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>
25 #include <net/inet_sock.h>
27 #include <net/pkt_cls.h>
29 #include <net/route.h>
30 #include <net/flow_dissector.h>
32 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
33 #include <net/netfilter/nf_conntrack.h>
37 struct list_head filters;
42 struct list_head list;
44 struct tcf_ematch_tree ematches;
46 struct timer_list perturb_timer;
63 static inline u32 addr_fold(void *addr)
65 unsigned long a = (unsigned long)addr;
67 return (a & 0xFFFFFFFF) ^ (BITS_PER_LONG > 32 ? a >> 32 : 0);
70 static u32 flow_get_src(const struct sk_buff *skb, const struct flow_keys *flow)
72 __be32 src = flow_get_u32_src(flow);
77 return addr_fold(skb->sk);
80 static u32 flow_get_dst(const struct sk_buff *skb, const struct flow_keys *flow)
82 __be32 dst = flow_get_u32_dst(flow);
87 return addr_fold(skb_dst(skb)) ^ (__force u16) tc_skb_protocol(skb);
90 static u32 flow_get_proto(const struct sk_buff *skb, const struct flow_keys *flow)
92 return flow->basic.ip_proto;
95 static u32 flow_get_proto_src(const struct sk_buff *skb, const struct flow_keys *flow)
97 if (flow->ports.ports)
98 return ntohs(flow->ports.src);
100 return addr_fold(skb->sk);
103 static u32 flow_get_proto_dst(const struct sk_buff *skb, const struct flow_keys *flow)
105 if (flow->ports.ports)
106 return ntohs(flow->ports.dst);
108 return addr_fold(skb_dst(skb)) ^ (__force u16) tc_skb_protocol(skb);
111 static u32 flow_get_iif(const struct sk_buff *skb)
116 static u32 flow_get_priority(const struct sk_buff *skb)
118 return skb->priority;
121 static u32 flow_get_mark(const struct sk_buff *skb)
126 static u32 flow_get_nfct(const struct sk_buff *skb)
128 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
129 return addr_fold(skb->nfct);
135 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
136 #define CTTUPLE(skb, member) \
138 enum ip_conntrack_info ctinfo; \
139 const struct nf_conn *ct = nf_ct_get(skb, &ctinfo); \
142 ct->tuplehash[CTINFO2DIR(ctinfo)].tuple.member; \
145 #define CTTUPLE(skb, member) \
152 static u32 flow_get_nfct_src(const struct sk_buff *skb, const struct flow_keys *flow)
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]));
161 return flow_get_src(skb, flow);
164 static u32 flow_get_nfct_dst(const struct sk_buff *skb, const struct flow_keys *flow)
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]));
173 return flow_get_dst(skb, flow);
176 static u32 flow_get_nfct_proto_src(const struct sk_buff *skb, const struct flow_keys *flow)
178 return ntohs(CTTUPLE(skb, src.u.all));
180 return flow_get_proto_src(skb, flow);
183 static u32 flow_get_nfct_proto_dst(const struct sk_buff *skb, const struct flow_keys *flow)
185 return ntohs(CTTUPLE(skb, dst.u.all));
187 return flow_get_proto_dst(skb, flow);
190 static u32 flow_get_rtclassid(const struct sk_buff *skb)
192 #ifdef CONFIG_IP_ROUTE_CLASSID
194 return skb_dst(skb)->tclassid;
199 static u32 flow_get_skuid(const struct sk_buff *skb)
201 struct sock *sk = skb_to_full_sk(skb);
203 if (sk && sk->sk_socket && sk->sk_socket->file) {
204 kuid_t skuid = sk->sk_socket->file->f_cred->fsuid;
206 return from_kuid(&init_user_ns, skuid);
211 static u32 flow_get_skgid(const struct sk_buff *skb)
213 struct sock *sk = skb_to_full_sk(skb);
215 if (sk && sk->sk_socket && sk->sk_socket->file) {
216 kgid_t skgid = sk->sk_socket->file->f_cred->fsgid;
218 return from_kgid(&init_user_ns, skgid);
223 static u32 flow_get_vlan_tag(const struct sk_buff *skb)
225 u16 uninitialized_var(tag);
227 if (vlan_get_tag(skb, &tag) < 0)
229 return tag & VLAN_VID_MASK;
232 static u32 flow_get_rxhash(struct sk_buff *skb)
234 return skb_get_hash(skb);
