Linux Audio

Check our new training course

Loading...
v5.4
  1// SPDX-License-Identifier: GPL-2.0-or-later
  2/*
  3 * net/sched/cls_flow.c		Generic flow classifier
  4 *
  5 * Copyright (c) 2007, 2008 Patrick McHardy <kaber@trash.net>
 
 
 
 
 
  6 */
  7
  8#include <linux/kernel.h>
  9#include <linux/init.h>
 10#include <linux/list.h>
 11#include <linux/jhash.h>
 12#include <linux/random.h>
 13#include <linux/pkt_cls.h>
 14#include <linux/skbuff.h>
 15#include <linux/in.h>
 16#include <linux/ip.h>
 17#include <linux/ipv6.h>
 18#include <linux/if_vlan.h>
 19#include <linux/slab.h>
 20#include <linux/module.h>
 21#include <net/inet_sock.h>
 22
 23#include <net/pkt_cls.h>
 24#include <net/ip.h>
 25#include <net/route.h>
 26#include <net/flow_dissector.h>
 27
 28#if IS_ENABLED(CONFIG_NF_CONNTRACK)
 29#include <net/netfilter/nf_conntrack.h>
 30#endif
 31
 32struct flow_head {
 33	struct list_head	filters;
 34	struct rcu_head		rcu;
 35};
 36
 37struct flow_filter {
 38	struct list_head	list;
 39	struct tcf_exts		exts;
 40	struct tcf_ematch_tree	ematches;
 41	struct tcf_proto	*tp;
 42	struct timer_list	perturb_timer;
 43	u32			perturb_period;
 44	u32			handle;
 45
 46	u32			nkeys;
 47	u32			keymask;
 48	u32			mode;
 49	u32			mask;
 50	u32			xor;
 51	u32			rshift;
 52	u32			addend;
 53	u32			divisor;
 54	u32			baseclass;
 55	u32			hashrnd;
 56	struct rcu_work		rwork;
 57};
 58
 59static inline u32 addr_fold(void *addr)
 60{
 61	unsigned long a = (unsigned long)addr;
 62
 63	return (a & 0xFFFFFFFF) ^ (BITS_PER_LONG > 32 ? a >> 32 : 0);
 64}
 65
 66static u32 flow_get_src(const struct sk_buff *skb, const struct flow_keys *flow)
 67{
 68	__be32 src = flow_get_u32_src(flow);
 69
 70	if (src)
 71		return ntohl(src);
 72
 73	return addr_fold(skb->sk);
 74}
 75
 76static u32 flow_get_dst(const struct sk_buff *skb, const struct flow_keys *flow)
 77{
 78	__be32 dst = flow_get_u32_dst(flow);
 79
 80	if (dst)
 81		return ntohl(dst);
 82
 83	return addr_fold(skb_dst(skb)) ^ (__force u16) tc_skb_protocol(skb);
 84}
 85
 86static u32 flow_get_proto(const struct sk_buff *skb,
 87			  const struct flow_keys *flow)
 88{
 89	return flow->basic.ip_proto;
 90}
 91
 92static u32 flow_get_proto_src(const struct sk_buff *skb,
 93			      const struct flow_keys *flow)
 94{
 95	if (flow->ports.ports)
 96		return ntohs(flow->ports.src);
 97
 98	return addr_fold(skb->sk);
 99}
100
101static u32 flow_get_proto_dst(const struct sk_buff *skb,
102			      const struct flow_keys *flow)
103{
104	if (flow->ports.ports)
105		return ntohs(flow->ports.dst);
106
107	return addr_fold(skb_dst(skb)) ^ (__force u16) tc_skb_protocol(skb);
108}
109
110static u32 flow_get_iif(const struct sk_buff *skb)
111{
112	return skb->skb_iif;
113}
114
115static u32 flow_get_priority(const struct sk_buff *skb)
116{
117	return skb->priority;
118}
119
120static u32 flow_get_mark(const struct sk_buff *skb)
121{
122	return skb->mark;
123}
124
125static u32 flow_get_nfct(const struct sk_buff *skb)
126{
127#if IS_ENABLED(CONFIG_NF_CONNTRACK)
128	return addr_fold(skb_nfct(skb));
129#else
130	return 0;
131#endif
132}
133
