Linux Audio

Check our new training course

Loading...
v4.6
  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 defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
 33#include <net/netfilter/nf_conntrack.h>
 34#endif
 35
 36struct flow_head {
 37	struct list_head	filters;
 38	struct rcu_head		rcu;
 39};
 40
 41struct 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
 63static 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
 70static 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
 80static 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
 90static u32 flow_get_proto(const struct sk_buff *skb, const struct flow_keys *flow)
 
 91{
 92	return flow->basic.ip_proto;
 93}
 94
 95static u32 flow_get_proto_src(const struct sk_buff *skb, const struct flow_keys *flow)
 
 96{
 97	if (flow->ports.ports)
 98		return ntohs(flow->ports.src);
 99
100	return addr_fold(skb->sk);
101}
102
103static u32 flow_get_proto_dst(const struct sk_buff *skb, const struct flow_keys *flow)
 
104{
105	if (flow->ports.ports)
106		return ntohs(flow->ports.dst);
107
108	return addr_fold(skb_dst(skb)) ^ (__force u16) tc_skb_protocol(skb);
109}
110
111static u32 flow_get_iif(const struct sk_buff *skb)
112{
113	return skb->skb_iif;
114}
115
116static u32 flow_get_priority(const struct sk_buff *skb)
117{
118	return skb->priority;
119}
120
121static u32 flow_get_mark(const struct sk_buff *skb)
122{
123	return skb->mark;
124}
125
126static u32 flow_get_nfct(const struct sk_buff *skb)
127{
128#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
129	return addr_fold(skb->nfct);
130#else
131	return 0;
132#endif
133}
134
135#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
136#define CTTUPLE(skb, member)						\
137({									\
138	enum ip_conntrack_info ctinfo;					\
139	const struct nf_conn *ct = nf_ct_get(skb, &ctinfo);		\
140	if (ct == NULL)							\
141		goto fallback;						\
142	ct->tuplehash[CTINFO2DIR(ctinfo)].tuple.member;			\
143})
144#else
145#define CTTUPLE(skb, member)						\
146({									\
147	goto fallback;							\
148	0;								\
149})
150#endif
151
152static u32 flow_get_nfct_src(const struct sk_buff *skb, const struct flow_keys *flow)
 
153{
154	switch (tc_skb_protocol(skb)) {
155	case htons(ETH_P_IP):
156		return ntohl(CTTUPLE(skb, src.u3.ip));
157	case htons(ETH_P_IPV6):
158		return ntohl(CTTUPLE(skb, src.u3.ip6[3]));
159	}
160fallback:
161	return flow_get_src(skb, flow);
162}
163
164static u32 flow_get_nfct_dst(const struct sk_buff *skb, const struct flow_keys *flow)
 
165{
166	switch (tc_skb_protocol(skb)) {
167	case htons(ETH_P_IP):
168		return ntohl(CTTUPLE(skb, dst.u3.ip));
169	case htons(ETH_P_IPV6):
170		return ntohl(CTTUPLE(skb, dst.u3.ip6[3]));
171	}
172fallback:
173	return flow_get_dst(skb, flow);
174}
175
176static u32 flow_get_nfct_proto_src(const struct sk_buff *skb, const struct flow_keys *flow)
 
177{
178	return ntohs(CTTUPLE(skb, src.u.all));
179fallback:
180	return flow_get_proto_src(skb, flow);
181}
182
183static u32 flow_get_nfct_proto_dst(const struct sk_buff *skb, const struct flow_keys *flow)
 
