MAINTAINERS: Add phy-miphy28lp.c and phy-miphy365x.c to ARCH/STI architecture
[deliverable/linux.git] / net / sched / ematch.c
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 * 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
84 #include <linux/module.h>
85 #include <linux/slab.h>
86 #include <linux/types.h>
87 #include <linux/kernel.h>
88 #include <linux/errno.h>
89 #include <linux/rtnetlink.h>
90 #include <linux/skbuff.h>
91 #include <net/pkt_cls.h>
92
93 static LIST_HEAD(ematch_ops);
94 static DEFINE_RWLOCK(ematch_mod_lock);
95
96 static struct tcf_ematch_ops *tcf_em_lookup(u16 kind)
97 {
98 struct tcf_ematch_ops *e = NULL;
99
100 read_lock(&ematch_mod_lock);
101 list_for_each_entry(e, &ematch_ops, link) {
102 if (kind == e->kind) {
103 if (!try_module_get(e->owner))
104 e = NULL;
105 read_unlock(&ematch_mod_lock);
106 return e;
107 }
108 }
109 read_unlock(&ematch_mod_lock);
110
111 return NULL;
112 }
113
114 /**
115 * tcf_em_register - register an extended match
116 *
117 * @ops: ematch operations lookup table
118 *
119 * This function must be called by ematches to announce their presence.
120 * The given @ops must have kind set to a unique identifier and the
121 * callback match() must be implemented. All other callbacks are optional
122 * and a fallback implementation is used instead.
123 *
124 * Returns -EEXISTS if an ematch of the same kind has already registered.
125 */
126 int tcf_em_register(struct tcf_ematch_ops *ops)
127 {
128 int err = -EEXIST;
129 struct tcf_ematch_ops *e;
130
131 if (ops->match == NULL)
132 return -EINVAL;
133
134 write_lock(&ematch_mod_lock);
135 list_for_each_entry(e, &ematch_ops, link)
136 if (ops->kind == e->kind)
137 goto errout;
138
139 list_add_tail(&ops->link, &ematch_ops);
140 err = 0;
141 errout:
142 write_unlock(&ematch_mod_lock);
143 return err;
144 }
145 EXPORT_SYMBOL(tcf_em_register);
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 */
158 void tcf_em_unregister(struct tcf_ematch_ops *ops)
159 {
160 write_lock(&ematch_mod_lock);
161 list_del(&ops->link);
162 write_unlock(&ematch_mod_lock);
163 }
164 EXPORT_SYMBOL(tcf_em_unregister);
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
172
173 static int tcf_em_validate(struct tcf_proto *tp,
174 struct tcf_ematch_tree_hdr *tree_hdr,
175 struct tcf_ematch *em, struct nlattr *nla, int idx)
176 {
177 int err = -EINVAL;
178 struct tcf_ematch_hdr *em_hdr = nla_data(nla);
179 int data_len = nla_len(nla) - sizeof(*em_hdr);
180 void *data = (void *) em_hdr + sizeof(*em_hdr);
181 struct net *net = dev_net(qdisc_dev(tp->q));
182
183 if (!TCF_EM_REL_VALID(em_hdr->flags))
184 goto errout;
185
186 if (em_hdr->kind == TCF_EM_CONTAINER) {
187 /* Special ematch called "container", carries an index
188 * referencing an external ematch sequence.
189 */
190 u32 ref;
191
192 if (data_len < sizeof(ref))
193 goto errout;
194 ref = *(u32 *) data;
195
196 if (ref >= tree_hdr->nmatches)
197 goto errout;
198
199 /* We do not allow backward jumps to avoid loops and jumps
200 * to our own position are of course illegal.
201 */
202 if (ref <= idx)
203 goto errout;
204
205
206 em->data = ref;
207 } else {
208 /* Note: This lookup will increase the module refcnt
209 * of the ematch module referenced. In case of a failure,
210 * a destroy function is called by the underlying layer
211 * which automatically releases the reference again, therefore
212 * the module MUST not be given back under any circumstances
213 * here. Be aware, the destroy function assumes that the
214 * module is held if the ops field is non zero.
