Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
[cris-mirror.git] / net / sched / cls_flow.c
blobcd5fe383afddd1ee3e82690a996fcb1fb8957155
1 /*
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>
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>
27 #include <net/pkt_cls.h>
28 #include <net/ip.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>
34 #endif
36 struct flow_head {
37 struct list_head filters;
38 struct rcu_head rcu;
41 struct 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;
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 union {
61 struct work_struct work;
62 struct rcu_head rcu;
66 static inline u32 addr_fold(void *addr)
68 unsigned long a = (unsigned long)addr;
70 return (a & 0xFFFFFFFF) ^ (BITS_PER_LONG > 32 ? a >> 32 : 0);
73 static u32 flow_get_src(const struct sk_buff *skb, const struct flow_keys *flow)
75 __be32 src = flow_get_u32_src(flow);
77 if (src)
78 return ntohl(src);
80 return addr_fold(skb->sk);
83 static u32 flow_get_dst(const struct sk_buff *skb, const struct flow_keys *flow)
85 __be32 dst = flow_get_u32_dst(flow);
87 if (dst)
88 return ntohl(dst);
90 return addr_fold(skb_dst(skb)) ^ (__force u16) tc_skb_protocol(skb);
93 static u32 flow_get_proto(const struct sk_buff *skb,
94 const struct flow_keys *flow)
96 return flow->basic.ip_proto;
99 static u32 flow_get_proto_src(const struct sk_buff *skb,
100 const struct flow_keys *flow)
102 if (flow->ports.ports)
103 return ntohs(flow->ports.src);
105 return addr_fold(skb->sk);
108 static u32 flow_get_proto_dst(const struct sk_buff *skb,
109 const struct flow_keys *flow)
111 if (flow->ports.ports)
112 return ntohs(flow->ports.dst);
114 return addr_fold(skb_dst(skb)) ^ (__force u16) tc_skb_protocol(skb);
117 static u32 flow_get_iif(const struct sk_buff *skb)
119 return skb->skb_iif;
122 static u32 flow_get_priority(const struct sk_buff *skb)
124 return skb->priority;
127 static u32 flow_get_mark(const struct sk_buff *skb)
129 return skb->mark;
132 static u32 flow_get_nfct(const struct sk_buff *skb)
134 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
135 return addr_fold(skb_nfct(skb));
136 #else
137 return 0;
138 #endif
141 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
142 #define CTTUPLE(skb, member) \
143 ({ \
144 enum ip_conntrack_info ctinfo; \
145 const struct nf_conn *ct = nf_ct_get(skb, &ctinfo); \
146 if (ct == NULL) \
147 goto fallback; \
148 ct->tuplehash[CTINFO2DIR(ctinfo)].tuple.member; \
150 #else
151 #define CTTUPLE(skb, member) \
152 ({ \
153 goto fallback; \
154 0; \
156 #endif
158 static u32 flow_get_nfct_src(const struct sk_buff *skb,
159 const struct flow_keys *flow)
161 switch (tc_skb_protocol(skb)) {
162 case htons(ETH_P_IP):
163 return ntohl(CTTUPLE(skb, src.u3.ip));
164 case htons(ETH_P_IPV6):
165 return ntohl(CTTUPLE(skb, src.u3.ip6[3]));
167 fallback:
