Loading...
1/*
2 * net/sched/cls_flow.c Generic flow classifier
3 *
4 * Copyright (c) 2007, 2008 Patrick McHardy <kaber@trash.net>
5 *
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.
10 */
11
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>
26
27#include <net/pkt_cls.h>
28#include <net/ip.h>
29#include <net/route.h>
30#include <net/flow_dissector.h>
31
32#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
33#include <net/netfilter/nf_conntrack.h>
34#endif
35
36struct flow_head {
37 struct list_head filters;
38 struct rcu_head rcu;
39};
40
41struct 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;
49
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 struct rcu_head rcu;
61};
62
63static inline u32 addr_fold(void *addr)
64{
65 unsigned long a = (unsigned long)addr;
66
67 return (a & 0xFFFFFFFF) ^ (BITS_PER_LONG > 32 ? a >> 32 : 0);
68}
69
70static u32 flow_get_src(const struct sk_buff *skb, const struct flow_keys *flow)
71{
72 __be32 src = flow_get_u32_src(flow);
73
74 if (src)
75 return ntohl(src);
76
77 return addr_fold(skb->sk);
78}
79
80static u32 flow_get_dst(const struct sk_buff *skb, const struct flow_keys *flow)
81{
82 __be32 dst = flow_get_u32_dst(flow);
83
84 if (dst)
85 return ntohl(dst);
86
87 return addr_fold(skb_dst(skb)) ^ (__force u16) tc_skb_protocol(skb);
88}
89
90static u32 flow_get_proto(const struct sk_buff *skb, const struct flow_keys *flow)
91{
92 return flow->basic.ip_proto;
93}
94
95static u32 flow_get_proto_src(const struct sk_buff *skb, const struct flow_keys *flow)
96{
97 if (flow->ports.ports)
98 return ntohs(flow->ports.src);
99
100 return addr_fold(skb->sk);
101}
102
103static u32 flow_get_proto_dst(const struct sk_buff *skb, const struct flow_keys *flow)
104{
105 if (flow->ports.ports)
106 return ntohs(flow->ports.dst);
107
108 return addr_fold(skb_dst(skb)) ^ (__force u16) tc_skb_protocol(skb);
109}
110
111static u32 flow_get_iif(const struct sk_buff *skb)
112{
113 return skb->skb_iif;
114}
115
116static u32 flow_get_priority(const struct sk_buff *skb)
117{
118 return skb->priority;
119}
120
121static u32 flow_get_mark(const struct sk_buff *skb)
122{
123 return skb->mark;
124}
125
126static u32 flow_get_nfct(const struct sk_buff *skb)
127{
128#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
129 return addr_fold(skb->nfct);
130#else
131 return 0;
132#endif
133}
134
135#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
136#define CTTUPLE(skb, member) \
137({ \
138 enum ip_conntrack_info ctinfo; \
139 const struct nf_conn *ct = nf_ct_get(skb, &ctinfo); \
140 if (ct == NULL) \
141 goto fallback; \
142 ct->tuplehash[CTINFO2DIR(ctinfo)].tuple.member; \
143})
144#else
145#define CTTUPLE(skb, member) \
146({ \
147 goto fallback; \
148 0; \
149})
150#endif
151
152static u32 flow_get_nfct_src(const struct sk_buff *skb, const struct flow_keys *flow)
153{
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]));
159 }
160fallback:
161 return flow_get_src(skb, flow);
162}
163
164static u32 flow_get_nfct_dst(const struct sk_buff *skb, const struct flow_keys *flow)
165{
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]));
171 }
172fallback:
173 return flow_get_dst(skb, flow);
174}
175
176static u32 flow_get_nfct_proto_src(const struct sk_buff *skb, const struct flow_keys *flow)
177{
178 return ntohs(CTTUPLE(skb, src.u.all));
179fallback:
180 return flow_get_proto_src(skb, flow);
181}
182
183static u32 flow_get_nfct_proto_dst(const struct sk_buff *skb, const struct flow_keys *flow)
184{
185 return ntohs(CTTUPLE(skb, dst.u.all));
186fallback:
187 return flow_get_proto_dst(skb, flow);
188}
189
190static u32 flow_get_rtclassid(const struct sk_buff *skb)