215 */
216 em->ops = tcf_em_lookup(em_hdr->kind);
217
218 if (em->ops == NULL) {
219 err = -ENOENT;
220 #ifdef CONFIG_MODULES
221 __rtnl_unlock();
222 request_module("ematch-kind-%u", em_hdr->kind);
223 rtnl_lock();
224 em->ops = tcf_em_lookup(em_hdr->kind);
225 if (em->ops) {
226 /* We dropped the RTNL mutex in order to
227 * perform the module load. Tell the caller
228 * to replay the request.
229 */
230 module_put(em->ops->owner);
231 err = -EAGAIN;
232 }
233 #endif
234 goto errout;
235 }
236
237 /* ematch module provides expected length of data, so we
238 * can do a basic sanity check.
239 */
240 if (em->ops->datalen && data_len < em->ops->datalen)
241 goto errout;
242
243 if (em->ops->change) {
244 err = em->ops->change(net, data, data_len, em);
245 if (err < 0)
246 goto errout;
247 } else if (data_len > 0) {
248 /* ematch module doesn't provide an own change
249 * procedure and expects us to allocate and copy
250 * the ematch data.
251 *
252 * TCF_EM_SIMPLE may be specified stating that the
253 * data only consists of a u32 integer and the module
254 * does not expected a memory reference but rather
255 * the value carried.
256 */
257 if (em_hdr->flags & TCF_EM_SIMPLE) {
258 if (data_len < sizeof(u32))
259 goto errout;
260 em->data = *(u32 *) data;
261 } else {
262 void *v = kmemdup(data, data_len, GFP_KERNEL);
263 if (v == NULL) {
264 err = -ENOBUFS;
265 goto errout;
266 }
267 em->data = (unsigned long) v;
268 }
269 }
270 }
271
272 em->matchid = em_hdr->matchid;
273 em->flags = em_hdr->flags;
274 em->datalen = data_len;
275 em->net = net;
276
277 err = 0;
278 errout:
279 return err;
280 }
281
282 static const struct nla_policy em_policy[TCA_EMATCH_TREE_MAX + 1] = {
283 [TCA_EMATCH_TREE_HDR] = { .len = sizeof(struct tcf_ematch_tree_hdr) },
284 [TCA_EMATCH_TREE_LIST] = { .type = NLA_NESTED },
285 };
286
287 /**
288 * tcf_em_tree_validate - validate ematch config TLV and build ematch tree
289 *
290 * @tp: classifier kind handle
291 * @nla: ematch tree configuration TLV
292 * @tree: destination ematch tree variable to store the resulting
293 * ematch tree.
294 *
295 * This function validates the given configuration TLV @nla and builds an
296 * ematch tree in @tree. The resulting tree must later be copied into
297 * the private classifier data using tcf_em_tree_change(). You MUST NOT
298 * provide the ematch tree variable of the private classifier data directly,
299 * the changes would not be locked properly.
300 *
301 * Returns a negative error code if the configuration TLV contains errors.
302 */
303 int tcf_em_tree_validate(struct tcf_proto *tp, struct nlattr *nla,
304 struct tcf_ematch_tree *tree)
305 {
306 int idx, list_len, matches_len, err;
307 struct nlattr *tb[TCA_EMATCH_TREE_MAX + 1];
308 struct nlattr *rt_match, *rt_hdr, *rt_list;
309 struct tcf_ematch_tree_hdr *tree_hdr;
310 struct tcf_ematch *em;
311
312 memset(tree, 0, sizeof(*tree));
313 if (!nla)
314 return 0;
315
316 err = nla_parse_nested(tb, TCA_EMATCH_TREE_MAX, nla, em_policy);
317 if (err < 0)
318 goto errout;
319
320 err = -EINVAL;
321 rt_hdr = tb[TCA_EMATCH_TREE_HDR];
322 rt_list = tb[TCA_EMATCH_TREE_LIST];
323
324 if (rt_hdr == NULL || rt_list == NULL)
325 goto errout;
326
327 tree_hdr = nla_data(rt_hdr);
328 memcpy(&tree->hdr, tree_hdr, sizeof(*tree_hdr));
329
330 rt_match = nla_data(rt_list);
331 list_len = nla_len(rt_list);
332 matches_len = tree_hdr->nmatches * sizeof(*em);
333
334 tree->matches = kzalloc(matches_len, GFP_KERNEL);
335 if (tree->matches == NULL)
336 goto errout;
337
338 /* We do not use nla_parse_nested here because the maximum
339 * number of attributes is unknown. This saves us the allocation
340 * for a tb buffer which would serve no purpose at all.
