xref: /openbmc/linux/net/sched/cls_api.c (revision 54cbac81)
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 <linux/err.h>
27 #include <linux/slab.h>
28 #include <net/net_namespace.h>
29 #include <net/sock.h>
30 #include <net/netlink.h>
31 #include <net/pkt_sched.h>
32 #include <net/pkt_cls.h>
33 
34 /* The list of all installed classifier types */
35 
36 static struct tcf_proto_ops *tcf_proto_base __read_mostly;
37 
38 /* Protects list of registered TC modules. It is pure SMP lock. */
39 static DEFINE_RWLOCK(cls_mod_lock);
40 
41 /* Find classifier type by string name */
42 
43 static const struct tcf_proto_ops *tcf_proto_lookup_ops(struct nlattr *kind)
44 {
45 	const struct tcf_proto_ops *t = NULL;
46 
47 	if (kind) {
48 		read_lock(&cls_mod_lock);
49 		for (t = tcf_proto_base; t; t = t->next) {
50 			if (nla_strcmp(kind, t->kind) == 0) {
51 				if (!try_module_get(t->owner))
52 					t = NULL;
53 				break;
54 			}
55 		}
56 		read_unlock(&cls_mod_lock);
57 	}
58 	return t;
59 }
60 
61 /* Register(unregister) new classifier type */
62 
63 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
64 {
65 	struct tcf_proto_ops *t, **tp;
66 	int rc = -EEXIST;
67 
68 	write_lock(&cls_mod_lock);
69 	for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
70 		if (!strcmp(ops->kind, t->kind))
71 			goto out;
72 
73 	ops->next = NULL;
74 	*tp = ops;
75 	rc = 0;
76 out:
77 	write_unlock(&cls_mod_lock);
78 	return rc;
79 }
80 EXPORT_SYMBOL(register_tcf_proto_ops);
81 
82 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
83 {
84 	struct tcf_proto_ops *t, **tp;
85 	int rc = -ENOENT;
86 
87 	write_lock(&cls_mod_lock);
88 	for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
89 		if (t == ops)
90 			break;
91 
92 	if (!t)
93 		goto out;
94 	*tp = t->next;
95 	rc = 0;
96 out:
97 	write_unlock(&cls_mod_lock);
98 	return rc;
99 }
100 EXPORT_SYMBOL(unregister_tcf_proto_ops);
101 
102 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
103 			  struct nlmsghdr *n, struct tcf_proto *tp,
104 			  unsigned long fh, int event);
105 
106 
107 /* Select new prio value from the range, managed by kernel. */
108 
109 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
110 {
111 	u32 first = TC_H_MAKE(0xC0000000U, 0U);
112 
113 	if (tp)
114 		first = tp->prio - 1;
115 
116 	return first;
117 }
118 
119 /* Add/change/delete/get a filter node */
120 
121 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
122 {
123 	struct net *net = sock_net(skb->sk);
124 	struct nlattr *tca[TCA_MAX + 1];
125 	spinlock_t *root_lock;
126 	struct tcmsg *t;
127 	u32 protocol;
128 	u32 prio;
129 	u32 nprio;
130 	u32 parent;
131 	struct net_device *dev;
132 	struct Qdisc  *q;
133 	struct tcf_proto **back, **chain;
134 	struct tcf_proto *tp;
135 	const struct tcf_proto_ops *tp_ops;
136 	const struct Qdisc_class_ops *cops;
137 	unsigned long cl;
138 	unsigned long fh;
139 	int err;
140 	int tp_created = 0;
141 
142 	if ((n->nlmsg_type != RTM_GETTFILTER) && !capable(CAP_NET_ADMIN))
143 		return -EPERM;
144 replay:
145 	t = nlmsg_data(n);
146 	protocol = TC_H_MIN(t->tcm_info);
147 	prio = TC_H_MAJ(t->tcm_info);
148 	nprio = prio;
149 	parent = t->tcm_parent;
150 	cl = 0;
151 
152 	if (prio == 0) {
153 		/* If no priority is given, user wants we allocated it. */
154 		if (n->nlmsg_type != RTM_NEWTFILTER ||
155 		    !(n->nlmsg_flags & NLM_F_CREATE))
156 			return -ENOENT;
157 		prio = TC_H_MAKE(0x80000000U, 0U);
158 	}
159 
160 	/* Find head of filter chain. */
161 
162 	/* Find link */
163 	dev = __dev_get_by_index(net, t->tcm_ifindex);
164 	if (dev == NULL)
165 		return -ENODEV;
166 
167 	err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
168 	if (err < 0)
169 		return err;
170 
171 	/* Find qdisc */
172 	if (!parent) {
173 		q = dev->qdisc;
174 		parent = q->handle;
175 	} else {
176 		q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
177 		if (q == NULL)
178 			return -EINVAL;
179 	}
180 
181 	/* Is it classful? */
182 	cops = q->ops->cl_ops;