184{
185	return ntohs(CTTUPLE(skb, dst.u.all));
186fallback:
187	return flow_get_proto_dst(skb, flow);
188}
189
190static u32 flow_get_rtclassid(const struct sk_buff *skb)
191{
192#ifdef CONFIG_IP_ROUTE_CLASSID
193	if (skb_dst(skb))
194		return skb_dst(skb)->tclassid;
195#endif
196	return 0;
197}
198
199static u32 flow_get_skuid(const struct sk_buff *skb)
200{
201	struct sock *sk = skb_to_full_sk(skb);
202
203	if (sk && sk->sk_socket && sk->sk_socket->file) {
204		kuid_t skuid = sk->sk_socket->file->f_cred->fsuid;
205
206		return from_kuid(&init_user_ns, skuid);
207	}
208	return 0;
209}
210
211static u32 flow_get_skgid(const struct sk_buff *skb)
212{
213	struct sock *sk = skb_to_full_sk(skb);
214
215	if (sk && sk->sk_socket && sk->sk_socket->file) {
216		kgid_t skgid = sk->sk_socket->file->f_cred->fsgid;
217
218		return from_kgid(&init_user_ns, skgid);
219	}
220	return 0;
221}
222
223static u32 flow_get_vlan_tag(const struct sk_buff *skb)
224{
225	u16 uninitialized_var(tag);
226
227	if (vlan_get_tag(skb, &tag) < 0)
228		return 0;
229	return tag & VLAN_VID_MASK;
230}
231
232static u32 flow_get_rxhash(struct sk_buff *skb)
233{
234	return skb_get_hash(skb);
235}
236
237static u32 flow_key_get(struct sk_buff *skb, int key, struct flow_keys *flow)
238{
239	switch (key) {
240	case FLOW_KEY_SRC:
241		return flow_get_src(skb, flow);
242	case FLOW_KEY_DST:
243		return flow_get_dst(skb, flow);
244	case FLOW_KEY_PROTO:
245		return flow_get_proto(skb, flow);
246	case FLOW_KEY_PROTO_SRC:
247		return flow_get_proto_src(skb, flow);
248	case FLOW_KEY_PROTO_DST:
249		return flow_get_proto_dst(skb, flow);
250	case FLOW_KEY_IIF:
251		return flow_get_iif(skb);
252	case FLOW_KEY_PRIORITY:
253		return flow_get_priority(skb);
254	case FLOW_KEY_MARK:
255		return flow_get_mark(skb);
256	case FLOW_KEY_NFCT:
257		return flow_get_nfct(skb);
258	case FLOW_KEY_NFCT_SRC:
259		return flow_get_nfct_src(skb, flow);
260	case FLOW_KEY_NFCT_DST:
261		return flow_get_nfct_dst(skb, flow);
262	case FLOW_KEY_NFCT_PROTO_SRC:
263		return flow_get_nfct_proto_src(skb, flow);
264	case FLOW_KEY_NFCT_PROTO_DST:
265		return flow_get_nfct_proto_dst(skb, flow);
266	case FLOW_KEY_RTCLASSID:
267		return flow_get_rtclassid(skb);
268	case FLOW_KEY_SKUID:
269		return flow_get_skuid(skb);
270	case FLOW_KEY_SKGID:
271		return flow_get_skgid(skb);
272	case FLOW_KEY_VLAN_TAG:
273		return flow_get_vlan_tag(skb);
274	case FLOW_KEY_RXHASH:
275		return flow_get_rxhash(skb);
276	default:
277		WARN_ON(1);
278		return 0;
279	}
280}
281
282#define FLOW_KEYS_NEEDED ((1 << FLOW_KEY_SRC) | 		\
283			  (1 << FLOW_KEY_DST) |			\
284			  (1 << FLOW_KEY_PROTO) |		\
285			  (1 << FLOW_KEY_PROTO_SRC) |		\
286			  (1 << FLOW_KEY_PROTO_DST) | 		\
287			  (1 << FLOW_KEY_NFCT_SRC) |		\
288			  (1 << FLOW_KEY_NFCT_DST) |		\
289			  (1 << FLOW_KEY_NFCT_PROTO_SRC) |	\
290			  (1 << FLOW_KEY_NFCT_PROTO_DST))
291
292static int flow_classify(struct sk_buff *skb, const struct tcf_proto *tp,
293			 struct tcf_result *res)
294{
295	struct flow_head *head = rcu_dereference_bh(tp->root);
296	struct flow_filter *f;
297	u32 keymask;
298	u32 classid;
299	unsigned int n, key;
300	int r;
301
302	list_for_each_entry_rcu(f, &head->filters, list) {
303		u32 keys[FLOW_KEY_MAX + 1];
304		struct flow_keys flow_keys;
305
306		if (!tcf_em_tree_match(skb, &f->ematches, NULL))
307			continue;
308
309		keymask = f->keymask;
310		if (keymask & FLOW_KEYS_NEEDED)
311			skb_flow_dissect_flow_keys(skb, &flow_keys, 0);
312
313		for (n = 0; n < f->nkeys; n++) {
314			key = ffs(keymask) - 1;
315			keymask &= ~(1 << key);
316			keys[n] = flow_key_get(skb, key, &flow_keys);
317		}
318
319		if (f->mode == FLOW_MODE_HASH)
320			classid = jhash2(keys, f->nkeys, f->hashrnd);
321		else {
322			classid = keys[0];
323			classid = (classid & f->mask) ^ f->xor;
324			classid = (classid >> f->rshift) + f->addend;
325		}
326
327		if (f->divisor)
328			classid %= f->divisor;
329
330		res->class   = 0;
331		res->classid = TC_H_MAKE(f->baseclass, f->baseclass + classid);
332
333		r = tcf_exts_exec(skb, &f->exts, res);
334		if (r < 0)
335			continue;
336		return r;
337	}
338	return -1;
339}
340
341static void flow_perturbation(unsigned long arg)
342{
343	struct flow_filter *f = (struct flow_filter *)arg;
344
345	get_random_bytes(&f->hashrnd, 4);
346	if (f->perturb_period)
347		mod_timer(&f->perturb_timer, jiffies + f->perturb_period);
348}
349
350static const struct nla_policy flow_policy[TCA_FLOW_MAX + 1] = {
351	[TCA_FLOW_KEYS]		= { .type = NLA_U32 },
352	[TCA_FLOW_MODE]		= { .type = NLA_U32 },
353	[TCA_FLOW_BASECLASS]	= { .type = NLA_U32 },
354	[TCA_FLOW_RSHIFT]	= { .type = NLA_U32 },
355	[TCA_FLOW_ADDEND]	= { .type = NLA_U32 },
356	[TCA_FLOW_MASK]		= { .type = NLA_U32 },
357	[TCA_FLOW_XOR]		= { .type = NLA_U32 },
358	[TCA_FLOW_DIVISOR]	= { .type = NLA_U32 },
359	[TCA_FLOW_ACT]		= { .type = NLA_NESTED },
360	[TCA_FLOW_POLICE]	= { .type = NLA_NESTED },
361	[TCA_FLOW_EMATCHES]	= { .type = NLA_NESTED },
362	[TCA_FLOW_PERTURB]	= { .type = NLA_U32 },
363};
364
365static void flow_destroy_filter(struct rcu_head *head)
366{
367	struct flow_filter *f = container_of(head, struct flow_filter, rcu);
368
369	del_timer_sync(&f->perturb_timer);
370	tcf_exts_destroy(&f->exts);
371	tcf_em_tree_destroy(&f->ematches);
 
