2 * net/sched/cls_flow.c Generic flow classifier
4 * Copyright (c) 2007, 2008 Patrick McHardy <kaber@trash.net>
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.
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>
21 #include <linux/ipv6.h>
22 #include <linux/if_vlan.h>
23 #include <linux/slab.h>
24 #include <linux/module.h>
26 #include <net/pkt_cls.h>
28 #include <net/route.h>
29 #include <net/flow_keys.h>
31 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
32 #include <net/netfilter/nf_conntrack.h>
36 struct list_head filters;
40 struct list_head list;
42 struct tcf_ematch_tree ematches;
43 struct timer_list perturb_timer;
59 static const struct tcf_ext_map flow_ext_map = {
60 .action = TCA_FLOW_ACT,
61 .police = TCA_FLOW_POLICE,
64 static inline u32 addr_fold(void *addr)
66 unsigned long a = (unsigned long)addr;
68 return (a & 0xFFFFFFFF) ^ (BITS_PER_LONG > 32 ? a >> 32 : 0);
71 static u32 flow_get_src(const struct sk_buff *skb, const struct flow_keys *flow)
74 return ntohl(flow->src);
75 return addr_fold(skb->sk);
78 static u32 flow_get_dst(const struct sk_buff *skb, const struct flow_keys *flow)
81 return ntohl(flow->dst);
82 return addr_fold(skb_dst(skb)) ^ (__force u16)skb->protocol;
85 static u32 flow_get_proto(const struct sk_buff *skb, const struct flow_keys *flow)
87 return flow->ip_proto;
90 static u32 flow_get_proto_src(const struct sk_buff *skb, const struct flow_keys *flow)
93 return ntohs(flow->port16[0]);
95 return addr_fold(skb->sk);
98 static u32 flow_get_proto_dst(const struct sk_buff *skb, const struct flow_keys *flow)
101 return ntohs(flow->port16[1]);
103 return addr_fold(skb_dst(skb)) ^ (__force u16)skb->protocol;
106 static u32 flow_get_iif(const struct sk_buff *skb)
111 static u32 flow_get_priority(const struct sk_buff *skb)
113 return skb->priority;
116 static u32 flow_get_mark(const struct sk_buff *skb)
121 static u32 flow_get_nfct(const struct sk_buff *skb)
123 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
124 return addr_fold(skb->nfct);
130 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
131 #define CTTUPLE(skb, member) \
133 enum ip_conntrack_info ctinfo; \
134 const struct nf_conn *ct = nf_ct_get(skb, &ctinfo); \
137 ct->tuplehash[CTINFO2DIR(ctinfo)].tuple.member; \
140 #define CTTUPLE(skb, member) \
147 static u32 flow_get_nfct_src(const struct sk_buff *skb, const struct flow_keys *flow)
149 switch (skb->protocol) {
150 case htons(ETH_P_IP):
151 return ntohl(CTTUPLE(skb, src.u3.ip));
152 case htons(ETH_P_IPV6):
153 return ntohl(CTTUPLE(skb, src.u3.ip6[3]));
156 return flow_get_src(skb, flow);
159 static u32 flow_get_nfct_dst(const struct sk_buff *skb, const struct flow_keys *flow)
161 switch (skb->protocol) {
162 case htons(ETH_P_IP):
163 return ntohl(CTTUPLE(skb, dst.u3.ip));
164 case htons(ETH_P_IPV6):
165 return ntohl(CTTUPLE(skb, dst.u3.ip6[3]));
168 return flow_get_dst(skb, flow);
171 static u32 flow_get_nfct_proto_src(const struct sk_buff *skb, const struct flow_keys *flow)
173 return ntohs(CTTUPLE(skb, src.u.all));
175 return flow_get_proto_src(skb, flow);
178 static u32 flow_get_nfct_proto_dst(const struct sk_buff *skb, const struct flow_keys *flow)
180 return ntohs(CTTUPLE(skb, dst.u.all));
182 return flow_get_proto_dst(skb, flow);