168 return flow_get_src(skb, flow);
171 static u32 flow_get_nfct_dst(const struct sk_buff *skb,
172 const struct flow_keys *flow)
174 switch (tc_skb_protocol(skb)) {
175 case htons(ETH_P_IP):
176 return ntohl(CTTUPLE(skb, dst.u3.ip));
177 case htons(ETH_P_IPV6):
178 return ntohl(CTTUPLE(skb, dst.u3.ip6[3]));
180 fallback:
181 return flow_get_dst(skb, flow);
184 static u32 flow_get_nfct_proto_src(const struct sk_buff *skb,
185 const struct flow_keys *flow)
187 return ntohs(CTTUPLE(skb, src.u.all));
188 fallback:
189 return flow_get_proto_src(skb, flow);
192 static u32 flow_get_nfct_proto_dst(const struct sk_buff *skb,
193 const struct flow_keys *flow)
195 return ntohs(CTTUPLE(skb, dst.u.all));
196 fallback:
197 return flow_get_proto_dst(skb, flow);
200 static u32 flow_get_rtclassid(const struct sk_buff *skb)
202 #ifdef CONFIG_IP_ROUTE_CLASSID
203 if (skb_dst(skb))
204 return skb_dst(skb)->tclassid;
205 #endif
206 return 0;
209 static u32 flow_get_skuid(const struct sk_buff *skb)
211 struct sock *sk = skb_to_full_sk(skb);
213 if (sk && sk->sk_socket && sk->sk_socket->file) {
214 kuid_t skuid = sk->sk_socket->file->f_cred->fsuid;
216 return from_kuid(&init_user_ns, skuid);
218 return 0;
221 static u32 flow_get_skgid(const struct sk_buff *skb)
223 struct sock *sk = skb_to_full_sk(skb);
225 if (sk && sk->sk_socket && sk->sk_socket->file) {
226 kgid_t skgid = sk->sk_socket->file->f_cred->fsgid;
228 return from_kgid(&init_user_ns, skgid);
230 return 0;
233 static u32 flow_get_vlan_tag(const struct sk_buff *skb)
235 u16 uninitialized_var(tag);
237 if (vlan_get_tag(skb, &tag) < 0)
238 return 0;
239 return tag & VLAN_VID_MASK;
242 static u32 flow_get_rxhash(struct sk_buff *skb)
244 return skb_get_hash(skb);
247 static u32 flow_key_get(struct sk_buff *skb, int key, struct flow_keys *flow)
249 switch (key) {
250 case FLOW_KEY_SRC:
251 return flow_get_src(skb, flow);
252 case FLOW_KEY_DST:
253 return flow_get_dst(skb, flow);
254 case FLOW_KEY_PROTO:
255 return flow_get_proto(skb, flow);
256 case FLOW_KEY_PROTO_SRC:
257 return flow_get_proto_src(skb, flow);
258 case FLOW_KEY_PROTO_DST:
259 return flow_get_proto_dst(skb, flow);
260 case FLOW_KEY_IIF:
261 return flow_get_iif(skb);
262 case FLOW_KEY_PRIORITY:
263 return flow_get_priority(skb);
264 case FLOW_KEY_MARK:
265 return flow_get_mark(skb);
266 case FLOW_KEY_NFCT:
267 return flow_get_nfct(skb);
268 case FLOW_KEY_NFCT_SRC:
269 return flow_get_nfct_src(skb, flow);
270 case FLOW_KEY_NFCT_DST:
271 return flow_get_nfct_dst(skb, flow);
272 case FLOW_KEY_NFCT_PROTO_SRC:
273 return flow_get_nfct_proto_src(skb, flow);
274 case FLOW_KEY_NFCT_PROTO_DST:
275 return flow_get_nfct_proto_dst(skb, flow);
276 case FLOW_KEY_RTCLASSID:
277 return flow_get_rtclassid(skb);
278 case FLOW_KEY_SKUID:
279 return flow_get_skuid(skb);
280 case FLOW_KEY_SKGID:
281 return flow_get_skgid(skb);