134#if IS_ENABLED(CONFIG_NF_CONNTRACK)
135#define CTTUPLE(skb, member)						\
136({									\
137	enum ip_conntrack_info ctinfo;					\
138	const struct nf_conn *ct = nf_ct_get(skb, &ctinfo);		\
139	if (ct == NULL)							\
140		goto fallback;						\
141	ct->tuplehash[CTINFO2DIR(ctinfo)].tuple.member;			\
142})
143#else
144#define CTTUPLE(skb, member)						\
145({									\
146	goto fallback;							\
147	0;								\
148})
149#endif
150
151static u32 flow_get_nfct_src(const struct sk_buff *skb,
152			     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,
165			     const struct flow_keys *flow)
166{
167	switch (tc_skb_protocol(skb)) {
168	case htons(ETH_P_IP):
169		return ntohl(CTTUPLE(skb, dst.u3.ip));
170	case htons(ETH_P_IPV6):
171		return ntohl(CTTUPLE(skb, dst.u3.ip6[3]));
172	}
173fallback:
174	return flow_get_dst(skb, flow);
175}
176
177static u32 flow_get_nfct_proto_src(const struct sk_buff *skb,
178				   const struct flow_keys *flow)
179{
180	return ntohs(CTTUPLE(skb, src.u.all));
181fallback:
182	return flow_get_proto_src(skb, flow);
183}
184
185static u32 flow_get_nfct_proto_dst(const struct sk_buff *skb,
186				   const struct flow_keys *flow)
187{
188	return ntohs(CTTUPLE(skb, dst.u.all));
189fallback:
190	return flow_get_proto_dst(skb, flow);
191}
192
193static u32 flow_get_rtclassid(const struct sk_buff *skb)
194{
195#ifdef CONFIG_IP_ROUTE_CLASSID
196	if (skb_dst(skb))
197		return skb_dst(skb)->tclassid;
198#endif
199	return 0;
200}
201
202static u32 flow_get_skuid(const struct sk_buff *skb)
203{
204	struct sock *sk = skb_to_full_sk(skb);
205
206	if (sk && sk->sk_socket && sk->sk_socket->file) {
207		kuid_t skuid = sk->sk_socket->file->f_cred->fsuid;
208
209		return from_kuid(&init_user_ns, skuid);
210	}
211	return 0;
212}
213
214static u32 flow_get_skgid(const struct sk_buff *skb)
215{
216	struct sock *sk = skb_to_full_sk(skb);
217
218	if (sk && sk->sk_socket && sk->sk_socket->file) {
219		kgid_t skgid = sk->sk_socket->file->f_cred->fsgid;
220
221		return from_kgid(&init_user_ns, skgid);
222	}
223	return 0;
224}
225
226static u32 flow_get_vlan_tag(const struct sk_buff *skb)
227{
228	u16 uninitialized_var(tag);
229
230	if (vlan_get_tag(skb, &tag) < 0)
231		return 0;
232	return tag & VLAN_VID_MASK;
233}
234
235static u32 flow_get_rxhash(struct sk_buff *skb)
236{
237	return skb_get_hash(skb);
238}
239
240static u32 flow_key_get(struct sk_buff *skb, int key, struct flow_keys *flow)
241{
242	switch (key) {
243	case FLOW_KEY_SRC:
244		return flow_get_src(skb, flow);
245	case FLOW_KEY_DST:
246		return flow_get_dst(skb, flow);
247	case FLOW_KEY_PROTO:
248		return flow_get_proto(skb, flow);
249	case FLOW_KEY_PROTO_SRC:
250		return flow_get_proto_src(skb, flow);
251	case FLOW_KEY_PROTO_DST:
252		return flow_get_proto_dst(skb, flow);
253	case FLOW_KEY_IIF:
254		return flow_get_iif(skb);
255	case FLOW_KEY_PRIORITY:
256		return flow_get_priority(skb);
257	case FLOW_KEY_MARK:
258		return flow_get_mark(skb);
259	case FLOW_KEY_NFCT:
260		return flow_get_nfct(skb);
261	case FLOW_KEY_NFCT_SRC:
262		return flow_get_nfct_src(skb, flow);
263	case FLOW_KEY_NFCT_DST:
264		return flow_get_nfct_dst(skb, flow);