191{
192#ifdef CONFIG_IP_ROUTE_CLASSID
193 if (skb_dst(skb))
194 return skb_dst(skb)->tclassid;
195#endif
196 return 0;
197}
198
199static u32 flow_get_skuid(const struct sk_buff *skb)
200{
201 struct sock *sk = skb_to_full_sk(skb);
202
203 if (sk && sk->sk_socket && sk->sk_socket->file) {
204 kuid_t skuid = sk->sk_socket->file->f_cred->fsuid;
205
206 return from_kuid(&init_user_ns, skuid);
207 }
208 return 0;
209}
210
211static u32 flow_get_skgid(const struct sk_buff *skb)
212{
213 struct sock *sk = skb_to_full_sk(skb);
214
215 if (sk && sk->sk_socket && sk->sk_socket->file) {
216 kgid_t skgid = sk->sk_socket->file->f_cred->fsgid;
217
218 return from_kgid(&init_user_ns, skgid);
219 }
220 return 0;
221}
222
223static u32 flow_get_vlan_tag(const struct sk_buff *skb)
224{
225 u16 uninitialized_var(tag);
226
227 if (vlan_get_tag(skb, &tag) < 0)
228 return 0;
229 return tag & VLAN_VID_MASK;
230}
231
232static u32 flow_get_rxhash(struct sk_buff *skb)
233{
234 return skb_get_hash(skb);
235}
236
237static u32 flow_key_get(struct sk_buff *skb, int key, struct flow_keys *flow)
238{
239 switch (key) {
240 case FLOW_KEY_SRC:
241 return flow_get_src(skb, flow);
242 case FLOW_KEY_DST:
243 return flow_get_dst(skb, flow);
244 case FLOW_KEY_PROTO:
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);
250 case FLOW_KEY_IIF:
251 return flow_get_iif(skb);
252 case FLOW_KEY_PRIORITY:
253 return flow_get_priority(skb);
254 case FLOW_KEY_MARK:
255 return flow_get_mark(skb);
256 case FLOW_KEY_NFCT:
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);
268 case FLOW_KEY_SKUID:
269 return flow_get_skuid(skb);
270 case FLOW_KEY_SKGID:
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);
276 default:
277 WARN_ON(1);
278 return 0;
279 }
280}
281
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))
291
292static int flow_classify(struct sk_buff *skb, const struct tcf_proto *tp,
293 struct tcf_result *res)
294{
295 struct flow_head *head = rcu_dereference_bh(tp->root);
296 struct flow_filter *f;
297 u32 keymask;
298 u32 classid;
299 unsigned int n, key;
300 int r;
301
302 list_for_each_entry_rcu(f, &head->filters, list) {
303 u32 keys[FLOW_KEY_MAX + 1];
304 struct flow_keys flow_keys;
305
306 if (!tcf_em_tree_match(skb, &f->ematches, NULL))
307 continue;
308
309 keymask = f->keymask;
310 if (keymask & FLOW_KEYS_NEEDED)
311 skb_flow_dissect_flow_keys(skb, &flow_keys, 0);
312
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);
317 }
318
319 if (f->mode == FLOW_MODE_HASH)
320 classid = jhash2(keys, f->nkeys, f->hashrnd);
321 else {
322 classid = keys[0];
323 classid = (classid & f->mask) ^ f->xor;
324 classid = (classid >> f->rshift) + f->addend;
325 }
326
327 if (f->divisor)
328 classid %= f->divisor;
329
330 res->class = 0;
331 res->classid = TC_H_MAKE(f->baseclass, f->baseclass + classid);
332
333 r = tcf_exts_exec(skb, &f->exts, res);
334 if (r < 0)
335 continue;
336 return r;
337 }
338 return -1;
339}
340
341static void flow_perturbation(unsigned long arg)
342{
343 struct flow_filter *f = (struct flow_filter *)arg;
344
345 get_random_bytes(&f->hashrnd, 4);
346 if (f->perturb_period)
347 mod_timer(&f->perturb_timer, jiffies + f->perturb_period);
348}
349
350static 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 },
363};
364
365static void flow_destroy_filter(struct rcu_head *head)
366{
367 struct flow_filter *f = container_of(head, struct flow_filter, rcu);
368
369 del_timer_sync(&f->perturb_timer);
370 tcf_exts_destroy(&f->exts);
371 tcf_em_tree_destroy(&f->ematches);
372 kfree(f);
373}
374
375static 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)
379{
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];
