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
25#include <net/pkt_cls.h>
26#include <net/ip.h>
27#include <net/route.h>
28#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
29#include <net/netfilter/nf_conntrack.h>
30#endif
31
32struct flow_head {
33 struct list_head filters;
34};
35
36struct flow_filter {
37 struct list_head list;
38 struct tcf_exts exts;
39 struct tcf_ematch_tree ematches;
40 struct timer_list perturb_timer;
41 u32 perturb_period;
42 u32 handle;
43
44 u32 nkeys;
45 u32 keymask;
46 u32 mode;
47 u32 mask;
48 u32 xor;
49 u32 rshift;
50 u32 addend;
51 u32 divisor;
52 u32 baseclass;
53 u32 hashrnd;
54};
55
56static const struct tcf_ext_map flow_ext_map = {
57 .action = TCA_FLOW_ACT,
58 .police = TCA_FLOW_POLICE,
59};
60
61static inline u32 addr_fold(void *addr)
62{
63 unsigned long a = (unsigned long)addr;
64
65 return (a & 0xFFFFFFFF) ^ (BITS_PER_LONG > 32 ? a >> 32 : 0);
66}
67
68static u32 flow_get_src(struct sk_buff *skb)
69{
70 switch (skb->protocol) {
71 case htons(ETH_P_IP):
72 if (pskb_network_may_pull(skb, sizeof(struct iphdr)))
73 return ntohl(ip_hdr(skb)->saddr);
74 break;
75 case htons(ETH_P_IPV6):
76 if (pskb_network_may_pull(skb, sizeof(struct ipv6hdr)))
77 return ntohl(ipv6_hdr(skb)->saddr.s6_addr32[3]);
78 break;
79 }
80
81 return addr_fold(skb->sk);
82}
83
84static u32 flow_get_dst(struct sk_buff *skb)
85{
86 switch (skb->protocol) {
87 case htons(ETH_P_IP):
88 if (pskb_network_may_pull(skb, sizeof(struct iphdr)))
89 return ntohl(ip_hdr(skb)->daddr);
90 break;
91 case htons(ETH_P_IPV6):
92 if (pskb_network_may_pull(skb, sizeof(struct ipv6hdr)))
93 return ntohl(ipv6_hdr(skb)->daddr.s6_addr32[3]);
94 break;
95 }
96
97 return addr_fold(skb_dst(skb)) ^ (__force u16)skb->protocol;
98}
99
100static u32 flow_get_proto(struct sk_buff *skb)
101{
102 switch (skb->protocol) {
103 case htons(ETH_P_IP):
104 return pskb_network_may_pull(skb, sizeof(struct iphdr)) ?
105 ip_hdr(skb)->protocol : 0;
106 case htons(ETH_P_IPV6):
107 return pskb_network_may_pull(skb, sizeof(struct ipv6hdr)) ?
108 ipv6_hdr(skb)->nexthdr : 0;
109 default:
110 return 0;
111 }
112}
113
114static u32 flow_get_proto_src(struct sk_buff *skb)
115{
116 switch (skb->protocol) {
117 case htons(ETH_P_IP): {
118 struct iphdr *iph;
119 int poff;
120
121 if (!pskb_network_may_pull(skb, sizeof(*iph)))
122 break;
123 iph = ip_hdr(skb);
124 if (ip_is_fragment(iph))
125 break;
126 poff = proto_ports_offset(iph->protocol);
127 if (poff >= 0 &&
128 pskb_network_may_pull(skb, iph->ihl * 4 + 2 + poff)) {
129 iph = ip_hdr(skb);
130 return ntohs(*(__be16 *)((void *)iph + iph->ihl * 4 +
131 poff));
132 }
133 break;
134 }
135 case htons(ETH_P_IPV6): {
136 struct ipv6hdr *iph;
137 int poff;
138
139 if (!pskb_network_may_pull(skb, sizeof(*iph)))
140 break;
141 iph = ipv6_hdr(skb);
142 poff = proto_ports_offset(iph->nexthdr);
143 if (poff >= 0 &&
144 pskb_network_may_pull(skb, sizeof(*iph) + poff + 2)) {