265	case FLOW_KEY_NFCT_PROTO_SRC:
266		return flow_get_nfct_proto_src(skb, flow);
267	case FLOW_KEY_NFCT_PROTO_DST:
268		return flow_get_nfct_proto_dst(skb, flow);
269	case FLOW_KEY_RTCLASSID:
270		return flow_get_rtclassid(skb);
271	case FLOW_KEY_SKUID:
272		return flow_get_skuid(skb);
273	case FLOW_KEY_SKGID:
274		return flow_get_skgid(skb);
275	case FLOW_KEY_VLAN_TAG:
276		return flow_get_vlan_tag(skb);
277	case FLOW_KEY_RXHASH:
278		return flow_get_rxhash(skb);
279	default:
280		WARN_ON(1);
281		return 0;
282	}
283}
284
285#define FLOW_KEYS_NEEDED ((1 << FLOW_KEY_SRC) | 		\
286			  (1 << FLOW_KEY_DST) |			\
287			  (1 << FLOW_KEY_PROTO) |		\
288			  (1 << FLOW_KEY_PROTO_SRC) |		\
289			  (1 << FLOW_KEY_PROTO_DST) | 		\
290			  (1 << FLOW_KEY_NFCT_SRC) |		\
291			  (1 << FLOW_KEY_NFCT_DST) |		\
292			  (1 << FLOW_KEY_NFCT_PROTO_SRC) |	\
293			  (1 << FLOW_KEY_NFCT_PROTO_DST))
294
295static int flow_classify(struct sk_buff *skb, const struct tcf_proto *tp,
296			 struct tcf_result *res)
297{
298	struct flow_head *head = rcu_dereference_bh(tp->root);
299	struct flow_filter *f;
300	u32 keymask;
301	u32 classid;
302	unsigned int n, key;
303	int r;
304
305	list_for_each_entry_rcu(f, &head->filters, list) {
306		u32 keys[FLOW_KEY_MAX + 1];
307		struct flow_keys flow_keys;
308
309		if (!tcf_em_tree_match(skb, &f->ematches, NULL))
310			continue;
311
312		keymask = f->keymask;
313		if (keymask & FLOW_KEYS_NEEDED)
314			skb_flow_dissect_flow_keys(skb, &flow_keys, 0);
315
316		for (n = 0; n < f->nkeys; n++) {
317			key = ffs(keymask) - 1;
318			keymask &= ~(1 << key);
319			keys[n] = flow_key_get(skb, key, &flow_keys);
320		}
321
322		if (f->mode == FLOW_MODE_HASH)
323			classid = jhash2(keys, f->nkeys, f->hashrnd);
324		else {
325			classid = keys[0];
326			classid = (classid & f->mask) ^ f->xor;
327			classid = (classid >> f->rshift) + f->addend;
328		}
329
330		if (f->divisor)
331			classid %= f->divisor;
332
333		res->class   = 0;
334		res->classid = TC_H_MAKE(f->baseclass, f->baseclass + classid);
335
336		r = tcf_exts_exec(skb, &f->exts, res);
337		if (r < 0)
338			continue;
339		return r;
340	}
341	return -1;
342}
343
344static void flow_perturbation(struct timer_list *t)
345{
346	struct flow_filter *f = from_timer(f, t, perturb_timer);
347
348	get_random_bytes(&f->hashrnd, 4);
349	if (f->perturb_period)
350		mod_timer(&f->perturb_timer, jiffies + f->perturb_period);
351}
352
353static const struct nla_policy flow_policy[TCA_FLOW_MAX + 1] = {
354	[TCA_FLOW_KEYS]		= { .type = NLA_U32 },
355	[TCA_FLOW_MODE]		= { .type = NLA_U32 },
356	[TCA_FLOW_BASECLASS]	= { .type = NLA_U32 },
357	[TCA_FLOW_RSHIFT]	= { .type = NLA_U32 },
358	[TCA_FLOW_ADDEND]	= { .type = NLA_U32 },
359	[TCA_FLOW_MASK]		= { .type = NLA_U32 },
360	[TCA_FLOW_XOR]		= { .type = NLA_U32 },
361	[TCA_FLOW_DIVISOR]	= { .type = NLA_U32 },
362	[TCA_FLOW_ACT]		= { .type = NLA_NESTED },
363	[TCA_FLOW_POLICE]	= { .type = NLA_NESTED },
364	[TCA_FLOW_EMATCHES]	= { .type = NLA_NESTED },
365	[TCA_FLOW_PERTURB]	= { .type = NLA_U32 },
366};
367
368static void __flow_destroy_filter(struct flow_filter *f)
369{
 