384 struct tcf_exts e;
385 struct tcf_ematch_tree t;
386 unsigned int nkeys = 0;
387 unsigned int perturb_period = 0;
388 u32 baseclass = 0;
389 u32 keymask = 0;
390 u32 mode;
391 int err;
392
393 if (opt == NULL)
394 return -EINVAL;
395
396 err = nla_parse_nested(tb, TCA_FLOW_MAX, opt, flow_policy);
397 if (err < 0)
398 return err;
399
400 if (tb[TCA_FLOW_BASECLASS]) {
401 baseclass = nla_get_u32(tb[TCA_FLOW_BASECLASS]);
402 if (TC_H_MIN(baseclass) == 0)
403 return -EINVAL;
404 }
405
406 if (tb[TCA_FLOW_KEYS]) {
407 keymask = nla_get_u32(tb[TCA_FLOW_KEYS]);
408
409 nkeys = hweight32(keymask);
410 if (nkeys == 0)
411 return -EINVAL;
412
413 if (fls(keymask) - 1 > FLOW_KEY_MAX)
414 return -EOPNOTSUPP;
415
416 if ((keymask & (FLOW_KEY_SKUID|FLOW_KEY_SKGID)) &&
417 sk_user_ns(NETLINK_CB(in_skb).sk) != &init_user_ns)
418 return -EOPNOTSUPP;
419 }
420
421 tcf_exts_init(&e, TCA_FLOW_ACT, TCA_FLOW_POLICE);
422 err = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &e, ovr);
423 if (err < 0)
424 return err;
425
426 err = tcf_em_tree_validate(tp, tb[TCA_FLOW_EMATCHES], &t);
427 if (err < 0)
428 goto err1;
429
430 err = -ENOBUFS;
431 fnew = kzalloc(sizeof(*fnew), GFP_KERNEL);
432 if (!fnew)
433 goto err2;
434
435 tcf_exts_init(&fnew->exts, TCA_FLOW_ACT, TCA_FLOW_POLICE);
436
437 fold = (struct flow_filter *)*arg;
438 if (fold) {
439 err = -EINVAL;
440 if (fold->handle != handle && handle)
441 goto err2;
442
443 /* Copy fold into fnew */
444 fnew->tp = fold->tp;
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;
456
457 mode = fold->mode;
458 if (tb[TCA_FLOW_MODE])
459 mode = nla_get_u32(tb[TCA_FLOW_MODE]);
460 if (mode != FLOW_MODE_HASH && nkeys > 1)
461 goto err2;
462
463 if (mode == FLOW_MODE_HASH)
464 perturb_period = fold->perturb_period;
465 if (tb[TCA_FLOW_PERTURB]) {
466 if (mode != FLOW_MODE_HASH)
467 goto err2;
468 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
469 }
470 } else {
471 err = -EINVAL;
472 if (!handle)
473 goto err2;
474 if (!tb[TCA_FLOW_KEYS])
475 goto err2;
476
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)
481 goto err2;
482
483 if (tb[TCA_FLOW_PERTURB]) {
484 if (mode != FLOW_MODE_HASH)
485 goto err2;
486 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
487 }
488
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);
493
494 fnew->handle = handle;
495 fnew->mask = ~0U;
496 fnew->tp = tp;
497 get_random_bytes(&fnew->hashrnd, 4);
498 }
499
500 fnew->perturb_timer.function = flow_perturbation;
501 fnew->perturb_timer.data = (unsigned long)fnew;
502 init_timer_deferrable(&fnew->perturb_timer);
503
504 tcf_exts_change(tp, &fnew->exts, &e);
505 tcf_em_tree_change(tp, &fnew->ematches, &t);
506
507 netif_keep_dst(qdisc_dev(tp->q));
508
509 if (tb[TCA_FLOW_KEYS]) {
510 fnew->keymask = keymask;
511 fnew->nkeys = nkeys;
512 }
513
514 fnew->mode = mode;
515
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]);
524
525 if (tb[TCA_FLOW_DIVISOR])
526 fnew->divisor = nla_get_u32(tb[TCA_FLOW_DIVISOR]);
527 if (baseclass)
528 fnew->baseclass = baseclass;
529
530 fnew->perturb_period = perturb_period;
531 if (perturb_period)
532 mod_timer(&fnew->perturb_timer, jiffies + perturb_period);
533
534 if (*arg == 0)
535 list_add_tail_rcu(&fnew->list, &head->filters);
536 else
537 list_replace_rcu(&fold->list, &fnew->list);
538
539 *arg = (unsigned long)fnew;
540
541 if (fold)
542 call_rcu(&fold->rcu, flow_destroy_filter);
543 return 0;
544
545err2:
546 tcf_em_tree_destroy(&t);
547 kfree(fnew);
548err1:
549 tcf_exts_destroy(&e);
550 return err;
551}
552
553static int flow_delete(struct tcf_proto *tp, unsigned long arg)
554{
555 struct flow_filter *f = (struct flow_filter *)arg;
