xref: /openbmc/linux/net/sched/cls_api.c (revision 96de0e252cedffad61b3cb5e05662c591898e69a)
1 /*
2  * net/sched/cls_api.c	Packet classifier 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:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  *
11  * Changes:
12  *
13  * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
14  *
15  */
16 
17 #include <linux/module.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/string.h>
21 #include <linux/errno.h>
22 #include <linux/skbuff.h>
23 #include <linux/init.h>
24 #include <linux/kmod.h>
25 #include <linux/netlink.h>
26 #include <net/netlink.h>
27 #include <net/pkt_sched.h>
28 #include <net/pkt_cls.h>
29 
30 #if 0 /* control */
31 #define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
32 #else
33 #define DPRINTK(format,args...)
34 #endif
35 
36 /* The list of all installed classifier types */
37 
38 static struct tcf_proto_ops *tcf_proto_base;
39 
40 /* Protects list of registered TC modules. It is pure SMP lock. */
41 static DEFINE_RWLOCK(cls_mod_lock);
42 
43 /* Find classifier type by string name */
44 
45 static struct tcf_proto_ops * tcf_proto_lookup_ops(struct rtattr *kind)
46 {
47 	struct tcf_proto_ops *t = NULL;
48 
49 	if (kind) {
50 		read_lock(&cls_mod_lock);
51 		for (t = tcf_proto_base; t; t = t->next) {
52 			if (rtattr_strcmp(kind, t->kind) == 0) {
53 				if (!try_module_get(t->owner))
54 					t = NULL;
55 				break;
56 			}
57 		}
58 		read_unlock(&cls_mod_lock);
59 	}
60 	return t;
61 }
62 
63 /* Register(unregister) new classifier type */
64 
65 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
66 {
67 	struct tcf_proto_ops *t, **tp;
68 	int rc = -EEXIST;
69 
70 	write_lock(&cls_mod_lock);
71 	for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
72 		if (!strcmp(ops->kind, t->kind))
73 			goto out;
74 
75 	ops->next = NULL;
76 	*tp = ops;
77 	rc = 0;
78 out:
79 	write_unlock(&cls_mod_lock);
80 	return rc;
81 }
82 
83 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
84 {
85 	struct tcf_proto_ops *t, **tp;
86 	int rc = -ENOENT;
87 
88 	write_lock(&cls_mod_lock);
89 	for (tp = &tcf_proto_base; (t=*tp) != NULL; tp = &t->next)
90 		if (t == ops)
91 			break;
92 
93 	if (!t)
94 		goto out;
95 	*tp = t->next;
96 	rc = 0;
97 out:
98 	write_unlock(&cls_mod_lock);
99 	return rc;
100 }
101 
102 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
103 			  struct tcf_proto *tp, unsigned long fh, int event);
104 
105 
106 /* Select new prio value from the range, managed by kernel. */
107 
108 static __inline__ u32 tcf_auto_prio(struct tcf_proto *tp)
109 {
110 	u32 first = TC_H_MAKE(0xC0000000U,0U);
111 
112 	if (tp)
113 		first = tp->prio-1;
114 
115 	return first;
116 }
117 
118 /* Add/change/delete/get a filter node */
119 
120 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
121 {
122 	struct rtattr **tca;
123 	struct tcmsg *t;
124 	u32 protocol;
125 	u32 prio;
126 	u32 nprio;
127 	u32 parent;
128 	struct net_device *dev;
129 	struct Qdisc  *q;
130 	struct tcf_proto **back, **chain;
131 	struct tcf_proto *tp;
132 	struct tcf_proto_ops *tp_ops;
133 	struct Qdisc_class_ops *cops;
134 	unsigned long cl;
135 	unsigned long fh;
136 	int err;
137 
138 replay:
139 	tca = arg;
140 	t = NLMSG_DATA(n);
141 	protocol = TC_H_MIN(t->tcm_info);
142 	prio = TC_H_MAJ(t->tcm_info);
143 	nprio = prio;
144 	parent = t->tcm_parent;
145 	cl = 0;
146 
147 	if (prio == 0) {
148 		/* If no priority is given, user wants we allocated it. */
149 		if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
150 			return -ENOENT;
151 		prio = TC_H_MAKE(0x80000000U,0U);
152 	}
153 
154 	/* Find head of filter chain. */
155 
156 	/* Find link */
157 	if ((dev = __dev_get_by_index(&init_net, t->tcm_ifindex)) == NULL)
158 		return -ENODEV;
159 
160 	/* Find qdisc */
161 	if (!parent) {
162 		q = dev->qdisc_sleeping;