 
370	del_timer_sync(&f->perturb_timer);
371	tcf_exts_destroy(&f->exts);
372	tcf_em_tree_destroy(&f->ematches);
373	tcf_exts_put_net(&f->exts);
374	kfree(f);
375}
376
377static void flow_destroy_filter_work(struct work_struct *work)
378{
379	struct flow_filter *f = container_of(to_rcu_work(work),
380					     struct flow_filter,
381					     rwork);
382	rtnl_lock();
383	__flow_destroy_filter(f);
384	rtnl_unlock();
385}
386
387static int flow_change(struct net *net, struct sk_buff *in_skb,
388		       struct tcf_proto *tp, unsigned long base,
389		       u32 handle, struct nlattr **tca,
390		       void **arg, bool ovr, bool rtnl_held,
391		       struct netlink_ext_ack *extack)
392{
393	struct flow_head *head = rtnl_dereference(tp->root);
394	struct flow_filter *fold, *fnew;
395	struct nlattr *opt = tca[TCA_OPTIONS];
396	struct nlattr *tb[TCA_FLOW_MAX + 1];
 
 
397	unsigned int nkeys = 0;
398	unsigned int perturb_period = 0;
399	u32 baseclass = 0;
400	u32 keymask = 0;
401	u32 mode;
402	int err;
403
404	if (opt == NULL)
405		return -EINVAL;
406
407	err = nla_parse_nested_deprecated(tb, TCA_FLOW_MAX, opt, flow_policy,
408					  NULL);
409	if (err < 0)
410		return err;
411
412	if (tb[TCA_FLOW_BASECLASS]) {
413		baseclass = nla_get_u32(tb[TCA_FLOW_BASECLASS]);
414		if (TC_H_MIN(baseclass) == 0)
415			return -EINVAL;
416	}
417
418	if (tb[TCA_FLOW_KEYS]) {
419		keymask = nla_get_u32(tb[TCA_FLOW_KEYS]);
420
421		nkeys = hweight32(keymask);
422		if (nkeys == 0)
423			return -EINVAL;
424
425		if (fls(keymask) - 1 > FLOW_KEY_MAX)
426			return -EOPNOTSUPP;
427
428		if ((keymask & (FLOW_KEY_SKUID|FLOW_KEY_SKGID)) &&
429		    sk_user_ns(NETLINK_CB(in_skb).sk) != &init_user_ns)
430			return -EOPNOTSUPP;
431	}
432
433	fnew = kzalloc(sizeof(*fnew), GFP_KERNEL);
434	if (!fnew)
435		return -ENOBUFS;
436
437	err = tcf_em_tree_validate(tp, tb[TCA_FLOW_EMATCHES], &fnew->ematches);
438	if (err < 0)
439		goto err1;
440
441	err = tcf_exts_init(&fnew->exts, net, TCA_FLOW_ACT, TCA_FLOW_POLICE);
442	if (err < 0)
 
 
 
 
 
443		goto err2;
444
445	err = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &fnew->exts, ovr,
446				true, extack);
447	if (err < 0)
448		goto err2;
449
450	fold = *arg;
451	if (fold) {
452		err = -EINVAL;
453		if (fold->handle != handle && handle)
454			goto err2;
455
456		/* Copy fold into fnew */
457		fnew->tp = fold->tp;
458		fnew->handle = fold->handle;
459		fnew->nkeys = fold->nkeys;
460		fnew->keymask = fold->keymask;
461		fnew->mode = fold->mode;
462		fnew->mask = fold->mask;
463		fnew->xor = fold->xor;
464		fnew->rshift = fold->rshift;
465		fnew->addend = fold->addend;
466		fnew->divisor = fold->divisor;
467		fnew->baseclass = fold->baseclass;
468		fnew->hashrnd = fold->hashrnd;
469
470		mode = fold->mode;
471		if (tb[TCA_FLOW_MODE])
472			mode = nla_get_u32(tb[TCA_FLOW_MODE]);
473		if (mode != FLOW_MODE_HASH && nkeys > 1)
474			goto err2;
475
476		if (mode == FLOW_MODE_HASH)
477			perturb_period = fold->perturb_period;
478		if (tb[TCA_FLOW_PERTURB]) {
479			if (mode != FLOW_MODE_HASH)
480				goto err2;
481			perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
482		}
483	} else {
484		err = -EINVAL;
485		if (!handle)
486			goto err2;
487		if (!tb[TCA_FLOW_KEYS])
488			goto err2;
489
490		mode = FLOW_MODE_MAP;
491		if (tb[TCA_FLOW_MODE])
492			mode = nla_get_u32(tb[TCA_FLOW_MODE]);
493		if (mode != FLOW_MODE_HASH && nkeys > 1)
494			goto err2;
495
496		if (tb[TCA_FLOW_PERTURB]) {
497			if (mode != FLOW_MODE_HASH)
498				goto err2;
499			perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
500		}
501
502		if (TC_H_MAJ(baseclass) == 0) {
503			struct Qdisc *q = tcf_block_q(tp->chain->block);
504
505			baseclass = TC_H_MAKE(q->handle, baseclass);
506		}
507		if (TC_H_MIN(baseclass) == 0)
508			baseclass = TC_H_MAKE(baseclass, 1);
509
510		fnew->handle = handle;
511		fnew->mask  = ~0U;
512		fnew->tp = tp;
513		get_random_bytes(&fnew->hashrnd, 4);
514	}
515
516	timer_setup(&fnew->perturb_timer, flow_perturbation, TIMER_DEFERRABLE);
 
