;
  165. }
  166. static inline struct tcf_ematch * tcf_em_get_match(struct tcf_ematch_tree *tree,
  167. int index)
  168. {
  169. return &tree->matches[index];
  170. }
  171. static int tcf_em_validate(struct tcf_proto *tp,
  172. struct tcf_ematch_tree_hdr *tree_hdr,
  173. struct tcf_ematch *em, struct rtattr *rta, int idx)
  174. {
  175. int err = -EINVAL;
  176. struct tcf_ematch_hdr *em_hdr = RTA_DATA(rta);
  177. int data_len = RTA_PAYLOAD(rta) - sizeof(*em_hdr);
  178. void *data = (void *) em_hdr + sizeof(*em_hdr);
  179. if (!TCF_EM_REL_VALID(em_hdr->flags))
  180. goto errout;
  181. if (em_hdr->kind == TCF_EM_CONTAINER) {
  182. /* Special ematch called "container", carries an index
  183. * referencing an external ematch sequence. */
  184. u32 ref;
  185. if (data_len < sizeof(ref))
  186. goto errout;
  187. ref = *(u32 *) data;
  188. if (ref >= tree_hdr->nmatches)
  189. goto errout;
  190. /* We do not allow backward jumps to avoid loops and jumps
  191. * to our own position are of course illegal. */
  192. if (ref <= idx)
  193. goto errout;
  194. em->data = ref;
  195. } else {
  196. /* Note: This lookup will increase the module refcnt
  197. * of the ematch module referenced. In case of a failure,
  198. * a destroy function is called by the underlying layer
  199. * which automatically releases the reference again, therefore
  200. * the module MUST not be given back under any circumstances
  201. * here. Be aware, the destroy function assumes that the
  202. * module is held if the ops field is non zero. */
  203. em->ops = tcf_em_lookup(em_hdr->kind);
  204. if (em->ops == NULL) {
  205. err = -ENOENT;
  206. goto errout;
  207. }
  208. /* ematch module provides expected length of data, so we
  209. * can do a basic sanity check. */
  210. if (em->ops->datalen && data_len < em->ops->datalen)
  211. goto errout;
  212. if (em->ops->change) {
  213. err = em->ops->change(tp, data, data_len, em);
  214. if (err < 0)
  215. goto errout;
  216. } else if (data_len > 0) {
  217. /* ematch module doesn't provide an own change
  218. * procedure and expects us to allocate and copy
  219. * the ematch data.
  220. *
  221. * TCF_EM_SIMPLE may be specified stating that the
  222. * data only consists of a u32 integer and the module
  223. * does not expected a memory reference but rather
  224. * the value carried. */
  225. if (em_hdr->flags & TCF_EM_SIMPLE) {
  226. if (data_len < sizeof(u32))
  227. goto errout;
  228. em->data = *(u32 *) data;
  229. } else {
  230. void *v = kmalloc(data_len, GFP_KERNEL);
  231. if (v == NULL) {
  232. err = -ENOBUFS;
  233. goto errout;
  234. }
  235. memcpy(v, data, data_len);
  236. em->data = (unsigned long) v;
  237. }
  238. }
  239. }
  240. em->matchid = em_hdr->matchid;
  241. em->flags = em_hdr->flags;
  242. em->datalen = data_len;
  243. err = 0;
  244. errout:
  245. return err;
  246. }
  247. /**
  248. * tcf_em_tree_validate - validate ematch config TLV and build ematch tree
  249. *
  250. * @tp: classifier kind handle
  251. * @rta: ematch tree configuration TLV
  252. * @tree: destination ematch tree variable to store the resulting
  253. * ematch tree.
  254. *
  255. * This function validates the given configuration TLV @rta and builds an
  256. * ematch tree in @tree. The resulting tree must later be copied into
  257. * the private classifier data using tcf_em_tree_change(). You MUST NOT
  258. * provide the ematch tree variable of the private classifier data directly,
  259. * the changes would not be locked properly.
  260. *
  261. * Returns a negative error code if the configuration TLV contains errors.
