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 IS_ENABLED(CONFIG_NF_CONNTRACK)
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;
60 struct rcu_work rwork;
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,
91 const struct flow_keys *flow)
93 return flow->basic.ip_proto;
96 static u32 flow_get_proto_src(const struct sk_buff *skb,
97 const struct flow_keys *flow)
99 if (flow->ports.ports)
100 return ntohs(flow->ports.src);
102 return addr_fold(skb->sk);
105 static u32 flow_get_proto_dst(const struct sk_buff *skb,
106 const struct flow_keys *flow)
108 if (flow->ports.ports)
109 return ntohs(flow->ports.dst);
111 return addr_fold(skb_dst(skb)) ^ (__force u16) tc_skb_protocol(skb);
114 static u32 flow_get_iif(const struct sk_buff *skb)
119 static u32 flow_get_priority(const struct sk_buff *skb)
121 return skb->priority;
124 static u32 flow_get_mark(const struct sk_buff *skb)
129 static u32 flow_get_nfct(const struct sk_buff *skb)
131 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
132 return addr_fold(skb_nfct(skb));
138 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
139 #define CTTUPLE(skb, member) \
141 enum ip_conntrack_info ctinfo; \
142 const struct nf_conn *ct = nf_ct_get(skb, &ctinfo); \
145 ct->tuplehash[CTINFO2DIR(ctinfo)].tuple.member; \
148 #define CTTUPLE(skb, member) \
155 static u32 flow_get_nfct_src(const struct sk_buff *skb,
156 const struct flow_keys *flow)
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]));
165 return flow_get_src(skb, flow);
168 static u32 flow_get_nfct_dst(const struct sk_buff *skb,
169 const struct flow_keys *flow)
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]));
178 return flow_get_dst(skb, flow);
181 static u32 flow_get_nfct_proto_src(const struct sk_buff *skb,
182 const struct flow_keys *flow)
184 return ntohs(CTTUPLE(skb, src.u.all));
186 return flow_get_proto_src(skb, flow);
189 static u32 flow_get_nfct_proto_dst(const struct sk_buff *skb,
190 const struct flow_keys *flow)
192 return ntohs(CTTUPLE(skb, dst.u.all));
194 return flow_get_proto_dst(skb, flow);
197 static u32 flow_get_rtclassid(const struct sk_buff *skb)
199 #ifdef CONFIG_IP_ROUTE_CLASSID
201 return skb_dst(skb)->tclassid;
206 static u32 flow_get_skuid(const struct sk_buff *skb)
208 struct sock *sk = skb_to_full_sk(skb);
210 if (sk && sk->sk_socket && sk->sk_socket->file) {
211 kuid_t skuid = sk->sk_socket->file->f_cred->fsuid;
213 return from_kuid(&init_user_ns, skuid);
218 static u32 flow_get_skgid(const struct sk_buff *skb)
220 struct sock *sk = skb_to_full_sk(skb);
222 if (sk && sk->sk_socket && sk->sk_socket->file) {
223 kgid_t skgid = sk->sk_socket->file->f_cred->fsgid;
225 return from_kgid(&init_user_ns, skgid);
230 static u32 flow_get_vlan_tag(const struct sk_buff *skb)
232 u16 uninitialized_var(tag);
234 if (vlan_get_tag(skb, &tag) < 0)
236 return tag & VLAN_VID_MASK;
239 static u32 flow_get_rxhash(struct sk_buff *skb)
241 return skb_get_hash(skb);