 
517
518	tcf_block_netif_keep_dst(tp->chain->block);
 
 
 
519
520	if (tb[TCA_FLOW_KEYS]) {
521		fnew->keymask = keymask;
522		fnew->nkeys   = nkeys;
523	}
524
525	fnew->mode = mode;
526
527	if (tb[TCA_FLOW_MASK])
528		fnew->mask = nla_get_u32(tb[TCA_FLOW_MASK]);
529	if (tb[TCA_FLOW_XOR])
530		fnew->xor = nla_get_u32(tb[TCA_FLOW_XOR]);
531	if (tb[TCA_FLOW_RSHIFT])
532		fnew->rshift = nla_get_u32(tb[TCA_FLOW_RSHIFT]);
533	if (tb[TCA_FLOW_ADDEND])
534		fnew->addend = nla_get_u32(tb[TCA_FLOW_ADDEND]);
535
536	if (tb[TCA_FLOW_DIVISOR])
537		fnew->divisor = nla_get_u32(tb[TCA_FLOW_DIVISOR]);
538	if (baseclass)
539		fnew->baseclass = baseclass;
540
541	fnew->perturb_period = perturb_period;
542	if (perturb_period)
543		mod_timer(&fnew->perturb_timer, jiffies + perturb_period);
544
545	if (!*arg)
546		list_add_tail_rcu(&fnew->list, &head->filters);
547	else
548		list_replace_rcu(&fold->list, &fnew->list);
549
550	*arg = fnew;
551
552	if (fold) {
553		tcf_exts_get_net(&fold->exts);
554		tcf_queue_work(&fold->rwork, flow_destroy_filter_work);
555	}
556	return 0;
557
558err2:
559	tcf_exts_destroy(&fnew->exts);
560	tcf_em_tree_destroy(&fnew->ematches);
561err1:
562	kfree(fnew);
 
 
563	return err;
564}
565
566static int flow_delete(struct tcf_proto *tp, void *arg, bool *last,
567		       bool rtnl_held, struct netlink_ext_ack *extack)
568{
569	struct flow_head *head = rtnl_dereference(tp->root);
570	struct flow_filter *f = arg;
571
572	list_del_rcu(&f->list);
573	tcf_exts_get_net(&f->exts);
574	tcf_queue_work(&f->rwork, flow_destroy_filter_work);
575	*last = list_empty(&head->filters);
576	return 0;
577}
578
579static int flow_init(struct tcf_proto *tp)
580{
581	struct flow_head *head;
582
583	head = kzalloc(sizeof(*head), GFP_KERNEL);
584	if (head == NULL)
585		return -ENOBUFS;
586	INIT_LIST_HEAD(&head->filters);
587	rcu_assign_pointer(tp->root, head);
588	return 0;
589}
590
591static void flow_destroy(struct tcf_proto *tp, bool rtnl_held,
592			 struct netlink_ext_ack *extack)
593{
594	struct flow_head *head = rtnl_dereference(tp->root);
595	struct flow_filter *f, *next;
596
 
 
 