237 static u32 flow_key_get(struct sk_buff *skb, int key, struct flow_keys *flow)
241 return flow_get_src(skb, flow);
243 return flow_get_dst(skb, flow);
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);
251 return flow_get_iif(skb);
252 case FLOW_KEY_PRIORITY:
253 return flow_get_priority(skb);
255 return flow_get_mark(skb);
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);
269 return flow_get_skuid(skb);
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);
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))
292 static int flow_classify(struct sk_buff *skb, const struct tcf_proto *tp,
293 struct tcf_result *res)
295 struct flow_head *head = rcu_dereference_bh(tp->root);
296 struct flow_filter *f;
302 list_for_each_entry_rcu(f, &head->filters, list) {
303 u32 keys[FLOW_KEY_MAX + 1];
304 struct flow_keys flow_keys;
306 if (!tcf_em_tree_match(skb, &f->ematches, NULL))
309 keymask = f->keymask;
310 if (keymask & FLOW_KEYS_NEEDED)
311 skb_flow_dissect_flow_keys(skb, &flow_keys, 0);
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);
319 if (f->mode == FLOW_MODE_HASH)
320 classid = jhash2(keys, f->nkeys, f->hashrnd);
323 classid = (classid & f->mask) ^ f->xor;
324 classid = (classid >> f->rshift) + f->addend;
328 classid %= f->divisor;
331 res->classid = TC_H_MAKE(f->baseclass, f->baseclass + classid);
333 r = tcf_exts_exec(skb, &f->exts, res);
341 static void flow_perturbation(unsigned long arg)
343 struct flow_filter *f = (struct flow_filter *)arg;
345 get_random_bytes(&f->hashrnd, 4);
346 if (f->perturb_period)
347 mod_timer(&f->perturb_timer, jiffies + f->perturb_period);
350 static 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 },
365 static void flow_destroy_filter(struct rcu_head *head)
367 struct flow_filter *f = container_of(head, struct flow_filter, rcu);
369 del_timer_sync(&f->perturb_timer);
370 tcf_exts_destroy(&f->exts);
371 tcf_em_tree_destroy(&f->ematches);
375 static 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)
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];
385 struct tcf_ematch_tree t;
386 unsigned int nkeys = 0;
387 unsigned int perturb_period = 0;
396 err = nla_parse_nested(tb, TCA_FLOW_MAX, opt, flow_policy);
400 if (tb[TCA_FLOW_BASECLASS]) {
401 baseclass = nla_get_u32(tb[TCA_FLOW_BASECLASS]);
402 if (TC_H_MIN(baseclass) == 0)
406 if (tb[TCA_FLOW_KEYS]) {
407 keymask = nla_get_u32(tb[TCA_FLOW_KEYS]);
409 nkeys = hweight32(keymask);
413 if (fls(keymask) - 1 > FLOW_KEY_MAX)
416 if ((keymask & (FLOW_KEY_SKUID|FLOW_KEY_SKGID)) &&
417 sk_user_ns(NETLINK_CB(in_skb).sk) != &init_user_ns)
421 tcf_exts_init(&e, TCA_FLOW_ACT, TCA_FLOW_POLICE);
422 err = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &e, ovr);
426 err = tcf_em_tree_validate(tp, tb[TCA_FLOW_EMATCHES], &t);
431 fnew = kzalloc(sizeof(*fnew), GFP_KERNEL);
435 tcf_exts_init(&fnew->exts, TCA_FLOW_ACT, TCA_FLOW_POLICE);
437 fold = (struct flow_filter *)*arg;
440 if (fold->handle != handle && handle)
443 /* Copy fold into fnew */
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;
458 if (tb[TCA_FLOW_MODE])
459 mode = nla_get_u32(tb[TCA_FLOW_MODE]);
460 if (mode != FLOW_MODE_HASH && nkeys > 1)
463 if (mode == FLOW_MODE_HASH)
464 perturb_period = fold->perturb_period;
465 if (tb[TCA_FLOW_PERTURB]) {
466 if (mode != FLOW_MODE_HASH)
468 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
474 if (!tb[TCA_FLOW_KEYS])
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)
483 if (tb[TCA_FLOW_PERTURB]) {
484 if (mode != FLOW_MODE_HASH)
486 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
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);
494 fnew->handle = handle;
497 get_random_bytes(&fnew->hashrnd, 4);