372	kfree(f);
373}
374
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
375static int flow_change(struct net *net, struct sk_buff *in_skb,
376		       struct tcf_proto *tp, unsigned long base,
377		       u32 handle, struct nlattr **tca,
378		       unsigned long *arg, bool ovr)
379{
380	struct flow_head *head = rtnl_dereference(tp->root);
381	struct flow_filter *fold, *fnew;
382	struct nlattr *opt = tca[TCA_OPTIONS];
383	struct nlattr *tb[TCA_FLOW_MAX + 1];
384	struct tcf_exts e;
385	struct tcf_ematch_tree t;
386	unsigned int nkeys = 0;
387	unsigned int perturb_period = 0;
388	u32 baseclass = 0;
389	u32 keymask = 0;
390	u32 mode;
391	int err;
392
393	if (opt == NULL)
394		return -EINVAL;
395
396	err = nla_parse_nested(tb, TCA_FLOW_MAX, opt, flow_policy);
397	if (err < 0)
398		return err;
399
400	if (tb[TCA_FLOW_BASECLASS]) {
401		baseclass = nla_get_u32(tb[TCA_FLOW_BASECLASS]);
402		if (TC_H_MIN(baseclass) == 0)
403			return -EINVAL;
404	}
405
406	if (tb[TCA_FLOW_KEYS]) {
407		keymask = nla_get_u32(tb[TCA_FLOW_KEYS]);
408
409		nkeys = hweight32(keymask);
410		if (nkeys == 0)
411			return -EINVAL;
412
413		if (fls(keymask) - 1 > FLOW_KEY_MAX)
414			return -EOPNOTSUPP;
415
416		if ((keymask & (FLOW_KEY_SKUID|FLOW_KEY_SKGID)) &&
417		    sk_user_ns(NETLINK_CB(in_skb).sk) != &init_user_ns)
418			return -EOPNOTSUPP;
419	}
420
421	tcf_exts_init(&e, TCA_FLOW_ACT, TCA_FLOW_POLICE);
422	err = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &e, ovr);
423	if (err < 0)
424		return err;
425
426	err = tcf_em_tree_validate(tp, tb[TCA_FLOW_EMATCHES], &t);
427	if (err < 0)
428		goto err1;
429
430	err = -ENOBUFS;
431	fnew = kzalloc(sizeof(*fnew), GFP_KERNEL);
432	if (!fnew)
433		goto err2;
434
435	tcf_exts_init(&fnew->exts, TCA_FLOW_ACT, TCA_FLOW_POLICE);
 
 
 
436
437	fold = (struct flow_filter *)*arg;
438	if (fold) {
439		err = -EINVAL;
440		if (fold->handle != handle && handle)
441			goto err2;
442
443		/* Copy fold into fnew */
444		fnew->tp = fold->tp;
445		fnew->handle = fold->handle;
446		fnew->nkeys = fold->nkeys;
447		fnew->keymask = fold->keymask;
448		fnew->mode = fold->mode;
449		fnew->mask = fold->mask;
450		fnew->xor = fold->xor;
451		fnew->rshift = fold->rshift;
452		fnew->addend = fold->addend;
453		fnew->divisor = fold->divisor;
454		fnew->baseclass = fold->baseclass;
455		fnew->hashrnd = fold->hashrnd;
456
457		mode = fold->mode;
458		if (tb[TCA_FLOW_MODE])
459			mode = nla_get_u32(tb[TCA_FLOW_MODE]);
460		if (mode != FLOW_MODE_HASH && nkeys > 1)
461			goto err2;
462
463		if (mode == FLOW_MODE_HASH)
464			perturb_period = fold->perturb_period;
465		if (tb[TCA_FLOW_PERTURB]) {
466			if (mode != FLOW_MODE_HASH)
467				goto err2;
468			perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
469		}
470	} else {
471		err = -EINVAL;
472		if (!handle)
473			goto err2;
474		if (!tb[TCA_FLOW_KEYS])
475			goto err2;
476
477		mode = FLOW_MODE_MAP;
478		if (tb[TCA_FLOW_MODE])
479			mode = nla_get_u32(tb[TCA_FLOW_MODE]);
480		if (mode != FLOW_MODE_HASH && nkeys > 1)
481			goto err2;
482
483		if (tb[TCA_FLOW_PERTURB]) {
484			if (mode != FLOW_MODE_HASH)
485				goto err2;
486			perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
487		}
488
489		if (TC_H_MAJ(baseclass) == 0)
490			baseclass = TC_H_MAKE(tp->q->handle, baseclass);
 
 
 
491		if (TC_H_MIN(baseclass) == 0)
492			baseclass = TC_H_MAKE(baseclass, 1);
493
494		fnew->handle = handle;
495		fnew->mask  = ~0U;
496		fnew->tp = tp;
497		get_random_bytes(&fnew->hashrnd, 4);
498	}
499
500	fnew->perturb_timer.function = flow_perturbation;
501	fnew->perturb_timer.data = (unsigned long)fnew;
502	init_timer_deferrable(&fnew->perturb_timer);
503
504	tcf_exts_change(tp, &fnew->exts, &e);
505	tcf_em_tree_change(tp, &fnew->ematches, &t);
506
507	netif_keep_dst(qdisc_dev(tp->q));
508
509	if (tb[TCA_FLOW_KEYS]) {
510		fnew->keymask = keymask;
511		fnew->nkeys   = nkeys;
512	}
513
514	fnew->mode = mode;
515
516	if (tb[TCA_FLOW_MASK])
517		fnew->mask = nla_get_u32(tb[TCA_FLOW_MASK]);
518	if (tb[TCA_FLOW_XOR])
519		fnew->xor = nla_get_u32(tb[TCA_FLOW_XOR]);
520	if (tb[TCA_FLOW_RSHIFT])
521		fnew->rshift = nla_get_u32(tb[TCA_FLOW_RSHIFT]);
522	if (tb[TCA_FLOW_ADDEND])
523		fnew->addend = nla_get_u32(tb[TCA_FLOW_ADDEND]);
524
525	if (tb[TCA_FLOW_DIVISOR])
526		fnew->divisor = nla_get_u32(tb[TCA_FLOW_DIVISOR]);
527	if (baseclass)
528		fnew->baseclass = baseclass;
529
530	fnew->perturb_period = perturb_period;
531	if (perturb_period)
532		mod_timer(&fnew->perturb_timer, jiffies + perturb_period);
533
534	if (*arg == 0)
535		list_add_tail_rcu(&fnew->list, &head->filters);
536	else
537		list_replace_rcu(&fold->list, &fnew->list);
538
539	*arg = (unsigned long)fnew;
540
541	if (fold)
 
542		call_rcu(&fold->rcu, flow_destroy_filter);
 
543	return 0;
544
545err2:
546	tcf_em_tree_destroy(&t);
547	kfree(fnew);
548err1:
549	tcf_exts_destroy(&e);
550	return err;
551}
552
553static int flow_delete(struct tcf_proto *tp, unsigned long arg)
 