163 		parent = q->handle;
164 	} else if ((q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent))) == NULL)
165 		return -EINVAL;
166 
167 	/* Is it classful? */
168 	if ((cops = q->ops->cl_ops) == NULL)
169 		return -EINVAL;
170 
171 	/* Do we search for filter, attached to class? */
172 	if (TC_H_MIN(parent)) {
173 		cl = cops->get(q, parent);
174 		if (cl == 0)
175 			return -ENOENT;
176 	}
177 
178 	/* And the last stroke */
179 	chain = cops->tcf_chain(q, cl);
180 	err = -EINVAL;
181 	if (chain == NULL)
182 		goto errout;
183 
184 	/* Check the chain for existence of proto-tcf with this priority */
185 	for (back = chain; (tp=*back) != NULL; back = &tp->next) {
186 		if (tp->prio >= prio) {
187 			if (tp->prio == prio) {
188 				if (!nprio || (tp->protocol != protocol && protocol))
189 					goto errout;
190 			} else
191 				tp = NULL;
192 			break;
193 		}
194 	}
195 
196 	if (tp == NULL) {
197 		/* Proto-tcf does not exist, create new one */
198 
199 		if (tca[TCA_KIND-1] == NULL || !protocol)
200 			goto errout;
201 
202 		err = -ENOENT;
203 		if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
204 			goto errout;
205 
206 
207 		/* Create new proto tcf */
208 
209 		err = -ENOBUFS;
210 		if ((tp = kzalloc(sizeof(*tp), GFP_KERNEL)) == NULL)
211 			goto errout;
212 		err = -EINVAL;
213 		tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND-1]);
214 		if (tp_ops == NULL) {
215 #ifdef CONFIG_KMOD
216 			struct rtattr *kind = tca[TCA_KIND-1];
217 			char name[IFNAMSIZ];
218 
219 			if (kind != NULL &&
220 			    rtattr_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
221 				rtnl_unlock();
222 				request_module("cls_%s", name);
223 				rtnl_lock();
224 				tp_ops = tcf_proto_lookup_ops(kind);
225 				/* We dropped the RTNL semaphore in order to
226 				 * perform the module load.  So, even if we
227 				 * succeeded in loading the module we have to
228 				 * replay the request.  We indicate this using
229 				 * -EAGAIN.
230 				 */
231 				if (tp_ops != NULL) {
232 					module_put(tp_ops->owner);
233 					err = -EAGAIN;
234 				}
235 			}
236 #endif
237 			kfree(tp);
238 			goto errout;
239 		}
240 		tp->ops = tp_ops;
241 		tp->protocol = protocol;
242 		tp->prio = nprio ? : tcf_auto_prio(*back);
243 		tp->q = q;
244 		tp->classify = tp_ops->classify;
245 		tp->classid = parent;
246 		if ((err = tp_ops->init(tp)) != 0) {
247 			module_put(tp_ops->owner);
248 			kfree(tp);
249 			goto errout;
250 		}
251 
252 		qdisc_lock_tree(dev);
253 		tp->next = *back;
254 		*back = tp;
255 		qdisc_unlock_tree(dev);
256 
257 	} else if (tca[TCA_KIND-1] && rtattr_strcmp(tca[TCA_KIND-1], tp->ops->kind))
258 		goto errout;
259 
260 	fh = tp->ops->get(tp, t->tcm_handle);
261 
262 	if (fh == 0) {
263 		if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
264 			qdisc_lock_tree(dev);
265 			*back = tp->next;
266 			qdisc_unlock_tree(dev);
267 
268 			tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
269 			tcf_destroy(tp);
270 			err = 0;
271 			goto errout;
272 		}
273 
274 		err = -ENOENT;
275 		if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
276 			goto errout;
277 	} else {
278 		switch (n->nlmsg_type) {
279 		case RTM_NEWTFILTER:
280 			err = -EEXIST;
281 			if (n->nlmsg_flags&NLM_F_EXCL)
282 				goto errout;
283 			break;
284 		case RTM_DELTFILTER:
285 			err = tp->ops->delete(tp, fh);
286 			if (err == 0)
287 				tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
288 			goto errout;
289 		case RTM_GETTFILTER:
290 			err = tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
291 			goto errout;
292 		default:
293 			err = -EINVAL;
294 			goto errout;
295 		}
296 	}
297 
298 	err = tp->ops->change(tp, cl, t->tcm_handle, tca, &fh);
299 	if (err == 0)
300 		tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
301 
302 errout:
303 	if (cl)