145 iph = ipv6_hdr(skb);
146 return ntohs(*(__be16 *)((void *)iph + sizeof(*iph) +
147 poff));
148 }
149 break;
150 }
151 }
152
153 return addr_fold(skb->sk);
154}
155
156static u32 flow_get_proto_dst(struct sk_buff *skb)
157{
158 switch (skb->protocol) {
159 case htons(ETH_P_IP): {
160 struct iphdr *iph;
161 int poff;
162
163 if (!pskb_network_may_pull(skb, sizeof(*iph)))
164 break;
165 iph = ip_hdr(skb);
166 if (ip_is_fragment(iph))
167 break;
168 poff = proto_ports_offset(iph->protocol);
169 if (poff >= 0 &&
170 pskb_network_may_pull(skb, iph->ihl * 4 + 4 + poff)) {
171 iph = ip_hdr(skb);
172 return ntohs(*(__be16 *)((void *)iph + iph->ihl * 4 +
173 2 + poff));
174 }
175 break;
176 }
177 case htons(ETH_P_IPV6): {
178 struct ipv6hdr *iph;
179 int poff;
180
181 if (!pskb_network_may_pull(skb, sizeof(*iph)))
182 break;
183 iph = ipv6_hdr(skb);
184 poff = proto_ports_offset(iph->nexthdr);
185 if (poff >= 0 &&
186 pskb_network_may_pull(skb, sizeof(*iph) + poff + 4)) {
187 iph = ipv6_hdr(skb);
188 return ntohs(*(__be16 *)((void *)iph + sizeof(*iph) +
189 poff + 2));
190 }
191 break;
192 }
193 }
194
195 return addr_fold(skb_dst(skb)) ^ (__force u16)skb->protocol;
196}
197
198static u32 flow_get_iif(const struct sk_buff *skb)
199{
200 return skb->skb_iif;
201}
202
203static u32 flow_get_priority(const struct sk_buff *skb)
204{
205 return skb->priority;
206}
207
208static u32 flow_get_mark(const struct sk_buff *skb)
209{
210 return skb->mark;
211}
212
213static u32 flow_get_nfct(const struct sk_buff *skb)
214{
215#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
216 return addr_fold(skb->nfct);
217#else
218 return 0;
219#endif
220}
221
222#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
223#define CTTUPLE(skb, member) \
224({ \
225 enum ip_conntrack_info ctinfo; \
226 struct nf_conn *ct = nf_ct_get(skb, &ctinfo); \
227 if (ct == NULL) \
228 goto fallback; \
229 ct->tuplehash[CTINFO2DIR(ctinfo)].tuple.member; \
230})
231#else
232#define CTTUPLE(skb, member) \
233({ \
234 goto fallback; \
235 0; \
236})
237#endif
238
239static u32 flow_get_nfct_src(struct sk_buff *skb)
240{
241 switch (skb->protocol) {
242 case htons(ETH_P_IP):
243 return ntohl(CTTUPLE(skb, src.u3.ip));
244 case htons(ETH_P_IPV6):
245 return ntohl(CTTUPLE(skb, src.u3.ip6[3]));
246 }
247fallback:
248 return flow_get_src(skb);
249}
250
251static u32 flow_get_nfct_dst(struct sk_buff *skb)
252{
253 switch (skb->protocol) {
254 case htons(ETH_P_IP):
255 return ntohl(CTTUPLE(skb, dst.u3.ip));
256 case htons(ETH_P_IPV6):
257 return ntohl(CTTUPLE(skb, dst.u3.ip6[3]));
258 }
259fallback:
260 return flow_get_dst(skb);
261}
262
263static u32 flow_get_nfct_proto_src(struct sk_buff *skb)
264{
265 return ntohs(CTTUPLE(skb, src.u.all));
266fallback:
267 return flow_get_proto_src(skb);
268}
269
270static u32 flow_get_nfct_proto_dst(struct sk_buff *skb)
271{
272 return ntohs(CTTUPLE(skb, dst.u.all));
273fallback:
274 return flow_get_proto_dst(skb);
