mm: always lock the root (oldest) anon_vma
[linux-2.6/libata-dev.git] / net / sched / ematch.c
blob5e37da961f804cfdf17fa30d6434332e32dc1332
1 /*
2 * net/sched/ematch.c Extended Match API
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.
9 * Authors: Thomas Graf <tgraf@suug.ch>
11 * ==========================================================================
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.
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.
27 * Example:
28 * A AND (B1 OR B2) AND C AND D
30 * ------->-PUSH-------
31 * -->-- / -->-- \ -->--
32 * / \ / / \ \ / \
33 * +-------+-------+-------+-------+-------+--------+
34 * | A AND | B AND | C AND | D END | B1 OR | B2 END |
35 * +-------+-------+-------+-------+-------+--------+
36 * \ /
37 * --------<-POP---------
39 * where B is a virtual ematch referencing to sequence starting with B1.
41 * ==========================================================================
43 * How to write an ematch in 60 seconds
44 * ------------------------------------
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;
52 * if (...matching goes here...)
53 * return 1;
54 * else
55 * return 0;
56 * }
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 * };
66 * 3) Register/Unregister your ematch:
67 * static int __init init_my_ematch(void)
68 * {
69 * return tcf_em_register(&my_ops);
70 * }
72 * static void __exit exit_my_ematch(void)
73 * {
74 * tcf_em_unregister(&my_ops);
75 * }
77 * module_init(init_my_ematch);
78 * module_exit(exit_my_ematch);
80 * 4) By now you should have two more seconds left, barely enough to
81 * open up a beer to watch the compilation going.
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>
93 static LIST_HEAD(ematch_ops);
94 static DEFINE_RWLOCK(ematch_mod_lock);
96 static inline struct tcf_ematch_ops * tcf_em_lookup(u16 kind)
98 struct tcf_ematch_ops *e = NULL;
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;
109 read_unlock(&ematch_mod_lock);
111 return NULL;
115 * tcf_em_register - register an extended match
117 * @ops: ematch operations lookup table
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.
124 * Returns -EEXISTS if an ematch of the same kind has already registered.
126 int tcf_em_register(struct tcf_ematch_ops *ops)
128 int err = -EEXIST;
129 struct tcf_ematch_ops *e;
131 if (ops->match == NULL)
132 return -EINVAL;
134 write_lock(&ematch_mod_lock);
135 list_for_each_entry(e, &ematch_ops, link)
136 if (ops->kind == e->kind)
137 goto errout;
139 list_add_tail(&ops->link, &ematch_ops);
140 err = 0;
141 errout:
142 write_unlock(&ematch_mod_lock);
143 return err;
145 EXPORT_SYMBOL(tcf_em_register);
148 * tcf_em_unregister - unregster and extended match
150 * @ops: ematch operations lookup table
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.
156 * Returns -ENOENT if no matching ematch was found.
158 void tcf_em_unregister(struct tcf_ematch_ops *ops)
160 write_lock(&ematch_mod_lock);
161 list_del(&ops->link);
162 write_unlock(&ematch_mod_lock);
164 EXPORT_SYMBOL(tcf_em_unregister);
166 static inline struct tcf_ematch * tcf_em_get_match(struct tcf_ematch_tree *tree,
167 int index)
169 return &tree->matches[index];
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)
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);
182 if (!TCF_EM_REL_VALID(em_hdr->flags))
183 goto errout;
185 if (em_hdr->kind == TCF_EM_CONTAINER) {
186 /* Special ematch called "container", carries an index
187 * referencing an external ematch sequence. */
188 u32 ref;
190 if (data_len < sizeof(ref))
191 goto errout;
192 ref = *(u32 *) data;
194 if (ref >= tree_hdr->nmatches)
195 goto errout;
197 /* We do not allow backward jumps to avoid loops and jumps
198 * to our own position are of course illegal. */
199 if (ref <= idx)
200 goto errout;
203 em->data = ref;
204 } else {
205 /* Note: This lookup will increase the module refcnt
206 * of the ematch module referenced. In case of a failure,
207 * a destroy function is called by the underlying layer
208 * which automatically releases the reference again, therefore
209 * the module MUST not be given back under any circumstances
210 * here. Be aware, the destroy function assumes that the
211 * module is held if the ops field is non zero. */
212 em->ops = tcf_em_lookup(em_hdr->kind);
214 if (em->ops == NULL) {
215 err = -ENOENT;
216 #ifdef CONFIG_MODULES
217 __rtnl_unlock();
218 request_module("ematch-kind-%u", em_hdr->kind);
219 rtnl_lock();
220 em->ops = tcf_em_lookup(em_hdr->kind);
221 if (em->ops) {
222 /* We dropped the RTNL mutex in order to
223 * perform the module load. Tell the caller
224 * to replay the request. */
225 module_put(em->ops->owner);
226 err = -EAGAIN;
228 #endif
229 goto errout;
232 /* ematch module provides expected length of data, so we
233 * can do a basic sanity check. */
234 if (em->ops->datalen && data_len < em->ops->datalen)
235 goto errout;
237 if (em->ops->change) {
238 err = em->ops->change(tp, data, data_len, em);
239 if (err < 0)
240 goto errout;
241 } else if (data_len > 0) {
242 /* ematch module doesn't provide an own change
243 * procedure and expects us to allocate and copy
244 * the ematch data.
