nfs: don't try to migrate pages with active requests
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / sched / ematch.c
blob88d93eb92507be2f80fd90db12a63b11a7e87aeb
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 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.
189 u32 ref;
191 if (data_len < sizeof(ref))
192 goto errout;
193 ref = *(u32 *) data;
195 if (ref >= tree_hdr->nmatches)
196 goto errout;
198 /* We do not allow backward jumps to avoid loops and jumps
199 * to our own position are of course illegal.
201 if (ref <= idx)
202 goto errout;
205 em->data = ref;
206 } else {
207 /* Note: This lookup will increase the module refcnt
208 * of the ematch module referenced. In case of a failure,
209 * a destroy function is called by the underlying layer
210 * which automatically releases the reference again, therefore
211 * the module MUST not be given back under any circumstances
212 * here. Be aware, the destroy function assumes that the
213 * module is held if the ops field is non zero.
215 em->ops = tcf_em_lookup(em_hdr->kind);
217 if (em->ops == NULL) {
218 err = -ENOENT;
219 #ifdef CONFIG_MODULES
220 __rtnl_unlock();
221 request_module("ematch-kind-%u", em_hdr->kind);
222 rtnl_lock();
223 em->ops = tcf_em_lookup(em_hdr->kind);
224 if (em->ops) {
225 /* We dropped the RTNL mutex in order to
226 * perform the module load. Tell the caller
227 * to replay the request.
229 module_put(em->ops->owner);
230 err = -EAGAIN;
232 #endif
233 goto errout;
236 /* ematch module provides expected length of data, so we
237 * can do a basic sanity check.
239 if (em->ops->datalen && data_len < em->ops->datalen)
240 goto errout;
242 if (em->ops->change) {
243 err = em->ops->change(tp, data, data_len, em);
244 if (err < 0)
245 goto errout;
246 } else if (data_len > 0) {
247 /* ematch module doesn't provide an own change
248 * procedure and expects us to allocate and copy
249 * the ematch data.
251 * TCF_EM_SIMPLE may be specified stating that the
252 * data only consists of a u32 integer and the module
253 * does not expected a memory reference but rather
254 * the value carried.
256 if (em_hdr->flags & TCF_EM_SIMPLE) {
257 if (data_len < sizeof(u32))
258 goto errout;
259 em->data = *(u32 *) data;
260 } else {
261 void *v = kmemdup(data, data_len, GFP_KERNEL);
262 if (v == NULL) {
263 err = -ENOBUFS;
264 goto errout;
266 em->data = (unsigned long) v;
271 em->matchid = em_hdr->matchid;
272 em->flags = em_hdr->flags;
273 em->datalen = data_len;
275 err = 0;
276 errout:
277 return err;
280 static const struct nla_policy em_policy[TCA_EMATCH_TREE_MAX + 1] = {
281 [TCA_EMATCH_TREE_HDR] = { .len = sizeof(struct tcf_ematch_tree_hdr) },
282 [TCA_EMATCH_TREE_LIST] = { .type = NLA_NESTED },
286 * tcf_em_tree_validate - validate ematch config TLV and build ematch tree
288 * @tp: classifier kind handle
289 * @nla: ematch tree configuration TLV
290 * @tree: destination ematch tree variable to store the resulting
291 * ematch tree.
293 * This function validates the given configuration TLV @nla and builds an
294 * ematch tree in @tree. The resulting tree must later be copied into
295 * the private classifier data using tcf_em_tree_change(). You MUST NOT
296 * provide the ematch tree variable of the private classifier data directly,
297 * the changes would not be locked properly.
299 * Returns a negative error code if the configuration TLV contains errors.
