xref: /linux/net/sched/cls_flow.c (revision 2fe05e1139a555ae91f00a812cb9520e7d3022ab)
1 /*
2  * net/sched/cls_flow.c		Generic flow classifier
3  *
4  * Copyright (c) 2007, 2008 Patrick McHardy <kaber@trash.net>
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License
8  * as published by the Free Software Foundation; either version 2
9  * of the License, or (at your option) any later version.
10  */
11 
12 #include <linux/kernel.h>
13 #include <linux/init.h>
14 #include <linux/list.h>
15 #include <linux/jhash.h>
16 #include <linux/random.h>
17 #include <linux/pkt_cls.h>
18 #include <linux/skbuff.h>
19 #include <linux/in.h>
20 #include <linux/ip.h>
21 #include <linux/ipv6.h>
22 #include <linux/if_vlan.h>
23 #include <linux/slab.h>
24 #include <linux/module.h>
25 #include <net/inet_sock.h>
26 
27 #include <net/pkt_cls.h>
28 #include <net/ip.h>
29 #include <net/route.h>
30 #include <net/flow_dissector.h>
31 
32 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
33 #include <net/netfilter/nf_conntrack.h>
34 #endif
35 
36 struct flow_head {
37 	struct list_head	filters;
38 	struct rcu_head		rcu;
39 };
40 
41 struct flow_filter {
42 	struct list_head	list;
43 	struct tcf_exts		exts;
44 	struct tcf_ematch_tree	ematches;
45 	struct tcf_proto	*tp;
46 	struct timer_list	perturb_timer;
47 	u32			perturb_period;
48 	u32			handle;
49 
50 	u32			nkeys;
51 	u32			keymask;
52 	u32			mode;
53 	u32			mask;
54 	u32			xor;
55 	u32			rshift;
56 	u32			addend;
57 	u32			divisor;
58 	u32			baseclass;
59 	u32			hashrnd;
60 	struct rcu_head		rcu;
61 };
62 
63 static inline u32 addr_fold(void *addr)
64 {
65 	unsigned long a = (unsigned long)addr;
66 
67 	return (a & 0xFFFFFFFF) ^ (BITS_PER_LONG > 32 ? a >> 32 : 0);
68 }
69 
70 static u32 flow_get_src(const struct sk_buff *skb, const struct flow_keys *flow)
71 {
72 	__be32 src = flow_get_u32_src(flow);
73 
74 	if (src)
75 		return ntohl(src);
76 
77 	return addr_fold(skb->sk);
78 }
79 
80 static u32 flow_get_dst(const struct sk_buff *skb, const struct flow_keys *flow)
81 {
82 	__be32 dst = flow_get_u32_dst(flow);
83 
84 	if (dst)
85 		return ntohl(dst);
86 
87 	return addr_fold(skb_dst(skb)) ^ (__force u16) tc_skb_protocol(skb);
88 }
89 
90 static u32 flow_get_proto(const struct sk_buff *skb,
91 			  const struct flow_keys *flow)
92 {
93 	return flow->basic.ip_proto;
94 }
95 
96 static u32 flow_get_proto_src(const struct sk_buff *skb,
97 			      const struct flow_keys *flow)
98 {
99 	if (flow->ports.ports)
100 		return ntohs(flow->ports.src);
101 
102 	return addr_fold(skb->sk);
103 }
104 
105 static u32 flow_get_proto_dst(const struct sk_buff *skb,
106 			      const struct flow_keys *flow)
107 {
108 	if (flow->ports.ports)
109 		return ntohs(flow->ports.dst);
110 
111 	return addr_fold(skb_dst(skb)) ^ (__force u16) tc_skb_protocol(skb);