282 case FLOW_KEY_VLAN_TAG:
283 return flow_get_vlan_tag(skb);
284 case FLOW_KEY_RXHASH:
285 return flow_get_rxhash(skb);
286 default:
287 WARN_ON(1);
288 return 0;
292 #define FLOW_KEYS_NEEDED ((1 << FLOW_KEY_SRC) | \
293 (1 << FLOW_KEY_DST) | \
294 (1 << FLOW_KEY_PROTO) | \
295 (1 << FLOW_KEY_PROTO_SRC) | \
296 (1 << FLOW_KEY_PROTO_DST) | \
297 (1 << FLOW_KEY_NFCT_SRC) | \
298 (1 << FLOW_KEY_NFCT_DST) | \
299 (1 << FLOW_KEY_NFCT_PROTO_SRC) | \
300 (1 << FLOW_KEY_NFCT_PROTO_DST))
302 static int flow_classify(struct sk_buff *skb, const struct tcf_proto *tp,
303 struct tcf_result *res)
305 struct flow_head *head = rcu_dereference_bh(tp->root);
306 struct flow_filter *f;
307 u32 keymask;
308 u32 classid;
309 unsigned int n, key;
310 int r;
312 list_for_each_entry_rcu(f, &head->filters, list) {
313 u32 keys[FLOW_KEY_MAX + 1];
314 struct flow_keys flow_keys;
316 if (!tcf_em_tree_match(skb, &f->ematches, NULL))
317 continue;
319 keymask = f->keymask;
320 if (keymask & FLOW_KEYS_NEEDED)
321 skb_flow_dissect_flow_keys(skb, &flow_keys, 0);
323 for (n = 0; n < f->nkeys; n++) {
324 key = ffs(keymask) - 1;
325 keymask &= ~(1 << key);
326 keys[n] = flow_key_get(skb, key, &flow_keys);
329 if (f->mode == FLOW_MODE_HASH)
330 classid = jhash2(keys, f->nkeys, f->hashrnd);
331 else {
332 classid = keys[0];
333 classid = (classid & f->mask) ^ f->xor;
334 classid = (classid >> f->rshift) + f->addend;
337 if (f->divisor)
338 classid %= f->divisor;
340 res->class = 0;
341 res->classid = TC_H_MAKE(f->baseclass, f->baseclass + classid);
343 r = tcf_exts_exec(skb, &f->exts, res);
344 if (r < 0)
345 continue;
346 return r;
348 return -1;
351 static void flow_perturbation(struct timer_list *t)
353 struct flow_filter *f = from_timer(f, t, perturb_timer);
355 get_random_bytes(&f->hashrnd, 4);
356 if (f->perturb_period)
357 mod_timer(&f->perturb_timer, jiffies + f->perturb_period);
360 static const struct nla_policy flow_policy[TCA_FLOW_MAX + 1] = {
361 [TCA_FLOW_KEYS] = { .type = NLA_U32 },
362 [TCA_FLOW_MODE] = { .type = NLA_U32 },
363 [TCA_FLOW_BASECLASS] = { .type = NLA_U32 },
364 [TCA_FLOW_RSHIFT] = { .type = NLA_U32 },
365 [TCA_FLOW_ADDEND] = { .type = NLA_U32 },
366 [TCA_FLOW_MASK] = { .type = NLA_U32 },
367 [TCA_FLOW_XOR] = { .type = NLA_U32 },
368 [TCA_FLOW_DIVISOR] = { .type = NLA_U32 },
369 [TCA_FLOW_ACT] = { .type = NLA_NESTED },
370 [TCA_FLOW_POLICE] = { .type = NLA_NESTED },
371 [TCA_FLOW_EMATCHES] = { .type = NLA_NESTED },
372 [TCA_FLOW_PERTURB] = { .type = NLA_U32 },
375 static void __flow_destroy_filter(struct flow_filter *f)
377 del_timer_sync(&f->perturb_timer);
378 tcf_exts_destroy(&f->exts);
379 tcf_em_tree_destroy(&f->ematches);
380 tcf_exts_put_net(&f->exts);
381 kfree(f);
384 static void flow_destroy_filter_work(struct work_struct *work)
386 struct flow_filter *f = container_of(work, struct flow_filter, work);