275}
276
277static u32 flow_get_rtclassid(const struct sk_buff *skb)
278{
279#ifdef CONFIG_IP_ROUTE_CLASSID
280 if (skb_dst(skb))
281 return skb_dst(skb)->tclassid;
282#endif
283 return 0;
284}
285
286static u32 flow_get_skuid(const struct sk_buff *skb)
287{
288 if (skb->sk && skb->sk->sk_socket && skb->sk->sk_socket->file)
289 return skb->sk->sk_socket->file->f_cred->fsuid;
290 return 0;
291}
292
293static u32 flow_get_skgid(const struct sk_buff *skb)
294{
295 if (skb->sk && skb->sk->sk_socket && skb->sk->sk_socket->file)
296 return skb->sk->sk_socket->file->f_cred->fsgid;
297 return 0;
298}
299
300static u32 flow_get_vlan_tag(const struct sk_buff *skb)
301{
302 u16 uninitialized_var(tag);
303
304 if (vlan_get_tag(skb, &tag) < 0)
305 return 0;
306 return tag & VLAN_VID_MASK;
307}
308
309static u32 flow_get_rxhash(struct sk_buff *skb)
310{
311 return skb_get_rxhash(skb);
312}
313
314static u32 flow_key_get(struct sk_buff *skb, int key)
315{
316 switch (key) {
317 case FLOW_KEY_SRC:
318 return flow_get_src(skb);
319 case FLOW_KEY_DST:
320 return flow_get_dst(skb);
321 case FLOW_KEY_PROTO:
322 return flow_get_proto(skb);
323 case FLOW_KEY_PROTO_SRC:
324 return flow_get_proto_src(skb);
325 case FLOW_KEY_PROTO_DST:
326 return flow_get_proto_dst(skb);
327 case FLOW_KEY_IIF:
328 return flow_get_iif(skb);
329 case FLOW_KEY_PRIORITY:
330 return flow_get_priority(skb);
331 case FLOW_KEY_MARK:
332 return flow_get_mark(skb);
333 case FLOW_KEY_NFCT:
334 return flow_get_nfct(skb);
335 case FLOW_KEY_NFCT_SRC:
336 return flow_get_nfct_src(skb);
337 case FLOW_KEY_NFCT_DST:
338 return flow_get_nfct_dst(skb);
339 case FLOW_KEY_NFCT_PROTO_SRC:
340 return flow_get_nfct_proto_src(skb);
341 case FLOW_KEY_NFCT_PROTO_DST:
342 return flow_get_nfct_proto_dst(skb);
343 case FLOW_KEY_RTCLASSID:
344 return flow_get_rtclassid(skb);
345 case FLOW_KEY_SKUID:
346 return flow_get_skuid(skb);
347 case FLOW_KEY_SKGID:
348 return flow_get_skgid(skb);
349 case FLOW_KEY_VLAN_TAG:
350 return flow_get_vlan_tag(skb);
351 case FLOW_KEY_RXHASH:
352 return flow_get_rxhash(skb);
353 default:
354 WARN_ON(1);
355 return 0;
356 }
357}
358
359static int flow_classify(struct sk_buff *skb, const struct tcf_proto *tp,
360 struct tcf_result *res)
361{
362 struct flow_head *head = tp->root;
363 struct flow_filter *f;
364 u32 keymask;
365 u32 classid;
366 unsigned int n, key;
367 int r;
368
369 list_for_each_entry(f, &head->filters, list) {
370 u32 keys[f->nkeys];
371
372 if (!tcf_em_tree_match(skb, &f->ematches, NULL))
373 continue;
374
375 keymask = f->keymask;
376
377 for (n = 0; n < f->nkeys; n++) {
378 key = ffs(keymask) - 1;
379 keymask &= ~(1 << key);
380 keys[n] = flow_key_get(skb, key);
381 }
382
383 if (f->mode == FLOW_MODE_HASH)
384 classid = jhash2(keys, f->nkeys, f->hashrnd);
385 else {
386 classid = keys[0];
387 classid = (classid & f->mask) ^ f->xor;
388 classid = (classid >> f->rshift) + f->addend;
389 }
390