500 fnew->perturb_timer.function = flow_perturbation;
501 fnew->perturb_timer.data = (unsigned long)fnew;
502 init_timer_deferrable(&fnew->perturb_timer);
504 tcf_exts_change(tp, &fnew->exts, &e);
505 tcf_em_tree_change(tp, &fnew->ematches, &t);
507 netif_keep_dst(qdisc_dev(tp->q));
509 if (tb[TCA_FLOW_KEYS]) {
510 fnew->keymask = keymask;
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]);
525 if (tb[TCA_FLOW_DIVISOR])
526 fnew->divisor = nla_get_u32(tb[TCA_FLOW_DIVISOR]);
528 fnew->baseclass = baseclass;
530 fnew->perturb_period = perturb_period;
532 mod_timer(&fnew->perturb_timer, jiffies + perturb_period);
535 list_add_tail_rcu(&fnew->list, &head->filters);
537 list_replace_rcu(&fold->list, &fnew->list);
539 *arg = (unsigned long)fnew;
542 call_rcu(&fold->rcu, flow_destroy_filter);
546 tcf_em_tree_destroy(&t);
549 tcf_exts_destroy(&e);
553 static int flow_delete(struct tcf_proto *tp, unsigned long arg)
555 struct flow_filter *f = (struct flow_filter *)arg;
557 list_del_rcu(&f->list);
558 call_rcu(&f->rcu, flow_destroy_filter);
562 static int flow_init(struct tcf_proto *tp)
564 struct flow_head *head;
566 head = kzalloc(sizeof(*head), GFP_KERNEL);
569 INIT_LIST_HEAD(&head->filters);
570 rcu_assign_pointer(tp->root, head);
574 static bool flow_destroy(struct tcf_proto *tp, bool force)
576 struct flow_head *head = rtnl_dereference(tp->root);
577 struct flow_filter *f, *next;
579 if (!force && !list_empty(&head->filters))
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);
586 RCU_INIT_POINTER(tp->root, NULL);
587 kfree_rcu(head, rcu);
591 static unsigned long flow_get(struct tcf_proto *tp, u32 handle)
593 struct flow_head *head = rtnl_dereference(tp->root);
594 struct flow_filter *f;
596 list_for_each_entry(f, &head->filters, list)
597 if (f->handle == handle)
598 return (unsigned long)f;
602 static int flow_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
603 struct sk_buff *skb, struct tcmsg *t)
605 struct flow_filter *f = (struct flow_filter *)fh;
611 t->tcm_handle = f->handle;
613 nest = nla_nest_start(skb, TCA_OPTIONS);
615 goto nla_put_failure;
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;
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;
627 nla_put_u32(skb, TCA_FLOW_RSHIFT, f->rshift))
628 goto nla_put_failure;
630 nla_put_u32(skb, TCA_FLOW_ADDEND, f->addend))
631 goto nla_put_failure;
634 nla_put_u32(skb, TCA_FLOW_DIVISOR, f->divisor))
635 goto nla_put_failure;
637 nla_put_u32(skb, TCA_FLOW_BASECLASS, f->baseclass))
638 goto nla_put_failure;
640 if (f->perturb_period &&
641 nla_put_u32(skb, TCA_FLOW_PERTURB, f->perturb_period / HZ))
642 goto nla_put_failure;
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;
651 nla_nest_end(skb, nest);
653 if (tcf_exts_dump_stats(skb, &f->exts) < 0)
654 goto nla_put_failure;
659 nla_nest_cancel(skb, nest);
663 static void flow_walk(struct tcf_proto *tp, struct tcf_walker *arg)
665 struct flow_head *head = rtnl_dereference(tp->root);
666 struct flow_filter *f;
668 list_for_each_entry(f, &head->filters, list) {
669 if (arg->count < arg->skip)
671 if (arg->fn(tp, (unsigned long)f, arg) < 0) {
680 static struct tcf_proto_ops cls_flow_ops __read_mostly = {
682 .classify = flow_classify,
684 .destroy = flow_destroy,
685 .change = flow_change,
686 .delete = flow_delete,
690 .owner = THIS_MODULE,
693 static int __init cls_flow_init(void)
695 return register_tcf_proto_ops(&cls_flow_ops);
698 static void __exit cls_flow_exit(void)
700 unregister_tcf_proto_ops(&cls_flow_ops);
703 module_init(cls_flow_init);
704 module_exit(cls_flow_exit);
706 MODULE_LICENSE("GPL");
707 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
708 MODULE_DESCRIPTION("TC flow classifier");