244 static u32 flow_key_get(struct sk_buff *skb, int key, struct flow_keys *flow)
248 return flow_get_src(skb, flow);
250 return flow_get_dst(skb, flow);
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);
258 return flow_get_iif(skb);
259 case FLOW_KEY_PRIORITY:
260 return flow_get_priority(skb);
262 return flow_get_mark(skb);
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);
276 return flow_get_skuid(skb);
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);
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))
299 static int flow_classify(struct sk_buff *skb, const struct tcf_proto *tp,
300 struct tcf_result *res)
302 struct flow_head *head = rcu_dereference_bh(tp->root);
303 struct flow_filter *f;
309 list_for_each_entry_rcu(f, &head->filters, list) {
310 u32 keys[FLOW_KEY_MAX + 1];
311 struct flow_keys flow_keys;
313 if (!tcf_em_tree_match(skb, &f->ematches, NULL))
316 keymask = f->keymask;
317 if (keymask & FLOW_KEYS_NEEDED)
318 skb_flow_dissect_flow_keys(skb, &flow_keys, 0);
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);
326 if (f->mode == FLOW_MODE_HASH)
327 classid = jhash2(keys, f->nkeys, f->hashrnd);
330 classid = (classid & f->mask) ^ f->xor;
331 classid = (classid >> f->rshift) + f->addend;
335 classid %= f->divisor;
338 res->classid = TC_H_MAKE(f->baseclass, f->baseclass + classid);
340 r = tcf_exts_exec(skb, &f->exts, res);
348 static void flow_perturbation(struct timer_list *t)
350 struct flow_filter *f = from_timer(f, t, perturb_timer);
352 get_random_bytes(&f->hashrnd, 4);
353 if (f->perturb_period)
354 mod_timer(&f->perturb_timer, jiffies + f->perturb_period);
357 static 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 },
372 static void __flow_destroy_filter(struct flow_filter *f)
374 del_timer_sync(&f->perturb_timer);
375 tcf_exts_destroy(&f->exts);
376 tcf_em_tree_destroy(&f->ematches);
377 tcf_exts_put_net(&f->exts);
381 static void flow_destroy_filter_work(struct work_struct *work)
383 struct flow_filter *f = container_of(to_rcu_work(work),
387 __flow_destroy_filter(f);
391 static int flow_change(struct net *net, struct sk_buff *in_skb,
392 struct tcf_proto *tp, unsigned long base,
393 u32 handle, struct nlattr **tca,
394 void **arg, bool ovr, bool rtnl_held,
395 struct netlink_ext_ack *extack)
397 struct flow_head *head = rtnl_dereference(tp->root);
398 struct flow_filter *fold, *fnew;
399 struct nlattr *opt = tca[TCA_OPTIONS];
400 struct nlattr *tb[TCA_FLOW_MAX + 1];
401 unsigned int nkeys = 0;
402 unsigned int perturb_period = 0;
411 err = nla_parse_nested(tb, TCA_FLOW_MAX, opt, flow_policy, NULL);
415 if (tb[TCA_FLOW_BASECLASS]) {
416 baseclass = nla_get_u32(tb[TCA_FLOW_BASECLASS]);
417 if (TC_H_MIN(baseclass) == 0)
421 if (tb[TCA_FLOW_KEYS]) {
422 keymask = nla_get_u32(tb[TCA_FLOW_KEYS]);
424 nkeys = hweight32(keymask);
428 if (fls(keymask) - 1 > FLOW_KEY_MAX)
431 if ((keymask & (FLOW_KEY_SKUID|FLOW_KEY_SKGID)) &&
432 sk_user_ns(NETLINK_CB(in_skb).sk) != &init_user_ns)
436 fnew = kzalloc(sizeof(*fnew), GFP_KERNEL);
