2 * net/sched/cls_flow.c Generic flow classifier
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 inline u32 addr_fold(void *addr)
61 unsigned long a = (unsigned long)addr;
63 return (a & 0xFFFFFFFF) ^ (BITS_PER_LONG > 32 ? a >> 32 : 0);
66 static u32 flow_get_src(const struct sk_buff *skb, const struct flow_keys *flow)
69 return ntohl(flow->src);
70 return addr_fold(skb->sk);
73 static u32 flow_get_dst(const struct sk_buff *skb, const struct flow_keys *flow)
76 return ntohl(flow->dst);
77 return addr_fold(skb_dst(skb)) ^ (__force u16)skb->protocol;
80 static u32 flow_get_proto(const struct sk_buff *skb, const struct flow_keys *flow)
82 return flow->ip_proto;
85 static u32 flow_get_proto_src(const struct sk_buff *skb, const struct flow_keys *flow)
88 return ntohs(flow->port16[0]);
90 return addr_fold(skb->sk);
93 static u32 flow_get_proto_dst(const struct sk_buff *skb, const struct flow_keys *flow)
96 return ntohs(flow->port16[1]);
98 return addr_fold(skb_dst(skb)) ^ (__force u16)skb->protocol;
101 static u32 flow_get_iif(const struct sk_buff *skb)
106 static u32 flow_get_priority(const struct sk_buff *skb)
108 return skb->priority;
111 static u32 flow_get_mark(const struct sk_buff *skb)
116 static u32 flow_get_nfct(const struct sk_buff *skb)
118 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
119 return addr_fold(skb->nfct);
125 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
126 #define CTTUPLE(skb, member) \
128 enum ip_conntrack_info ctinfo; \
129 const struct nf_conn *ct = nf_ct_get(skb, &ctinfo); \
132 ct->tuplehash[CTINFO2DIR(ctinfo)].tuple.member; \
135 #define CTTUPLE(skb, member) \
142 static u32 flow_get_nfct_src(const struct sk_buff *skb, const struct flow_keys *flow)
144 switch (skb->protocol) {
145 case htons(ETH_P_IP):
146 return ntohl(CTTUPLE(skb, src.u3.ip));
147 case htons(ETH_P_IPV6):
148 return ntohl(CTTUPLE(skb, src.u3.ip6[3]));
151 return flow_get_src(skb, flow);
154 static u32 flow_get_nfct_dst(const struct sk_buff *skb, const struct flow_keys *flow)
156 switch (skb->protocol) {
157 case htons(ETH_P_IP):
158 return ntohl(CTTUPLE(skb, dst.u3.ip));
159 case htons(ETH_P_IPV6):
160 return ntohl(CTTUPLE(skb, dst.u3.ip6[3]));
163 return flow_get_dst(skb, flow);
166 static u32 flow_get_nfct_proto_src(const struct sk_buff *skb, const struct flow_keys *flow)
168 return ntohs(CTTUPLE(skb, src.u.all));
170 return flow_get_proto_src(skb, flow);
173 static u32 flow_get_nfct_proto_dst(const struct sk_buff *skb, const struct flow_keys *flow)
175 return ntohs(CTTUPLE(skb, dst.u.all));
177 return flow_get_proto_dst(skb, flow);
180 static u32 flow_get_rtclassid(const struct sk_buff *skb)
182 #ifdef CONFIG_IP_ROUTE_CLASSID
184 return skb_dst(skb)->tclassid;
189 static u32 flow_get_skuid(const struct sk_buff *skb)
191 if (skb->sk && skb->sk->sk_socket && skb->sk->sk_socket->file) {
192 kuid_t skuid = skb->sk->sk_socket->file->f_cred->fsuid;
193 return from_kuid(&init_user_ns, skuid);
198 static u32 flow_get_skgid(const struct sk_buff *skb)