185 static u32 flow_get_rtclassid(const struct sk_buff *skb)
187 #ifdef CONFIG_IP_ROUTE_CLASSID
189 return skb_dst(skb)->tclassid;
194 static u32 flow_get_skuid(const struct sk_buff *skb)
196 if (skb->sk && skb->sk->sk_socket && skb->sk->sk_socket->file)
197 return skb->sk->sk_socket->file->f_cred->fsuid;
201 static u32 flow_get_skgid(const struct sk_buff *skb)
203 if (skb->sk && skb->sk->sk_socket && skb->sk->sk_socket->file)
204 return skb->sk->sk_socket->file->f_cred->fsgid;
208 static u32 flow_get_vlan_tag(const struct sk_buff *skb)
210 u16 uninitialized_var(tag);
212 if (vlan_get_tag(skb, &tag) < 0)
214 return tag & VLAN_VID_MASK;
217 static u32 flow_get_rxhash(struct sk_buff *skb)
219 return skb_get_rxhash(skb);
222 static u32 flow_key_get(struct sk_buff *skb, int key, struct flow_keys *flow)
226 return flow_get_src(skb, flow);
228 return flow_get_dst(skb, flow);
230 return flow_get_proto(skb, flow);
231 case FLOW_KEY_PROTO_SRC:
232 return flow_get_proto_src(skb, flow);
233 case FLOW_KEY_PROTO_DST:
234 return flow_get_proto_dst(skb, flow);
236 return flow_get_iif(skb);
237 case FLOW_KEY_PRIORITY:
238 return flow_get_priority(skb);
240 return flow_get_mark(skb);
242 return flow_get_nfct(skb);
243 case FLOW_KEY_NFCT_SRC:
244 return flow_get_nfct_src(skb, flow);
245 case FLOW_KEY_NFCT_DST:
246 return flow_get_nfct_dst(skb, flow);
247 case FLOW_KEY_NFCT_PROTO_SRC:
248 return flow_get_nfct_proto_src(skb, flow);
249 case FLOW_KEY_NFCT_PROTO_DST:
250 return flow_get_nfct_proto_dst(skb, flow);
251 case FLOW_KEY_RTCLASSID:
252 return flow_get_rtclassid(skb);
254 return flow_get_skuid(skb);
256 return flow_get_skgid(skb);
257 case FLOW_KEY_VLAN_TAG:
258 return flow_get_vlan_tag(skb);
259 case FLOW_KEY_RXHASH:
260 return flow_get_rxhash(skb);
267 #define FLOW_KEYS_NEEDED ((1 << FLOW_KEY_SRC) | \
268 (1 << FLOW_KEY_DST) | \
269 (1 << FLOW_KEY_PROTO) | \
270 (1 << FLOW_KEY_PROTO_SRC) | \
271 (1 << FLOW_KEY_PROTO_DST) | \
272 (1 << FLOW_KEY_NFCT_SRC) | \
273 (1 << FLOW_KEY_NFCT_DST) | \
274 (1 << FLOW_KEY_NFCT_PROTO_SRC) | \
275 (1 << FLOW_KEY_NFCT_PROTO_DST))
277 static int flow_classify(struct sk_buff *skb, const struct tcf_proto *tp,
278 struct tcf_result *res)
280 struct flow_head *head = tp->root;
281 struct flow_filter *f;
287 list_for_each_entry(f, &head->filters, list) {
288 u32 keys[FLOW_KEY_MAX + 1];
289 struct flow_keys flow_keys;
291 if (!tcf_em_tree_match(skb, &f->ematches, NULL))
294 keymask = f->keymask;
295 if (keymask & FLOW_KEYS_NEEDED)
296 skb_flow_dissect(skb, &flow_keys);
298 for (n = 0; n < f->nkeys; n++) {
299 key = ffs(keymask) - 1;
300 keymask &= ~(1 << key);
301 keys[n] = flow_key_get(skb, key, &flow_keys);
304 if (f->mode == FLOW_MODE_HASH)
305 classid = jhash2(keys, f->nkeys, f->hashrnd);
308 classid = (classid & f->mask) ^ f->xor;
309 classid = (classid >> f->rshift) + f->addend;
313 classid %= f->divisor;
316 res->classid = TC_H_MAKE(f->baseclass, f->baseclass + classid);
318 r = tcf_exts_exec(skb, &f->exts, res);
326 static void flow_perturbation(unsigned long arg)
328 struct flow_filter *f = (struct flow_filter *)arg;
330 get_random_bytes(&f->hashrnd, 4);
331 if (f->perturb_period)
332 mod_timer(&f->perturb_timer, jiffies + f->perturb_period);
335 static const struct nla_policy flow_policy[TCA_FLOW_MAX + 1] = {