440 err = tcf_em_tree_validate(tp, tb[TCA_FLOW_EMATCHES], &fnew->ematches);
444 err = tcf_exts_init(&fnew->exts, net, TCA_FLOW_ACT, TCA_FLOW_POLICE);
448 err = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &fnew->exts, ovr,
456 if (fold->handle != handle && handle)
459 /* Copy fold into fnew */
461 fnew->handle = fold->handle;
462 fnew->nkeys = fold->nkeys;
463 fnew->keymask = fold->keymask;
464 fnew->mode = fold->mode;
465 fnew->mask = fold->mask;
466 fnew->xor = fold->xor;
467 fnew->rshift = fold->rshift;
468 fnew->addend = fold->addend;
469 fnew->divisor = fold->divisor;
470 fnew->baseclass = fold->baseclass;
471 fnew->hashrnd = fold->hashrnd;
474 if (tb[TCA_FLOW_MODE])
475 mode = nla_get_u32(tb[TCA_FLOW_MODE]);
476 if (mode != FLOW_MODE_HASH && nkeys > 1)
479 if (mode == FLOW_MODE_HASH)
480 perturb_period = fold->perturb_period;
481 if (tb[TCA_FLOW_PERTURB]) {
482 if (mode != FLOW_MODE_HASH)
484 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
490 if (!tb[TCA_FLOW_KEYS])
493 mode = FLOW_MODE_MAP;
494 if (tb[TCA_FLOW_MODE])
495 mode = nla_get_u32(tb[TCA_FLOW_MODE]);
496 if (mode != FLOW_MODE_HASH && nkeys > 1)
499 if (tb[TCA_FLOW_PERTURB]) {
500 if (mode != FLOW_MODE_HASH)
502 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
505 if (TC_H_MAJ(baseclass) == 0) {
506 struct Qdisc *q = tcf_block_q(tp->chain->block);
508 baseclass = TC_H_MAKE(q->handle, baseclass);
510 if (TC_H_MIN(baseclass) == 0)
511 baseclass = TC_H_MAKE(baseclass, 1);
513 fnew->handle = handle;
516 get_random_bytes(&fnew->hashrnd, 4);
519 timer_setup(&fnew->perturb_timer, flow_perturbation, TIMER_DEFERRABLE);
521 tcf_block_netif_keep_dst(tp->chain->block);
523 if (tb[TCA_FLOW_KEYS]) {
524 fnew->keymask = keymask;
530 if (tb[TCA_FLOW_MASK])
531 fnew->mask = nla_get_u32(tb[TCA_FLOW_MASK]);
532 if (tb[TCA_FLOW_XOR])
533 fnew->xor = nla_get_u32(tb[TCA_FLOW_XOR]);
534 if (tb[TCA_FLOW_RSHIFT])
535 fnew->rshift = nla_get_u32(tb[TCA_FLOW_RSHIFT]);
536 if (tb[TCA_FLOW_ADDEND])
537 fnew->addend = nla_get_u32(tb[TCA_FLOW_ADDEND]);
539 if (tb[TCA_FLOW_DIVISOR])
540 fnew->divisor = nla_get_u32(tb[TCA_FLOW_DIVISOR]);
542 fnew->baseclass = baseclass;
544 fnew->perturb_period = perturb_period;
546 mod_timer(&fnew->perturb_timer, jiffies + perturb_period);
549 list_add_tail_rcu(&fnew->list, &head->filters);
551 list_replace_rcu(&fold->list, &fnew->list);
556 tcf_exts_get_net(&fold->exts);
557 tcf_queue_work(&fold->rwork, flow_destroy_filter_work);
562 tcf_exts_destroy(&fnew->exts);
563 tcf_em_tree_destroy(&fnew->ematches);
569 static int flow_delete(struct tcf_proto *tp, void *arg, bool *last,
570 bool rtnl_held, struct netlink_ext_ack *extack)
572 struct flow_head *head = rtnl_dereference(tp->root);
573 struct flow_filter *f = arg;
575 list_del_rcu(&f->list);
576 tcf_exts_get_net(&f->exts);
577 tcf_queue_work(&f->rwork, flow_destroy_filter_work);
578 *last = list_empty(&head->filters);
582 static int flow_init(struct tcf_proto *tp)
584 struct flow_head *head;