554{
555	struct flow_filter *f = (struct flow_filter *)arg;
 
556
557	list_del_rcu(&f->list);
 
558	call_rcu(&f->rcu, flow_destroy_filter);
 
559	return 0;
560}
561
562static int flow_init(struct tcf_proto *tp)
563{
564	struct flow_head *head;
565
566	head = kzalloc(sizeof(*head), GFP_KERNEL);
567	if (head == NULL)
568		return -ENOBUFS;
569	INIT_LIST_HEAD(&head->filters);
570	rcu_assign_pointer(tp->root, head);
571	return 0;
572}
573
574static bool flow_destroy(struct tcf_proto *tp, bool force)
575{
576	struct flow_head *head = rtnl_dereference(tp->root);
577	struct flow_filter *f, *next;
578
579	if (!force && !list_empty(&head->filters))
580		return false;
581
582	list_for_each_entry_safe(f, next, &head->filters, list) {
583		list_del_rcu(&f->list);
584		call_rcu(&f->rcu, flow_destroy_filter);
 
 
 
585	}
586	RCU_INIT_POINTER(tp->root, NULL);
587	kfree_rcu(head, rcu);
588	return true;
589}
590
591static unsigned long flow_get(struct tcf_proto *tp, u32 handle)
592{
593	struct flow_head *head = rtnl_dereference(tp->root);
594	struct flow_filter *f;
595
596	list_for_each_entry(f, &head->filters, list)
597		if (f->handle == handle)
598			return (unsigned long)f;
599	return 0;
600}
601
602static int flow_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
603		     struct sk_buff *skb, struct tcmsg *t)
604{
605	struct flow_filter *f = (struct flow_filter *)fh;
606	struct nlattr *nest;
607
608	if (f == NULL)
609		return skb->len;
610
611	t->tcm_handle = f->handle;
612
613	nest = nla_nest_start(skb, TCA_OPTIONS);
614	if (nest == NULL)
615		goto nla_put_failure;
616
617	if (nla_put_u32(skb, TCA_FLOW_KEYS, f->keymask) ||
618	    nla_put_u32(skb, TCA_FLOW_MODE, f->mode))
619		goto nla_put_failure;
620
621	if (f->mask != ~0 || f->xor != 0) {
622		if (nla_put_u32(skb, TCA_FLOW_MASK, f->mask) ||
623		    nla_put_u32(skb, TCA_FLOW_XOR, f->xor))
624			goto nla_put_failure;
625	}
626	if (f->rshift &&
627	    nla_put_u32(skb, TCA_FLOW_RSHIFT, f->rshift))
628		goto nla_put_failure;
629	if (f->addend &&
630	    nla_put_u32(skb, TCA_FLOW_ADDEND, f->addend))
631		goto nla_put_failure;
632
633	if (f->divisor &&
634	    nla_put_u32(skb, TCA_FLOW_DIVISOR, f->divisor))
635		goto nla_put_failure;
636	if (f->baseclass &&
637	    nla_put_u32(skb, TCA_FLOW_BASECLASS, f->baseclass))
638		goto nla_put_failure;
639
640	if (f->perturb_period &&
641	    nla_put_u32(skb, TCA_FLOW_PERTURB, f->perturb_period / HZ))
642		goto nla_put_failure;
643
644	if (tcf_exts_dump(skb, &f->exts) < 0)
645		goto nla_put_failure;
646#ifdef CONFIG_NET_EMATCH
647	if (f->ematches.hdr.nmatches &&
648	    tcf_em_tree_dump(skb, &f->ematches, TCA_FLOW_EMATCHES) < 0)
649		goto nla_put_failure;
650#endif
651	nla_nest_end(skb, nest);
652
653	if (tcf_exts_dump_stats(skb, &f->exts) < 0)
654		goto nla_put_failure;
655
656	return skb->len;
657
658nla_put_failure:
659	nla_nest_cancel(skb, nest);
660	return -1;
661}
662
663static void flow_walk(struct tcf_proto *tp, struct tcf_walker *arg)
664{
665	struct flow_head *head = rtnl_dereference(tp->root);
666	struct flow_filter *f;
667
668	list_for_each_entry(f, &head->filters, list) {
669		if (arg->count < arg->skip)
670			goto skip;
671		if (arg->fn(tp, (unsigned long)f, arg) < 0) {
672			arg->stop = 1;
673			break;
674		}
675skip:
676		arg->count++;
677	}
678}
679
680static struct tcf_proto_ops cls_flow_ops __read_mostly = {
681	.kind		= "flow",
682	.classify	= flow_classify,
683	.init		= flow_init,
684	.destroy	= flow_destroy,
685	.change		= flow_change,
686	.delete		= flow_delete,
687	.get		= flow_get,
688	.dump		= flow_dump,
689	.walk		= flow_walk,
690	.owner		= THIS_MODULE,
691};
692
693static int __init cls_flow_init(void)
694{
695	return register_tcf_proto_ops(&cls_flow_ops);
696}
697
698static void __exit cls_flow_exit(void)
699{
700	unregister_tcf_proto_ops(&cls_flow_ops);
701}
702
703module_init(cls_flow_init);
704module_exit(cls_flow_exit);
705
706MODULE_LICENSE("GPL");
707MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
708MODULE_DESCRIPTION("TC flow classifier");
v4.17
  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
 36struct flow_head {
 37	struct list_head	filters;
 38	struct rcu_head		rcu;
 39};
 40
 41struct 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	union {
 61		struct work_struct	work;
 62		struct rcu_head		rcu;
 63	};
 64};
 65
 66static inline u32 addr_fold(void *addr)
 67{
 68	unsigned long a = (unsigned long)addr;
 69
 70	return (a & 0xFFFFFFFF) ^ (BITS_PER_LONG > 32 ? a >> 32 : 0);
 71}
 72
 73static u32 flow_get_src(const struct sk_buff *skb, const struct flow_keys *flow)
 74{
 75	__be32 src = flow_get_u32_src(flow);
 76
 77	if (src)
 78		return ntohl(src);
 79
 80	return addr_fold(skb->sk);
 81}
 82
 83static u32 flow_get_dst(const struct sk_buff *skb, const struct flow_keys *flow)
 84{
 85	__be32 dst = flow_get_u32_dst(flow);
 86
 87	if (dst)
 88		return ntohl(dst);
 89
 90	return addr_fold(skb_dst(skb)) ^ (__force u16) tc_skb_protocol(skb);
 91}
 92
 93static u32 flow_get_proto(const struct sk_buff *skb,
 94			  const struct flow_keys *flow)
 95{
 96	return flow->basic.ip_proto;
 97}
 98
 99static u32 flow_get_proto_src(const struct sk_buff *skb,
100			      const struct flow_keys *flow)
101{
102	if (flow->ports.ports)
103		return ntohs(flow->ports.src);
104
105	return addr_fold(skb->sk);
106}
107
108static u32 flow_get_proto_dst(const struct sk_buff *skb,
109			      const struct flow_keys *flow)
110{
111	if (flow->ports.ports)
112		return ntohs(flow->ports.dst);
113
114	return addr_fold(skb_dst(skb)) ^ (__force u16) tc_skb_protocol(skb);
115}
116
117static u32 flow_get_iif(const struct sk_buff *skb)
118{
119	return skb->skb_iif;
120}
121
122static u32 flow_get_priority(const struct sk_buff *skb)
123{
124	return skb->priority;
125}
126
127static u32 flow_get_mark(const struct sk_buff *skb)
128{
129	return skb->mark;
130}
131
132static u32 flow_get_nfct(const struct sk_buff *skb)
133{
134#if IS_ENABLED(CONFIG_NF_CONNTRACK)
135	return addr_fold(skb_nfct(skb));
136#else
137	return 0;
138#endif
139}
140
141#if IS_ENABLED(CONFIG_NF_CONNTRACK)
142#define CTTUPLE(skb, member)						\
143({									\
144	enum ip_conntrack_info ctinfo;					\
145	const struct nf_conn *ct = nf_ct_get(skb, &ctinfo);		\
146	if (ct == NULL)							\
147		goto fallback;						\