336 [TCA_FLOW_KEYS] = { .type = NLA_U32 },
337 [TCA_FLOW_MODE] = { .type = NLA_U32 },
338 [TCA_FLOW_BASECLASS] = { .type = NLA_U32 },
339 [TCA_FLOW_RSHIFT] = { .type = NLA_U32 },
340 [TCA_FLOW_ADDEND] = { .type = NLA_U32 },
341 [TCA_FLOW_MASK] = { .type = NLA_U32 },
342 [TCA_FLOW_XOR] = { .type = NLA_U32 },
343 [TCA_FLOW_DIVISOR] = { .type = NLA_U32 },
344 [TCA_FLOW_ACT] = { .type = NLA_NESTED },
345 [TCA_FLOW_POLICE] = { .type = NLA_NESTED },
346 [TCA_FLOW_EMATCHES] = { .type = NLA_NESTED },
347 [TCA_FLOW_PERTURB] = { .type = NLA_U32 },
350 static int flow_change(struct tcf_proto *tp, unsigned long base,
351 u32 handle, struct nlattr **tca,
354 struct flow_head *head = tp->root;
355 struct flow_filter *f;
356 struct nlattr *opt = tca[TCA_OPTIONS];
357 struct nlattr *tb[TCA_FLOW_MAX + 1];
359 struct tcf_ematch_tree t;
360 unsigned int nkeys = 0;
361 unsigned int perturb_period = 0;
370 err = nla_parse_nested(tb, TCA_FLOW_MAX, opt, flow_policy);
374 if (tb[TCA_FLOW_BASECLASS]) {
375 baseclass = nla_get_u32(tb[TCA_FLOW_BASECLASS]);
376 if (TC_H_MIN(baseclass) == 0)
380 if (tb[TCA_FLOW_KEYS]) {
381 keymask = nla_get_u32(tb[TCA_FLOW_KEYS]);
383 nkeys = hweight32(keymask);
387 if (fls(keymask) - 1 > FLOW_KEY_MAX)
391 err = tcf_exts_validate(tp, tb, tca[TCA_RATE], &e, &flow_ext_map);
395 err = tcf_em_tree_validate(tp, tb[TCA_FLOW_EMATCHES], &t);
399 f = (struct flow_filter *)*arg;
402 if (f->handle != handle && handle)
406 if (tb[TCA_FLOW_MODE])
407 mode = nla_get_u32(tb[TCA_FLOW_MODE]);
408 if (mode != FLOW_MODE_HASH && nkeys > 1)
411 if (mode == FLOW_MODE_HASH)
412 perturb_period = f->perturb_period;
413 if (tb[TCA_FLOW_PERTURB]) {
414 if (mode != FLOW_MODE_HASH)
416 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
422 if (!tb[TCA_FLOW_KEYS])
425 mode = FLOW_MODE_MAP;
426 if (tb[TCA_FLOW_MODE])
427 mode = nla_get_u32(tb[TCA_FLOW_MODE]);
428 if (mode != FLOW_MODE_HASH && nkeys > 1)
431 if (tb[TCA_FLOW_PERTURB]) {
432 if (mode != FLOW_MODE_HASH)
434 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
437 if (TC_H_MAJ(baseclass) == 0)
438 baseclass = TC_H_MAKE(tp->q->handle, baseclass);
439 if (TC_H_MIN(baseclass) == 0)
440 baseclass = TC_H_MAKE(baseclass, 1);
443 f = kzalloc(sizeof(*f), GFP_KERNEL);
450 get_random_bytes(&f->hashrnd, 4);
451 f->perturb_timer.function = flow_perturbation;
452 f->perturb_timer.data = (unsigned long)f;
453 init_timer_deferrable(&f->perturb_timer);
456 tcf_exts_change(tp, &f->exts, &e);
457 tcf_em_tree_change(tp, &f->ematches, &t);
461 if (tb[TCA_FLOW_KEYS]) {
462 f->keymask = keymask;
468 if (tb[TCA_FLOW_MASK])
469 f->mask = nla_get_u32(tb[TCA_FLOW_MASK]);
470 if (tb[TCA_FLOW_XOR])
471 f->xor = nla_get_u32(tb[TCA_FLOW_XOR]);
472 if (tb[TCA_FLOW_RSHIFT])
473 f->rshift = nla_get_u32(tb[TCA_FLOW_RSHIFT]);
474 if (tb[TCA_FLOW_ADDEND])
475 f->addend = nla_get_u32(tb[TCA_FLOW_ADDEND]);
477 if (tb[TCA_FLOW_DIVISOR])
478 f->divisor = nla_get_u32(tb[TCA_FLOW_DIVISOR]);
480 f->baseclass = baseclass;
482 f->perturb_period = perturb_period;
483 del_timer(&f->perturb_timer);
485 mod_timer(&f->perturb_timer, jiffies + perturb_period);
488 list_add_tail(&f->list, &head->filters);
492 *arg = (unsigned long)f;
496 tcf_em_tree_destroy(tp, &t);
498 tcf_exts_destroy(tp, &e);