388 rtnl_lock();
389 __flow_destroy_filter(f);
390 rtnl_unlock();
393 static void flow_destroy_filter(struct rcu_head *head)
395 struct flow_filter *f = container_of(head, struct flow_filter, rcu);
397 INIT_WORK(&f->work, flow_destroy_filter_work);
398 tcf_queue_work(&f->work);
401 static int flow_change(struct net *net, struct sk_buff *in_skb,
402 struct tcf_proto *tp, unsigned long base,
403 u32 handle, struct nlattr **tca,
404 void **arg, bool ovr, struct netlink_ext_ack *extack)
406 struct flow_head *head = rtnl_dereference(tp->root);
407 struct flow_filter *fold, *fnew;
408 struct nlattr *opt = tca[TCA_OPTIONS];
409 struct nlattr *tb[TCA_FLOW_MAX + 1];
410 unsigned int nkeys = 0;
411 unsigned int perturb_period = 0;
412 u32 baseclass = 0;
413 u32 keymask = 0;
414 u32 mode;
415 int err;
417 if (opt == NULL)
418 return -EINVAL;
420 err = nla_parse_nested(tb, TCA_FLOW_MAX, opt, flow_policy, NULL);
421 if (err < 0)
422 return err;
424 if (tb[TCA_FLOW_BASECLASS]) {
425 baseclass = nla_get_u32(tb[TCA_FLOW_BASECLASS]);
426 if (TC_H_MIN(baseclass) == 0)
427 return -EINVAL;
430 if (tb[TCA_FLOW_KEYS]) {
431 keymask = nla_get_u32(tb[TCA_FLOW_KEYS]);
433 nkeys = hweight32(keymask);
434 if (nkeys == 0)
435 return -EINVAL;
437 if (fls(keymask) - 1 > FLOW_KEY_MAX)
438 return -EOPNOTSUPP;
440 if ((keymask & (FLOW_KEY_SKUID|FLOW_KEY_SKGID)) &&
441 sk_user_ns(NETLINK_CB(in_skb).sk) != &init_user_ns)
442 return -EOPNOTSUPP;
445 fnew = kzalloc(sizeof(*fnew), GFP_KERNEL);
446 if (!fnew)
447 return -ENOBUFS;
449 err = tcf_em_tree_validate(tp, tb[TCA_FLOW_EMATCHES], &fnew->ematches);
450 if (err < 0)
451 goto err1;
453 err = tcf_exts_init(&fnew->exts, TCA_FLOW_ACT, TCA_FLOW_POLICE);
454 if (err < 0)
455 goto err2;
457 err = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &fnew->exts, ovr,
458 extack);
459 if (err < 0)
460 goto err2;
462 fold = *arg;
463 if (fold) {
464 err = -EINVAL;
465 if (fold->handle != handle && handle)
466 goto err2;
468 /* Copy fold into fnew */
469 fnew->tp = fold->tp;
470 fnew->handle = fold->handle;
471 fnew->nkeys = fold->nkeys;
472 fnew->keymask = fold->keymask;
473 fnew->mode = fold->mode;
474 fnew->mask = fold->mask;
475 fnew->xor = fold->xor;
476 fnew->rshift = fold->rshift;
477 fnew->addend = fold->addend;
478 fnew->divisor = fold->divisor;
479 fnew->baseclass = fold->baseclass;
480 fnew->hashrnd = fold->hashrnd;
482 mode = fold->mode;
483 if (tb[TCA_FLOW_MODE])
484 mode = nla_get_u32(tb[TCA_FLOW_MODE]);
485 if (mode != FLOW_MODE_HASH && nkeys > 1)
486 goto err2;
488 if (mode == FLOW_MODE_HASH)
489 perturb_period = fold->perturb_period;
490 if (tb[TCA_FLOW_PERTURB]) {
491 if (mode != FLOW_MODE_HASH)
492 goto err2;
493 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
495 } else {
496 err = -EINVAL;
497 if (!handle)
498 goto err2;
499 if (!tb[TCA_FLOW_KEYS])