246 * TCF_EM_SIMPLE may be specified stating that the
247 * data only consists of a u32 integer and the module
248 * does not expected a memory reference but rather
249 * the value carried. */
250 if (em_hdr->flags & TCF_EM_SIMPLE) {
251 if (data_len < sizeof(u32))
252 goto errout;
253 em->data = *(u32 *) data;
254 } else {
255 void *v = kmemdup(data, data_len, GFP_KERNEL);
256 if (v == NULL) {
257 err = -ENOBUFS;
258 goto errout;
260 em->data = (unsigned long) v;
265 em->matchid = em_hdr->matchid;
266 em->flags = em_hdr->flags;
267 em->datalen = data_len;
269 err = 0;
270 errout:
271 return err;
274 static const struct nla_policy em_policy[TCA_EMATCH_TREE_MAX + 1] = {
275 [TCA_EMATCH_TREE_HDR] = { .len = sizeof(struct tcf_ematch_tree_hdr) },
276 [TCA_EMATCH_TREE_LIST] = { .type = NLA_NESTED },
280 * tcf_em_tree_validate - validate ematch config TLV and build ematch tree
282 * @tp: classifier kind handle
283 * @nla: ematch tree configuration TLV
284 * @tree: destination ematch tree variable to store the resulting
285 * ematch tree.
287 * This function validates the given configuration TLV @nla and builds an
288 * ematch tree in @tree. The resulting tree must later be copied into
289 * the private classifier data using tcf_em_tree_change(). You MUST NOT
290 * provide the ematch tree variable of the private classifier data directly,
291 * the changes would not be locked properly.
293 * Returns a negative error code if the configuration TLV contains errors.
295 int tcf_em_tree_validate(struct tcf_proto *tp, struct nlattr *nla,
296 struct tcf_ematch_tree *tree)
298 int idx, list_len, matches_len, err;
299 struct nlattr *tb[TCA_EMATCH_TREE_MAX + 1];
300 struct nlattr *rt_match, *rt_hdr, *rt_list;
301 struct tcf_ematch_tree_hdr *tree_hdr;
302 struct tcf_ematch *em;
304 memset(tree, 0, sizeof(*tree));
305 if (!nla)
306 return 0;
308 err = nla_parse_nested(tb, TCA_EMATCH_TREE_MAX, nla, em_policy);
309 if (err < 0)
310 goto errout;
312 err = -EINVAL;
313 rt_hdr = tb[TCA_EMATCH_TREE_HDR];
314 rt_list = tb[TCA_EMATCH_TREE_LIST];
316 if (rt_hdr == NULL || rt_list == NULL)
317 goto errout;
319 tree_hdr = nla_data(rt_hdr);
320 memcpy(&tree->hdr, tree_hdr, sizeof(*tree_hdr));
322 rt_match = nla_data(rt_list);
323 list_len = nla_len(rt_list);
324 matches_len = tree_hdr->nmatches * sizeof(*em);
326 tree->matches = kzalloc(matches_len, GFP_KERNEL);
327 if (tree->matches == NULL)
328 goto errout;
330 /* We do not use nla_parse_nested here because the maximum
331 * number of attributes is unknown. This saves us the allocation
332 * for a tb buffer which would serve no purpose at all.
334 * The array of rt attributes is parsed in the order as they are
335 * provided, their type must be incremental from 1 to n. Even
336 * if it does not serve any real purpose, a failure of sticking
337 * to this policy will result in parsing failure. */
338 for (idx = 0; nla_ok(rt_match, list_len); idx++) {
339 err = -EINVAL;
341 if (rt_match->nla_type != (idx + 1))
342 goto errout_abort;
344 if (idx >= tree_hdr->nmatches)
345 goto errout_abort;
347 if (nla_len(rt_match) < sizeof(struct tcf_ematch_hdr))
348 goto errout_abort;
350 em = tcf_em_get_match(tree, idx);
352 err = tcf_em_validate(tp, tree_hdr, em, rt_match, idx);
353 if (err < 0)
354 goto errout_abort;
356 rt_match = nla_next(rt_match, &list_len);
359 /* Check if the number of matches provided by userspace actually
360 * complies with the array of matches. The number was used for
361 * the validation of references and a mismatch could lead to
362 * undefined references during the matching process. */
363 if (idx != tree_hdr->nmatches) {
364 err = -EINVAL;
365 goto errout_abort;
368 err = 0;
369 errout:
370 return err;
372 errout_abort:
373 tcf_em_tree_destroy(tp, tree);
374 return err;
376 EXPORT_SYMBOL(tcf_em_tree_validate);
379 * tcf_em_tree_destroy - destroy an ematch tree
381 * @tp: classifier kind handle
382 * @tree: ematch tree to be deleted
384 * This functions destroys an ematch tree previously created by
385 * tcf_em_tree_validate()/tcf_em_tree_change(). You must ensure that
386 * the ematch tree is not in use before calling this function.