112 }
113 
114 static u32 flow_get_iif(const struct sk_buff *skb)
115 {
116 	return skb->skb_iif;
117 }
118 
119 static u32 flow_get_priority(const struct sk_buff *skb)
120 {
121 	return skb->priority;
122 }
123 
124 static u32 flow_get_mark(const struct sk_buff *skb)
125 {
126 	return skb->mark;
127 }
128 
129 static u32 flow_get_nfct(const struct sk_buff *skb)
130 {
131 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
132 	return addr_fold(skb_nfct(skb));
133 #else
134 	return 0;
135 #endif
136 }
137 
138 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
139 #define CTTUPLE(skb, member)						\
140 ({									\
141 	enum ip_conntrack_info ctinfo;					\
142 	const struct nf_conn *ct = nf_ct_get(skb, &ctinfo);		\
143 	if (ct == NULL)							\
144 		goto fallback;						\
145 	ct->tuplehash[CTINFO2DIR(ctinfo)].tuple.member;			\
146 })
147 #else
148 #define CTTUPLE(skb, member)						\
149 ({									\
150 	goto fallback;							\
151 	0;								\
152 })
153 #endif
154 
155 static u32 flow_get_nfct_src(const struct sk_buff *skb,
156 			     const struct flow_keys *flow)
157 {
158 	switch (tc_skb_protocol(skb)) {
159 	case htons(ETH_P_IP):
160 		return ntohl(CTTUPLE(skb, src.u3.ip));
161 	case htons(ETH_P_IPV6):
162 		return ntohl(CTTUPLE(skb, src.u3.ip6[3]));
163 	}
164 fallback:
165 	return flow_get_src(skb, flow);
166 }
167 
168 static u32 flow_get_nfct_dst(const struct sk_buff *skb,
169 			     const struct flow_keys *flow)
170 {
171 	switch (tc_skb_protocol(skb)) {
172 	case htons(ETH_P_IP):
173 		return ntohl(CTTUPLE(skb, dst.u3.ip));
174 	case htons(ETH_P_IPV6):
175 		return ntohl(CTTUPLE(skb, dst.u3.ip6[3]));
176 	}
177 fallback:
178 	return flow_get_dst(skb, flow);
179 }
180 
181 static u32 flow_get_nfct_proto_src(const struct sk_buff *skb,
182 				   const struct flow_keys *flow)
183 {
184 	return ntohs(CTTUPLE(skb, src.u.all));
185 fallback:
186 	return flow_get_proto_src(skb, flow);
187 }
188 
189 static u32 flow_get_nfct_proto_dst(const struct sk_buff *skb,
190 				   const struct flow_keys *flow)
191 {
192 	return ntohs(CTTUPLE(skb, dst.u.all));
193 fallback:
194 	return flow_get_proto_dst(skb, flow);
195 }
196 
197 static u32 flow_get_rtclassid(const struct sk_buff *skb)
198 {
199 #ifdef CONFIG_IP_ROUTE_CLASSID
200 	if (skb_dst(skb))
201 		return skb_dst(skb)->tclassid;
202 #endif
203 	return 0;
204 }
205 
206 static u32 flow_get_skuid(const struct sk_buff *skb)
207 {
208 	struct sock *sk = skb_to_full_sk(skb);
209 
210 	if (sk && sk->sk_socket && sk->sk_socket->file) {
211 		kuid_t skuid = sk->sk_socket->file->f_cred->fsuid;
212 
213 		return from_kuid(&init_user_ns, skuid);
214 	}
215 	return 0;
216 }
217 
218 static u32 flow_get_skgid(const struct sk_buff *skb)
219 {
220 	struct sock *sk = skb_to_full_sk(skb);
221 