556
557 list_del_rcu(&f->list);
558 call_rcu(&f->rcu, flow_destroy_filter);
559 return 0;
560}
561
562static int flow_init(struct tcf_proto *tp)
563{
564 struct flow_head *head;
565
566 head = kzalloc(sizeof(*head), GFP_KERNEL);
567 if (head == NULL)
568 return -ENOBUFS;
569 INIT_LIST_HEAD(&head->filters);
570 rcu_assign_pointer(tp->root, head);
571 return 0;
572}
573
574static bool flow_destroy(struct tcf_proto *tp, bool force)
575{
576 struct flow_head *head = rtnl_dereference(tp->root);
577 struct flow_filter *f, *next;
578
579 if (!force && !list_empty(&head->filters))
580 return false;
581
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);
585 }
586 RCU_INIT_POINTER(tp->root, NULL);
587 kfree_rcu(head, rcu);
588 return true;
589}
590
591static unsigned long flow_get(struct tcf_proto *tp, u32 handle)
592{
593 struct flow_head *head = rtnl_dereference(tp->root);
594 struct flow_filter *f;
595
596 list_for_each_entry(f, &head->filters, list)
597 if (f->handle == handle)
598 return (unsigned long)f;
599 return 0;
600}
601
602static int flow_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
603 struct sk_buff *skb, struct tcmsg *t)
604{
605 struct flow_filter *f = (struct flow_filter *)fh;
606 struct nlattr *nest;
607
608 if (f == NULL)
609 return skb->len;
610
611 t->tcm_handle = f->handle;
612
613 nest = nla_nest_start(skb, TCA_OPTIONS);
614 if (nest == NULL)
615 goto nla_put_failure;
616
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;
620
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;
625 }
626 if (f->rshift &&
627 nla_put_u32(skb, TCA_FLOW_RSHIFT, f->rshift))
628 goto nla_put_failure;
629 if (f->addend &&
630 nla_put_u32(skb, TCA_FLOW_ADDEND, f->addend))
631 goto nla_put_failure;
632
633 if (f->divisor &&
634 nla_put_u32(skb, TCA_FLOW_DIVISOR, f->divisor))
635 goto nla_put_failure;
636 if (f->baseclass &&
637 nla_put_u32(skb, TCA_FLOW_BASECLASS, f->baseclass))
638 goto nla_put_failure;
639
640 if (f->perturb_period &&
641 nla_put_u32(skb, TCA_FLOW_PERTURB, f->perturb_period / HZ))
642 goto nla_put_failure;
643
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;
650#endif
651 nla_nest_end(skb, nest);
652
653 if (tcf_exts_dump_stats(skb, &f->exts) < 0)
654 goto nla_put_failure;
655
656 return skb->len;
657
658nla_put_failure:
659 nla_nest_cancel(skb, nest);
660 return -1;
661}
662
663static void flow_walk(struct tcf_proto *tp, struct tcf_walker *arg)
664{
665 struct flow_head *head = rtnl_dereference(tp->root);
666 struct flow_filter *f;
667
668 list_for_each_entry(f, &head->filters, list) {
669 if (arg->count < arg->skip)
670 goto skip;
671 if (arg->fn(tp, (unsigned long)f, arg) < 0) {
672 arg->stop = 1;
673 break;
674 }
675skip:
676 arg->count++;
677 }
678}
679
680static struct tcf_proto_ops cls_flow_ops __read_mostly = {
681 .kind = "flow",
682 .classify = flow_classify,
683 .init = flow_init,
684 .destroy = flow_destroy,
685 .change = flow_change,
686 .delete = flow_delete,
687 .get = flow_get,
688 .dump = flow_dump,
689 .walk = flow_walk,
690 .owner = THIS_MODULE,
691};
692
693static int __init cls_flow_init(void)
694{
695 return register_tcf_proto_ops(&cls_flow_ops);
696}
697
698static void __exit cls_flow_exit(void)
699{
700 unregister_tcf_proto_ops(&cls_flow_ops);
701}
702
703module_init(cls_flow_init);
704module_exit(cls_flow_exit);
705
706MODULE_LICENSE("GPL");
707MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
708MODULE_DESCRIPTION("TC flow classifier");
1/*
2 * net/sched/cls_flow.c Generic flow classifier
3 *
4 * Copyright (c) 2007, 2008 Patrick McHardy <kaber@trash.net>
5 *
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.