391 if (f->divisor)
392 classid %= f->divisor;
393
394 res->class = 0;
395 res->classid = TC_H_MAKE(f->baseclass, f->baseclass + classid);
396
397 r = tcf_exts_exec(skb, &f->exts, res);
398 if (r < 0)
399 continue;
400 return r;
401 }
402 return -1;
403}
404
405static void flow_perturbation(unsigned long arg)
406{
407 struct flow_filter *f = (struct flow_filter *)arg;
408
409 get_random_bytes(&f->hashrnd, 4);
410 if (f->perturb_period)
411 mod_timer(&f->perturb_timer, jiffies + f->perturb_period);
412}
413
414static const struct nla_policy flow_policy[TCA_FLOW_MAX + 1] = {
415 [TCA_FLOW_KEYS] = { .type = NLA_U32 },
416 [TCA_FLOW_MODE] = { .type = NLA_U32 },
417 [TCA_FLOW_BASECLASS] = { .type = NLA_U32 },
418 [TCA_FLOW_RSHIFT] = { .type = NLA_U32 },
419 [TCA_FLOW_ADDEND] = { .type = NLA_U32 },
420 [TCA_FLOW_MASK] = { .type = NLA_U32 },
421 [TCA_FLOW_XOR] = { .type = NLA_U32 },
422 [TCA_FLOW_DIVISOR] = { .type = NLA_U32 },
423 [TCA_FLOW_ACT] = { .type = NLA_NESTED },
424 [TCA_FLOW_POLICE] = { .type = NLA_NESTED },
425 [TCA_FLOW_EMATCHES] = { .type = NLA_NESTED },
426 [TCA_FLOW_PERTURB] = { .type = NLA_U32 },
427};
428
429static int flow_change(struct tcf_proto *tp, unsigned long base,
430 u32 handle, struct nlattr **tca,
431 unsigned long *arg)
432{
433 struct flow_head *head = tp->root;
434 struct flow_filter *f;
435 struct nlattr *opt = tca[TCA_OPTIONS];
436 struct nlattr *tb[TCA_FLOW_MAX + 1];
437 struct tcf_exts e;
438 struct tcf_ematch_tree t;
439 unsigned int nkeys = 0;
440 unsigned int perturb_period = 0;
441 u32 baseclass = 0;
442 u32 keymask = 0;
443 u32 mode;
444 int err;
445
446 if (opt == NULL)
447 return -EINVAL;
448
449 err = nla_parse_nested(tb, TCA_FLOW_MAX, opt, flow_policy);
450 if (err < 0)
451 return err;
452
453 if (tb[TCA_FLOW_BASECLASS]) {
454 baseclass = nla_get_u32(tb[TCA_FLOW_BASECLASS]);
455 if (TC_H_MIN(baseclass) == 0)
456 return -EINVAL;
457 }
458
459 if (tb[TCA_FLOW_KEYS]) {
460 keymask = nla_get_u32(tb[TCA_FLOW_KEYS]);
461
462 nkeys = hweight32(keymask);
463 if (nkeys == 0)
464 return -EINVAL;
465
466 if (fls(keymask) - 1 > FLOW_KEY_MAX)
467 return -EOPNOTSUPP;
468 }
469
470 err = tcf_exts_validate(tp, tb, tca[TCA_RATE], &e, &flow_ext_map);
471 if (err < 0)
472 return err;
473
474 err = tcf_em_tree_validate(tp, tb[TCA_FLOW_EMATCHES], &t);
475 if (err < 0)
476 goto err1;
477
478 f = (struct flow_filter *)*arg;
479 if (f != NULL) {
480 err = -EINVAL;
481 if (f->handle != handle && handle)
482 goto err2;
483
484 mode = f->mode;
485 if (tb[TCA_FLOW_MODE])
486 mode = nla_get_u32(tb[TCA_FLOW_MODE]);
487 if (mode != FLOW_MODE_HASH && nkeys > 1)
488 goto err2;
489
490 if (mode == FLOW_MODE_HASH)
491 perturb_period = f->perturb_period;
492 if (tb[TCA_FLOW_PERTURB]) {
493 if (mode != FLOW_MODE_HASH)
494 goto err2;
495 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
496 }
497 } else {
498 err = -EINVAL;
499 if (!handle)
500 goto err2;
501 if (!tb[TCA_FLOW_KEYS])