183 	if (!cops)
184 		return -EINVAL;
185 
186 	if (cops->tcf_chain == NULL)
187 		return -EOPNOTSUPP;
188 
189 	/* Do we search for filter, attached to class? */
190 	if (TC_H_MIN(parent)) {
191 		cl = cops->get(q, parent);
192 		if (cl == 0)
193 			return -ENOENT;
194 	}
195 
196 	/* And the last stroke */
197 	chain = cops->tcf_chain(q, cl);
198 	err = -EINVAL;
199 	if (chain == NULL)
200 		goto errout;
201 
202 	/* Check the chain for existence of proto-tcf with this priority */
203 	for (back = chain; (tp = *back) != NULL; back = &tp->next) {
204 		if (tp->prio >= prio) {
205 			if (tp->prio == prio) {
206 				if (!nprio ||
207 				    (tp->protocol != protocol && protocol))
208 					goto errout;
209 			} else
210 				tp = NULL;
211 			break;
212 		}
213 	}
214 
215 	root_lock = qdisc_root_sleeping_lock(q);
216 
217 	if (tp == NULL) {
218 		/* Proto-tcf does not exist, create new one */
219 
220 		if (tca[TCA_KIND] == NULL || !protocol)
221 			goto errout;
222 
223 		err = -ENOENT;
224 		if (n->nlmsg_type != RTM_NEWTFILTER ||
225 		    !(n->nlmsg_flags & NLM_F_CREATE))
226 			goto errout;
227 
228 
229 		/* Create new proto tcf */
230 
231 		err = -ENOBUFS;
232 		tp = kzalloc(sizeof(*tp), GFP_KERNEL);
233 		if (tp == NULL)
234 			goto errout;
235 		err = -ENOENT;
236 		tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
237 		if (tp_ops == NULL) {
238 #ifdef CONFIG_MODULES
239 			struct nlattr *kind = tca[TCA_KIND];
240 			char name[IFNAMSIZ];
241 
242 			if (kind != NULL &&
243 			    nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
244 				rtnl_unlock();
245 				request_module("cls_%s", name);
246 				rtnl_lock();
247 				tp_ops = tcf_proto_lookup_ops(kind);
248 				/* We dropped the RTNL semaphore in order to
249 				 * perform the module load.  So, even if we
250 				 * succeeded in loading the module we have to
251 				 * replay the request.  We indicate this using
252 				 * -EAGAIN.
253 				 */
254 				if (tp_ops != NULL) {
255 					module_put(tp_ops->owner);
256 					err = -EAGAIN;
257 				}
258 			}
259 #endif
260 			kfree(tp);
261 			goto errout;
262 		}
263 		tp->ops = tp_ops;
264 		tp->protocol = protocol;
265 		tp->prio = nprio ? : TC_H_MAJ(tcf_auto_prio(*back));
266 		tp->q = q;
267 		tp->classify = tp_ops->classify;
268 		tp->classid = parent;
269 
270 		err = tp_ops->init(tp);
271 		if (err != 0) {
272 			module_put(tp_ops->owner);
273 			kfree(tp);
274 			goto errout;
275 		}
276 
277 		tp_created = 1;
278 
279 	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
280 		goto errout;
281 
282 	fh = tp->ops->get(tp, t->tcm_handle);
283 
284 	if (fh == 0) {
285 		if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
286 			spin_lock_bh(root_lock);
287 			*back = tp->next;
288 			spin_unlock_bh(root_lock);
289 
290 			tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
291 			tcf_destroy(tp);
292 			err = 0;
293 			goto errout;
294 		}
295 
296 		err = -ENOENT;
297 		if (n->nlmsg_type != RTM_NEWTFILTER ||
298 		    !(n->nlmsg_flags & NLM_F_CREATE))
299 			goto errout;
300 	} else {
301 		switch (n->nlmsg_type) {
302 		case RTM_NEWTFILTER:
303 			err = -EEXIST;
304 			if (n->nlmsg_flags & NLM_F_EXCL) {
305 				if (tp_created)
306 					tcf_destroy(tp);
307 				goto errout;
308 			}
309 			break;
310 		case RTM_DELTFILTER:
311 			err = tp->ops->delete(tp, fh);
312 			if (err == 0)
313 				tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
314 			goto errout;
315 		case RTM_GETTFILTER:
316 			err = tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
317 			goto errout;
318 		default:
319 			err = -EINVAL;
320 			goto errout;
321 		}
322 	}
323 
324 	err = tp->ops->change(skb, tp, cl, t->tcm_handle, tca, &fh);
325 	if (err == 0) {
326 		if (tp_created) {
327 			spin_lock_bh(root_lock);
328 			tp->next = *back;
329 			*back = tp;
330 			spin_unlock_bh(root_lock);
331 		}
332 		tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
333 	} else {
334 		if (tp_created)