341 *
342 * The array of rt attributes is parsed in the order as they are
343 * provided, their type must be incremental from 1 to n. Even
344 * if it does not serve any real purpose, a failure of sticking
345 * to this policy will result in parsing failure.
346 */
347 for (idx = 0; nla_ok(rt_match, list_len); idx++) {
348 err = -EINVAL;
349
350 if (rt_match->nla_type != (idx + 1))
351 goto errout_abort;
352
353 if (idx >= tree_hdr->nmatches)
354 goto errout_abort;
355
356 if (nla_len(rt_match) < sizeof(struct tcf_ematch_hdr))
357 goto errout_abort;
358
359 em = tcf_em_get_match(tree, idx);
360
361 err = tcf_em_validate(tp, tree_hdr, em, rt_match, idx);
362 if (err < 0)
363 goto errout_abort;
364
365 rt_match = nla_next(rt_match, &list_len);
366 }
367
368 /* Check if the number of matches provided by userspace actually
369 * complies with the array of matches. The number was used for
370 * the validation of references and a mismatch could lead to
371 * undefined references during the matching process.
372 */
373 if (idx != tree_hdr->nmatches) {
374 err = -EINVAL;
375 goto errout_abort;
376 }
377
378 err = 0;
379 errout:
380 return err;
381
382 errout_abort:
383 tcf_em_tree_destroy(tree);
384 return err;
385 }
386 EXPORT_SYMBOL(tcf_em_tree_validate);
387
388 /**
389 * tcf_em_tree_destroy - destroy an ematch tree
390 *
391 * @tp: classifier kind handle
392 * @tree: ematch tree to be deleted
393 *
394 * This functions destroys an ematch tree previously created by
395 * tcf_em_tree_validate()/tcf_em_tree_change(). You must ensure that
396 * the ematch tree is not in use before calling this function.
397 */
398 void tcf_em_tree_destroy(struct tcf_ematch_tree *tree)
399 {
400 int i;
401
402 if (tree->matches == NULL)
403 return;
404
405 for (i = 0; i < tree->hdr.nmatches; i++) {
406 struct tcf_ematch *em = tcf_em_get_match(tree, i);
407
408 if (em->ops) {
409 if (em->ops->destroy)
410 em->ops->destroy(em);
411 else if (!tcf_em_is_simple(em))
412 kfree((void *) em->data);
413 module_put(em->ops->owner);
414 }
415 }
416
417 tree->hdr.nmatches = 0;
418 kfree(tree->matches);
419 tree->matches = NULL;
420 }
421 EXPORT_SYMBOL(tcf_em_tree_destroy);
422
423 /**
424 * tcf_em_tree_dump - dump ematch tree into a rtnl message
425 *
426 * @skb: skb holding the rtnl message
427 * @t: ematch tree to be dumped
428 * @tlv: TLV type to be used to encapsulate the tree
429 *
430 * This function dumps a ematch tree into a rtnl message. It is valid to
431 * call this function while the ematch tree is in use.
432 *
433 * Returns -1 if the skb tailroom is insufficient.