222 	if (sk && sk->sk_socket && sk->sk_socket->file) {
223 		kgid_t skgid = sk->sk_socket->file->f_cred->fsgid;
224 
225 		return from_kgid(&init_user_ns, skgid);
226 	}
227 	return 0;
228 }
229 
230 static u32 flow_get_vlan_tag(const struct sk_buff *skb)
231 {
232 	u16 uninitialized_var(tag);
233 
234 	if (vlan_get_tag(skb, &tag) < 0)
235 		return 0;
236 	return tag & VLAN_VID_MASK;
237 }
238 
239 static u32 flow_get_rxhash(struct sk_buff *skb)
240 {
241 	return skb_get_hash(skb);
242 }
243 
244 static u32 flow_key_get(struct sk_buff *skb, int key, struct flow_keys *flow)
245 {
246 	switch (key) {
247 	case FLOW_KEY_SRC:
248 		return flow_get_src(skb, flow);
249 	case FLOW_KEY_DST:
250 		return flow_get_dst(skb, flow);
251 	case FLOW_KEY_PROTO:
252 		return flow_get_proto(skb, flow);
253 	case FLOW_KEY_PROTO_SRC:
254 		return flow_get_proto_src(skb, flow);
255 	case FLOW_KEY_PROTO_DST:
256 		return flow_get_proto_dst(skb, flow);
257 	case FLOW_KEY_IIF:
258 		return flow_get_iif(skb);
259 	case FLOW_KEY_PRIORITY:
260 		return flow_get_priority(skb);
261 	case FLOW_KEY_MARK:
262 		return flow_get_mark(skb);
263 	case FLOW_KEY_NFCT:
264 		return flow_get_nfct(skb);
265 	case FLOW_KEY_NFCT_SRC:
266 		return flow_get_nfct_src(skb, flow);
267 	case FLOW_KEY_NFCT_DST:
268 		return flow_get_nfct_dst(skb, flow);
269 	case FLOW_KEY_NFCT_PROTO_SRC:
270 		return flow_get_nfct_proto_src(skb, flow);
271 	case FLOW_KEY_NFCT_PROTO_DST:
272 		return flow_get_nfct_proto_dst(skb, flow);
273 	case FLOW_KEY_RTCLASSID:
274 		return flow_get_rtclassid(skb);
275 	case FLOW_KEY_SKUID:
276 		return flow_get_skuid(skb);
277 	case FLOW_KEY_SKGID:
278 		return flow_get_skgid(skb);
279 	case FLOW_KEY_VLAN_TAG:
280 		return flow_get_vlan_tag(skb);
281 	case FLOW_KEY_RXHASH:
282 		return flow_get_rxhash(skb);
283 	default:
284 		WARN_ON(1);
285 		return 0;
286 	}
287 }
288 
289 #define FLOW_KEYS_NEEDED ((1 << FLOW_KEY_SRC) | 		\
290 			  (1 << FLOW_KEY_DST) |			\
291 			  (1 << FLOW_KEY_PROTO) |		\
292 			  (1 << FLOW_KEY_PROTO_SRC) |		\
293 			  (1 << FLOW_KEY_PROTO_DST) | 		\
294 			  (1 << FLOW_KEY_NFCT_SRC) |		\
295 			  (1 << FLOW_KEY_NFCT_DST) |		\
296 			  (1 << FLOW_KEY_NFCT_PROTO_SRC) |	\
297 			  (1 << FLOW_KEY_NFCT_PROTO_DST))
298 
299 static int flow_classify(struct sk_buff *skb, const struct tcf_proto *tp,
300 			 struct tcf_result *res)
301 {
302 	struct flow_head *head = rcu_dereference_bh(tp->root);
303 	struct flow_filter *f;
304 	u32 keymask;
305 	u32 classid;
306 	unsigned int n, key;
307 	int r;
308 
309 	list_for_each_entry_rcu(f, &head->filters, list) {
310 		u32 keys[FLOW_KEY_MAX + 1];
311 		struct flow_keys flow_keys;
312 
313 		if (!tcf_em_tree_match(skb, &f->ematches, NULL))
314 			continue;
315 
316 		keymask = f->keymask;
317 		if (keymask & FLOW_KEYS_NEEDED)