335 			tcf_destroy(tp);
336 	}
337 
338 errout:
339 	if (cl)
340 		cops->put(q, cl);
341 	if (err == -EAGAIN)
342 		/* Replay the request. */
343 		goto replay;
344 	return err;
345 }
346 
347 static int tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp,
348 			 unsigned long fh, u32 portid, u32 seq, u16 flags, int event)
349 {
350 	struct tcmsg *tcm;
351 	struct nlmsghdr  *nlh;
352 	unsigned char *b = skb_tail_pointer(skb);
353 
354 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
355 	if (!nlh)
356 		goto out_nlmsg_trim;
357 	tcm = nlmsg_data(nlh);
358 	tcm->tcm_family = AF_UNSPEC;
359 	tcm->tcm__pad1 = 0;
360 	tcm->tcm__pad2 = 0;
361 	tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
362 	tcm->tcm_parent = tp->classid;
363 	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
364 	if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
365 		goto nla_put_failure;
366 	tcm->tcm_handle = fh;
367 	if (RTM_DELTFILTER != event) {
368 		tcm->tcm_handle = 0;
369 		if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
370 			goto nla_put_failure;
371 	}
372 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
373 	return skb->len;
374 
375 out_nlmsg_trim:
376 nla_put_failure:
377 	nlmsg_trim(skb, b);
378 	return -1;
379 }
380 
381 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
382 			  struct nlmsghdr *n, struct tcf_proto *tp,
383 			  unsigned long fh, int event)
384 {
385 	struct sk_buff *skb;
386 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
387 
388 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
389 	if (!skb)
390 		return -ENOBUFS;
391 
392 	if (tcf_fill_node(skb, tp, fh, portid, n->nlmsg_seq, 0, event) <= 0) {
393 		kfree_skb(skb);
394 		return -EINVAL;
395 	}
396 
397 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
398 			      n->nlmsg_flags & NLM_F_ECHO);
399 }
400 
401 struct tcf_dump_args {
402 	struct tcf_walker w;
403 	struct sk_buff *skb;
404 	struct netlink_callback *cb;
405 };
406 
407 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
408 			 struct tcf_walker *arg)
409 {
410 	struct tcf_dump_args *a = (void *)arg;
411 
412 	return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).portid,
413 			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
414 }
415 
416 /* called with RTNL */
417 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
418 {
419 	struct net *net = sock_net(skb->sk);
420 	int t;
421 	int s_t;
422 	struct net_device *dev;
423 	struct Qdisc *q;
424 	struct tcf_proto *tp, **chain;
425 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
426 	unsigned long cl = 0;
427 	const struct Qdisc_class_ops *cops;
428 	struct tcf_dump_args arg;
429 
430 	if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
431 		return skb->len;
432 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
433 	if (!dev)
434 		return skb->len;
435 
436 	if (!tcm->tcm_parent)
437 		q = dev->qdisc;
438 	else
439 		q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
440 	if (!q)
441 		goto out;
442 	cops = q->ops->cl_ops;
443 	if (!cops)
444 		goto errout;
445 	if (cops->tcf_chain == NULL)
446 		goto errout;
447 	if (TC_H_MIN(tcm->tcm_parent)) {
448 		cl = cops->get(q, tcm->tcm_parent);
449 		if (cl == 0)
450 			goto errout;
451 	}
452 	chain = cops->tcf_chain(q, cl);
453 	if (chain == NULL)
454 		goto errout;
455 
456 	s_t = cb->args[0];
457 
458 	for (tp = *chain, t = 0; tp; tp = tp->next, t++) {
459 		if (t < s_t)
460 			continue;
461 		if (TC_H_MAJ(tcm->tcm_info) &&
462 		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
463 			continue;
464 		if (TC_H_MIN(tcm->tcm_info) &&
465 		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
466 			continue;
467 		if (t > s_t)
468 			memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
469 		if (cb->args[1] == 0) {
470 			if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).portid,
471 					  cb->nlh->nlmsg_seq, NLM_F_MULTI,
472 					  RTM_NEWTFILTER) <= 0)
473 				break;
474 
475 			cb->args[1] = 1;