597	list_for_each_entry_safe(f, next, &head->filters, list) {
598		list_del_rcu(&f->list);
599		if (tcf_exts_get_net(&f->exts))
600			tcf_queue_work(&f->rwork, flow_destroy_filter_work);
601		else
602			__flow_destroy_filter(f);
603	}
604	kfree_rcu(head, rcu);
 
605}
606
607static void *flow_get(struct tcf_proto *tp, u32 handle)
608{
609	struct flow_head *head = rtnl_dereference(tp->root);
610	struct flow_filter *f;
611
612	list_for_each_entry(f, &head->filters, list)
613		if (f->handle == handle)
614			return f;
615	return NULL;
616}
617
618static int flow_dump(struct net *net, struct tcf_proto *tp, void *fh,
619		     struct sk_buff *skb, struct tcmsg *t, bool rtnl_held)
620{
621	struct flow_filter *f = fh;
622	struct nlattr *nest;
623
624	if (f == NULL)
625		return skb->len;
626
627	t->tcm_handle = f->handle;
628
629	nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
630	if (nest == NULL)
631		goto nla_put_failure;
632
633	if (nla_put_u32(skb, TCA_FLOW_KEYS, f->keymask) ||
634	    nla_put_u32(skb, TCA_FLOW_MODE, f->mode))
635		goto nla_put_failure;
636
637	if (f->mask != ~0 || f->xor != 0) {
638		if (nla_put_u32(skb, TCA_FLOW_MASK, f->mask) ||
639		    nla_put_u32(skb, TCA_FLOW_XOR, f->xor))
640			goto nla_put_failure;
641	}
642	if (f->rshift &&
643	    nla_put_u32(skb, TCA_FLOW_RSHIFT, f->rshift))
644		goto nla_put_failure;
645	if (f->addend &&
646	    nla_put_u32(skb, TCA_FLOW_ADDEND, f->addend))
647		goto nla_put_failure;
648
649	if (f->divisor &&
650	    nla_put_u32(skb, TCA_FLOW_DIVISOR, f->divisor))
651		goto nla_put_failure;
652	if (f->baseclass &&
653	    nla_put_u32(skb, TCA_FLOW_BASECLASS, f->baseclass))
654		goto nla_put_failure;
655
656	if (f->perturb_period &&
657	    nla_put_u32(skb, TCA_FLOW_PERTURB, f->perturb_period / HZ))
658		goto nla_put_failure;
659
660	if (tcf_exts_dump(skb, &f->exts) < 0)
661		goto nla_put_failure;
662#ifdef CONFIG_NET_EMATCH
663	if (f->ematches.hdr.nmatches &&
664	    tcf_em_tree_dump(skb, &f->ematches, TCA_FLOW_EMATCHES) < 0)
665		goto nla_put_failure;
666#endif
667	nla_nest_end(skb, nest);
668
669	if (tcf_exts_dump_stats(skb, &f->exts) < 0)
670		goto nla_put_failure;
671
672	return skb->len;
673
674nla_put_failure:
675	nla_nest_cancel(skb, nest);
676	return -1;
677}
678
679static void flow_walk(struct tcf_proto *tp, struct tcf_walker *arg,
680		      bool rtnl_held)
681{
682	struct flow_head *head = rtnl_dereference(tp->root);
683	struct flow_filter *f;
684
685	list_for_each_entry(f, &head->filters, list) {
686		if (arg->count < arg->skip)
687			goto skip;
688		if (arg->fn(tp, f, arg) < 0) {
689			arg->stop = 1;
690			break;
691		}
692skip:
693		arg->count++;
694	}
695}
696
697static struct tcf_proto_ops cls_flow_ops __read_mostly = {
698	.kind		= "flow",
699	.classify	= flow_classify,
700	.init		= flow_init,
701	.destroy	= flow_destroy,
702	.change		= flow_change,
703	.delete		= flow_delete,
704	.get		= flow_get,
705	.dump		= flow_dump,
706	.walk		= flow_walk,
707	.owner		= THIS_MODULE,
708};
709
710static int __init cls_flow_init(void)
711{
712	return register_tcf_proto_ops(&cls_flow_ops);
713}
714
715static void __exit cls_flow_exit(void)
716{
717	unregister_tcf_proto_ops(&cls_flow_ops);
718}
719
720module_init(cls_flow_init);
721module_exit(cls_flow_exit);
722
723MODULE_LICENSE("GPL");
724MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
725MODULE_DESCRIPTION("TC flow classifier");
v4.10.11
 