318 			skb_flow_dissect_flow_keys(skb, &flow_keys, 0);
319 
320 		for (n = 0; n < f->nkeys; n++) {
321 			key = ffs(keymask) - 1;
322 			keymask &= ~(1 << key);
323 			keys[n] = flow_key_get(skb, key, &flow_keys);
324 		}
325 
326 		if (f->mode == FLOW_MODE_HASH)
327 			classid = jhash2(keys, f->nkeys, f->hashrnd);
328 		else {
329 			classid = keys[0];
330 			classid = (classid & f->mask) ^ f->xor;
331 			classid = (classid >> f->rshift) + f->addend;
332 		}
333 
334 		if (f->divisor)
335 			classid %= f->divisor;
336 
337 		res->class   = 0;
338 		res->classid = TC_H_MAKE(f->baseclass, f->baseclass + classid);
339 
340 		r = tcf_exts_exec(skb, &f->exts, res);
341 		if (r < 0)
342 			continue;
343 		return r;
344 	}
345 	return -1;
346 }
347 
348 static void flow_perturbation(unsigned long arg)
349 {
350 	struct flow_filter *f = (struct flow_filter *)arg;
351 
352 	get_random_bytes(&f->hashrnd, 4);
353 	if (f->perturb_period)
354 		mod_timer(&f->perturb_timer, jiffies + f->perturb_period);
355 }
356 
357 static const struct nla_policy flow_policy[TCA_FLOW_MAX + 1] = {
358 	[TCA_FLOW_KEYS]		= { .type = NLA_U32 },
359 	[TCA_FLOW_MODE]		= { .type = NLA_U32 },
360 	[TCA_FLOW_BASECLASS]	= { .type = NLA_U32 },
361 	[TCA_FLOW_RSHIFT]	= { .type = NLA_U32 },
362 	[TCA_FLOW_ADDEND]	= { .type = NLA_U32 },
363 	[TCA_FLOW_MASK]		= { .type = NLA_U32 },
364 	[TCA_FLOW_XOR]		= { .type = NLA_U32 },
365 	[TCA_FLOW_DIVISOR]	= { .type = NLA_U32 },
366 	[TCA_FLOW_ACT]		= { .type = NLA_NESTED },
367 	[TCA_FLOW_POLICE]	= { .type = NLA_NESTED },
368 	[TCA_FLOW_EMATCHES]	= { .type = NLA_NESTED },
369 	[TCA_FLOW_PERTURB]	= { .type = NLA_U32 },
370 };
371 
372 static void flow_destroy_filter(struct rcu_head *head)
373 {
374 	struct flow_filter *f = container_of(head, struct flow_filter, rcu);
375 
376 	del_timer_sync(&f->perturb_timer);
377 	tcf_exts_destroy(&f->exts);
378 	tcf_em_tree_destroy(&f->ematches);
379 	kfree(f);
380 }
381 
382 static int flow_change(struct net *net, struct sk_buff *in_skb,
383 		       struct tcf_proto *tp, unsigned long base,
384 		       u32 handle, struct nlattr **tca,
385 		       unsigned long *arg, bool ovr)
386 {
387 	struct flow_head *head = rtnl_dereference(tp->root);
388 	struct flow_filter *fold, *fnew;
389 	struct nlattr *opt = tca[TCA_OPTIONS];
390 	struct nlattr *tb[TCA_FLOW_MAX + 1];
391 	struct tcf_exts e;
392 	struct tcf_ematch_tree t;
393 	unsigned int nkeys = 0;
394 	unsigned int perturb_period = 0;
395 	u32 baseclass = 0;
396 	u32 keymask = 0;
397 	u32 mode;
398 	int err;
399 
400 	if (opt == NULL)
401 		return -EINVAL;
402 
403 	err = nla_parse_nested(tb, TCA_FLOW_MAX, opt, flow_policy, NULL);
404 	if (err < 0)
405 		return err;
406 
407 	if (tb[TCA_FLOW_BASECLASS]) {
408 		baseclass = nla_get_u32(tb[TCA_FLOW_BASECLASS]);