502 goto err2;
503
504 mode = FLOW_MODE_MAP;
505 if (tb[TCA_FLOW_MODE])
506 mode = nla_get_u32(tb[TCA_FLOW_MODE]);
507 if (mode != FLOW_MODE_HASH && nkeys > 1)
508 goto err2;
509
510 if (tb[TCA_FLOW_PERTURB]) {
511 if (mode != FLOW_MODE_HASH)
512 goto err2;
513 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
514 }
515
516 if (TC_H_MAJ(baseclass) == 0)
517 baseclass = TC_H_MAKE(tp->q->handle, baseclass);
518 if (TC_H_MIN(baseclass) == 0)
519 baseclass = TC_H_MAKE(baseclass, 1);
520
521 err = -ENOBUFS;
522 f = kzalloc(sizeof(*f), GFP_KERNEL);
523 if (f == NULL)
524 goto err2;
525
526 f->handle = handle;
527 f->mask = ~0U;
528
529 get_random_bytes(&f->hashrnd, 4);
530 f->perturb_timer.function = flow_perturbation;
531 f->perturb_timer.data = (unsigned long)f;
532 init_timer_deferrable(&f->perturb_timer);
533 }
534
535 tcf_exts_change(tp, &f->exts, &e);
536 tcf_em_tree_change(tp, &f->ematches, &t);
537
538 tcf_tree_lock(tp);
539
540 if (tb[TCA_FLOW_KEYS]) {
541 f->keymask = keymask;
542 f->nkeys = nkeys;
543 }
544
545 f->mode = mode;
546
547 if (tb[TCA_FLOW_MASK])
548 f->mask = nla_get_u32(tb[TCA_FLOW_MASK]);
549 if (tb[TCA_FLOW_XOR])
550 f->xor = nla_get_u32(tb[TCA_FLOW_XOR]);
551 if (tb[TCA_FLOW_RSHIFT])
552 f->rshift = nla_get_u32(tb[TCA_FLOW_RSHIFT]);
553 if (tb[TCA_FLOW_ADDEND])
554 f->addend = nla_get_u32(tb[TCA_FLOW_ADDEND]);
555
556 if (tb[TCA_FLOW_DIVISOR])
557 f->divisor = nla_get_u32(tb[TCA_FLOW_DIVISOR]);
558 if (baseclass)
559 f->baseclass = baseclass;
560
561 f->perturb_period = perturb_period;
562 del_timer(&f->perturb_timer);
563 if (perturb_period)
564 mod_timer(&f->perturb_timer, jiffies + perturb_period);
565
566 if (*arg == 0)
567 list_add_tail(&f->list, &head->filters);
568
569 tcf_tree_unlock(tp);
570
571 *arg = (unsigned long)f;
572 return 0;
573
574err2:
575 tcf_em_tree_destroy(tp, &t);
576err1:
577 tcf_exts_destroy(tp, &e);
578 return err;
579}
580
581static void flow_destroy_filter(struct tcf_proto *tp, struct flow_filter *f)
582{
583 del_timer_sync(&f->perturb_timer);
584 tcf_exts_destroy(tp, &f->exts);
585 tcf_em_tree_destroy(tp, &f->ematches);
586 kfree(f);
587}
588
589static int flow_delete(struct tcf_proto *tp, unsigned long arg)
590{
591 struct flow_filter *f = (struct flow_filter *)arg;
592
593 tcf_tree_lock(tp);
594 list_del(&f->list);
595 tcf_tree_unlock(tp);
596 flow_destroy_filter(tp, f);
597 return 0;
598}
599
600static int flow_init(struct tcf_proto *tp)
601{
602 struct flow_head *head;
603
604 head = kzalloc(sizeof(*head), GFP_KERNEL);
605 if (head == NULL)
606 return -ENOBUFS;
607 INIT_LIST_HEAD(&head->filters);
608 tp->root = head;
609 return 0;
610}
611
612static void flow_destroy(struct tcf_proto *tp)
613{
614 struct flow_head *head = tp->root;
615 struct flow_filter *f, *next;
616
617 list_for_each_entry_safe(f, next, &head->filters, list) {
618 list_del(&f->list);
619 flow_destroy_filter(tp, f);
620 }
621 kfree(head);
622}
623