304 		cops->put(q, cl);
305 	if (err == -EAGAIN)
306 		/* Replay the request. */
307 		goto replay;
308 	return err;
309 }
310 
311 static int
312 tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp, unsigned long fh,
313 	      u32 pid, u32 seq, u16 flags, int event)
314 {
315 	struct tcmsg *tcm;
316 	struct nlmsghdr  *nlh;
317 	unsigned char *b = skb_tail_pointer(skb);
318 
319 	nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
320 	tcm = NLMSG_DATA(nlh);
321 	tcm->tcm_family = AF_UNSPEC;
322 	tcm->tcm__pad1 = 0;
323 	tcm->tcm__pad1 = 0;
324 	tcm->tcm_ifindex = tp->q->dev->ifindex;
325 	tcm->tcm_parent = tp->classid;
326 	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
327 	RTA_PUT(skb, TCA_KIND, IFNAMSIZ, tp->ops->kind);
328 	tcm->tcm_handle = fh;
329 	if (RTM_DELTFILTER != event) {
330 		tcm->tcm_handle = 0;
331 		if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
332 			goto rtattr_failure;
333 	}
334 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
335 	return skb->len;
336 
337 nlmsg_failure:
338 rtattr_failure:
339 	nlmsg_trim(skb, b);
340 	return -1;
341 }
342 
343 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
344 			  struct tcf_proto *tp, unsigned long fh, int event)
345 {
346 	struct sk_buff *skb;
347 	u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
348 
349 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
350 	if (!skb)
351 		return -ENOBUFS;
352 
353 	if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
354 		kfree_skb(skb);
355 		return -EINVAL;
356 	}
357 
358 	return rtnetlink_send(skb, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
359 }
360 
361 struct tcf_dump_args
362 {
363 	struct tcf_walker w;
364 	struct sk_buff *skb;
365 	struct netlink_callback *cb;
366 };
367 
368 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n, struct tcf_walker *arg)
369 {
370 	struct tcf_dump_args *a = (void*)arg;
371 
372 	return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
373 			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
374 }
375 
376 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
377 {
378 	int t;
379 	int s_t;
380 	struct net_device *dev;
381 	struct Qdisc *q;
382 	struct tcf_proto *tp, **chain;
383 	struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
384 	unsigned long cl = 0;
385 	struct Qdisc_class_ops *cops;
386 	struct tcf_dump_args arg;
387 
388 	if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
389 		return skb->len;
390 	if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
391 		return skb->len;
392 
393 	if (!tcm->tcm_parent)
394 		q = dev->qdisc_sleeping;
395 	else
396 		q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
397 	if (!q)
398 		goto out;
399 	if ((cops = q->ops->cl_ops) == NULL)
400 		goto errout;
401 	if (TC_H_MIN(tcm->tcm_parent)) {
402 		cl = cops->get(q, tcm->tcm_parent);
403 		if (cl == 0)
404 			goto errout;
405 	}
406 	chain = cops->tcf_chain(q, cl);
407 	if (chain == NULL)
408 		goto errout;
409 
410 	s_t = cb->args[0];
411 
412 	for (tp=*chain, t=0; tp; tp = tp->next, t++) {
413 		if (t < s_t) continue;
414 		if (TC_H_MAJ(tcm->tcm_info) &&
415 		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
416 			continue;
417 		if (TC_H_MIN(tcm->tcm_info) &&
418 		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
419 			continue;
420 		if (t > s_t)
421 			memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
422 		if (cb->args[1] == 0) {
423 			if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
424 					  cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER) <= 0) {
425 				break;
426 			}
427 			cb->args[1] = 1;
428 		}
429 		if (tp->ops->walk == NULL)
430 			continue;
431 		arg.w.fn = tcf_node_dump;
432 		arg.skb = skb;
433 		arg.cb = cb;
434 		arg.w.stop = 0;
435 		arg.w.skip = cb->args[1]-1;
436 		arg.w.count = 0;