  262. */
  263. int tcf_em_tree_validate(struct tcf_proto *tp, struct rtattr *rta,
  264. struct tcf_ematch_tree *tree)
  265. {
  266. int idx, list_len, matches_len, err = -EINVAL;
  267. struct rtattr *tb[TCA_EMATCH_TREE_MAX];
  268. struct rtattr *rt_match, *rt_hdr, *rt_list;
  269. struct tcf_ematch_tree_hdr *tree_hdr;
  270. struct tcf_ematch *em;
  271. if (!rta) {
  272. memset(tree, 0, sizeof(*tree));
  273. return 0;
  274. }
  275. if (rtattr_parse_nested(tb, TCA_EMATCH_TREE_MAX, rta) < 0)
  276. goto errout;
  277. rt_hdr = tb[TCA_EMATCH_TREE_HDR-1];
  278. rt_list = tb[TCA_EMATCH_TREE_LIST-1];
  279. if (rt_hdr == NULL || rt_list == NULL)
  280. goto errout;
  281. if (RTA_PAYLOAD(rt_hdr) < sizeof(*tree_hdr) ||
  282. RTA_PAYLOAD(rt_list) < sizeof(*rt_match))
  283. goto errout;
  284. tree_hdr = RTA_DATA(rt_hdr);
  285. memcpy(&tree->hdr, tree_hdr, sizeof(*tree_hdr));
  286. rt_match = RTA_DATA(rt_list);
  287. list_len = RTA_PAYLOAD(rt_list);
  288. matches_len = tree_hdr->nmatches * sizeof(*em);
  289. tree->matches = kmalloc(matches_len, GFP_KERNEL);
  290. if (tree->matches == NULL)
  291. goto errout;
  292. memset(tree->matches, 0, matches_len);
  293. /* We do not use rtattr_parse_nested here because the maximum
  294. * number of attributes is unknown. This saves us the allocation
  295. * for a tb buffer which would serve no purpose at all.
  296. *
  297. * The array of rt attributes is parsed in the order as they are
  298. * provided, their type must be incremental from 1 to n. Even
  299. * if it does not serve any real purpose, a failure of sticking
  300. * to this policy will result in parsing failure. */
  301. for (idx = 0; RTA_OK(rt_match, list_len); idx++) {
  302. err = -EINVAL;
  303. if (rt_match->rta_type != (idx + 1))
  304. goto errout_abort;
  305. if (idx >= tree_hdr->nmatches)
  306. goto errout_abort;
  307. if (RTA_PAYLOAD(rt_match) < sizeof(struct tcf_ematch_hdr))
  308. goto errout_abort;
  309. em = tcf_em_get_match(tree, idx);
  310. err = tcf_em_validate(tp, tree_hdr, em, rt_match, idx);
  311. if (err < 0)
  312. goto errout_abort;
  313. rt_match = RTA_NEXT(rt_match, list_len);
  314. }
  315. /* Check if the number of matches provided by userspace actually
  316. * complies with the array of matches. The number was used for
  317. * the validation of references and a mismatch could lead to
  318. * undefined references during the matching process. */
  319. if (idx != tree_hdr->nmatches) {
  320. err = -EINVAL;
  321. goto errout_abort;
  322. }
  323. err = 0;
  324. errout:
  325. return err;
  326. errout_abort:
  327. tcf_em_tree_destroy(tp, tree);
  328. return err;
  329. }
  330. /**
  331. * tcf_em_tree_destroy - destroy an ematch tree
  332. *
  333. * @tp: classifier kind handle
  334. * @tree: ematch tree to be deleted
  335. *
  336. * This functions destroys an ematch tree previously created by
  337. * tcf_em_tree_validate()/tcf_em_tree_change(). You must ensure that
  338. * the ematch tree is not in use before calling this function.
  339. */
  340. void tcf_em_tree_destroy(struct tcf_proto *tp, struct tcf_ematch_tree *tree)
  341. {
  342. int i;
  343. if (tree->matches == NULL)
  344. return;
  345. for (i = 0; i < tree->hdr.nmatches; i++) {
  346. struct tcf_ematch *em = tcf_em_get_match(tree, i);
  347. if (em->ops) {
  348. if (em->ops->destroy)
  349. em->ops->destroy(tp, em);
  350. else if (!tcf_em_is_simple(em) && em->data)
  351. kfree((void *) em->data);
  352. module_put(em->ops->owner);
  353. }
  354. }
  355. tree->hdr.nmatches = 0;
  356. kfree(tree->matches);
  357. }
  358. /**
  359. * tcf_em_tree_dump - dump ematch tree into a rtnl message
  360. *
  361. * @skb: skb holding the rtnl message
  362. * @t: ematch tree to be dumped
  363. * @tlv: TLV type to be used to encapsulate the tree
  364. *
  365. * This function dumps a ematch tree into a rtnl message. It is valid to
  366. * call this function while the ematch tree is in use.