624static unsigned long flow_get(struct tcf_proto *tp, u32 handle)
625{
626 struct flow_head *head = tp->root;
627 struct flow_filter *f;
628
629 list_for_each_entry(f, &head->filters, list)
630 if (f->handle == handle)
631 return (unsigned long)f;
632 return 0;
633}
634
635static void flow_put(struct tcf_proto *tp, unsigned long f)
636{
637}
638
639static int flow_dump(struct tcf_proto *tp, unsigned long fh,
640 struct sk_buff *skb, struct tcmsg *t)
641{
642 struct flow_filter *f = (struct flow_filter *)fh;
643 struct nlattr *nest;
644
645 if (f == NULL)
646 return skb->len;
647
648 t->tcm_handle = f->handle;
649
650 nest = nla_nest_start(skb, TCA_OPTIONS);
651 if (nest == NULL)
652 goto nla_put_failure;
653
654 NLA_PUT_U32(skb, TCA_FLOW_KEYS, f->keymask);
655 NLA_PUT_U32(skb, TCA_FLOW_MODE, f->mode);
656
657 if (f->mask != ~0 || f->xor != 0) {
658 NLA_PUT_U32(skb, TCA_FLOW_MASK, f->mask);
659 NLA_PUT_U32(skb, TCA_FLOW_XOR, f->xor);
660 }
661 if (f->rshift)
662 NLA_PUT_U32(skb, TCA_FLOW_RSHIFT, f->rshift);
663 if (f->addend)
664 NLA_PUT_U32(skb, TCA_FLOW_ADDEND, f->addend);
665
666 if (f->divisor)
667 NLA_PUT_U32(skb, TCA_FLOW_DIVISOR, f->divisor);
668 if (f->baseclass)
669 NLA_PUT_U32(skb, TCA_FLOW_BASECLASS, f->baseclass);
670
671 if (f->perturb_period)
672 NLA_PUT_U32(skb, TCA_FLOW_PERTURB, f->perturb_period / HZ);
673
674 if (tcf_exts_dump(skb, &f->exts, &flow_ext_map) < 0)
675 goto nla_put_failure;
676#ifdef CONFIG_NET_EMATCH
677 if (f->ematches.hdr.nmatches &&
678 tcf_em_tree_dump(skb, &f->ematches, TCA_FLOW_EMATCHES) < 0)
679 goto nla_put_failure;
680#endif
681 nla_nest_end(skb, nest);
682
683 if (tcf_exts_dump_stats(skb, &f->exts, &flow_ext_map) < 0)
684 goto nla_put_failure;
685
686 return skb->len;
687
688nla_put_failure:
689 nlmsg_trim(skb, nest);
690 return -1;
691}
692
693static void flow_walk(struct tcf_proto *tp, struct tcf_walker *arg)
694{
695 struct flow_head *head = tp->root;
696 struct flow_filter *f;
697
698 list_for_each_entry(f, &head->filters, list) {
699 if (arg->count < arg->skip)
700 goto skip;
701 if (arg->fn(tp, (unsigned long)f, arg) < 0) {
702 arg->stop = 1;
703 break;
704 }
705skip:
706 arg->count++;
707 }
708}
709
710static struct tcf_proto_ops cls_flow_ops __read_mostly = {
711 .kind = "flow",
712 .classify = flow_classify,
713 .init = flow_init,
714 .destroy = flow_destroy,
715 .change = flow_change,
716 .delete = flow_delete,
717 .get = flow_get,
718 .put = flow_put,
719 .dump = flow_dump,
720 .walk = flow_walk,
721 .owner = THIS_MODULE,
722};
723
724static int __init cls_flow_init(void)
725{
726 return register_tcf_proto_ops(&cls_flow_ops);
727}
728
729static void __exit cls_flow_exit(void)
730{
731 unregister_tcf_proto_ops(&cls_flow_ops);
732}
733
734module_init(cls_flow_init);
735module_exit(cls_flow_exit);
736
737MODULE_LICENSE("GPL");
738MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
739MODULE_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");