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 IS_ENABLED(CONFIG_NF_CONNTRACK)
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 union {
61 struct work_struct work;
62 struct rcu_head rcu;
63 };
64};
65
66static inline u32 addr_fold(void *addr)
67{
68 unsigned long a = (unsigned long)addr;
69
70 return (a & 0xFFFFFFFF) ^ (BITS_PER_LONG > 32 ? a >> 32 : 0);
71}
72
73static u32 flow_get_src(const struct sk_buff *skb, const struct flow_keys *flow)
74{
75 __be32 src = flow_get_u32_src(flow);
76
77 if (src)
78 return ntohl(src);
79
80 return addr_fold(skb->sk);
81}
82
83static u32 flow_get_dst(const struct sk_buff *skb, const struct flow_keys *flow)
84{
85 __be32 dst = flow_get_u32_dst(flow);
86
87 if (dst)
88 return ntohl(dst);
89
90 return addr_fold(skb_dst(skb)) ^ (__force u16) tc_skb_protocol(skb);
91}
92
93static u32 flow_get_proto(const struct sk_buff *skb,
94 const struct flow_keys *flow)
95{
96 return flow->basic.ip_proto;
97}
98
99static u32 flow_get_proto_src(const struct sk_buff *skb,
100 const struct flow_keys *flow)
101{
102 if (flow->ports.ports)
103 return ntohs(flow->ports.src);
104
105 return addr_fold(skb->sk);
106}
107
108static u32 flow_get_proto_dst(const struct sk_buff *skb,
109 const struct flow_keys *flow)
110{
111 if (flow->ports.ports)
112 return ntohs(flow->ports.dst);
113
114 return addr_fold(skb_dst(skb)) ^ (__force u16) tc_skb_protocol(skb);
115}
116
117static u32 flow_get_iif(const struct sk_buff *skb)
118{
119 return skb->skb_iif;
120}
121
122static u32 flow_get_priority(const struct sk_buff *skb)
123{
124 return skb->priority;
125}
126
127static u32 flow_get_mark(const struct sk_buff *skb)
128{
129 return skb->mark;
130}
131
132static u32 flow_get_nfct(const struct sk_buff *skb)
133{
134#if IS_ENABLED(CONFIG_NF_CONNTRACK)
135 return addr_fold(skb_nfct(skb));
136#else
137 return 0;
138#endif
139}
140
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; \
149})
150#else
151#define CTTUPLE(skb, member) \
152({ \
153 goto fallback; \
154 0; \
155})
156#endif
157
158static u32 flow_get_nfct_src(const struct sk_buff *skb,
159 const struct flow_keys *flow)
160{
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]));
166 }
167fallback:
168 return flow_get_src(skb, flow);
169}
170
171static u32 flow_get_nfct_dst(const struct sk_buff *skb,
172 const struct flow_keys *flow)
173{
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]));
179 }
180fallback:
181 return flow_get_dst(skb, flow);
182}
183
184static u32 flow_get_nfct_proto_src(const struct sk_buff *skb,
185 const struct flow_keys *flow)
186{
187 return ntohs(CTTUPLE(skb, src.u.all));
188fallback:
189 return flow_get_proto_src(skb, flow);
190}
191
192static u32 flow_get_nfct_proto_dst(const struct sk_buff *skb,
193 const struct flow_keys *flow)
194{
195 return ntohs(CTTUPLE(skb, dst.u.all));
196fallback:
197 return flow_get_proto_dst(skb, flow);
198}
199
200static u32 flow_get_rtclassid(const struct sk_buff *skb)
201{
202#ifdef CONFIG_IP_ROUTE_CLASSID
203 if (skb_dst(skb))
204 return skb_dst(skb)->tclassid;
205#endif
206 return 0;
207}
208
209static u32 flow_get_skuid(const struct sk_buff *skb)
210{
211 struct sock *sk = skb_to_full_sk(skb);
212
213 if (sk && sk->sk_socket && sk->sk_socket->file) {
214 kuid_t skuid = sk->sk_socket->file->f_cred->fsuid;
215
216 return from_kuid(&init_user_ns, skuid);
217 }
218 return 0;
219}
220
221static u32 flow_get_skgid(const struct sk_buff *skb)
222{
223 struct sock *sk = skb_to_full_sk(skb);
224
225 if (sk && sk->sk_socket && sk->sk_socket->file) {
226 kgid_t skgid = sk->sk_socket->file->f_cred->fsgid;
227
228 return from_kgid(&init_user_ns, skgid);
229 }
230 return 0;
231}
232
233static u32 flow_get_vlan_tag(const struct sk_buff *skb)
234{
235 u16 uninitialized_var(tag);
236
237 if (vlan_get_tag(skb, &tag) < 0)
238 return 0;
239 return tag & VLAN_VID_MASK;
240}
241
242static u32 flow_get_rxhash(struct sk_buff *skb)
243{
244 return skb_get_hash(skb);
245}
246
247static u32 flow_key_get(struct sk_buff *skb, int key, struct flow_keys *flow)
248{
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;
289 }
290}
291
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))
301
302static int flow_classify(struct sk_buff *skb, const struct tcf_proto *tp,
303 struct tcf_result *res)
304{
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;
311
312 list_for_each_entry_rcu(f, &head->filters, list) {
313 u32 keys[FLOW_KEY_MAX + 1];
314 struct flow_keys flow_keys;
315
316 if (!tcf_em_tree_match(skb, &f->ematches, NULL))
