blob: 63146d339d814e09a19cd26ba6909e3c9c9e8b6b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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.
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +090040 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070041 * ==========================================================================
42 *
43 * How to write an ematch in 60 seconds
44 * ------------------------------------
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +090045 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070046 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070084#include <linux/module.h>
85#include <linux/types.h>
86#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070087#include <linux/mm.h>
88#include <linux/errno.h>
89#include <linux/interrupt.h>
90#include <linux/rtnetlink.h>
91#include <linux/skbuff.h>
92#include <net/pkt_cls.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070093
94static LIST_HEAD(ematch_ops);
95static DEFINE_RWLOCK(ematch_mod_lock);
96
97static inline struct tcf_ematch_ops * tcf_em_lookup(u16 kind)
98{
99 struct tcf_ematch_ops *e = NULL;
100
101 read_lock(&ematch_mod_lock);
102 list_for_each_entry(e, &ematch_ops, link) {
103 if (kind == e->kind) {
104 if (!try_module_get(e->owner))
105 e = NULL;
106 read_unlock(&ematch_mod_lock);
107 return e;
108 }
109 }
110 read_unlock(&ematch_mod_lock);
111
112 return NULL;
113}
114
115/**
116 * tcf_em_register - register an extended match
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900117 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 * @ops: ematch operations lookup table
119 *
120 * This function must be called by ematches to announce their presence.
121 * The given @ops must have kind set to a unique identifier and the
122 * callback match() must be implemented. All other callbacks are optional
123 * and a fallback implementation is used instead.
124 *
125 * Returns -EEXISTS if an ematch of the same kind has already registered.
126 */
127int tcf_em_register(struct tcf_ematch_ops *ops)
128{
129 int err = -EEXIST;
130 struct tcf_ematch_ops *e;
131
132 if (ops->match == NULL)
133 return -EINVAL;
134
135 write_lock(&ematch_mod_lock);
136 list_for_each_entry(e, &ematch_ops, link)
137 if (ops->kind == e->kind)
138 goto errout;
139
140 list_add_tail(&ops->link, &ematch_ops);
141 err = 0;
142errout:
143 write_unlock(&ematch_mod_lock);
144 return err;
145}
146
147/**
148 * tcf_em_unregister - unregster and extended match
149 *
150 * @ops: ematch operations lookup table
151 *
152 * This function must be called by ematches to announce their disappearance
153 * for examples when the module gets unloaded. The @ops parameter must be
154 * the same as the one used for registration.
155 *
156 * Returns -ENOENT if no matching ematch was found.
157 */
158int tcf_em_unregister(struct tcf_ematch_ops *ops)
159{
160 int err = 0;
161 struct tcf_ematch_ops *e;
162
163 write_lock(&ematch_mod_lock);
164 list_for_each_entry(e, &ematch_ops, link) {
165 if (e == ops) {
166 list_del(&e->link);
167 goto out;
168 }
169 }
170
171 err = -ENOENT;
172out:
173 write_unlock(&ematch_mod_lock);
174 return err;
175}
176
177static inline struct tcf_ematch * tcf_em_get_match(struct tcf_ematch_tree *tree,
178 int index)
179{
180 return &tree->matches[index];
181}
182
183
184static int tcf_em_validate(struct tcf_proto *tp,
185 struct tcf_ematch_tree_hdr *tree_hdr,
186 struct tcf_ematch *em, struct rtattr *rta, int idx)
187{
188 int err = -EINVAL;
189 struct tcf_ematch_hdr *em_hdr = RTA_DATA(rta);
190 int data_len = RTA_PAYLOAD(rta) - sizeof(*em_hdr);
191 void *data = (void *) em_hdr + sizeof(*em_hdr);
192
193 if (!TCF_EM_REL_VALID(em_hdr->flags))
194 goto errout;
195
196 if (em_hdr->kind == TCF_EM_CONTAINER) {
197 /* Special ematch called "container", carries an index
198 * referencing an external ematch sequence. */
199 u32 ref;
200
201 if (data_len < sizeof(ref))
202 goto errout;
203 ref = *(u32 *) data;
204
205 if (ref >= tree_hdr->nmatches)
206 goto errout;
207
208 /* We do not allow backward jumps to avoid loops and jumps
209 * to our own position are of course illegal. */