148	ct->tuplehash[CTINFO2DIR(ctinfo)].tuple.member;			\
149})
150#else
151#define CTTUPLE(skb, member)						\
152({									\
153	goto fallback;							\
154	0;								\
155})
156#endif
157
158static u32 flow_get_nfct_src(const struct sk_buff *skb,
159			     const struct flow_keys *flow)
160{
161	switch (tc_skb_protocol(skb)) {
162	case htons(ETH_P_IP):
163		return ntohl(CTTUPLE(skb, src.u3.ip));
164	case htons(ETH_P_IPV6):
165		return ntohl(CTTUPLE(skb, src.u3.ip6[3]));
166	}
167fallback:
168	return flow_get_src(skb, flow);
169}
170
171static u32 flow_get_nfct_dst(const struct sk_buff *skb,
172			     const struct flow_keys *flow)
173{
174	switch (tc_skb_protocol(skb)) {
175	case htons(ETH_P_IP):
176		return ntohl(CTTUPLE(skb, dst.u3.ip));
177	case htons(ETH_P_IPV6):
178		return ntohl(CTTUPLE(skb, dst.u3.ip6[3]));
179	}
180fallback:
181	return flow_get_dst(skb, flow);
182}
183
184static u32 flow_get_nfct_proto_src(const struct sk_buff *skb,
185				   const struct flow_keys *flow)
186{
187	return ntohs(CTTUPLE(skb, src.u.all));
188fallback:
189	return flow_get_proto_src(skb, flow);
190}
191
192static u32 flow_get_nfct_proto_dst(const struct sk_buff *skb,
193				   const struct flow_keys *flow)
194{
195	return ntohs(CTTUPLE(skb, dst.u.all));
196fallback:
197	return flow_get_proto_dst(skb, flow);
198}
199
200static u32 flow_get_rtclassid(const struct sk_buff *skb)
201{
202#ifdef CONFIG_IP_ROUTE_CLASSID
203	if (skb_dst(skb))
204		return skb_dst(skb)->tclassid;
205#endif
206	return 0;
207}
208
209static u32 flow_get_skuid(const struct sk_buff *skb)
210{
211	struct sock *sk = skb_to_full_sk(skb);
212
213	if (sk && sk->sk_socket && sk->sk_socket->file) {
214		kuid_t skuid = sk->sk_socket->file->f_cred->fsuid;
215
216		return from_kuid(&init_user_ns, skuid);
217	}
218	return 0;
219}
220
221static u32 flow_get_skgid(const struct sk_buff *skb)
222{
223	struct sock *sk = skb_to_full_sk(skb);
224
225	if (sk && sk->sk_socket && sk->sk_socket->file) {
226		kgid_t skgid = sk->sk_socket->file->f_cred->fsgid;
227
228		return from_kgid(&init_user_ns, skgid);
229	}
230	return 0;
231}
232
233static u32 flow_get_vlan_tag(const struct sk_buff *skb)
234{
235	u16 uninitialized_var(tag);
236
237	if (vlan_get_tag(skb, &tag) < 0)
238		return 0;
239	return tag & VLAN_VID_MASK;
240}
241
242static u32 flow_get_rxhash(struct sk_buff *skb)
243{
244	return skb_get_hash(skb);
245}
246
247static u32 flow_key_get(struct sk_buff *skb, int key, struct flow_keys *flow)
248{
249	switch (key) {
250	case FLOW_KEY_SRC:
251		return flow_get_src(skb, flow);
252	case FLOW_KEY_DST:
253		return flow_get_dst(skb, flow);
254	case FLOW_KEY_PROTO:
255		return flow_get_proto(skb, flow);
256	case FLOW_KEY_PROTO_SRC:
257		return flow_get_proto_src(skb, flow);
258	case FLOW_KEY_PROTO_DST:
259		return flow_get_proto_dst(skb, flow);
260	case FLOW_KEY_IIF:
261		return flow_get_iif(skb);
262	case FLOW_KEY_PRIORITY:
263		return flow_get_priority(skb);
264	case FLOW_KEY_MARK:
265		return flow_get_mark(skb);
266	case FLOW_KEY_NFCT:
267		return flow_get_nfct(skb);
268	case FLOW_KEY_NFCT_SRC:
269		return flow_get_nfct_src(skb, flow);
270	case FLOW_KEY_NFCT_DST:
271		return flow_get_nfct_dst(skb, flow);
272	case FLOW_KEY_NFCT_PROTO_SRC:
273		return flow_get_nfct_proto_src(skb, flow);
274	case FLOW_KEY_NFCT_PROTO_DST:
275		return flow_get_nfct_proto_dst(skb, flow);
276	case FLOW_KEY_RTCLASSID:
277		return flow_get_rtclassid(skb);
278	case FLOW_KEY_SKUID:
279		return flow_get_skuid(skb);
280	case FLOW_KEY_SKGID:
281		return flow_get_skgid(skb);
282	case FLOW_KEY_VLAN_TAG:
283		return flow_get_vlan_tag(skb);
284	case FLOW_KEY_RXHASH:
285		return flow_get_rxhash(skb);
286	default:
287		WARN_ON(1);
288		return 0;
289	}
290}
291
292#define FLOW_KEYS_NEEDED ((1 << FLOW_KEY_SRC) | 		\
293			  (1 << FLOW_KEY_DST) |			\
294			  (1 << FLOW_KEY_PROTO) |		\
295			  (1 << FLOW_KEY_PROTO_SRC) |		\
296			  (1 << FLOW_KEY_PROTO_DST) | 		\
297			  (1 << FLOW_KEY_NFCT_SRC) |		\
298			  (1 << FLOW_KEY_NFCT_DST) |		\
299			  (1 << FLOW_KEY_NFCT_PROTO_SRC) |	\
300			  (1 << FLOW_KEY_NFCT_PROTO_DST))
301
302static int flow_classify(struct sk_buff *skb, const struct tcf_proto *tp,
303			 struct tcf_result *res)
304{
305	struct flow_head *head = rcu_dereference_bh(tp->root);
306	struct flow_filter *f;
307	u32 keymask;
308	u32 classid;
309	unsigned int n, key;
310	int r;
311
312	list_for_each_entry_rcu(f, &head->filters, list) {
313		u32 keys[FLOW_KEY_MAX + 1];
314		struct flow_keys flow_keys;
315
316		if (!tcf_em_tree_match(skb, &f->ematches, NULL))
317			continue;
318
319		keymask = f->keymask;
320		if (keymask & FLOW_KEYS_NEEDED)
321			skb_flow_dissect_flow_keys(skb, &flow_keys, 0);
322
323		for (n = 0; n < f->nkeys; n++) {
324			key = ffs(keymask) - 1;
325			keymask &= ~(1 << key);
326			keys[n] = flow_key_get(skb, key, &flow_keys);
327		}
328
329		if (f->mode == FLOW_MODE_HASH)
330			classid = jhash2(keys, f->nkeys, f->hashrnd);
331		else {
332			classid = keys[0];
333			classid = (classid & f->mask) ^ f->xor;
334			classid = (classid >> f->rshift) + f->addend;
335		}
336
337		if (f->divisor)
338			classid %= f->divisor;
339
340		res->class   = 0;
341		res->classid = TC_H_MAKE(f->baseclass, f->baseclass + classid);
342
343		r = tcf_exts_exec(skb, &f->exts, res);
344		if (r < 0)
345			continue;
346		return r;
347	}
348	return -1;
349}
350
351static void flow_perturbation(struct timer_list *t)
352{
353	struct flow_filter *f = from_timer(f, t, perturb_timer);
354
355	get_random_bytes(&f->hashrnd, 4);
356	if (f->perturb_period)
357		mod_timer(&f->perturb_timer, jiffies + f->perturb_period);
358}
359
360static const struct nla_policy flow_policy[TCA_FLOW_MAX + 1] = {
361	[TCA_FLOW_KEYS]		= { .type = NLA_U32 },
362	[TCA_FLOW_MODE]		= { .type = NLA_U32 },
363	[TCA_FLOW_BASECLASS]	= { .type = NLA_U32 },
364	[TCA_FLOW_RSHIFT]	= { .type = NLA_U32 },
365	[TCA_FLOW_ADDEND]	= { .type = NLA_U32 },
366	[TCA_FLOW_MASK]		= { .type = NLA_U32 },
367	[TCA_FLOW_XOR]		= { .type = NLA_U32 },
368	[TCA_FLOW_DIVISOR]	= { .type = NLA_U32 },
369	[TCA_FLOW_ACT]		= { .type = NLA_NESTED },
370	[TCA_FLOW_POLICE]	= { .type = NLA_NESTED },
371	[TCA_FLOW_EMATCHES]	= { .type = NLA_NESTED },
372	[TCA_FLOW_PERTURB]	= { .type = NLA_U32 },
373};
374
375static void __flow_destroy_filter(struct flow_filter *f)
376{
 