502 static void flow_destroy_filter(struct tcf_proto *tp, struct flow_filter *f)
504 del_timer_sync(&f->perturb_timer);
505 tcf_exts_destroy(tp, &f->exts);
506 tcf_em_tree_destroy(tp, &f->ematches);
510 static int flow_delete(struct tcf_proto *tp, unsigned long arg)
512 struct flow_filter *f = (struct flow_filter *)arg;
517 flow_destroy_filter(tp, f);
521 static int flow_init(struct tcf_proto *tp)
523 struct flow_head *head;
525 head = kzalloc(sizeof(*head), GFP_KERNEL);
528 INIT_LIST_HEAD(&head->filters);
533 static void flow_destroy(struct tcf_proto *tp)
535 struct flow_head *head = tp->root;
536 struct flow_filter *f, *next;
538 list_for_each_entry_safe(f, next, &head->filters, list) {
540 flow_destroy_filter(tp, f);
545 static unsigned long flow_get(struct tcf_proto *tp, u32 handle)
547 struct flow_head *head = tp->root;
548 struct flow_filter *f;
550 list_for_each_entry(f, &head->filters, list)
551 if (f->handle == handle)
552 return (unsigned long)f;
556 static void flow_put(struct tcf_proto *tp, unsigned long f)
560 static int flow_dump(struct tcf_proto *tp, unsigned long fh,
561 struct sk_buff *skb, struct tcmsg *t)
563 struct flow_filter *f = (struct flow_filter *)fh;
569 t->tcm_handle = f->handle;
571 nest = nla_nest_start(skb, TCA_OPTIONS);
573 goto nla_put_failure;
575 if (nla_put_u32(skb, TCA_FLOW_KEYS, f->keymask) ||
576 nla_put_u32(skb, TCA_FLOW_MODE, f->mode))
577 goto nla_put_failure;
579 if (f->mask != ~0 || f->xor != 0) {
580 if (nla_put_u32(skb, TCA_FLOW_MASK, f->mask) ||
581 nla_put_u32(skb, TCA_FLOW_XOR, f->xor))
582 goto nla_put_failure;
585 nla_put_u32(skb, TCA_FLOW_RSHIFT, f->rshift))
586 goto nla_put_failure;
588 nla_put_u32(skb, TCA_FLOW_ADDEND, f->addend))
589 goto nla_put_failure;
592 nla_put_u32(skb, TCA_FLOW_DIVISOR, f->divisor))
593 goto nla_put_failure;
595 nla_put_u32(skb, TCA_FLOW_BASECLASS, f->baseclass))
596 goto nla_put_failure;
598 if (f->perturb_period &&
599 nla_put_u32(skb, TCA_FLOW_PERTURB, f->perturb_period / HZ))
600 goto nla_put_failure;
602 if (tcf_exts_dump(skb, &f->exts, &flow_ext_map) < 0)
603 goto nla_put_failure;
604 #ifdef CONFIG_NET_EMATCH
605 if (f->ematches.hdr.nmatches &&
606 tcf_em_tree_dump(skb, &f->ematches, TCA_FLOW_EMATCHES) < 0)
607 goto nla_put_failure;
609 nla_nest_end(skb, nest);
611 if (tcf_exts_dump_stats(skb, &f->exts, &flow_ext_map) < 0)
612 goto nla_put_failure;
617 nlmsg_trim(skb, nest);
621 static void flow_walk(struct tcf_proto *tp, struct tcf_walker *arg)
623 struct flow_head *head = tp->root;
624 struct flow_filter *f;
626 list_for_each_entry(f, &head->filters, list) {
627 if (arg->count < arg->skip)
629 if (arg->fn(tp, (unsigned long)f, arg) < 0) {
638 static struct tcf_proto_ops cls_flow_ops __read_mostly = {
640 .classify = flow_classify,
642 .destroy = flow_destroy,
643 .change = flow_change,
644 .delete = flow_delete,
649 .owner = THIS_MODULE,
652 static int __init cls_flow_init(void)
654 return register_tcf_proto_ops(&cls_flow_ops);
657 static void __exit cls_flow_exit(void)
659 unregister_tcf_proto_ops(&cls_flow_ops);
662 module_init(cls_flow_init);
663 module_exit(cls_flow_exit);
665 MODULE_LICENSE("GPL");
666 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
667 MODULE_DESCRIPTION("TC flow classifier");