xref: /linux/net/sched/cls_route.c (revision a8b70ccf10e38775785d9cb12ead916474549f99) 
1 /*
2  * net/sched/cls_route.c	ROUTE4 classifier.
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 
12 #include <linux/module.h>
13 #include <linux/slab.h>
14 #include <linux/types.h>
15 #include <linux/kernel.h>
16 #include <linux/string.h>
17 #include <linux/errno.h>
18 #include <linux/skbuff.h>
19 #include <net/dst.h>
20 #include <net/route.h>
21 #include <net/netlink.h>
22 #include <net/act_api.h>
23 #include <net/pkt_cls.h>
24 
25 /*
26  * 1. For now we assume that route tags < 256.
27  *    It allows to use direct table lookups, instead of hash tables.
28  * 2. For now we assume that "from TAG" and "fromdev DEV" statements
29  *    are mutually  exclusive.
30  * 3. "to TAG from ANY" has higher priority, than "to ANY from XXX"
31  */
32 struct route4_fastmap {
33 	struct route4_filter		*filter;
34 	u32				id;
35 	int				iif;
36 };
37 
38 struct route4_head {
39 	struct route4_fastmap		fastmap[16];
40 	struct route4_bucket __rcu	*table[256 + 1];
41 	struct rcu_head			rcu;
42 };
43 
44 struct route4_bucket {
45 	/* 16 FROM buckets + 16 IIF buckets + 1 wildcard bucket */
46 	struct route4_filter __rcu	*ht[16 + 16 + 1];
47 	struct rcu_head			rcu;
48 };
49 
50 struct route4_filter {
51 	struct route4_filter __rcu	*next;
52 	u32			id;
53 	int			iif;
54 
55 	struct tcf_result	res;
56 	struct tcf_exts		exts;
57 	u32			handle;
58 	struct route4_bucket	*bkt;
59 	struct tcf_proto	*tp;
60 	union {
61 		struct work_struct	work;
62 		struct rcu_head		rcu;
63 	};
64 };
65 
66 #define ROUTE4_FAILURE ((struct route4_filter *)(-1L))
67 
68 static inline int route4_fastmap_hash(u32 id, int iif)
69 {
70 	return id & 0xF;
71 }
72 
73 static DEFINE_SPINLOCK(fastmap_lock);
74 static void
75 route4_reset_fastmap(struct route4_head *head)
76 {
77 	spin_lock_bh(&fastmap_lock);
78 	memset(head->fastmap, 0, sizeof(head->fastmap));
79 	spin_unlock_bh(&fastmap_lock);
80 }
81 
82 static void
83 route4_set_fastmap(struct route4_head *head, u32 id, int iif,
84 		   struct route4_filter *f)
85 {
86 	int h = route4_fastmap_hash(id, iif);
87 
88 	/* fastmap updates must look atomic to aling id, iff, filter */
89 	spin_lock_bh(&fastmap_lock);
90 	head->fastmap[h].id = id;
91 	head->fastmap[h].iif = iif;
92 	head->fastmap[h].filter = f;
93 	spin_unlock_bh(&fastmap_lock);
94 }
95 
96 static inline int route4_hash_to(u32 id)
97 {
98 	return id & 0xFF;
99 }
100 
101 static inline int route4_hash_from(u32 id)
102 {
103 	return (id >> 16) & 0xF;
104 }
105 
106 static inline int route4_hash_iif(int iif)
107 {
108 	return 16 + ((iif >> 16) & 0xF);
109 }
110 
111 static inline int route4_hash_wild(void)
112 {
113 	return 32;
114 }
115 
116 #define ROUTE4_APPLY_RESULT()					\
117 {								\
118 	*res = f->res;						\
119 	if (tcf_exts_has_actions(&f->exts)) {			\