10 */
11
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
26#include <net/pkt_cls.h>
27#include <net/ip.h>
28#include <net/route.h>
29#include <net/flow_keys.h>
30
31#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
32#include <net/netfilter/nf_conntrack.h>
33#endif
34
35struct flow_head {
36 struct list_head filters;
37};
38
39struct flow_filter {
40 struct list_head list;
41 struct tcf_exts exts;
42 struct tcf_ematch_tree ematches;
43 struct timer_list perturb_timer;
44 u32 perturb_period;
45 u32 handle;
46
47 u32 nkeys;
48 u32 keymask;
49 u32 mode;
50 u32 mask;
51 u32 xor;
52 u32 rshift;
53 u32 addend;
54 u32 divisor;
55 u32 baseclass;
56 u32 hashrnd;
57};
58
59static inline u32 addr_fold(void *addr)
60{
61 unsigned long a = (unsigned long)addr;
62
63 return (a & 0xFFFFFFFF) ^ (BITS_PER_LONG > 32 ? a >> 32 : 0);
64}
65
66static u32 flow_get_src(const struct sk_buff *skb, const struct flow_keys *flow)
67{
68 if (flow->src)
69 return ntohl(flow->src);
70 return addr_fold(skb->sk);
71}
72
73static u32 flow_get_dst(const struct sk_buff *skb, const struct flow_keys *flow)
74{
75 if (flow->dst)
76 return ntohl(flow->dst);
77 return addr_fold(skb_dst(skb)) ^ (__force u16)skb->protocol;
78}
79
80static u32 flow_get_proto(const struct sk_buff *skb, const struct flow_keys *flow)
81{
82 return flow->ip_proto;
83}
84
85static u32 flow_get_proto_src(const struct sk_buff *skb, const struct flow_keys *flow)
86{
87 if (flow->ports)
88 return ntohs(flow->port16[0]);
89
90 return addr_fold(skb->sk);
91}
92
93static u32 flow_get_proto_dst(const struct sk_buff *skb, const struct flow_keys *flow)
94{
95 if (flow->ports)
96 return ntohs(flow->port16[1]);
97
98 return addr_fold(skb_dst(skb)) ^ (__force u16)skb->protocol;
99}
100
101static u32 flow_get_iif(const struct sk_buff *skb)
102{
103 return skb->skb_iif;
104}
105
106static u32 flow_get_priority(const struct sk_buff *skb)
107{
108 return skb->priority;
109}
110
111static u32 flow_get_mark(const struct sk_buff *skb)
112{
113 return skb->mark;
114}
115
116static u32 flow_get_nfct(const struct sk_buff *skb)
117{
118#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
119 return addr_fold(skb->nfct);
120#else
121 return 0;
122#endif
123}
124
125#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
126#define CTTUPLE(skb, member) \
127({ \
128 enum ip_conntrack_info ctinfo; \
129 const struct nf_conn *ct = nf_ct_get(skb, &ctinfo); \
130 if (ct == NULL) \
131 goto fallback; \
132 ct->tuplehash[CTINFO2DIR(ctinfo)].tuple.member; \
133})
134#else
135#define CTTUPLE(skb, member) \
136({ \
137 goto fallback; \
138 0; \
139})
140#endif
141
142static u32 flow_get_nfct_src(const struct sk_buff *skb, const struct flow_keys *flow)
143{
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]));
149 }
150fallback:
151 return flow_get_src(skb, flow);
152}
153
154static u32 flow_get_nfct_dst(const struct sk_buff *skb, const struct flow_keys *flow)
155{
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]));
161 }
162fallback:
163 return flow_get_dst(skb, flow);
164}
165
166static u32 flow_get_nfct_proto_src(const struct sk_buff *skb, const struct flow_keys *flow)
167{
168 return ntohs(CTTUPLE(skb, src.u.all));
169fallback:
170 return flow_get_proto_src(skb, flow);
171}
172
173static u32 flow_get_nfct_proto_dst(const struct sk_buff *skb, const struct flow_keys *flow)
174{