434 */
435 int tcf_em_tree_dump(struct sk_buff *skb, struct tcf_ematch_tree *tree, int tlv)
436 {
437 int i;
438 u8 *tail;
439 struct nlattr *top_start;
440 struct nlattr *list_start;
441
442 top_start = nla_nest_start(skb, tlv);
443 if (top_start == NULL)
444 goto nla_put_failure;
445
446 if (nla_put(skb, TCA_EMATCH_TREE_HDR, sizeof(tree->hdr), &tree->hdr))
447 goto nla_put_failure;
448
449 list_start = nla_nest_start(skb, TCA_EMATCH_TREE_LIST);
450 if (list_start == NULL)
451 goto nla_put_failure;
452
453 tail = skb_tail_pointer(skb);
454 for (i = 0; i < tree->hdr.nmatches; i++) {
455 struct nlattr *match_start = (struct nlattr *)tail;
456 struct tcf_ematch *em = tcf_em_get_match(tree, i);
457 struct tcf_ematch_hdr em_hdr = {
458 .kind = em->ops ? em->ops->kind : TCF_EM_CONTAINER,
459 .matchid = em->matchid,
460 .flags = em->flags
461 };
462
463 if (nla_put(skb, i + 1, sizeof(em_hdr), &em_hdr))
464 goto nla_put_failure;
465
466 if (em->ops && em->ops->dump) {
467 if (em->ops->dump(skb, em) < 0)
468 goto nla_put_failure;
469 } else if (tcf_em_is_container(em) || tcf_em_is_simple(em)) {
470 u32 u = em->data;
471 nla_put_nohdr(skb, sizeof(u), &u);
472 } else if (em->datalen > 0)
473 nla_put_nohdr(skb, em->datalen, (void *) em->data);
474
475 tail = skb_tail_pointer(skb);
476 match_start->nla_len = tail - (u8 *)match_start;
477 }
478
479 nla_nest_end(skb, list_start);
480 nla_nest_end(skb, top_start);
481
482 return 0;
483
484 nla_put_failure:
485 return -1;
486 }
487 EXPORT_SYMBOL(tcf_em_tree_dump);
488
489 static inline int tcf_em_match(struct sk_buff *skb, struct tcf_ematch *em,
490 struct tcf_pkt_info *info)
491 {
492 int r = em->ops->match(skb, em, info);
493
494 return tcf_em_is_inverted(em) ? !r : r;
495 }
496
497 /* Do not use this function directly, use tcf_em_tree_match instead */
498 int __tcf_em_tree_match(struct sk_buff *skb, struct tcf_ematch_tree *tree,
499 struct tcf_pkt_info *info)
500 {
501 int stackp = 0, match_idx = 0, res = 0;
502 struct tcf_ematch *cur_match;
503 int stack[CONFIG_NET_EMATCH_STACK];
504
505 proceed:
506 while (match_idx < tree->hdr.nmatches) {
507 cur_match = tcf_em_get_match(tree, match_idx);
508
509 if (tcf_em_is_container(cur_match)) {
510 if (unlikely(stackp >= CONFIG_NET_EMATCH_STACK))
511 goto stack_overflow;
512
513 stack[stackp++] = match_idx;
514 match_idx = cur_match->data;
515 goto proceed;
516 }
517
518 res = tcf_em_match(skb, cur_match, info);
519
520 if (tcf_em_early_end(cur_match, res))
521 break;
522
523 match_idx++;
524 }
525
526 pop_stack:
527 if (stackp > 0) {
528 match_idx = stack[--stackp];
529 cur_match = tcf_em_get_match(tree, match_idx);
530
531 if (tcf_em_is_inverted(cur_match))
532 res = !res;
533
534 if (tcf_em_early_end(cur_match, res)) {
535 goto pop_stack;
536 } else {
537 match_idx++;
538 goto proceed;
539 }
540 }
541
542 return res;
543
544 stack_overflow:
545 net_warn_ratelimited("tc ematch: local stack overflow, increase NET_EMATCH_STACK\n");
546 return -1;
547 }
548 EXPORT_SYMBOL(__tcf_em_tree_match);
This page took 0.070458 seconds and 5 git commands to generate.