200 if (skb->sk && skb->sk->sk_socket && skb->sk->sk_socket->file) {
201 kgid_t skgid = skb->sk->sk_socket->file->f_cred->fsgid;
202 return from_kgid(&init_user_ns, skgid);
207 static u32 flow_get_vlan_tag(const struct sk_buff *skb)
209 u16 uninitialized_var(tag);
211 if (vlan_get_tag(skb, &tag) < 0)
213 return tag & VLAN_VID_MASK;
216 static u32 flow_get_rxhash(struct sk_buff *skb)
218 return skb_get_hash(skb);
221 static u32 flow_key_get(struct sk_buff *skb, int key, struct flow_keys *flow)
225 return flow_get_src(skb, flow);
227 return flow_get_dst(skb, flow);
229 return flow_get_proto(skb, flow);
230 case FLOW_KEY_PROTO_SRC:
231 return flow_get_proto_src(skb, flow);
232 case FLOW_KEY_PROTO_DST:
233 return flow_get_proto_dst(skb, flow);
235 return flow_get_iif(skb);
236 case FLOW_KEY_PRIORITY:
237 return flow_get_priority(skb);
239 return flow_get_mark(skb);
241 return flow_get_nfct(skb);
242 case FLOW_KEY_NFCT_SRC:
243 return flow_get_nfct_src(skb, flow);
244 case FLOW_KEY_NFCT_DST:
245 return flow_get_nfct_dst(skb, flow);
246 case FLOW_KEY_NFCT_PROTO_SRC:
247 return flow_get_nfct_proto_src(skb, flow);
248 case FLOW_KEY_NFCT_PROTO_DST:
249 return flow_get_nfct_proto_dst(skb, flow);
250 case FLOW_KEY_RTCLASSID:
251 return flow_get_rtclassid(skb);
253 return flow_get_skuid(skb);
255 return flow_get_skgid(skb);
256 case FLOW_KEY_VLAN_TAG:
257 return flow_get_vlan_tag(skb);
258 case FLOW_KEY_RXHASH:
259 return flow_get_rxhash(skb);
266 #define FLOW_KEYS_NEEDED ((1 << FLOW_KEY_SRC) | \
267 (1 << FLOW_KEY_DST) | \
268 (1 << FLOW_KEY_PROTO) | \
269 (1 << FLOW_KEY_PROTO_SRC) | \
270 (1 << FLOW_KEY_PROTO_DST) | \
271 (1 << FLOW_KEY_NFCT_SRC) | \
272 (1 << FLOW_KEY_NFCT_DST) | \
273 (1 << FLOW_KEY_NFCT_PROTO_SRC) | \
274 (1 << FLOW_KEY_NFCT_PROTO_DST))
276 static int flow_classify(struct sk_buff *skb, const struct tcf_proto *tp,
277 struct tcf_result *res)
279 struct flow_head *head = tp->root;
280 struct flow_filter *f;
286 list_for_each_entry(f, &head->filters, list) {
287 u32 keys[FLOW_KEY_MAX + 1];
288 struct flow_keys flow_keys;
290 if (!tcf_em_tree_match(skb, &f->ematches, NULL))
293 keymask = f->keymask;
294 if (keymask & FLOW_KEYS_NEEDED)
295 skb_flow_dissect(skb, &flow_keys);
297 for (n = 0; n < f->nkeys; n++) {
298 key = ffs(keymask) - 1;
299 keymask &= ~(1 << key);
300 keys[n] = flow_key_get(skb, key, &flow_keys);
303 if (f->mode == FLOW_MODE_HASH)
304 classid = jhash2(keys, f->nkeys, f->hashrnd);
307 classid = (classid & f->mask) ^ f->xor;
308 classid = (classid >> f->rshift) + f->addend;
312 classid %= f->divisor;
315 res->classid = TC_H_MAKE(f->baseclass, f->baseclass + classid);
317 r = tcf_exts_exec(skb, &f->exts, res);
325 static void flow_perturbation(unsigned long arg)
327 struct flow_filter *f = (struct flow_filter *)arg;
329 get_random_bytes(&f->hashrnd, 4);
330 if (f->perturb_period)
331 mod_timer(&f->perturb_timer, jiffies + f->perturb_period);
334 static const struct nla_policy flow_policy[TCA_FLOW_MAX + 1] = {
335 [TCA_FLOW_KEYS] = { .type = NLA_U32 },