175 return ntohs(CTTUPLE(skb, dst.u.all));
176fallback:
177 return flow_get_proto_dst(skb, flow);
178}
179
180static u32 flow_get_rtclassid(const struct sk_buff *skb)
181{
182#ifdef CONFIG_IP_ROUTE_CLASSID
183 if (skb_dst(skb))
184 return skb_dst(skb)->tclassid;
185#endif
186 return 0;
187}
188
189static u32 flow_get_skuid(const struct sk_buff *skb)
190{
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);
194 }
195 return 0;
196}
197
198static u32 flow_get_skgid(const struct sk_buff *skb)
199{
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);
203 }
204 return 0;
205}
206
207static u32 flow_get_vlan_tag(const struct sk_buff *skb)
208{
209 u16 uninitialized_var(tag);
210
211 if (vlan_get_tag(skb, &tag) < 0)
212 return 0;
213 return tag & VLAN_VID_MASK;
214}
215
216static u32 flow_get_rxhash(struct sk_buff *skb)
217{
218 return skb_get_hash(skb);
219}
220
221static u32 flow_key_get(struct sk_buff *skb, int key, struct flow_keys *flow)
222{
223 switch (key) {
224 case FLOW_KEY_SRC:
225 return flow_get_src(skb, flow);
226 case FLOW_KEY_DST:
227 return flow_get_dst(skb, flow);
228 case FLOW_KEY_PROTO:
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);
234 case FLOW_KEY_IIF:
235 return flow_get_iif(skb);
236 case FLOW_KEY_PRIORITY:
237 return flow_get_priority(skb);
238 case FLOW_KEY_MARK:
239 return flow_get_mark(skb);
240 case FLOW_KEY_NFCT:
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);
252 case FLOW_KEY_SKUID:
253 return flow_get_skuid(skb);
254 case FLOW_KEY_SKGID:
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);
260 default:
261 WARN_ON(1);
262 return 0;
263 }
264}
265
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))
275
276static int flow_classify(struct sk_buff *skb, const struct tcf_proto *tp,
277 struct tcf_result *res)
278{
279 struct flow_head *head = tp->root;
280 struct flow_filter *f;
281 u32 keymask;
282 u32 classid;
283 unsigned int n, key;
284 int r;
285
286 list_for_each_entry(f, &head->filters, list) {
287 u32 keys[FLOW_KEY_MAX + 1];
288 struct flow_keys flow_keys;
289
290 if (!tcf_em_tree_match(skb, &f->ematches, NULL))
291 continue;
292
293 keymask = f->keymask;
294 if (keymask & FLOW_KEYS_NEEDED)
295 skb_flow_dissect(skb, &flow_keys);
296
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);
301 }
302
303 if (f->mode == FLOW_MODE_HASH)
304 classid = jhash2(keys, f->nkeys, f->hashrnd);
305 else {
306 classid = keys[0];
307 classid = (classid & f->mask) ^ f->xor;
308 classid = (classid >> f->rshift) + f->addend;
309 }
310
311 if (f->divisor)
312 classid %= f->divisor;
313
314 res->class = 0;
315 res->classid = TC_H_MAKE(f->baseclass, f->baseclass + classid);
316
317 r = tcf_exts_exec(skb, &f->exts, res);
318 if (r < 0)
319 continue;
320 return r;
321 }
322 return -1;
323}
324
325static void flow_perturbation(unsigned long arg)
326{
327 struct flow_filter *f = (struct flow_filter *)arg;
328
329 get_random_bytes(&f->hashrnd, 4);
330 if (f->perturb_period)
331 mod_timer(&f->perturb_timer, jiffies + f->perturb_period);
332}
333
334static 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 },
347};
348
349static 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,
352 unsigned long *arg)
353{
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];
358 struct tcf_exts e;
359 struct tcf_ematch_tree t;
360 unsigned int nkeys = 0;
361 unsigned int perturb_period = 0;
362 u32 baseclass = 0;