 
377	del_timer_sync(&f->perturb_timer);
378	tcf_exts_destroy(&f->exts);
379	tcf_em_tree_destroy(&f->ematches);
380	tcf_exts_put_net(&f->exts);
381	kfree(f);
382}
383
384static void flow_destroy_filter_work(struct work_struct *work)
385{
386	struct flow_filter *f = container_of(work, struct flow_filter, work);
387
388	rtnl_lock();
389	__flow_destroy_filter(f);
390	rtnl_unlock();
391}
392
393static void flow_destroy_filter(struct rcu_head *head)
394{
395	struct flow_filter *f = container_of(head, struct flow_filter, rcu);
396
397	INIT_WORK(&f->work, flow_destroy_filter_work);
398	tcf_queue_work(&f->work);
399}
400
401static int flow_change(struct net *net, struct sk_buff *in_skb,
402		       struct tcf_proto *tp, unsigned long base,
403		       u32 handle, struct nlattr **tca,
404		       void **arg, bool ovr, struct netlink_ext_ack *extack)
405{
406	struct flow_head *head = rtnl_dereference(tp->root);
407	struct flow_filter *fold, *fnew;
408	struct nlattr *opt = tca[TCA_OPTIONS];
409	struct nlattr *tb[TCA_FLOW_MAX + 1];
 