301 int tcf_em_tree_validate(struct tcf_proto *tp, struct nlattr *nla,
302 struct tcf_ematch_tree *tree)
304 int idx, list_len, matches_len, err;
305 struct nlattr *tb[TCA_EMATCH_TREE_MAX + 1];
306 struct nlattr *rt_match, *rt_hdr, *rt_list;
307 struct tcf_ematch_tree_hdr *tree_hdr;
308 struct tcf_ematch *em;
310 memset(tree, 0, sizeof(*tree));
311 if (!nla)
312 return 0;
314 err = nla_parse_nested(tb, TCA_EMATCH_TREE_MAX, nla, em_policy);
315 if (err < 0)
316 goto errout;
318 err = -EINVAL;
319 rt_hdr = tb[TCA_EMATCH_TREE_HDR];
320 rt_list = tb[TCA_EMATCH_TREE_LIST];
322 if (rt_hdr == NULL || rt_list == NULL)
323 goto errout;
325 tree_hdr = nla_data(rt_hdr);
326 memcpy(&tree->hdr, tree_hdr, sizeof(*tree_hdr));
328 rt_match = nla_data(rt_list);
329 list_len = nla_len(rt_list);
330 matches_len = tree_hdr->nmatches * sizeof(*em);
332 tree->matches = kzalloc(matches_len, GFP_KERNEL);
333 if (tree->matches == NULL)
334 goto errout;
336 /* We do not use nla_parse_nested here because the maximum
337 * number of attributes is unknown. This saves us the allocation
338 * for a tb buffer which would serve no purpose at all.
340 * The array of rt attributes is parsed in the order as they are
341 * provided, their type must be incremental from 1 to n. Even
342 * if it does not serve any real purpose, a failure of sticking
343 * to this policy will result in parsing failure.
345 for (idx = 0; nla_ok(rt_match, list_len); idx++) {
346 err = -EINVAL;
348 if (rt_match->nla_type != (idx + 1))
349 goto errout_abort;
351 if (idx >= tree_hdr->nmatches)
352 goto errout_abort;
354 if (nla_len(rt_match) < sizeof(struct tcf_ematch_hdr))
355 goto errout_abort;
357 em = tcf_em_get_match(tree, idx);
359 err = tcf_em_validate(tp, tree_hdr, em, rt_match, idx);
360 if (err < 0)
361 goto errout_abort;
363 rt_match = nla_next(rt_match, &list_len);
366 /* Check if the number of matches provided by userspace actually
367 * complies with the array of matches. The number was used for
368 * the validation of references and a mismatch could lead to
369 * undefined references during the matching process.
371 if (idx != tree_hdr->nmatches) {
372 err = -EINVAL;
373 goto errout_abort;
376 err = 0;
377 errout:
378 return err;
380 errout_abort:
381 tcf_em_tree_destroy(tp, tree);
382 return err;
384 EXPORT_SYMBOL(tcf_em_tree_validate);
387 * tcf_em_tree_destroy - destroy an ematch tree
389 * @tp: classifier kind handle
390 * @tree: ematch tree to be deleted
392 * This functions destroys an ematch tree previously created by
393 * tcf_em_tree_validate()/tcf_em_tree_change(). You must ensure that
394 * the ematch tree is not in use before calling this function.
396 void tcf_em_tree_destroy(struct tcf_proto *tp, struct tcf_ematch_tree *tree)
398 int i;
400 if (tree->matches == NULL)
401 return;
403 for (i = 0; i < tree->hdr.nmatches; i++) {
404 struct tcf_ematch *em = tcf_em_get_match(tree, i);
406 if (em->ops) {
407 if (em->ops->destroy)
408 em->ops->destroy(tp, em);
409 else if (!tcf_em_is_simple(em))
410 kfree((void *) em->data);
411 module_put(em->ops->owner);
415 tree->hdr.nmatches = 0;
416 kfree(tree->matches);
417 tree->matches = NULL;
419 EXPORT_SYMBOL(tcf_em_tree_destroy);
422 * tcf_em_tree_dump - dump ematch tree into a rtnl message
424 * @skb: skb holding the rtnl message
425 * @t: ematch tree to be dumped
426 * @tlv: TLV type to be used to encapsulate the tree
428 * This function dumps a ematch tree into a rtnl message. It is valid to
429 * call this function while the ematch tree is in use.