363 u32 keymask = 0;
364 u32 mode;
365 int err;
366
367 if (opt == NULL)
368 return -EINVAL;
369
370 err = nla_parse_nested(tb, TCA_FLOW_MAX, opt, flow_policy);
371 if (err < 0)
372 return err;
373
374 if (tb[TCA_FLOW_BASECLASS]) {
375 baseclass = nla_get_u32(tb[TCA_FLOW_BASECLASS]);
376 if (TC_H_MIN(baseclass) == 0)
377 return -EINVAL;
378 }
379
380 if (tb[TCA_FLOW_KEYS]) {
381 keymask = nla_get_u32(tb[TCA_FLOW_KEYS]);
382
383 nkeys = hweight32(keymask);
384 if (nkeys == 0)
385 return -EINVAL;
386
387 if (fls(keymask) - 1 > FLOW_KEY_MAX)
388 return -EOPNOTSUPP;
389
390 if ((keymask & (FLOW_KEY_SKUID|FLOW_KEY_SKGID)) &&
391 sk_user_ns(NETLINK_CB(in_skb).sk) != &init_user_ns)
392 return -EOPNOTSUPP;
393 }
394
395 tcf_exts_init(&e, TCA_FLOW_ACT, TCA_FLOW_POLICE);
396 err = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &e);
397 if (err < 0)
398 return err;
399
400 err = tcf_em_tree_validate(tp, tb[TCA_FLOW_EMATCHES], &t);
401 if (err < 0)
402 goto err1;
403
404 f = (struct flow_filter *)*arg;
405 if (f != NULL) {
406 err = -EINVAL;
407 if (f->handle != handle && handle)
408 goto err2;
409
410 mode = f->mode;
411 if (tb[TCA_FLOW_MODE])
412 mode = nla_get_u32(tb[TCA_FLOW_MODE]);
413 if (mode != FLOW_MODE_HASH && nkeys > 1)
414 goto err2;
415
416 if (mode == FLOW_MODE_HASH)
417 perturb_period = f->perturb_period;
418 if (tb[TCA_FLOW_PERTURB]) {
419 if (mode != FLOW_MODE_HASH)
420 goto err2;
421 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
422 }
423 } else {
424 err = -EINVAL;
425 if (!handle)
426 goto err2;
427 if (!tb[TCA_FLOW_KEYS])
428 goto err2;
429
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)
434 goto err2;
435
436 if (tb[TCA_FLOW_PERTURB]) {
437 if (mode != FLOW_MODE_HASH)
438 goto err2;
439 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
440 }
441
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);
446
447 err = -ENOBUFS;
448 f = kzalloc(sizeof(*f), GFP_KERNEL);
449 if (f == NULL)
450 goto err2;
451
452 f->handle = handle;
453 f->mask = ~0U;
454 tcf_exts_init(&f->exts, TCA_FLOW_ACT, TCA_FLOW_POLICE);
455
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);
460 }
461
462 tcf_exts_change(tp, &f->exts, &e);
463 tcf_em_tree_change(tp, &f->ematches, &t);
464
465 tcf_tree_lock(tp);
466
467 if (tb[TCA_FLOW_KEYS]) {
468 f->keymask = keymask;
469 f->nkeys = nkeys;
470 }
471
472 f->mode = mode;
473
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]);
482
483 if (tb[TCA_FLOW_DIVISOR])
484 f->divisor = nla_get_u32(tb[TCA_FLOW_DIVISOR]);
485 if (baseclass)
486 f->baseclass = baseclass;
487
488 f->perturb_period = perturb_period;
489 del_timer(&f->perturb_timer);
490 if (perturb_period)
491 mod_timer(&f->perturb_timer, jiffies + perturb_period);
492
493 if (*arg == 0)
494 list_add_tail(&f->list, &head->filters);
495
496 tcf_tree_unlock(tp);
497
498 *arg = (unsigned long)f;
499 return 0;
500
501err2:
502 tcf_em_tree_destroy(tp, &t);
503err1:
504 tcf_exts_destroy(tp, &e);
505 return err;
506}
507
508static void flow_destroy_filter(struct tcf_proto *tp, struct flow_filter *f)
509{
510 del_timer_sync(&f->perturb_timer);
511 tcf_exts_destroy(tp, &f->exts);
512 tcf_em_tree_destroy(tp, &f->ematches);
513 kfree(f);
514}
515
516static int flow_delete(struct tcf_proto *tp, unsigned long arg)
517{
518 struct flow_filter *f = (struct flow_filter *)arg;
519
520 tcf_tree_lock(tp);
521 list_del(&f->list);