476 		}
477 		if (tp->ops->walk == NULL)
478 			continue;
479 		arg.w.fn = tcf_node_dump;
480 		arg.skb = skb;
481 		arg.cb = cb;
482 		arg.w.stop = 0;
483 		arg.w.skip = cb->args[1] - 1;
484 		arg.w.count = 0;
485 		tp->ops->walk(tp, &arg.w);
486 		cb->args[1] = arg.w.count + 1;
487 		if (arg.w.stop)
488 			break;
489 	}
490 
491 	cb->args[0] = t;
492 
493 errout:
494 	if (cl)
495 		cops->put(q, cl);
496 out:
497 	return skb->len;
498 }
499 
500 void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
501 {
502 #ifdef CONFIG_NET_CLS_ACT
503 	if (exts->action) {
504 		tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
505 		exts->action = NULL;
506 	}
507 #endif
508 }
509 EXPORT_SYMBOL(tcf_exts_destroy);
510 
511 int tcf_exts_validate(struct tcf_proto *tp, struct nlattr **tb,
512 		  struct nlattr *rate_tlv, struct tcf_exts *exts,
513 		  const struct tcf_ext_map *map)
514 {
515 	memset(exts, 0, sizeof(*exts));
516 
517 #ifdef CONFIG_NET_CLS_ACT
518 	{
519 		struct tc_action *act;
520 
521 		if (map->police && tb[map->police]) {
522 			act = tcf_action_init_1(tb[map->police], rate_tlv,
523 						"police", TCA_ACT_NOREPLACE,
524 						TCA_ACT_BIND);
525 			if (IS_ERR(act))
526 				return PTR_ERR(act);
527 
528 			act->type = TCA_OLD_COMPAT;
529 			exts->action = act;
530 		} else if (map->action && tb[map->action]) {
531 			act = tcf_action_init(tb[map->action], rate_tlv, NULL,
532 					      TCA_ACT_NOREPLACE, TCA_ACT_BIND);
533 			if (IS_ERR(act))
534 				return PTR_ERR(act);
535 
536 			exts->action = act;
537 		}
538 	}
539 #else
540 	if ((map->action && tb[map->action]) ||
541 	    (map->police && tb[map->police]))
542 		return -EOPNOTSUPP;
543 #endif
544 
545 	return 0;
546 }
547 EXPORT_SYMBOL(tcf_exts_validate);
548 
549 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
550 		     struct tcf_exts *src)
551 {
552 #ifdef CONFIG_NET_CLS_ACT
553 	if (src->action) {
554 		struct tc_action *act;
555 		tcf_tree_lock(tp);
556 		act = dst->action;
557 		dst->action = src->action;
558 		tcf_tree_unlock(tp);
559 		if (act)
560 			tcf_action_destroy(act, TCA_ACT_UNBIND);
561 	}
562 #endif
563 }
564 EXPORT_SYMBOL(tcf_exts_change);
565 
566 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
567 		  const struct tcf_ext_map *map)
568 {
569 #ifdef CONFIG_NET_CLS_ACT
570 	if (map->action && exts->action) {
571 		/*
572 		 * again for backward compatible mode - we want
573 		 * to work with both old and new modes of entering
574 		 * tc data even if iproute2  was newer - jhs
575 		 */
576 		struct nlattr *nest;
577 
578 		if (exts->action->type != TCA_OLD_COMPAT) {
579 			nest = nla_nest_start(skb, map->action);
580 			if (nest == NULL)
581 				goto nla_put_failure;
582 			if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
583 				goto nla_put_failure;
584 			nla_nest_end(skb, nest);
585 		} else if (map->police) {
586 			nest = nla_nest_start(skb, map->police);
587 			if (nest == NULL)
588 				goto nla_put_failure;
589 			if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
590 				goto nla_put_failure;
591 			nla_nest_end(skb, nest);
592 		}
593 	}
594 #endif
595 	return 0;
596 nla_put_failure: __attribute__ ((unused))
597 	return -1;
598 }
599 EXPORT_SYMBOL(tcf_exts_dump);
600 
601 
602 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
603 			const struct tcf_ext_map *map)
604 {
605 #ifdef CONFIG_NET_CLS_ACT
606 	if (exts->action)
607 		if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
608 			goto nla_put_failure;
609 #endif
610 	return 0;
611 nla_put_failure: __attribute__ ((unused))
612 	return -1;
613 }
614 EXPORT_SYMBOL(tcf_exts_dump_stats);
615 
616 static int __init tc_filter_init(void)
617 {
618 	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
619 	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
620 	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
621 		      tc_dump_tfilter, NULL);
622 
623 	return 0;
624 }
625 
626 subsys_initcall(tc_filter_init);
627