409 		if (TC_H_MIN(baseclass) == 0)
410 			return -EINVAL;
411 	}
412 
413 	if (tb[TCA_FLOW_KEYS]) {
414 		keymask = nla_get_u32(tb[TCA_FLOW_KEYS]);
415 
416 		nkeys = hweight32(keymask);
417 		if (nkeys == 0)
418 			return -EINVAL;
419 
420 		if (fls(keymask) - 1 > FLOW_KEY_MAX)
421 			return -EOPNOTSUPP;
422 
423 		if ((keymask & (FLOW_KEY_SKUID|FLOW_KEY_SKGID)) &&
424 		    sk_user_ns(NETLINK_CB(in_skb).sk) != &init_user_ns)
425 			return -EOPNOTSUPP;
426 	}
427 
428 	err = tcf_exts_init(&e, TCA_FLOW_ACT, TCA_FLOW_POLICE);
429 	if (err < 0)
430 		goto err1;
431 	err = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &e, ovr);
432 	if (err < 0)
433 		goto err1;
434 
435 	err = tcf_em_tree_validate(tp, tb[TCA_FLOW_EMATCHES], &t);
436 	if (err < 0)
437 		goto err1;
438 
439 	err = -ENOBUFS;
440 	fnew = kzalloc(sizeof(*fnew), GFP_KERNEL);
441 	if (!fnew)
442 		goto err2;
443 
444 	err = tcf_exts_init(&fnew->exts, TCA_FLOW_ACT, TCA_FLOW_POLICE);
445 	if (err < 0)
446 		goto err3;
447 
448 	fold = (struct flow_filter *)*arg;
449 	if (fold) {
450 		err = -EINVAL;
451 		if (fold->handle != handle && handle)
452 			goto err3;
453 
454 		/* Copy fold into fnew */
455 		fnew->tp = fold->tp;
456 		fnew->handle = fold->handle;
457 		fnew->nkeys = fold->nkeys;
458 		fnew->keymask = fold->keymask;
459 		fnew->mode = fold->mode;
460 		fnew->mask = fold->mask;
461 		fnew->xor = fold->xor;
462 		fnew->rshift = fold->rshift;
463 		fnew->addend = fold->addend;
464 		fnew->divisor = fold->divisor;
465 		fnew->baseclass = fold->baseclass;
466 		fnew->hashrnd = fold->hashrnd;
467 
468 		mode = fold->mode;
469 		if (tb[TCA_FLOW_MODE])
470 			mode = nla_get_u32(tb[TCA_FLOW_MODE]);
471 		if (mode != FLOW_MODE_HASH && nkeys > 1)
472 			goto err3;
473 
474 		if (mode == FLOW_MODE_HASH)
475 			perturb_period = fold->perturb_period;
476 		if (tb[TCA_FLOW_PERTURB]) {
477 			if (mode != FLOW_MODE_HASH)
478 				goto err3;
479 			perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
480 		}
481 	} else {
482 		err = -EINVAL;
483 		if (!handle)
484 			goto err3;
485 		if (!tb[TCA_FLOW_KEYS])
486 			goto err3;
487 
488 		mode = FLOW_MODE_MAP;
489 		if (tb[TCA_FLOW_MODE])
490 			mode = nla_get_u32(tb[TCA_FLOW_MODE]);
491 		if (mode != FLOW_MODE_HASH && nkeys > 1)
492 			goto err3;
493 
494 		if (tb[TCA_FLOW_PERTURB]) {
495 			if (mode != FLOW_MODE_HASH)
496 				goto err3;
497 			perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
498 		}
499 
500 		if (TC_H_MAJ(baseclass) == 0)
501 			baseclass = TC_H_MAKE(tp->q->handle, baseclass);
502 		if (TC_H_MIN(baseclass) == 0)
503 			baseclass = TC_H_MAKE(baseclass, 1);
504 
505 		fnew->handle = handle;
506 		fnew->mask  = ~0U;
507 		fnew->tp = tp;
508 		get_random_bytes(&fnew->hashrnd, 4);
509 	}
510 
511 	setup_deferrable_timer(&fnew->perturb_timer, flow_perturbation,