 
410	unsigned int nkeys = 0;
411	unsigned int perturb_period = 0;
412	u32 baseclass = 0;
413	u32 keymask = 0;
414	u32 mode;
415	int err;
416
417	if (opt == NULL)
418		return -EINVAL;
419
420	err = nla_parse_nested(tb, TCA_FLOW_MAX, opt, flow_policy, NULL);
421	if (err < 0)
422		return err;
423
424	if (tb[TCA_FLOW_BASECLASS]) {
425		baseclass = nla_get_u32(tb[TCA_FLOW_BASECLASS]);
426		if (TC_H_MIN(baseclass) == 0)
427			return -EINVAL;
428	}
429
430	if (tb[TCA_FLOW_KEYS]) {
431		keymask = nla_get_u32(tb[TCA_FLOW_KEYS]);
432
433		nkeys = hweight32(keymask);
434		if (nkeys == 0)
435			return -EINVAL;
436
437		if (fls(keymask) - 1 > FLOW_KEY_MAX)
438			return -EOPNOTSUPP;
439
440		if ((keymask & (FLOW_KEY_SKUID|FLOW_KEY_SKGID)) &&
441		    sk_user_ns(NETLINK_CB(in_skb).sk) != &init_user_ns)
442			return -EOPNOTSUPP;
443	}
444
445	fnew = kzalloc(sizeof(*fnew), GFP_KERNEL);
446	if (!fnew)
447		return -ENOBUFS;
 
448
449	err = tcf_em_tree_validate(tp, tb[TCA_FLOW_EMATCHES], &fnew->ematches);
450	if (err < 0)
451		goto err1;
452
453	err = tcf_exts_init(&fnew->exts, TCA_FLOW_ACT, TCA_FLOW_POLICE);
454	if (err < 0)
 
455		goto err2;
456
457	err = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &fnew->exts, ovr,
458				extack);
459	if (err < 0)
460		goto err2;
461
462	fold = *arg;
463	if (fold) {
464		err = -EINVAL;
465		if (fold->handle != handle && handle)
466			goto err2;
467
468		/* Copy fold into fnew */
469		fnew->tp = fold->tp;
470		fnew->handle = fold->handle;
471		fnew->nkeys = fold->nkeys;
472		fnew->keymask = fold->keymask;
473		fnew->mode = fold->mode;
474		fnew->mask = fold->mask;
475		fnew->xor = fold->xor;
476		fnew->rshift = fold->rshift;
477		fnew->addend = fold->addend;
478		fnew->divisor = fold->divisor;
479		fnew->baseclass = fold->baseclass;
480		fnew->hashrnd = fold->hashrnd;
481
482		mode = fold->mode;
483		if (tb[TCA_FLOW_MODE])
484			mode = nla_get_u32(tb[TCA_FLOW_MODE]);
485		if (mode != FLOW_MODE_HASH && nkeys > 1)
486			goto err2;
487
488		if (mode == FLOW_MODE_HASH)
489			perturb_period = fold->perturb_period;
490		if (tb[TCA_FLOW_PERTURB]) {
491			if (mode != FLOW_MODE_HASH)
492				goto err2;
493			perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
494		}
495	} else {
496		err = -EINVAL;
497		if (!handle)
498			goto err2;
499		if (!tb[TCA_FLOW_KEYS])
500			goto err2;
501
502		mode = FLOW_MODE_MAP;
503		if (tb[TCA_FLOW_MODE])
504			mode = nla_get_u32(tb[TCA_FLOW_MODE]);
505		if (mode != FLOW_MODE_HASH && nkeys > 1)
506			goto err2;
507
508		if (tb[TCA_FLOW_PERTURB]) {
509			if (mode != FLOW_MODE_HASH)
510				goto err2;
511			perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
512		}
513
514		if (TC_H_MAJ(baseclass) == 0) {
515			struct Qdisc *q = tcf_block_q(tp->chain->block);
516
517			baseclass = TC_H_MAKE(q->handle, baseclass);
518		}
519		if (TC_H_MIN(baseclass) == 0)
520			baseclass = TC_H_MAKE(baseclass, 1);
521
522		fnew->handle = handle;
523		fnew->mask  = ~0U;
524		fnew->tp = tp;
525		get_random_bytes(&fnew->hashrnd, 4);
526	}
527
528	timer_setup(&fnew->perturb_timer, flow_perturbation, TIMER_DEFERRABLE);
 
 
529
530	tcf_block_netif_keep_dst(tp->chain->block);
 
 
 