  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	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,
 91			  const struct flow_keys *flow)
 92{
 93	return flow->basic.ip_proto;
 94}
 95
 96static 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
105static 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
114static u32 flow_get_iif(const struct sk_buff *skb)
115{
116	return skb->skb_iif;
117}
118
119static u32 flow_get_priority(const struct sk_buff *skb)
120{
121	return skb->priority;
122}
123
124static u32 flow_get_mark(const struct sk_buff *skb)
125{
126	return skb->mark;
127}
128
129static u32 flow_get_nfct(const struct sk_buff *skb)
130{
131#if IS_ENABLED(CONFIG_NF_CONNTRACK)
132	return addr_fold(skb->nfct);
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
155static 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	}
164fallback:
165	return flow_get_src(skb, flow);
166}
167
168static 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	}
177fallback:
178	return flow_get_dst(skb, flow);
179}
180
181static 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));
185fallback:
186	return flow_get_proto_src(skb, flow);
187}
188
189static 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));
193fallback:
194	return flow_get_proto_dst(skb, flow);
195}
196
197static 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
206static 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
218static 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
230static 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
239static u32 flow_get_rxhash(struct sk_buff *skb)
240{
241	return skb_get_hash(skb);
242}
243
244static 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
299static 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
348static 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
357static 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
372static 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
 
 
 
 
 
 
 
 
 
 
382static 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);
 
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	fnew->perturb_timer.function = flow_perturbation;
512	fnew->perturb_timer.data = (unsigned long)fnew;
513	init_timer_deferrable(&fnew->perturb_timer);
514
515	tcf_exts_change(tp, &fnew->exts, &e);
516	tcf_em_tree_change(tp, &fnew->ematches, &t);
517
518	netif_keep_dst(qdisc_dev(tp->q));
519
520	if (tb[TCA_FLOW_KEYS]) {
521		fnew->keymask = keymask;
522		fnew->nkeys   = nkeys;
523	}
524
525	fnew->mode = mode;
526
527	if (tb[TCA_FLOW_MASK])
528		fnew->mask = nla_get_u32(tb[TCA_FLOW_MASK]);
529	if (tb[TCA_FLOW_XOR])
530		fnew->xor = nla_get_u32(tb[TCA_FLOW_XOR]);
531	if (tb[TCA_FLOW_RSHIFT])
532		fnew->rshift = nla_get_u32(tb[TCA_FLOW_RSHIFT]);
533	if (tb[TCA_FLOW_ADDEND])
534		fnew->addend = nla_get_u32(tb[TCA_FLOW_ADDEND]);
535
536	if (tb[TCA_FLOW_DIVISOR])
537		fnew->divisor = nla_get_u32(tb[TCA_FLOW_DIVISOR]);
538	if (baseclass)
539		fnew->baseclass = baseclass;
540
541	fnew->perturb_period = perturb_period;
542	if (perturb_period)
543		mod_timer(&fnew->perturb_timer, jiffies + perturb_period);
544
545	if (*arg == 0)
546		list_add_tail_rcu(&fnew->list, &head->filters);
547	else
548		list_replace_rcu(&fold->list, &fnew->list);
549
550	*arg = (unsigned long)fnew;
551
552	if (fold)
553		call_rcu(&fold->rcu, flow_destroy_filter);
 
 
554	return 0;
555
556err3:
557	tcf_exts_destroy(&fnew->exts);
558err2:
559	tcf_em_tree_destroy(&t);
560	kfree(fnew);
561err1:
562	tcf_exts_destroy(&e);
563	return err;
564}
565
566static int flow_delete(struct tcf_proto *tp, unsigned long arg)
 
567{
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	return 0;
573}
574
575static int flow_init(struct tcf_proto *tp)
576{
577	struct flow_head *head;
578
579	head = kzalloc(sizeof(*head), GFP_KERNEL);
580	if (head == NULL)
581		return -ENOBUFS;
582	INIT_LIST_HEAD(&head->filters);
583	rcu_assign_pointer(tp->root, head);
584	return 0;
585}
586
587static bool flow_destroy(struct tcf_proto *tp, bool force)
 
588{
589	struct flow_head *head = rtnl_dereference(tp->root);
590	struct flow_filter *f, *next;
591
592	if (!force && !list_empty(&head->filters))
593		return false;
594
595	list_for_each_entry_safe(f, next, &head->filters, list) {
596		list_del_rcu(&f->list);
597		call_rcu(&f->rcu, flow_destroy_filter);
 