431 * Returns -1 if the skb tailroom is insufficient.
433 int tcf_em_tree_dump(struct sk_buff *skb, struct tcf_ematch_tree *tree, int tlv)
435 int i;
436 u8 *tail;
437 struct nlattr *top_start;
438 struct nlattr *list_start;
440 top_start = nla_nest_start(skb, tlv);
441 if (top_start == NULL)
442 goto nla_put_failure;
444 NLA_PUT(skb, TCA_EMATCH_TREE_HDR, sizeof(tree->hdr), &tree->hdr);
446 list_start = nla_nest_start(skb, TCA_EMATCH_TREE_LIST);
447 if (list_start == NULL)
448 goto nla_put_failure;
450 tail = skb_tail_pointer(skb);
451 for (i = 0; i < tree->hdr.nmatches; i++) {
452 struct nlattr *match_start = (struct nlattr *)tail;
453 struct tcf_ematch *em = tcf_em_get_match(tree, i);
454 struct tcf_ematch_hdr em_hdr = {
455 .kind = em->ops ? em->ops->kind : TCF_EM_CONTAINER,
456 .matchid = em->matchid,
457 .flags = em->flags
460 NLA_PUT(skb, i + 1, sizeof(em_hdr), &em_hdr);
462 if (em->ops && em->ops->dump) {
463 if (em->ops->dump(skb, em) < 0)
464 goto nla_put_failure;
465 } else if (tcf_em_is_container(em) || tcf_em_is_simple(em)) {
466 u32 u = em->data;
467 nla_put_nohdr(skb, sizeof(u), &u);
468 } else if (em->datalen > 0)
469 nla_put_nohdr(skb, em->datalen, (void *) em->data);
471 tail = skb_tail_pointer(skb);
472 match_start->nla_len = tail - (u8 *)match_start;
475 nla_nest_end(skb, list_start);
476 nla_nest_end(skb, top_start);
478 return 0;
480 nla_put_failure:
481 return -1;
483 EXPORT_SYMBOL(tcf_em_tree_dump);
485 static inline int tcf_em_match(struct sk_buff *skb, struct tcf_ematch *em,
486 struct tcf_pkt_info *info)
488 int r = em->ops->match(skb, em, info);
490 return tcf_em_is_inverted(em) ? !r : r;
493 /* Do not use this function directly, use tcf_em_tree_match instead */
494 int __tcf_em_tree_match(struct sk_buff *skb, struct tcf_ematch_tree *tree,
495 struct tcf_pkt_info *info)
497 int stackp = 0, match_idx = 0, res = 0;
498 struct tcf_ematch *cur_match;
499 int stack[CONFIG_NET_EMATCH_STACK];
501 proceed:
502 while (match_idx < tree->hdr.nmatches) {
503 cur_match = tcf_em_get_match(tree, match_idx);
505 if (tcf_em_is_container(cur_match)) {
506 if (unlikely(stackp >= CONFIG_NET_EMATCH_STACK))
507 goto stack_overflow;
509 stack[stackp++] = match_idx;
510 match_idx = cur_match->data;
511 goto proceed;
514 res = tcf_em_match(skb, cur_match, info);
516 if (tcf_em_early_end(cur_match, res))
517 break;
519 match_idx++;
522 pop_stack:
523 if (stackp > 0) {
524 match_idx = stack[--stackp];
525 cur_match = tcf_em_get_match(tree, match_idx);
527 if (tcf_em_early_end(cur_match, res))
528 goto pop_stack;
529 else {
530 match_idx++;
531 goto proceed;
535 return res;
537 stack_overflow:
538 if (net_ratelimit())
539 pr_warning("tc ematch: local stack overflow,"
540 " increase NET_EMATCH_STACK\n");
541 return -1;
543 EXPORT_SYMBOL(__tcf_em_tree_match);