388 void tcf_em_tree_destroy(struct tcf_proto *tp, struct tcf_ematch_tree *tree)
390 int i;
392 if (tree->matches == NULL)
393 return;
395 for (i = 0; i < tree->hdr.nmatches; i++) {
396 struct tcf_ematch *em = tcf_em_get_match(tree, i);
398 if (em->ops) {
399 if (em->ops->destroy)
400 em->ops->destroy(tp, em);
401 else if (!tcf_em_is_simple(em))
402 kfree((void *) em->data);
403 module_put(em->ops->owner);
407 tree->hdr.nmatches = 0;
408 kfree(tree->matches);
409 tree->matches = NULL;
411 EXPORT_SYMBOL(tcf_em_tree_destroy);
414 * tcf_em_tree_dump - dump ematch tree into a rtnl message
416 * @skb: skb holding the rtnl message
417 * @t: ematch tree to be dumped
418 * @tlv: TLV type to be used to encapsulate the tree
420 * This function dumps a ematch tree into a rtnl message. It is valid to
421 * call this function while the ematch tree is in use.
423 * Returns -1 if the skb tailroom is insufficient.
425 int tcf_em_tree_dump(struct sk_buff *skb, struct tcf_ematch_tree *tree, int tlv)
427 int i;
428 u8 *tail;
429 struct nlattr *top_start;
430 struct nlattr *list_start;
432 top_start = nla_nest_start(skb, tlv);
433 if (top_start == NULL)
434 goto nla_put_failure;
436 NLA_PUT(skb, TCA_EMATCH_TREE_HDR, sizeof(tree->hdr), &tree->hdr);
438 list_start = nla_nest_start(skb, TCA_EMATCH_TREE_LIST);
439 if (list_start == NULL)
440 goto nla_put_failure;
442 tail = skb_tail_pointer(skb);
443 for (i = 0; i < tree->hdr.nmatches; i++) {
444 struct nlattr *match_start = (struct nlattr *)tail;
445 struct tcf_ematch *em = tcf_em_get_match(tree, i);
446 struct tcf_ematch_hdr em_hdr = {
447 .kind = em->ops ? em->ops->kind : TCF_EM_CONTAINER,
448 .matchid = em->matchid,
449 .flags = em->flags
452 NLA_PUT(skb, i+1, sizeof(em_hdr), &em_hdr);
454 if (em->ops && em->ops->dump) {
455 if (em->ops->dump(skb, em) < 0)
456 goto nla_put_failure;
457 } else if (tcf_em_is_container(em) || tcf_em_is_simple(em)) {
458 u32 u = em->data;
459 nla_put_nohdr(skb, sizeof(u), &u);
460 } else if (em->datalen > 0)
461 nla_put_nohdr(skb, em->datalen, (void *) em->data);
463 tail = skb_tail_pointer(skb);
464 match_start->nla_len = tail - (u8 *)match_start;
467 nla_nest_end(skb, list_start);
468 nla_nest_end(skb, top_start);
470 return 0;
472 nla_put_failure:
473 return -1;
475 EXPORT_SYMBOL(tcf_em_tree_dump);
477 static inline int tcf_em_match(struct sk_buff *skb, struct tcf_ematch *em,
478 struct tcf_pkt_info *info)
480 int r = em->ops->match(skb, em, info);
481 return tcf_em_is_inverted(em) ? !r : r;
484 /* Do not use this function directly, use tcf_em_tree_match instead */
485 int __tcf_em_tree_match(struct sk_buff *skb, struct tcf_ematch_tree *tree,
486 struct tcf_pkt_info *info)
488 int stackp = 0, match_idx = 0, res = 0;
489 struct tcf_ematch *cur_match;
490 int stack[CONFIG_NET_EMATCH_STACK];
492 proceed:
493 while (match_idx < tree->hdr.nmatches) {
494 cur_match = tcf_em_get_match(tree, match_idx);
496 if (tcf_em_is_container(cur_match)) {
497 if (unlikely(stackp >= CONFIG_NET_EMATCH_STACK))
498 goto stack_overflow;
500 stack[stackp++] = match_idx;
501 match_idx = cur_match->data;
502 goto proceed;
505 res = tcf_em_match(skb, cur_match, info);
507 if (tcf_em_early_end(cur_match, res))
508 break;
510 match_idx++;
513 pop_stack:
514 if (stackp > 0) {
515 match_idx = stack[--stackp];
516 cur_match = tcf_em_get_match(tree, match_idx);
518 if (tcf_em_early_end(cur_match, res))
519 goto pop_stack;
520 else {
521 match_idx++;
522 goto proceed;
526 return res;
528 stack_overflow:
529 if (net_ratelimit())
530 printk(KERN_WARNING "tc ematch: local stack overflow,"
531 " increase NET_EMATCH_STACK\n");
532 return -1;
534 EXPORT_SYMBOL(__tcf_em_tree_match);