 
 
598	}
599	kfree_rcu(head, rcu);
600	return true;
601}
602
603static unsigned long flow_get(struct tcf_proto *tp, u32 handle)
604{
605	struct flow_head *head = rtnl_dereference(tp->root);
606	struct flow_filter *f;
607
608	list_for_each_entry(f, &head->filters, list)
609		if (f->handle == handle)
610			return (unsigned long)f;
611	return 0;
612}
613
614static int flow_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
615		     struct sk_buff *skb, struct tcmsg *t)
616{
617	struct flow_filter *f = (struct flow_filter *)fh;
618	struct nlattr *nest;
619
620	if (f == NULL)
621		return skb->len;
622
623	t->tcm_handle = f->handle;
624
625	nest = nla_nest_start(skb, TCA_OPTIONS);
626	if (nest == NULL)
627		goto nla_put_failure;
628
629	if (nla_put_u32(skb, TCA_FLOW_KEYS, f->keymask) ||
630	    nla_put_u32(skb, TCA_FLOW_MODE, f->mode))
631		goto nla_put_failure;
632
633	if (f->mask != ~0 || f->xor != 0) {
634		if (nla_put_u32(skb, TCA_FLOW_MASK, f->mask) ||
635		    nla_put_u32(skb, TCA_FLOW_XOR, f->xor))
636			goto nla_put_failure;
637	}
638	if (f->rshift &&
639	    nla_put_u32(skb, TCA_FLOW_RSHIFT, f->rshift))
640		goto nla_put_failure;
641	if (f->addend &&
642	    nla_put_u32(skb, TCA_FLOW_ADDEND, f->addend))
643		goto nla_put_failure;
644
645	if (f->divisor &&
646	    nla_put_u32(skb, TCA_FLOW_DIVISOR, f->divisor))
647		goto nla_put_failure;
648	if (f->baseclass &&
649	    nla_put_u32(skb, TCA_FLOW_BASECLASS, f->baseclass))
650		goto nla_put_failure;
651
652	if (f->perturb_period &&
653	    nla_put_u32(skb, TCA_FLOW_PERTURB, f->perturb_period / HZ))
654		goto nla_put_failure;
655
656	if (tcf_exts_dump(skb, &f->exts) < 0)
657		goto nla_put_failure;
658#ifdef CONFIG_NET_EMATCH
659	if (f->ematches.hdr.nmatches &&
660	    tcf_em_tree_dump(skb, &f->ematches, TCA_FLOW_EMATCHES) < 0)
661		goto nla_put_failure;
662#endif
663	nla_nest_end(skb, nest);
664
665	if (tcf_exts_dump_stats(skb, &f->exts) < 0)
666		goto nla_put_failure;
667
668	return skb->len;
669
670nla_put_failure:
671	nla_nest_cancel(skb, nest);
672	return -1;
673}
674
675static void flow_walk(struct tcf_proto *tp, struct tcf_walker *arg)
 
676{
677	struct flow_head *head = rtnl_dereference(tp->root);
678	struct flow_filter *f;
679
680	list_for_each_entry(f, &head->filters, list) {
681		if (arg->count < arg->skip)
682			goto skip;
683		if (arg->fn(tp, (unsigned long)f, arg) < 0) {
684			arg->stop = 1;
685			break;
686		}
687skip:
688		arg->count++;
689	}
690}
691
692static struct tcf_proto_ops cls_flow_ops __read_mostly = {
693	.kind		= "flow",
694	.classify	= flow_classify,
695	.init		= flow_init,
696	.destroy	= flow_destroy,
697	.change		= flow_change,
698	.delete		= flow_delete,
699	.get		= flow_get,
700	.dump		= flow_dump,
701	.walk		= flow_walk,
702	.owner		= THIS_MODULE,
703};
704
705static int __init cls_flow_init(void)
706{
707	return register_tcf_proto_ops(&cls_flow_ops);
708}
709
710static void __exit cls_flow_exit(void)
711{
712	unregister_tcf_proto_ops(&cls_flow_ops);
713}
714
715module_init(cls_flow_init);
716module_exit(cls_flow_exit);
717
718MODULE_LICENSE("GPL");
719MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
720MODULE_DESCRIPTION("TC flow classifier");