120 		int r = tcf_exts_exec(skb, &f->exts, res);	\
121 		if (r < 0) {					\
122 			dont_cache = 1;				\
123 			continue;				\
124 		}						\
125 		return r;					\
126 	} else if (!dont_cache)					\
127 		route4_set_fastmap(head, id, iif, f);		\
128 	return 0;						\
129 }
130 
131 static int route4_classify(struct sk_buff *skb, const struct tcf_proto *tp,
132 			   struct tcf_result *res)
133 {
134 	struct route4_head *head = rcu_dereference_bh(tp->root);
135 	struct dst_entry *dst;
136 	struct route4_bucket *b;
137 	struct route4_filter *f;
138 	u32 id, h;
139 	int iif, dont_cache = 0;
140 
141 	dst = skb_dst(skb);
142 	if (!dst)
143 		goto failure;
144 
145 	id = dst->tclassid;
146 
147 	iif = inet_iif(skb);
148 
149 	h = route4_fastmap_hash(id, iif);
150 
151 	spin_lock(&fastmap_lock);
152 	if (id == head->fastmap[h].id &&
153 	    iif == head->fastmap[h].iif &&
154 	    (f = head->fastmap[h].filter) != NULL) {
155 		if (f == ROUTE4_FAILURE) {
156 			spin_unlock(&fastmap_lock);
157 			goto failure;
158 		}
159 
160 		*res = f->res;
161 		spin_unlock(&fastmap_lock);
162 		return 0;
163 	}
164 	spin_unlock(&fastmap_lock);
165 
166 	h = route4_hash_to(id);
167 
168 restart:
169 	b = rcu_dereference_bh(head->table[h]);
170 	if (b) {
171 		for (f = rcu_dereference_bh(b->ht[route4_hash_from(id)]);
172 		     f;
173 		     f = rcu_dereference_bh(f->next))
174 			if (f->id == id)
175 				ROUTE4_APPLY_RESULT();
176 
177 		for (f = rcu_dereference_bh(b->ht[route4_hash_iif(iif)]);
178 		     f;
179 		     f = rcu_dereference_bh(f->next))
180 			if (f->iif == iif)
181 				ROUTE4_APPLY_RESULT();
182 
183 		for (f = rcu_dereference_bh(b->ht[route4_hash_wild()]);
184 		     f;
185 		     f = rcu_dereference_bh(f->next))
186 			ROUTE4_APPLY_RESULT();
187 	}
188 	if (h < 256) {
189 		h = 256;
190 		id &= ~0xFFFF;
191 		goto restart;
192 	}
193 
194 	if (!dont_cache)
195 		route4_set_fastmap(head, id, iif, ROUTE4_FAILURE);
196 failure:
197 	return -1;
198 }
199 
200 static inline u32 to_hash(u32 id)
201 {
202 	u32 h = id & 0xFF;
203 
204 	if (id & 0x8000)
205 		h += 256;
206 	return h;
207 }
208 
209 static inline u32 from_hash(u32 id)
210 {
211 	id &= 0xFFFF;
212 	if (id == 0xFFFF)
213 		return 32;
214 	if (!(id & 0x8000)) {
215 		if (id > 255)
216 			return 256;
217 		return id & 0xF;
218 	}
219 	return 16 + (id & 0xF);
220 }
221 
222 static void *route4_get(struct tcf_proto *tp, u32 handle)
223 {
224 	struct route4_head *head = rtnl_dereference(tp->root);
225 	struct route4_bucket *b;
226 	struct route4_filter *f;
227 	unsigned int h1, h2;
228 
229 	h1 = to_hash(handle);
230 	if (h1 > 256)
231 		return NULL;
232 
233 	h2 = from_hash(handle >> 16);
234 	if (h2 > 32)
235 		return NULL;
236 
237 	b = rtnl_dereference(head->table[h1]);
238 	if (b) {
239 		for (f = rtnl_dereference(b->ht[h2]);