437 		tp->ops->walk(tp, &arg.w);
438 		cb->args[1] = arg.w.count+1;
439 		if (arg.w.stop)
440 			break;
441 	}
442 
443 	cb->args[0] = t;
444 
445 errout:
446 	if (cl)
447 		cops->put(q, cl);
448 out:
449 	dev_put(dev);
450 	return skb->len;
451 }
452 
453 void
454 tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
455 {
456 #ifdef CONFIG_NET_CLS_ACT
457 	if (exts->action) {
458 		tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
459 		exts->action = NULL;
460 	}
461 #endif
462 }
463 
464 
465 int
466 tcf_exts_validate(struct tcf_proto *tp, struct rtattr **tb,
467 		  struct rtattr *rate_tlv, struct tcf_exts *exts,
468 		  struct tcf_ext_map *map)
469 {
470 	memset(exts, 0, sizeof(*exts));
471 
472 #ifdef CONFIG_NET_CLS_ACT
473 	{
474 		int err;
475 		struct tc_action *act;
476 
477 		if (map->police && tb[map->police-1]) {
478 			act = tcf_action_init_1(tb[map->police-1], rate_tlv, "police",
479 				TCA_ACT_NOREPLACE, TCA_ACT_BIND, &err);
480 			if (act == NULL)
481 				return err;
482 
483 			act->type = TCA_OLD_COMPAT;
484 			exts->action = act;
485 		} else if (map->action && tb[map->action-1]) {
486 			act = tcf_action_init(tb[map->action-1], rate_tlv, NULL,
487 				TCA_ACT_NOREPLACE, TCA_ACT_BIND, &err);
488 			if (act == NULL)
489 				return err;
490 
491 			exts->action = act;
492 		}
493 	}
494 #else
495 	if ((map->action && tb[map->action-1]) ||
496 	    (map->police && tb[map->police-1]))
497 		return -EOPNOTSUPP;
498 #endif
499 
500 	return 0;
501 }
502 
503 void
504 tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
505 		struct tcf_exts *src)
506 {
507 #ifdef CONFIG_NET_CLS_ACT
508 	if (src->action) {
509 		struct tc_action *act;
510 		tcf_tree_lock(tp);
511 		act = xchg(&dst->action, src->action);
512 		tcf_tree_unlock(tp);
513 		if (act)
514 			tcf_action_destroy(act, TCA_ACT_UNBIND);
515 	}
516 #endif
517 }
518 
519 int
520 tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
521 	      struct tcf_ext_map *map)
522 {
523 #ifdef CONFIG_NET_CLS_ACT
524 	if (map->action && exts->action) {
525 		/*
526 		 * again for backward compatible mode - we want
527 		 * to work with both old and new modes of entering
528 		 * tc data even if iproute2  was newer - jhs
529 		 */
530 		struct rtattr *p_rta = (struct rtattr *)skb_tail_pointer(skb);
531 
532 		if (exts->action->type != TCA_OLD_COMPAT) {
533 			RTA_PUT(skb, map->action, 0, NULL);
534 			if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
535 				goto rtattr_failure;
536 			p_rta->rta_len = skb_tail_pointer(skb) - (u8 *)p_rta;
537 		} else if (map->police) {
538 			RTA_PUT(skb, map->police, 0, NULL);
539 			if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
540 				goto rtattr_failure;
541 			p_rta->rta_len = skb_tail_pointer(skb) - (u8 *)p_rta;
542 		}
543 	}
544 #endif
545 	return 0;
546 rtattr_failure: __attribute__ ((unused))
547 	return -1;
548 }
549 
550 int
551 tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
552 		    struct tcf_ext_map *map)
553 {
554 #ifdef CONFIG_NET_CLS_ACT
555 	if (exts->action)
556 		if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
557 			goto rtattr_failure;
558 #endif
559 	return 0;
560 rtattr_failure: __attribute__ ((unused))
561 	return -1;
562 }
563 
564 static int __init tc_filter_init(void)
565 {
566 	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL);
567 	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL);
568 	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
569 						 tc_dump_tfilter);
570 
571 	return 0;
572 }
573 
574 subsys_initcall(tc_filter_init);
575 
576 EXPORT_SYMBOL(register_tcf_proto_ops);
577 EXPORT_SYMBOL(unregister_tcf_proto_ops);
578 EXPORT_SYMBOL(tcf_exts_validate);
579 EXPORT_SYMBOL(tcf_exts_destroy);
580 EXPORT_SYMBOL(tcf_exts_change);
581 EXPORT_SYMBOL(tcf_exts_dump);
582 EXPORT_SYMBOL(tcf_exts_dump_stats);
583