210 if (ref <= idx)
211 goto errout;
212
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900213
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 em->data = ref;
215 } else {
216 /* Note: This lookup will increase the module refcnt
217 * of the ematch module referenced. In case of a failure,
218 * a destroy function is called by the underlying layer
219 * which automatically releases the reference again, therefore
220 * the module MUST not be given back under any circumstances
221 * here. Be aware, the destroy function assumes that the
222 * module is held if the ops field is non zero. */
223 em->ops = tcf_em_lookup(em_hdr->kind);
224
225 if (em->ops == NULL) {
226 err = -ENOENT;
227 goto errout;
228 }
229
230 /* ematch module provides expected length of data, so we
231 * can do a basic sanity check. */
232 if (em->ops->datalen && data_len < em->ops->datalen)
233 goto errout;
234
235 if (em->ops->change) {
236 err = em->ops->change(tp, data, data_len, em);
237 if (err < 0)
238 goto errout;
239 } else if (data_len > 0) {
240 /* ematch module doesn't provide an own change
241 * procedure and expects us to allocate and copy
242 * the ematch data.
243 *
244 * TCF_EM_SIMPLE may be specified stating that the
245 * data only consists of a u32 integer and the module
246 * does not expected a memory reference but rather
247 * the value carried. */
248 if (em_hdr->flags & TCF_EM_SIMPLE) {
249 if (data_len < sizeof(u32))
250 goto errout;
251 em->data = *(u32 *) data;
252 } else {
Arnaldo Carvalho de Meloc7b1b242006-11-21 01:19:40 -0200253 void *v = kmemdup(data, data_len, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 if (v == NULL) {
255 err = -ENOBUFS;
256 goto errout;
257 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 em->data = (unsigned long) v;
259 }
260 }
261 }
262
263 em->matchid = em_hdr->matchid;
264 em->flags = em_hdr->flags;
265 em->datalen = data_len;
266
267 err = 0;
268errout:
269 return err;
270}
271
272/**
273 * tcf_em_tree_validate - validate ematch config TLV and build ematch tree
274 *
275 * @tp: classifier kind handle
276 * @rta: ematch tree configuration TLV
277 * @tree: destination ematch tree variable to store the resulting
278 * ematch tree.
279 *
280 * This function validates the given configuration TLV @rta and builds an
281 * ematch tree in @tree. The resulting tree must later be copied into
282 * the private classifier data using tcf_em_tree_change(). You MUST NOT
283 * provide the ematch tree variable of the private classifier data directly,
284 * the changes would not be locked properly.
285 *
286 * Returns a negative error code if the configuration TLV contains errors.
287 */
288int tcf_em_tree_validate(struct tcf_proto *tp, struct rtattr *rta,
289 struct tcf_ematch_tree *tree)
290{
291 int idx, list_len, matches_len, err = -EINVAL;
292 struct rtattr *tb[TCA_EMATCH_TREE_MAX];
293 struct rtattr *rt_match, *rt_hdr, *rt_list;
294 struct tcf_ematch_tree_hdr *tree_hdr;
295 struct tcf_ematch *em;
296
Thomas Grafb541ca22005-11-08 09:39:17 -0800297 if (!rta) {
298 memset(tree, 0, sizeof(*tree));
299 return 0;
300 }
301
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 if (rtattr_parse_nested(tb, TCA_EMATCH_TREE_MAX, rta) < 0)
303 goto errout;
304
305 rt_hdr = tb[TCA_EMATCH_TREE_HDR-1];
306 rt_list = tb[TCA_EMATCH_TREE_LIST-1];
307
308 if (rt_hdr == NULL || rt_list == NULL)
309 goto errout;
310
311 if (RTA_PAYLOAD(rt_hdr) < sizeof(*tree_hdr) ||
312 RTA_PAYLOAD(rt_list) < sizeof(*rt_match))
313 goto errout;
314
315 tree_hdr = RTA_DATA(rt_hdr);
316 memcpy(&tree->hdr, tree_hdr, sizeof(*tree_hdr));
317
318 rt_match = RTA_DATA(rt_list);
319 list_len = RTA_PAYLOAD(rt_list);
320 matches_len = tree_hdr->nmatches * sizeof(*em);
321
Panagiotis Issaris0da974f2006-07-21 14:51:30 -0700322 tree->matches = kzalloc(matches_len, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 if (tree->matches == NULL)
324 goto errout;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325
326 /* We do not use rtattr_parse_nested here because the maximum
327 * number of attributes is unknown. This saves us the allocation
328 * for a tb buffer which would serve no purpose at all.