240 		     f;
241 		     f = rtnl_dereference(f->next))
242 			if (f->handle == handle)
243 				return f;
244 	}
245 	return NULL;
246 }
247 
248 static int route4_init(struct tcf_proto *tp)
249 {
250 	struct route4_head *head;
251 
252 	head = kzalloc(sizeof(struct route4_head), GFP_KERNEL);
253 	if (head == NULL)
254 		return -ENOBUFS;
255 
256 	rcu_assign_pointer(tp->root, head);
257 	return 0;
258 }
259 
260 static void __route4_delete_filter(struct route4_filter *f)
261 {
262 	tcf_exts_destroy(&f->exts);
263 	tcf_exts_put_net(&f->exts);
264 	kfree(f);
265 }
266 
267 static void route4_delete_filter_work(struct work_struct *work)
268 {
269 	struct route4_filter *f = container_of(work, struct route4_filter, work);
270 
271 	rtnl_lock();
272 	__route4_delete_filter(f);
273 	rtnl_unlock();
274 }
275 
276 static void route4_delete_filter(struct rcu_head *head)
277 {
278 	struct route4_filter *f = container_of(head, struct route4_filter, rcu);
279 
280 	INIT_WORK(&f->work, route4_delete_filter_work);
281 	tcf_queue_work(&f->work);
282 }
283 
284 static void route4_destroy(struct tcf_proto *tp, struct netlink_ext_ack *extack)
285 {
286 	struct route4_head *head = rtnl_dereference(tp->root);
287 	int h1, h2;
288 
289 	if (head == NULL)
290 		return;
291 
292 	for (h1 = 0; h1 <= 256; h1++) {
293 		struct route4_bucket *b;
294 
295 		b = rtnl_dereference(head->table[h1]);
296 		if (b) {
297 			for (h2 = 0; h2 <= 32; h2++) {
298 				struct route4_filter *f;
299 
300 				while ((f = rtnl_dereference(b->ht[h2])) != NULL) {
301 					struct route4_filter *next;
302 
303 					next = rtnl_dereference(f->next);
304 					RCU_INIT_POINTER(b->ht[h2], next);
305 					tcf_unbind_filter(tp, &f->res);
306 					if (tcf_exts_get_net(&f->exts))
307 						call_rcu(&f->rcu, route4_delete_filter);
308 					else
309 						__route4_delete_filter(f);
310 				}
311 			}
312 			RCU_INIT_POINTER(head->table[h1], NULL);
313 			kfree_rcu(b, rcu);
314 		}
315 	}
316 	kfree_rcu(head, rcu);
317 }
318 
319 static int route4_delete(struct tcf_proto *tp, void *arg, bool *last,
320 			 struct netlink_ext_ack *extack)
321 {
322 	struct route4_head *head = rtnl_dereference(tp->root);
323 	struct route4_filter *f = arg;
324 	struct route4_filter __rcu **fp;
325 	struct route4_filter *nf;
326 	struct route4_bucket *b;
327 	unsigned int h = 0;
328 	int i, h1;
329 
330 	if (!head || !f)
331 		return -EINVAL;
332 
333 	h = f->handle;
334 	b = f->bkt;
335 
336 	fp = &b->ht[from_hash(h >> 16)];
337 	for (nf = rtnl_dereference(*fp); nf;
338 	     fp = &nf->next, nf = rtnl_dereference(*fp)) {
339 		if (nf == f) {
340 			/* unlink it */
341 			RCU_INIT_POINTER(*fp, rtnl_dereference(f->next));
342 
343 			/* Remove any fastmap lookups that might ref filter
344 			 * notice we unlink'd the filter so we can't get it
345 			 * back in the fastmap.