500 goto err2;
502 mode = FLOW_MODE_MAP;
503 if (tb[TCA_FLOW_MODE])
504 mode = nla_get_u32(tb[TCA_FLOW_MODE]);
505 if (mode != FLOW_MODE_HASH && nkeys > 1)
506 goto err2;
508 if (tb[TCA_FLOW_PERTURB]) {
509 if (mode != FLOW_MODE_HASH)
510 goto err2;
511 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
514 if (TC_H_MAJ(baseclass) == 0) {
515 struct Qdisc *q = tcf_block_q(tp->chain->block);
517 baseclass = TC_H_MAKE(q->handle, baseclass);
519 if (TC_H_MIN(baseclass) == 0)
520 baseclass = TC_H_MAKE(baseclass, 1);
522 fnew->handle = handle;
523 fnew->mask = ~0U;
524 fnew->tp = tp;
525 get_random_bytes(&fnew->hashrnd, 4);
528 timer_setup(&fnew->perturb_timer, flow_perturbation, TIMER_DEFERRABLE);
530 tcf_block_netif_keep_dst(tp->chain->block);
532 if (tb[TCA_FLOW_KEYS]) {
533 fnew->keymask = keymask;
534 fnew->nkeys = nkeys;
537 fnew->mode = mode;
539 if (tb[TCA_FLOW_MASK])
540 fnew->mask = nla_get_u32(tb[TCA_FLOW_MASK]);
541 if (tb[TCA_FLOW_XOR])
542 fnew->xor = nla_get_u32(tb[TCA_FLOW_XOR]);
543 if (tb[TCA_FLOW_RSHIFT])
544 fnew->rshift = nla_get_u32(tb[TCA_FLOW_RSHIFT]);
545 if (tb[TCA_FLOW_ADDEND])
546 fnew->addend = nla_get_u32(tb[TCA_FLOW_ADDEND]);
548 if (tb[TCA_FLOW_DIVISOR])
549 fnew->divisor = nla_get_u32(tb[TCA_FLOW_DIVISOR]);
550 if (baseclass)
551 fnew->baseclass = baseclass;
553 fnew->perturb_period = perturb_period;
554 if (perturb_period)
555 mod_timer(&fnew->perturb_timer, jiffies + perturb_period);
557 if (!*arg)
558 list_add_tail_rcu(&fnew->list, &head->filters);
559 else
560 list_replace_rcu(&fold->list, &fnew->list);
562 *arg = fnew;
564 if (fold) {
565 tcf_exts_get_net(&fold->exts);
566 call_rcu(&fold->rcu, flow_destroy_filter);
568 return 0;
570 err2:
571 tcf_exts_destroy(&fnew->exts);
572 tcf_em_tree_destroy(&fnew->ematches);
573 err1:
574 kfree(fnew);
575 return err;
578 static int flow_delete(struct tcf_proto *tp, void *arg, bool *last,
579 struct netlink_ext_ack *extack)
581 struct flow_head *head = rtnl_dereference(tp->root);
582 struct flow_filter *f = arg;
584 list_del_rcu(&f->list);
585 tcf_exts_get_net(&f->exts);
586 call_rcu(&f->rcu, flow_destroy_filter);
587 *last = list_empty(&head->filters);
588 return 0;
591 static int flow_init(struct tcf_proto *tp)
593 struct flow_head *head;
595 head = kzalloc(sizeof(*head), GFP_KERNEL);
596 if (head == NULL)
597 return -ENOBUFS;
598 INIT_LIST_HEAD(&head->filters);
599 rcu_assign_pointer(tp->root, head);
600 return 0;
603 static void flow_destroy(struct tcf_proto *tp, struct netlink_ext_ack *extack)
605 struct flow_head *head = rtnl_dereference(tp->root);
606 struct flow_filter *f, *next;
608 list_for_each_entry_safe(f, next, &head->filters, list) {
609 list_del_rcu(&f->list);
610 if (tcf_exts_get_net(&f->exts))
611 call_rcu(&f->rcu, flow_destroy_filter);
612 else
613 __flow_destroy_filter(f);
615 kfree_rcu(head, rcu);