336 [TCA_FLOW_MODE] = { .type = NLA_U32 },
337 [TCA_FLOW_BASECLASS] = { .type = NLA_U32 },
338 [TCA_FLOW_RSHIFT] = { .type = NLA_U32 },
339 [TCA_FLOW_ADDEND] = { .type = NLA_U32 },
340 [TCA_FLOW_MASK] = { .type = NLA_U32 },
341 [TCA_FLOW_XOR] = { .type = NLA_U32 },
342 [TCA_FLOW_DIVISOR] = { .type = NLA_U32 },
343 [TCA_FLOW_ACT] = { .type = NLA_NESTED },
344 [TCA_FLOW_POLICE] = { .type = NLA_NESTED },
345 [TCA_FLOW_EMATCHES] = { .type = NLA_NESTED },
346 [TCA_FLOW_PERTURB] = { .type = NLA_U32 },
349 static int flow_change(struct net *net, struct sk_buff *in_skb,
350 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)
390 if ((keymask & (FLOW_KEY_SKUID|FLOW_KEY_SKGID)) &&
391 sk_user_ns(NETLINK_CB(in_skb).sk) != &init_user_ns)
395 tcf_exts_init(&e, TCA_FLOW_ACT, TCA_FLOW_POLICE);
396 err = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &e);
400 err = tcf_em_tree_validate(tp, tb[TCA_FLOW_EMATCHES], &t);
404 f = (struct flow_filter *)*arg;
407 if (f->handle != handle && handle)
411 if (tb[TCA_FLOW_MODE])
412 mode = nla_get_u32(tb[TCA_FLOW_MODE]);
413 if (mode != FLOW_MODE_HASH && nkeys > 1)
416 if (mode == FLOW_MODE_HASH)
417 perturb_period = f->perturb_period;
418 if (tb[TCA_FLOW_PERTURB]) {
419 if (mode != FLOW_MODE_HASH)
421 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
427 if (!tb[TCA_FLOW_KEYS])
430 mode = FLOW_MODE_MAP;
431 if (tb[TCA_FLOW_MODE])
432 mode = nla_get_u32(tb[TCA_FLOW_MODE]);
433 if (mode != FLOW_MODE_HASH && nkeys > 1)
436 if (tb[TCA_FLOW_PERTURB]) {
437 if (mode != FLOW_MODE_HASH)
439 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
442 if (TC_H_MAJ(baseclass) == 0)
443 baseclass = TC_H_MAKE(tp->q->handle, baseclass);
444 if (TC_H_MIN(baseclass) == 0)
445 baseclass = TC_H_MAKE(baseclass, 1);
448 f = kzalloc(sizeof(*f), GFP_KERNEL);
454 tcf_exts_init(&f->exts, TCA_FLOW_ACT, TCA_FLOW_POLICE);
456 get_random_bytes(&f->hashrnd, 4);
457 f->perturb_timer.function = flow_perturbation;
458 f->perturb_timer.data = (unsigned long)f;
459 init_timer_deferrable(&f->perturb_timer);
462 tcf_exts_change(tp, &f->exts, &e);
463 tcf_em_tree_change(tp, &f->ematches, &t);
467 if (tb[TCA_FLOW_KEYS]) {
468 f->keymask = keymask;
474 if (tb[TCA_FLOW_MASK])
475 f->mask = nla_get_u32(tb[TCA_FLOW_MASK]);
476 if (tb[TCA_FLOW_XOR])
477 f->xor = nla_get_u32(tb[TCA_FLOW_XOR]);
478 if (tb[TCA_FLOW_RSHIFT])
479 f->rshift = nla_get_u32(tb[TCA_FLOW_RSHIFT]);
480 if (tb[TCA_FLOW_ADDEND])
481 f->addend = nla_get_u32(tb[TCA_FLOW_ADDEND]);
483 if (tb[TCA_FLOW_DIVISOR])
484 f->divisor = nla_get_u32(tb[TCA_FLOW_DIVISOR]);
486 f->baseclass = baseclass;
488 f->perturb_period = perturb_period;
489 del_timer(&f->perturb_timer);
491 mod_timer(&f->perturb_timer, jiffies + perturb_period);
494 list_add_tail(&f->list, &head->filters);
498 *arg = (unsigned long)f;
502 tcf_em_tree_destroy(tp, &t);
504 tcf_exts_destroy(tp, &e);
508 static void flow_destroy_filter(struct tcf_proto *tp, struct flow_filter *f)
510 del_timer_sync(&f->perturb_timer);
511 tcf_exts_destroy(tp, &f->exts);