522 tcf_tree_unlock(tp);
523 flow_destroy_filter(tp, f);
524 return 0;
525}
526
527static int flow_init(struct tcf_proto *tp)
528{
529 struct flow_head *head;
530
531 head = kzalloc(sizeof(*head), GFP_KERNEL);
532 if (head == NULL)
533 return -ENOBUFS;
534 INIT_LIST_HEAD(&head->filters);
535 tp->root = head;
536 return 0;
537}
538
539static void flow_destroy(struct tcf_proto *tp)
540{
541 struct flow_head *head = tp->root;
542 struct flow_filter *f, *next;
543
544 list_for_each_entry_safe(f, next, &head->filters, list) {
545 list_del(&f->list);
546 flow_destroy_filter(tp, f);
547 }
548 kfree(head);
549}
550
551static unsigned long flow_get(struct tcf_proto *tp, u32 handle)
552{
553 struct flow_head *head = tp->root;
554 struct flow_filter *f;
555
556 list_for_each_entry(f, &head->filters, list)
557 if (f->handle == handle)
558 return (unsigned long)f;
559 return 0;
560}
561
562static void flow_put(struct tcf_proto *tp, unsigned long f)
563{
564}
565
566static int flow_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
567 struct sk_buff *skb, struct tcmsg *t)
568{
569 struct flow_filter *f = (struct flow_filter *)fh;
570 struct nlattr *nest;
571
572 if (f == NULL)
573 return skb->len;
574
575 t->tcm_handle = f->handle;
576
577 nest = nla_nest_start(skb, TCA_OPTIONS);
578 if (nest == NULL)
579 goto nla_put_failure;
580
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;
584
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;
589 }
590 if (f->rshift &&
591 nla_put_u32(skb, TCA_FLOW_RSHIFT, f->rshift))
592 goto nla_put_failure;
593 if (f->addend &&
594 nla_put_u32(skb, TCA_FLOW_ADDEND, f->addend))
595 goto nla_put_failure;
596
597 if (f->divisor &&
598 nla_put_u32(skb, TCA_FLOW_DIVISOR, f->divisor))
599 goto nla_put_failure;
600 if (f->baseclass &&
601 nla_put_u32(skb, TCA_FLOW_BASECLASS, f->baseclass))
602 goto nla_put_failure;
603
604 if (f->perturb_period &&
605 nla_put_u32(skb, TCA_FLOW_PERTURB, f->perturb_period / HZ))
606 goto nla_put_failure;
607
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;
614#endif
615 nla_nest_end(skb, nest);
616
617 if (tcf_exts_dump_stats(skb, &f->exts) < 0)
618 goto nla_put_failure;
619
620 return skb->len;
621
622nla_put_failure:
623 nlmsg_trim(skb, nest);
624 return -1;
625}
626
627static void flow_walk(struct tcf_proto *tp, struct tcf_walker *arg)
628{
629 struct flow_head *head = tp->root;
630 struct flow_filter *f;
631
632 list_for_each_entry(f, &head->filters, list) {
633 if (arg->count < arg->skip)
634 goto skip;
635 if (arg->fn(tp, (unsigned long)f, arg) < 0) {
636 arg->stop = 1;
637 break;
638 }
639skip:
640 arg->count++;
641 }
642}
643
644static struct tcf_proto_ops cls_flow_ops __read_mostly = {
645 .kind = "flow",
646 .classify = flow_classify,
647 .init = flow_init,
648 .destroy = flow_destroy,
649 .change = flow_change,
650 .delete = flow_delete,
651 .get = flow_get,
652 .put = flow_put,
653 .dump = flow_dump,
654 .walk = flow_walk,
655 .owner = THIS_MODULE,
656};
657
658static int __init cls_flow_init(void)
659{
660 return register_tcf_proto_ops(&cls_flow_ops);
661}
662
663static void __exit cls_flow_exit(void)
664{
665 unregister_tcf_proto_ops(&cls_flow_ops);
666}
667
668module_init(cls_flow_init);
669module_exit(cls_flow_exit);
670
671MODULE_LICENSE("GPL");
672MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
673MODULE_DESCRIPTION("TC flow classifier");