346 			 */
347 			route4_reset_fastmap(head);
348 
349 			/* Delete it */
350 			tcf_unbind_filter(tp, &f->res);
351 			tcf_exts_get_net(&f->exts);
352 			call_rcu(&f->rcu, route4_delete_filter);
353 
354 			/* Strip RTNL protected tree */
355 			for (i = 0; i <= 32; i++) {
356 				struct route4_filter *rt;
357 
358 				rt = rtnl_dereference(b->ht[i]);
359 				if (rt)
360 					goto out;
361 			}
362 
363 			/* OK, session has no flows */
364 			RCU_INIT_POINTER(head->table[to_hash(h)], NULL);
365 			kfree_rcu(b, rcu);
366 			break;
367 		}
368 	}
369 
370 out:
371 	*last = true;
372 	for (h1 = 0; h1 <= 256; h1++) {
373 		if (rcu_access_pointer(head->table[h1])) {
374 			*last = false;
375 			break;
376 		}
377 	}
378 
379 	return 0;
380 }
381 
382 static const struct nla_policy route4_policy[TCA_ROUTE4_MAX + 1] = {
383 	[TCA_ROUTE4_CLASSID]	= { .type = NLA_U32 },
384 	[TCA_ROUTE4_TO]		= { .type = NLA_U32 },
385 	[TCA_ROUTE4_FROM]	= { .type = NLA_U32 },
386 	[TCA_ROUTE4_IIF]	= { .type = NLA_U32 },
387 };
388 
389 static int route4_set_parms(struct net *net, struct tcf_proto *tp,
390 			    unsigned long base, struct route4_filter *f,
391 			    u32 handle, struct route4_head *head,
392 			    struct nlattr **tb, struct nlattr *est, int new,
393 			    bool ovr, struct netlink_ext_ack *extack)
394 {
395 	u32 id = 0, to = 0, nhandle = 0x8000;
396 	struct route4_filter *fp;
397 	unsigned int h1;
398 	struct route4_bucket *b;
399 	int err;
400 
401 	err = tcf_exts_validate(net, tp, tb, est, &f->exts, ovr, extack);
402 	if (err < 0)
403 		return err;
404 
405 	if (tb[TCA_ROUTE4_TO]) {
406 		if (new && handle & 0x8000)
407 			return -EINVAL;
408 		to = nla_get_u32(tb[TCA_ROUTE4_TO]);
409 		if (to > 0xFF)
410 			return -EINVAL;
411 		nhandle = to;
412 	}
413 
414 	if (tb[TCA_ROUTE4_FROM]) {
415 		if (tb[TCA_ROUTE4_IIF])
416 			return -EINVAL;
417 		id = nla_get_u32(tb[TCA_ROUTE4_FROM]);
418 		if (id > 0xFF)
419 			return -EINVAL;
420 		nhandle |= id << 16;
421 	} else if (tb[TCA_ROUTE4_IIF]) {
422 		id = nla_get_u32(tb[TCA_ROUTE4_IIF]);
423 		if (id > 0x7FFF)
424 			return -EINVAL;
425 		nhandle |= (id | 0x8000) << 16;
426 	} else
427 		nhandle |= 0xFFFF << 16;
428 
429 	if (handle && new) {
430 		nhandle |= handle & 0x7F00;
431 		if (nhandle != handle)
432 			return -EINVAL;
433 	}
434 
435 	h1 = to_hash(nhandle);
436 	b = rtnl_dereference(head->table[h1]);
437 	if (!b) {
438 		b = kzalloc(sizeof(struct route4_bucket), GFP_KERNEL);
439 		if (b == NULL)
440 			return -ENOBUFS;
441 
442 		rcu_assign_pointer(head->table[h1], b);
443 	} else {
444 		unsigned int h2 = from_hash(nhandle >> 16);
445 
446 		for (fp = rtnl_dereference(b->ht[h2]);
447 		     fp;
448 		     fp = rtnl_dereference(fp->next))
449 			if (fp->handle == f->handle)
450 				return -EEXIST;
451 	}
452 
453 	if (tb[TCA_ROUTE4_TO])
454 		f->id = to;
455 
456 	if (tb[TCA_ROUTE4_FROM])
457 		f->id = to | id<<16;