  367. *
  368. * Returns -1 if the skb tailroom is insufficient.
  369. */
  370. int tcf_em_tree_dump(struct sk_buff *skb, struct tcf_ematch_tree *tree, int tlv)
  371. {
  372. int i;
  373. struct rtattr * top_start = (struct rtattr*) skb->tail;
  374. struct rtattr * list_start;
  375. RTA_PUT(skb, tlv, 0, NULL);
  376. RTA_PUT(skb, TCA_EMATCH_TREE_HDR, sizeof(tree->hdr), &tree->hdr);
  377. list_start = (struct rtattr *) skb->tail;
  378. RTA_PUT(skb, TCA_EMATCH_TREE_LIST, 0, NULL);
  379. for (i = 0; i < tree->hdr.nmatches; i++) {
  380. struct rtattr *match_start = (struct rtattr*) skb->tail;
  381. struct tcf_ematch *em = tcf_em_get_match(tree, i);
  382. struct tcf_ematch_hdr em_hdr = {
  383. .kind = em->ops ? em->ops->kind : TCF_EM_CONTAINER,
  384. .matchid = em->matchid,
  385. .flags = em->flags
  386. };
  387. RTA_PUT(skb, i+1, sizeof(em_hdr), &em_hdr);
  388. if (em->ops && em->ops->dump) {
  389. if (em->ops->dump(skb, em) < 0)
  390. goto rtattr_failure;
  391. } else if (tcf_em_is_container(em) || tcf_em_is_simple(em)) {
  392. u32 u = em->data;
  393. RTA_PUT_NOHDR(skb, sizeof(u), &u);
  394. } else if (em->datalen > 0)
  395. RTA_PUT_NOHDR(skb, em->datalen, (void *) em->data);
  396. match_start->rta_len = skb->tail - (u8*) match_start;
  397. }
  398. list_start->rta_len = skb->tail - (u8 *) list_start;
  399. top_start->rta_len = skb->tail - (u8 *) top_start;
  400. return 0;
  401. rtattr_failure:
  402. return -1;
  403. }
  404. static inline int tcf_em_match(struct sk_buff *skb, struct tcf_ematch *em,
  405. struct tcf_pkt_info *info)
  406. {
  407. int r = em->ops->match(skb, em, info);
  408. return tcf_em_is_inverted(em) ? !r : r;
  409. }
  410. /* Do not use this function directly, use tcf_em_tree_match instead */
  411. int __tcf_em_tree_match(struct sk_buff *skb, struct tcf_ematch_tree *tree,
  412. struct tcf_pkt_info *info)
  413. {
  414. int stackp = 0, match_idx = 0, res = 0;
  415. struct tcf_ematch *cur_match;
  416. int stack[CONFIG_NET_EMATCH_STACK];
  417. proceed:
  418. while (match_idx < tree->hdr.nmatches) {
  419. cur_match = tcf_em_get_match(tree, match_idx);
  420. if (tcf_em_is_container(cur_match)) {
  421. if (unlikely(stackp >= CONFIG_NET_EMATCH_STACK))
  422. goto stack_overflow;
  423. stack[stackp++] = match_idx;
  424. match_idx = cur_match->data;
  425. goto proceed;
  426. }
  427. res = tcf_em_match(skb, cur_match, info);
  428. if (tcf_em_early_end(cur_match, res))
  429. break;
  430. match_idx++;
  431. }
  432. pop_stack:
  433. if (stackp > 0) {
  434. match_idx = stack[--stackp];
  435. cur_match = tcf_em_get_match(tree, match_idx);
  436. if (tcf_em_early_end(cur_match, res))
  437. goto pop_stack;
  438. else {
  439. match_idx++;
  440. goto proceed;
  441. }
  442. }
  443. return res;
  444. stack_overflow:
  445. if (net_ratelimit())
  446. printk("Local stack overflow, increase NET_EMATCH_STACK\n");
  447. return -1;
  448. }
  449. EXPORT_SYMBOL(tcf_em_register);
  450. EXPORT_SYMBOL(tcf_em_unregister);
  451. EXPORT_SYMBOL(tcf_em_tree_validate);
  452. EXPORT_SYMBOL(tcf_em_tree_destroy);
  453. EXPORT_SYMBOL(tcf_em_tree_dump);
  454. EXPORT_SYMBOL(__tcf_em_tree_match);