512 			       (unsigned long)fnew);
513 
514 	tcf_exts_change(tp, &fnew->exts, &e);
515 	tcf_em_tree_change(tp, &fnew->ematches, &t);
516 
517 	netif_keep_dst(qdisc_dev(tp->q));
518 
519 	if (tb[TCA_FLOW_KEYS]) {
520 		fnew->keymask = keymask;
521 		fnew->nkeys   = nkeys;
522 	}
523 
524 	fnew->mode = mode;
525 
526 	if (tb[TCA_FLOW_MASK])
527 		fnew->mask = nla_get_u32(tb[TCA_FLOW_MASK]);
528 	if (tb[TCA_FLOW_XOR])
529 		fnew->xor = nla_get_u32(tb[TCA_FLOW_XOR]);
530 	if (tb[TCA_FLOW_RSHIFT])
531 		fnew->rshift = nla_get_u32(tb[TCA_FLOW_RSHIFT]);
532 	if (tb[TCA_FLOW_ADDEND])
533 		fnew->addend = nla_get_u32(tb[TCA_FLOW_ADDEND]);
534 
535 	if (tb[TCA_FLOW_DIVISOR])
536 		fnew->divisor = nla_get_u32(tb[TCA_FLOW_DIVISOR]);
537 	if (baseclass)
538 		fnew->baseclass = baseclass;
539 
540 	fnew->perturb_period = perturb_period;
541 	if (perturb_period)
542 		mod_timer(&fnew->perturb_timer, jiffies + perturb_period);
543 
544 	if (*arg == 0)
545 		list_add_tail_rcu(&fnew->list, &head->filters);
546 	else
547 		list_replace_rcu(&fold->list, &fnew->list);
548 
549 	*arg = (unsigned long)fnew;
550 
551 	if (fold)
552 		call_rcu(&fold->rcu, flow_destroy_filter);
553 	return 0;
554 
555 err3:
556 	tcf_exts_destroy(&fnew->exts);
557 err2:
558 	tcf_em_tree_destroy(&t);
559 	kfree(fnew);
560 err1:
561 	tcf_exts_destroy(&e);
562 	return err;
563 }
564 
565 static int flow_delete(struct tcf_proto *tp, unsigned long arg, bool *last)
566 {
567 	struct flow_head *head = rtnl_dereference(tp->root);
568 	struct flow_filter *f = (struct flow_filter *)arg;
569 
570 	list_del_rcu(&f->list);
571 	call_rcu(&f->rcu, flow_destroy_filter);
572 	*last = list_empty(&head->filters);
573 	return 0;
574 }
575 
576 static int flow_init(struct tcf_proto *tp)
577 {
578 	struct flow_head *head;
579 
580 	head = kzalloc(sizeof(*head), GFP_KERNEL);
581 	if (head == NULL)
582 		return -ENOBUFS;
583 	INIT_LIST_HEAD(&head->filters);
584 	rcu_assign_pointer(tp->root, head);
585 	return 0;
586 }
587 
588 static void flow_destroy(struct tcf_proto *tp)
589 {
590 	struct flow_head *head = rtnl_dereference(tp->root);
591 	struct flow_filter *f, *next;
592 
593 	list_for_each_entry_safe(f, next, &head->filters, list) {
594 		list_del_rcu(&f->list);
595 		call_rcu(&f->rcu, flow_destroy_filter);
596 	}
597 	kfree_rcu(head, rcu);
598 }
599 
600 static unsigned long flow_get(struct tcf_proto *tp, u32 handle)
601 {
602 	struct flow_head *head = rtnl_dereference(tp->root);
603 	struct flow_filter *f;
604 
605 	list_for_each_entry(f, &head->filters, list)
606 		if (f->handle == handle)
607 			return (unsigned long)f;
608 	return 0;
609 }
610 
611 static int flow_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
612 		     struct sk_buff *skb, struct tcmsg *t)
613 {