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900329 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 * The array of rt attributes is parsed in the order as they are
331 * provided, their type must be incremental from 1 to n. Even
332 * if it does not serve any real purpose, a failure of sticking
333 * to this policy will result in parsing failure. */
334 for (idx = 0; RTA_OK(rt_match, list_len); idx++) {
335 err = -EINVAL;
336
337 if (rt_match->rta_type != (idx + 1))
338 goto errout_abort;
339
340 if (idx >= tree_hdr->nmatches)
341 goto errout_abort;
342
343 if (RTA_PAYLOAD(rt_match) < sizeof(struct tcf_ematch_hdr))
344 goto errout_abort;
345
346 em = tcf_em_get_match(tree, idx);
347
348 err = tcf_em_validate(tp, tree_hdr, em, rt_match, idx);
349 if (err < 0)
350 goto errout_abort;
351
352 rt_match = RTA_NEXT(rt_match, list_len);
353 }
354
355 /* Check if the number of matches provided by userspace actually
356 * complies with the array of matches. The number was used for
357 * the validation of references and a mismatch could lead to
358 * undefined references during the matching process. */
359 if (idx != tree_hdr->nmatches) {
360 err = -EINVAL;
361 goto errout_abort;
362 }
363
364 err = 0;
365errout:
366 return err;
367
368errout_abort:
369 tcf_em_tree_destroy(tp, tree);
370 return err;
371}
372
373/**
374 * tcf_em_tree_destroy - destroy an ematch tree
375 *
376 * @tp: classifier kind handle
377 * @tree: ematch tree to be deleted
378 *
379 * This functions destroys an ematch tree previously created by
380 * tcf_em_tree_validate()/tcf_em_tree_change(). You must ensure that
381 * the ematch tree is not in use before calling this function.
382 */
383void tcf_em_tree_destroy(struct tcf_proto *tp, struct tcf_ematch_tree *tree)
384{
385 int i;
386
387 if (tree->matches == NULL)
388 return;
389
390 for (i = 0; i < tree->hdr.nmatches; i++) {
391 struct tcf_ematch *em = tcf_em_get_match(tree, i);
392
393 if (em->ops) {
394 if (em->ops->destroy)
395 em->ops->destroy(tp, em);
396 else if (!tcf_em_is_simple(em) && em->data)
397 kfree((void *) em->data);
398 module_put(em->ops->owner);
399 }
400 }
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900401
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 tree->hdr.nmatches = 0;
403 kfree(tree->matches);
404}
405
406/**
407 * tcf_em_tree_dump - dump ematch tree into a rtnl message
408 *
409 * @skb: skb holding the rtnl message
410 * @t: ematch tree to be dumped
411 * @tlv: TLV type to be used to encapsulate the tree
412 *
413 * This function dumps a ematch tree into a rtnl message. It is valid to
414 * call this function while the ematch tree is in use.
415 *
416 * Returns -1 if the skb tailroom is insufficient.