586 head = kzalloc(sizeof(*head), GFP_KERNEL);
589 INIT_LIST_HEAD(&head->filters);
590 rcu_assign_pointer(tp->root, head);
594 static void flow_destroy(struct tcf_proto *tp, bool rtnl_held,
595 struct netlink_ext_ack *extack)
597 struct flow_head *head = rtnl_dereference(tp->root);
598 struct flow_filter *f, *next;
600 list_for_each_entry_safe(f, next, &head->filters, list) {
601 list_del_rcu(&f->list);
602 if (tcf_exts_get_net(&f->exts))
603 tcf_queue_work(&f->rwork, flow_destroy_filter_work);
605 __flow_destroy_filter(f);
607 kfree_rcu(head, rcu);
610 static void *flow_get(struct tcf_proto *tp, u32 handle)
612 struct flow_head *head = rtnl_dereference(tp->root);
613 struct flow_filter *f;
615 list_for_each_entry(f, &head->filters, list)
616 if (f->handle == handle)
621 static int flow_dump(struct net *net, struct tcf_proto *tp, void *fh,
622 struct sk_buff *skb, struct tcmsg *t, bool rtnl_held)
624 struct flow_filter *f = fh;
630 t->tcm_handle = f->handle;
632 nest = nla_nest_start(skb, TCA_OPTIONS);
634 goto nla_put_failure;
636 if (nla_put_u32(skb, TCA_FLOW_KEYS, f->keymask) ||
637 nla_put_u32(skb, TCA_FLOW_MODE, f->mode))
638 goto nla_put_failure;
640 if (f->mask != ~0 || f->xor != 0) {
641 if (nla_put_u32(skb, TCA_FLOW_MASK, f->mask) ||
642 nla_put_u32(skb, TCA_FLOW_XOR, f->xor))
643 goto nla_put_failure;
646 nla_put_u32(skb, TCA_FLOW_RSHIFT, f->rshift))
647 goto nla_put_failure;
649 nla_put_u32(skb, TCA_FLOW_ADDEND, f->addend))
650 goto nla_put_failure;
653 nla_put_u32(skb, TCA_FLOW_DIVISOR, f->divisor))
654 goto nla_put_failure;
656 nla_put_u32(skb, TCA_FLOW_BASECLASS, f->baseclass))
657 goto nla_put_failure;
659 if (f->perturb_period &&
660 nla_put_u32(skb, TCA_FLOW_PERTURB, f->perturb_period / HZ))
661 goto nla_put_failure;
663 if (tcf_exts_dump(skb, &f->exts) < 0)
664 goto nla_put_failure;
665 #ifdef CONFIG_NET_EMATCH
666 if (f->ematches.hdr.nmatches &&
667 tcf_em_tree_dump(skb, &f->ematches, TCA_FLOW_EMATCHES) < 0)
668 goto nla_put_failure;
670 nla_nest_end(skb, nest);
672 if (tcf_exts_dump_stats(skb, &f->exts) < 0)
673 goto nla_put_failure;
678 nla_nest_cancel(skb, nest);
682 static void flow_walk(struct tcf_proto *tp, struct tcf_walker *arg,
685 struct flow_head *head = rtnl_dereference(tp->root);
686 struct flow_filter *f;
688 list_for_each_entry(f, &head->filters, list) {
689 if (arg->count < arg->skip)
691 if (arg->fn(tp, f, arg) < 0) {
700 static struct tcf_proto_ops cls_flow_ops __read_mostly = {
702 .classify = flow_classify,
704 .destroy = flow_destroy,
705 .change = flow_change,
706 .delete = flow_delete,
710 .owner = THIS_MODULE,
713 static int __init cls_flow_init(void)
715 return register_tcf_proto_ops(&cls_flow_ops);
718 static void __exit cls_flow_exit(void)
720 unregister_tcf_proto_ops(&cls_flow_ops);
723 module_init(cls_flow_init);
724 module_exit(cls_flow_exit);
726 MODULE_LICENSE("GPL");
727 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
728 MODULE_DESCRIPTION("TC flow classifier");