458 	else if (tb[TCA_ROUTE4_IIF])
459 		f->iif = id;
460 
461 	f->handle = nhandle;
462 	f->bkt = b;
463 	f->tp = tp;
464 
465 	if (tb[TCA_ROUTE4_CLASSID]) {
466 		f->res.classid = nla_get_u32(tb[TCA_ROUTE4_CLASSID]);
467 		tcf_bind_filter(tp, &f->res, base);
468 	}
469 
470 	return 0;
471 }
472 
473 static int route4_change(struct net *net, struct sk_buff *in_skb,
474 			 struct tcf_proto *tp, unsigned long base, u32 handle,
475 			 struct nlattr **tca, void **arg, bool ovr,
476 			 struct netlink_ext_ack *extack)
477 {
478 	struct route4_head *head = rtnl_dereference(tp->root);
479 	struct route4_filter __rcu **fp;
480 	struct route4_filter *fold, *f1, *pfp, *f = NULL;
481 	struct route4_bucket *b;
482 	struct nlattr *opt = tca[TCA_OPTIONS];
483 	struct nlattr *tb[TCA_ROUTE4_MAX + 1];
484 	unsigned int h, th;
485 	int err;
486 	bool new = true;
487 
488 	if (opt == NULL)
489 		return handle ? -EINVAL : 0;
490 
491 	err = nla_parse_nested(tb, TCA_ROUTE4_MAX, opt, route4_policy, NULL);
492 	if (err < 0)
493 		return err;
494 
495 	fold = *arg;
496 	if (fold && handle && fold->handle != handle)
497 			return -EINVAL;
498 
499 	err = -ENOBUFS;
500 	f = kzalloc(sizeof(struct route4_filter), GFP_KERNEL);
501 	if (!f)
502 		goto errout;
503 
504 	err = tcf_exts_init(&f->exts, TCA_ROUTE4_ACT, TCA_ROUTE4_POLICE);
505 	if (err < 0)
506 		goto errout;
507 
508 	if (fold) {
509 		f->id = fold->id;
510 		f->iif = fold->iif;
511 		f->res = fold->res;
512 		f->handle = fold->handle;
513 
514 		f->tp = fold->tp;
515 		f->bkt = fold->bkt;
516 		new = false;
517 	}
518 
519 	err = route4_set_parms(net, tp, base, f, handle, head, tb,
520 			       tca[TCA_RATE], new, ovr, extack);
521 	if (err < 0)
522 		goto errout;
523 
524 	h = from_hash(f->handle >> 16);
525 	fp = &f->bkt->ht[h];
526 	for (pfp = rtnl_dereference(*fp);
527 	     (f1 = rtnl_dereference(*fp)) != NULL;
528 	     fp = &f1->next)
529 		if (f->handle < f1->handle)
530 			break;
531 
532 	tcf_block_netif_keep_dst(tp->chain->block);
533 	rcu_assign_pointer(f->next, f1);
534 	rcu_assign_pointer(*fp, f);
535 
536 	if (fold && fold->handle && f->handle != fold->handle) {
537 		th = to_hash(fold->handle);
538 		h = from_hash(fold->handle >> 16);
539 		b = rtnl_dereference(head->table[th]);
540 		if (b) {
541 			fp = &b->ht[h];
542 			for (pfp = rtnl_dereference(*fp); pfp;
543 			     fp = &pfp->next, pfp = rtnl_dereference(*fp)) {
544 				if (pfp == f) {
545 					*fp = f->next;
546 					break;
547 				}
548 			}
549 		}
550 	}
551 
552 	route4_reset_fastmap(head);
553 	*arg = f;
554 	if (fold) {
555 		tcf_unbind_filter(tp, &fold->res);
556 		tcf_exts_get_net(&fold->exts);
557 		call_rcu(&fold->rcu, route4_delete_filter);
558 	}
559 	return 0;
560 
561 errout:
562 	if (f)
563 		tcf_exts_destroy(&f->exts);
564 	kfree(f);
565 	return err;
566 }
567 
568 static void route4_walk(struct tcf_proto *tp, struct tcf_walker *arg)