417 */
418int tcf_em_tree_dump(struct sk_buff *skb, struct tcf_ematch_tree *tree, int tlv)
419{
420 int i;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700421 u8 *tail;
422 struct rtattr *top_start = (struct rtattr *)skb_tail_pointer(skb);
423 struct rtattr *list_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424
425 RTA_PUT(skb, tlv, 0, NULL);
426 RTA_PUT(skb, TCA_EMATCH_TREE_HDR, sizeof(tree->hdr), &tree->hdr);
427
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700428 list_start = (struct rtattr *)skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 RTA_PUT(skb, TCA_EMATCH_TREE_LIST, 0, NULL);
430
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700431 tail = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 for (i = 0; i < tree->hdr.nmatches; i++) {
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700433 struct rtattr *match_start = (struct rtattr *)tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 struct tcf_ematch *em = tcf_em_get_match(tree, i);
435 struct tcf_ematch_hdr em_hdr = {
436 .kind = em->ops ? em->ops->kind : TCF_EM_CONTAINER,
437 .matchid = em->matchid,
438 .flags = em->flags
439 };
440
441 RTA_PUT(skb, i+1, sizeof(em_hdr), &em_hdr);
442
443 if (em->ops && em->ops->dump) {
444 if (em->ops->dump(skb, em) < 0)
445 goto rtattr_failure;
446 } else if (tcf_em_is_container(em) || tcf_em_is_simple(em)) {
447 u32 u = em->data;
448 RTA_PUT_NOHDR(skb, sizeof(u), &u);
449 } else if (em->datalen > 0)
450 RTA_PUT_NOHDR(skb, em->datalen, (void *) em->data);
451
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700452 tail = skb_tail_pointer(skb);
453 match_start->rta_len = tail - (u8 *)match_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 }
455
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700456 list_start->rta_len = tail - (u8 *)list_start;
457 top_start->rta_len = tail - (u8 *)top_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
459 return 0;
460
461rtattr_failure:
462 return -1;
463}
464
465static inline int tcf_em_match(struct sk_buff *skb, struct tcf_ematch *em,
466 struct tcf_pkt_info *info)
467{
468 int r = em->ops->match(skb, em, info);
469 return tcf_em_is_inverted(em) ? !r : r;
470}
471
472/* Do not use this function directly, use tcf_em_tree_match instead */
473int __tcf_em_tree_match(struct sk_buff *skb, struct tcf_ematch_tree *tree,
474 struct tcf_pkt_info *info)
475{
476 int stackp = 0, match_idx = 0, res = 0;
477 struct tcf_ematch *cur_match;
478 int stack[CONFIG_NET_EMATCH_STACK];
479
480proceed:
481 while (match_idx < tree->hdr.nmatches) {
482 cur_match = tcf_em_get_match(tree, match_idx);
483
484 if (tcf_em_is_container(cur_match)) {
485 if (unlikely(stackp >= CONFIG_NET_EMATCH_STACK))
486 goto stack_overflow;
487
488 stack[stackp++] = match_idx;
489 match_idx = cur_match->data;
490 goto proceed;
491 }
492
493 res = tcf_em_match(skb, cur_match, info);
494
495 if (tcf_em_early_end(cur_match, res))
496 break;
497
498 match_idx++;
499 }
500
501pop_stack:
502 if (stackp > 0) {
503 match_idx = stack[--stackp];
504 cur_match = tcf_em_get_match(tree, match_idx);
505
506 if (tcf_em_early_end(cur_match, res))
507 goto pop_stack;
508 else {
509 match_idx++;
510 goto proceed;
511 }
512 }
513
514 return res;
515
516stack_overflow:
517 if (net_ratelimit())
518 printk("Local stack overflow, increase NET_EMATCH_STACK\n");
519 return -1;
520}
521
522EXPORT_SYMBOL(tcf_em_register);
523EXPORT_SYMBOL(tcf_em_unregister);
524EXPORT_SYMBOL(tcf_em_tree_validate);
525EXPORT_SYMBOL(tcf_em_tree_destroy);
526EXPORT_SYMBOL(tcf_em_tree_dump);
527EXPORT_SYMBOL(__tcf_em_tree_match);