531
532	if (tb[TCA_FLOW_KEYS]) {
533		fnew->keymask = keymask;
534		fnew->nkeys   = nkeys;
535	}
536
537	fnew->mode = mode;
538
539	if (tb[TCA_FLOW_MASK])
540		fnew->mask = nla_get_u32(tb[TCA_FLOW_MASK]);
541	if (tb[TCA_FLOW_XOR])
542		fnew->xor = nla_get_u32(tb[TCA_FLOW_XOR]);
543	if (tb[TCA_FLOW_RSHIFT])
544		fnew->rshift = nla_get_u32(tb[TCA_FLOW_RSHIFT]);
545	if (tb[TCA_FLOW_ADDEND])
546		fnew->addend = nla_get_u32(tb[TCA_FLOW_ADDEND]);
547
548	if (tb[TCA_FLOW_DIVISOR])
549		fnew->divisor = nla_get_u32(tb[TCA_FLOW_DIVISOR]);
550	if (baseclass)
551		fnew->baseclass = baseclass;
552
553	fnew->perturb_period = perturb_period;
554	if (perturb_period)
555		mod_timer(&fnew->perturb_timer, jiffies + perturb_period);
556
557	if (!*arg)
558		list_add_tail_rcu(&fnew->list, &head->filters);
559	else
560		list_replace_rcu(&fold->list, &fnew->list);
561
562	*arg = fnew;
563
564	if (fold) {
565		tcf_exts_get_net(&fold->exts);
566		call_rcu(&fold->rcu, flow_destroy_filter);
567	}
568	return 0;
569
570err2:
571	tcf_exts_destroy(&fnew->exts);
572	tcf_em_tree_destroy(&fnew->ematches);
573err1:
574	kfree(fnew);
575	return err;
576}
577
578static int flow_delete(struct tcf_proto *tp, void *arg, bool *last,
579		       struct netlink_ext_ack *extack)
580{
581	struct flow_head *head = rtnl_dereference(tp->root);
582	struct flow_filter *f = arg;
583
584	list_del_rcu(&f->list);
585	tcf_exts_get_net(&f->exts);
586	call_rcu(&f->rcu, flow_destroy_filter);
587	*last = list_empty(&head->filters);
588	return 0;
589}
590
591static int flow_init(struct tcf_proto *tp)
592{
593	struct flow_head *head;
594
595	head = kzalloc(sizeof(*head), GFP_KERNEL);
596	if (head == NULL)
597		return -ENOBUFS;
598	INIT_LIST_HEAD(&head->filters);
599	rcu_assign_pointer(tp->root, head);
600	return 0;
601}
602
603static void flow_destroy(struct tcf_proto *tp, struct netlink_ext_ack *extack)
604{
605	struct flow_head *head = rtnl_dereference(tp->root);
606	struct flow_filter *f, *next;
607
 
 
 
608	list_for_each_entry_safe(f, next, &head->filters, list) {
609		list_del_rcu(&f->list);
610		if (tcf_exts_get_net(&f->exts))
611			call_rcu(&f->rcu, flow_destroy_filter);
612		else
613			__flow_destroy_filter(f);
614	}
 
615	kfree_rcu(head, rcu);
 
616}
617
618static void *flow_get(struct tcf_proto *tp, u32 handle)
619{
620	struct flow_head *head = rtnl_dereference(tp->root);
621	struct flow_filter *f;
622
623	list_for_each_entry(f, &head->filters, list)
624		if (f->handle == handle)
625			return f;
626	return NULL;
627}
628
629static int flow_dump(struct net *net, struct tcf_proto *tp, void *fh,
630		     struct sk_buff *skb, struct tcmsg *t)
631{
632	struct flow_filter *f = fh;
633	struct nlattr *nest;
634
635	if (f == NULL)
636		return skb->len;
637
638	t->tcm_handle = f->handle;
639
640	nest = nla_nest_start(skb, TCA_OPTIONS);
641	if (nest == NULL)
642		goto nla_put_failure;
643
644	if (nla_put_u32(skb, TCA_FLOW_KEYS, f->keymask) ||
645	    nla_put_u32(skb, TCA_FLOW_MODE, f->mode))
646		goto nla_put_failure;
647
648	if (f->mask != ~0 || f->xor != 0) {
649		if (nla_put_u32(skb, TCA_FLOW_MASK, f->mask) ||
650		    nla_put_u32(skb, TCA_FLOW_XOR, f->xor))
651			goto nla_put_failure;
652	}
653	if (f->rshift &&
654	    nla_put_u32(skb, TCA_FLOW_RSHIFT, f->rshift))
655		goto nla_put_failure;
656	if (f->addend &&
657	    nla_put_u32(skb, TCA_FLOW_ADDEND, f->addend))
658		goto nla_put_failure;
659
660	if (f->divisor &&
661	    nla_put_u32(skb, TCA_FLOW_DIVISOR, f->divisor))
662		goto nla_put_failure;
663	if (f->baseclass &&
664	    nla_put_u32(skb, TCA_FLOW_BASECLASS, f->baseclass))
665		goto nla_put_failure;
666
667	if (f->perturb_period &&
668	    nla_put_u32(skb, TCA_FLOW_PERTURB, f->perturb_period / HZ))
669		goto nla_put_failure;
670
671	if (tcf_exts_dump(skb, &f->exts) < 0)
672		goto nla_put_failure;
673#ifdef CONFIG_NET_EMATCH
674	if (f->ematches.hdr.nmatches &&
675	    tcf_em_tree_dump(skb, &f->ematches, TCA_FLOW_EMATCHES) < 0)
676		goto nla_put_failure;
677#endif
678	nla_nest_end(skb, nest);
679
680	if (tcf_exts_dump_stats(skb, &f->exts) < 0)
681		goto nla_put_failure;
682
683	return skb->len;
684
685nla_put_failure:
686	nla_nest_cancel(skb, nest);
687	return -1;
688}
689
690static void flow_walk(struct tcf_proto *tp, struct tcf_walker *arg)
691{
692	struct flow_head *head = rtnl_dereference(tp->root);
693	struct flow_filter *f;
694
695	list_for_each_entry(f, &head->filters, list) {
696		if (arg->count < arg->skip)
697			goto skip;
698		if (arg->fn(tp, f, arg) < 0) {
699			arg->stop = 1;
700			break;
701		}
702skip:
703		arg->count++;
704	}
705}
706
707static struct tcf_proto_ops cls_flow_ops __read_mostly = {
708	.kind		= "flow",
709	.classify	= flow_classify,
710	.init		= flow_init,
711	.destroy	= flow_destroy,
712	.change		= flow_change,
713	.delete		= flow_delete,
714	.get		= flow_get,
715	.dump		= flow_dump,
716	.walk		= flow_walk,
717	.owner		= THIS_MODULE,
718};
719
720static int __init cls_flow_init(void)
721{
722	return register_tcf_proto_ops(&cls_flow_ops);
723}
724
725static void __exit cls_flow_exit(void)
726{
727	unregister_tcf_proto_ops(&cls_flow_ops);
728}
729
730module_init(cls_flow_init);
731module_exit(cls_flow_exit);
732
733MODULE_LICENSE("GPL");
734MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
735MODULE_DESCRIPTION("TC flow classifier");