569 {
570 	struct route4_head *head = rtnl_dereference(tp->root);
571 	unsigned int h, h1;
572 
573 	if (head == NULL)
574 		arg->stop = 1;
575 
576 	if (arg->stop)
577 		return;
578 
579 	for (h = 0; h <= 256; h++) {
580 		struct route4_bucket *b = rtnl_dereference(head->table[h]);
581 
582 		if (b) {
583 			for (h1 = 0; h1 <= 32; h1++) {
584 				struct route4_filter *f;
585 
586 				for (f = rtnl_dereference(b->ht[h1]);
587 				     f;
588 				     f = rtnl_dereference(f->next)) {
589 					if (arg->count < arg->skip) {
590 						arg->count++;
591 						continue;
592 					}
593 					if (arg->fn(tp, f, arg) < 0) {
594 						arg->stop = 1;
595 						return;
596 					}
597 					arg->count++;
598 				}
599 			}
600 		}
601 	}
602 }
603 
604 static int route4_dump(struct net *net, struct tcf_proto *tp, void *fh,
605 		       struct sk_buff *skb, struct tcmsg *t)
606 {
607 	struct route4_filter *f = fh;
608 	struct nlattr *nest;
609 	u32 id;
610 
611 	if (f == NULL)
612 		return skb->len;
613 
614 	t->tcm_handle = f->handle;
615 
616 	nest = nla_nest_start(skb, TCA_OPTIONS);
617 	if (nest == NULL)
618 		goto nla_put_failure;
619 
620 	if (!(f->handle & 0x8000)) {
621 		id = f->id & 0xFF;
622 		if (nla_put_u32(skb, TCA_ROUTE4_TO, id))
623 			goto nla_put_failure;
624 	}
625 	if (f->handle & 0x80000000) {
626 		if ((f->handle >> 16) != 0xFFFF &&
627 		    nla_put_u32(skb, TCA_ROUTE4_IIF, f->iif))
628 			goto nla_put_failure;
629 	} else {
630 		id = f->id >> 16;
631 		if (nla_put_u32(skb, TCA_ROUTE4_FROM, id))
632 			goto nla_put_failure;
633 	}
634 	if (f->res.classid &&
635 	    nla_put_u32(skb, TCA_ROUTE4_CLASSID, f->res.classid))
636 		goto nla_put_failure;
637 
638 	if (tcf_exts_dump(skb, &f->exts) < 0)
639 		goto nla_put_failure;
640 
641 	nla_nest_end(skb, nest);
642 
643 	if (tcf_exts_dump_stats(skb, &f->exts) < 0)
644 		goto nla_put_failure;
645 
646 	return skb->len;
647 
648 nla_put_failure:
649 	nla_nest_cancel(skb, nest);
650 	return -1;
651 }
652 
653 static void route4_bind_class(void *fh, u32 classid, unsigned long cl)
654 {
655 	struct route4_filter *f = fh;
656 
657 	if (f && f->res.classid == classid)
658 		f->res.class = cl;
659 }
660 
661 static struct tcf_proto_ops cls_route4_ops __read_mostly = {
662 	.kind		=	"route",
663 	.classify	=	route4_classify,
664 	.init		=	route4_init,
665 	.destroy	=	route4_destroy,
666 	.get		=	route4_get,
667 	.change		=	route4_change,
668 	.delete		=	route4_delete,
669 	.walk		=	route4_walk,
670 	.dump		=	route4_dump,
671 	.bind_class	=	route4_bind_class,
672 	.owner		=	THIS_MODULE,
673 };
674 
675 static int __init init_route4(void)
676 {
677 	return register_tcf_proto_ops(&cls_route4_ops);
678 }
679 
680 static void __exit exit_route4(void)
681 {
682 	unregister_tcf_proto_ops(&cls_route4_ops);
683 }
684 
685 module_init(init_route4)
686 module_exit(exit_route4)
687 MODULE_LICENSE("GPL");
688