618 static void *flow_get(struct tcf_proto *tp, u32 handle)
620 struct flow_head *head = rtnl_dereference(tp->root);
621 struct flow_filter *f;
623 list_for_each_entry(f, &head->filters, list)
624 if (f->handle == handle)
625 return f;
626 return NULL;
629 static int flow_dump(struct net *net, struct tcf_proto *tp, void *fh,
630 struct sk_buff *skb, struct tcmsg *t)
632 struct flow_filter *f = fh;
633 struct nlattr *nest;
635 if (f == NULL)
636 return skb->len;
638 t->tcm_handle = f->handle;
640 nest = nla_nest_start(skb, TCA_OPTIONS);
641 if (nest == NULL)
642 goto nla_put_failure;
644 if (nla_put_u32(skb, TCA_FLOW_KEYS, f->keymask) ||
645 nla_put_u32(skb, TCA_FLOW_MODE, f->mode))
646 goto nla_put_failure;
648 if (f->mask != ~0 || f->xor != 0) {
649 if (nla_put_u32(skb, TCA_FLOW_MASK, f->mask) ||
650 nla_put_u32(skb, TCA_FLOW_XOR, f->xor))
651 goto nla_put_failure;
653 if (f->rshift &&
654 nla_put_u32(skb, TCA_FLOW_RSHIFT, f->rshift))
655 goto nla_put_failure;
656 if (f->addend &&
657 nla_put_u32(skb, TCA_FLOW_ADDEND, f->addend))
658 goto nla_put_failure;
660 if (f->divisor &&
661 nla_put_u32(skb, TCA_FLOW_DIVISOR, f->divisor))
662 goto nla_put_failure;
663 if (f->baseclass &&
664 nla_put_u32(skb, TCA_FLOW_BASECLASS, f->baseclass))
665 goto nla_put_failure;
667 if (f->perturb_period &&
668 nla_put_u32(skb, TCA_FLOW_PERTURB, f->perturb_period / HZ))
669 goto nla_put_failure;
671 if (tcf_exts_dump(skb, &f->exts) < 0)
672 goto nla_put_failure;
673 #ifdef CONFIG_NET_EMATCH
674 if (f->ematches.hdr.nmatches &&
675 tcf_em_tree_dump(skb, &f->ematches, TCA_FLOW_EMATCHES) < 0)
676 goto nla_put_failure;
677 #endif
678 nla_nest_end(skb, nest);
680 if (tcf_exts_dump_stats(skb, &f->exts) < 0)
681 goto nla_put_failure;
683 return skb->len;
685 nla_put_failure:
686 nla_nest_cancel(skb, nest);
687 return -1;
690 static void flow_walk(struct tcf_proto *tp, struct tcf_walker *arg)
692 struct flow_head *head = rtnl_dereference(tp->root);
693 struct flow_filter *f;
695 list_for_each_entry(f, &head->filters, list) {
696 if (arg->count < arg->skip)
697 goto skip;
698 if (arg->fn(tp, f, arg) < 0) {
699 arg->stop = 1;
700 break;
702 skip:
703 arg->count++;
707 static struct tcf_proto_ops cls_flow_ops __read_mostly = {
708 .kind = "flow",
709 .classify = flow_classify,
710 .init = flow_init,
711 .destroy = flow_destroy,
712 .change = flow_change,
713 .delete = flow_delete,
714 .get = flow_get,
715 .dump = flow_dump,
716 .walk = flow_walk,
717 .owner = THIS_MODULE,
720 static int __init cls_flow_init(void)
722 return register_tcf_proto_ops(&cls_flow_ops);
725 static void __exit cls_flow_exit(void)
727 unregister_tcf_proto_ops(&cls_flow_ops);
730 module_init(cls_flow_init);
731 module_exit(cls_flow_exit);
733 MODULE_LICENSE("GPL");
734 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
735 MODULE_DESCRIPTION("TC flow classifier");