614 	struct flow_filter *f = (struct flow_filter *)fh;
615 	struct nlattr *nest;
616 
617 	if (f == NULL)
618 		return skb->len;
619 
620 	t->tcm_handle = f->handle;
621 
622 	nest = nla_nest_start(skb, TCA_OPTIONS);
623 	if (nest == NULL)
624 		goto nla_put_failure;
625 
626 	if (nla_put_u32(skb, TCA_FLOW_KEYS, f->keymask) ||
627 	    nla_put_u32(skb, TCA_FLOW_MODE, f->mode))
628 		goto nla_put_failure;
629 
630 	if (f->mask != ~0 || f->xor != 0) {
631 		if (nla_put_u32(skb, TCA_FLOW_MASK, f->mask) ||
632 		    nla_put_u32(skb, TCA_FLOW_XOR, f->xor))
633 			goto nla_put_failure;
634 	}
635 	if (f->rshift &&
636 	    nla_put_u32(skb, TCA_FLOW_RSHIFT, f->rshift))
637 		goto nla_put_failure;
638 	if (f->addend &&
639 	    nla_put_u32(skb, TCA_FLOW_ADDEND, f->addend))
640 		goto nla_put_failure;
641 
642 	if (f->divisor &&
643 	    nla_put_u32(skb, TCA_FLOW_DIVISOR, f->divisor))
644 		goto nla_put_failure;
645 	if (f->baseclass &&
646 	    nla_put_u32(skb, TCA_FLOW_BASECLASS, f->baseclass))
647 		goto nla_put_failure;
648 
649 	if (f->perturb_period &&
650 	    nla_put_u32(skb, TCA_FLOW_PERTURB, f->perturb_period / HZ))
651 		goto nla_put_failure;
652 
653 	if (tcf_exts_dump(skb, &f->exts) < 0)
654 		goto nla_put_failure;
655 #ifdef CONFIG_NET_EMATCH
656 	if (f->ematches.hdr.nmatches &&
657 	    tcf_em_tree_dump(skb, &f->ematches, TCA_FLOW_EMATCHES) < 0)
658 		goto nla_put_failure;
659 #endif
660 	nla_nest_end(skb, nest);
661 
662 	if (tcf_exts_dump_stats(skb, &f->exts) < 0)
663 		goto nla_put_failure;
664 
665 	return skb->len;
666 
667 nla_put_failure:
668 	nla_nest_cancel(skb, nest);
669 	return -1;
670 }
671 
672 static void flow_walk(struct tcf_proto *tp, struct tcf_walker *arg)
673 {
674 	struct flow_head *head = rtnl_dereference(tp->root);
675 	struct flow_filter *f;
676 
677 	list_for_each_entry(f, &head->filters, list) {
678 		if (arg->count < arg->skip)
679 			goto skip;
680 		if (arg->fn(tp, (unsigned long)f, arg) < 0) {
681 			arg->stop = 1;
682 			break;
683 		}
684 skip:
685 		arg->count++;
686 	}
687 }
688 
689 static struct tcf_proto_ops cls_flow_ops __read_mostly = {
690 	.kind		= "flow",
691 	.classify	= flow_classify,
692 	.init		= flow_init,
693 	.destroy	= flow_destroy,
694 	.change		= flow_change,
695 	.delete		= flow_delete,
696 	.get		= flow_get,
697 	.dump		= flow_dump,
698 	.walk		= flow_walk,
699 	.owner		= THIS_MODULE,
700 };
701 
702 static int __init cls_flow_init(void)
703 {
704 	return register_tcf_proto_ops(&cls_flow_ops);
705 }
706 
707 static void __exit cls_flow_exit(void)
708 {
709 	unregister_tcf_proto_ops(&cls_flow_ops);
710 }
711 
712 module_init(cls_flow_init);
713 module_exit(cls_flow_exit);
714 
715 MODULE_LICENSE("GPL");
716 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
717 MODULE_DESCRIPTION("TC flow classifier");
718