512 tcf_em_tree_destroy(tp, &f->ematches);
516 static int flow_delete(struct tcf_proto *tp, unsigned long arg)
518 struct flow_filter *f = (struct flow_filter *)arg;
523 flow_destroy_filter(tp, f);
527 static int flow_init(struct tcf_proto *tp)
529 struct flow_head *head;
531 head = kzalloc(sizeof(*head), GFP_KERNEL);
534 INIT_LIST_HEAD(&head->filters);
539 static void flow_destroy(struct tcf_proto *tp)
541 struct flow_head *head = tp->root;
542 struct flow_filter *f, *next;
544 list_for_each_entry_safe(f, next, &head->filters, list) {
546 flow_destroy_filter(tp, f);
551 static unsigned long flow_get(struct tcf_proto *tp, u32 handle)
553 struct flow_head *head = tp->root;
554 struct flow_filter *f;
556 list_for_each_entry(f, &head->filters, list)
557 if (f->handle == handle)
558 return (unsigned long)f;
562 static void flow_put(struct tcf_proto *tp, unsigned long f)
566 static int flow_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
567 struct sk_buff *skb, struct tcmsg *t)
569 struct flow_filter *f = (struct flow_filter *)fh;
575 t->tcm_handle = f->handle;
577 nest = nla_nest_start(skb, TCA_OPTIONS);
579 goto nla_put_failure;
581 if (nla_put_u32(skb, TCA_FLOW_KEYS, f->keymask) ||
582 nla_put_u32(skb, TCA_FLOW_MODE, f->mode))
583 goto nla_put_failure;
585 if (f->mask != ~0 || f->xor != 0) {
586 if (nla_put_u32(skb, TCA_FLOW_MASK, f->mask) ||
587 nla_put_u32(skb, TCA_FLOW_XOR, f->xor))
588 goto nla_put_failure;
591 nla_put_u32(skb, TCA_FLOW_RSHIFT, f->rshift))
592 goto nla_put_failure;
594 nla_put_u32(skb, TCA_FLOW_ADDEND, f->addend))
595 goto nla_put_failure;
598 nla_put_u32(skb, TCA_FLOW_DIVISOR, f->divisor))
599 goto nla_put_failure;
601 nla_put_u32(skb, TCA_FLOW_BASECLASS, f->baseclass))
602 goto nla_put_failure;
604 if (f->perturb_period &&
605 nla_put_u32(skb, TCA_FLOW_PERTURB, f->perturb_period / HZ))
606 goto nla_put_failure;
608 if (tcf_exts_dump(skb, &f->exts) < 0)
609 goto nla_put_failure;
610 #ifdef CONFIG_NET_EMATCH
611 if (f->ematches.hdr.nmatches &&
612 tcf_em_tree_dump(skb, &f->ematches, TCA_FLOW_EMATCHES) < 0)
613 goto nla_put_failure;
615 nla_nest_end(skb, nest);
617 if (tcf_exts_dump_stats(skb, &f->exts) < 0)
618 goto nla_put_failure;
623 nlmsg_trim(skb, nest);
627 static void flow_walk(struct tcf_proto *tp, struct tcf_walker *arg)
629 struct flow_head *head = tp->root;
630 struct flow_filter *f;
632 list_for_each_entry(f, &head->filters, list) {
633 if (arg->count < arg->skip)
635 if (arg->fn(tp, (unsigned long)f, arg) < 0) {
644 static struct tcf_proto_ops cls_flow_ops __read_mostly = {
646 .classify = flow_classify,
648 .destroy = flow_destroy,
649 .change = flow_change,
650 .delete = flow_delete,
655 .owner = THIS_MODULE,
658 static int __init cls_flow_init(void)
660 return register_tcf_proto_ops(&cls_flow_ops);
663 static void __exit cls_flow_exit(void)
665 unregister_tcf_proto_ops(&cls_flow_ops);
668 module_init(cls_flow_init);
669 module_exit(cls_flow_exit);
671 MODULE_LICENSE("GPL");
673 MODULE_DESCRIPTION("TC flow classifier");