317 continue;
318
319 keymask = f->keymask;
320 if (keymask & FLOW_KEYS_NEEDED)
321 skb_flow_dissect_flow_keys(skb, &flow_keys, 0);
322
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);
327 }
328
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;
335 }
336
337 if (f->divisor)
338 classid %= f->divisor;
339
340 res->class = 0;
341 res->classid = TC_H_MAKE(f->baseclass, f->baseclass + classid);
342
343 r = tcf_exts_exec(skb, &f->exts, res);
344 if (r < 0)
345 continue;
346 return r;
347 }
348 return -1;
349}
350
351static void flow_perturbation(struct timer_list *t)
352{
353 struct flow_filter *f = from_timer(f, t, perturb_timer);
354
355 get_random_bytes(&f->hashrnd, 4);
356 if (f->perturb_period)
357 mod_timer(&f->perturb_timer, jiffies + f->perturb_period);
358}
359
360static 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 },
373};
374
375static void __flow_destroy_filter(struct flow_filter *f)
376{
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);
382}
383
384static void flow_destroy_filter_work(struct work_struct *work)
385{
386 struct flow_filter *f = container_of(work, struct flow_filter, work);
387
388 rtnl_lock();
389 __flow_destroy_filter(f);
390 rtnl_unlock();
391}
392
393static void flow_destroy_filter(struct rcu_head *head)
394{
395 struct flow_filter *f = container_of(head, struct flow_filter, rcu);
396
397 INIT_WORK(&f->work, flow_destroy_filter_work);
398 tcf_queue_work(&f->work);
399}
400
401static 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)
405{
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;
416
417 if (opt == NULL)
418 return -EINVAL;
419
420 err = nla_parse_nested(tb, TCA_FLOW_MAX, opt, flow_policy, NULL);
421 if (err < 0)
422 return err;
423
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;
428 }
429
430 if (tb[TCA_FLOW_KEYS]) {
431 keymask = nla_get_u32(tb[TCA_FLOW_KEYS]);
432
433 nkeys = hweight32(keymask);
434 if (nkeys == 0)
435 return -EINVAL;
436
437 if (fls(keymask) - 1 > FLOW_KEY_MAX)
438 return -EOPNOTSUPP;
439
440 if ((keymask & (FLOW_KEY_SKUID|FLOW_KEY_SKGID)) &&
441 sk_user_ns(NETLINK_CB(in_skb).sk) != &init_user_ns)
442 return -EOPNOTSUPP;
443 }
444
445 fnew = kzalloc(sizeof(*fnew), GFP_KERNEL);
446 if (!fnew)
447 return -ENOBUFS;
448
449 err = tcf_em_tree_validate(tp, tb[TCA_FLOW_EMATCHES], &fnew->ematches);
450 if (err < 0)
451 goto err1;
452
453 err = tcf_exts_init(&fnew->exts, TCA_FLOW_ACT, TCA_FLOW_POLICE);
454 if (err < 0)
455 goto err2;
456
457 err = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &fnew->exts, ovr,
458 extack);
459 if (err < 0)
460 goto err2;
461
462 fold = *arg;
463 if (fold) {
464 err = -EINVAL;
465 if (fold->handle != handle && handle)
466 goto err2;
467
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;
481
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;
487
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;
494 }
495 } else {
496 err = -EINVAL;
497 if (!handle)
498 goto err2;
499 if (!tb[TCA_FLOW_KEYS])
500 goto err2;
501
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;
507
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;
512 }
513
514 if (TC_H_MAJ(baseclass) == 0) {
515 struct Qdisc *q = tcf_block_q(tp->chain->block);
516
517 baseclass = TC_H_MAKE(q->handle, baseclass);
518 }
519 if (TC_H_MIN(baseclass) == 0)
520 baseclass = TC_H_MAKE(baseclass, 1);
521
522 fnew->handle = handle;
523 fnew->mask = ~0U;
524 fnew->tp = tp;
525 get_random_bytes(&fnew->hashrnd, 4);
526 }
527
528 timer_setup(&fnew->perturb_timer, flow_perturbation, TIMER_DEFERRABLE);
529
530 tcf_block_netif_keep_dst(tp->chain->block);
531
532 if (tb[TCA_FLOW_KEYS]) {
533 fnew->keymask = keymask;
534 fnew->nkeys = nkeys;
535 }
536
537 fnew->mode = mode;
538
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]);
547
548 if (tb[TCA_FLOW_DIVISOR])
549 fnew->divisor = nla_get_u32(tb[TCA_FLOW_DIVISOR]);
550 if (baseclass)
551 fnew->baseclass = baseclass;
552
553 fnew->perturb_period = perturb_period;
554 if (perturb_period)
555 mod_timer(&fnew->perturb_timer, jiffies + perturb_period);
556
557 if (!*arg)
558 list_add_tail_rcu(&fnew->list, &head->filters);
559 else
560 list_replace_rcu(&fold->list, &fnew->list);
561
562 *arg = fnew;
563
564 if (fold) {
565 tcf_exts_get_net(&fold->exts);
566 call_rcu(&fold->rcu, flow_destroy_filter);
567 }
568 return 0;
569
570err2:
571 tcf_exts_destroy(&fnew->exts);
572 tcf_em_tree_destroy(&fnew->ematches);
573err1:
574 kfree(fnew);
575 return err;
576}
577
578static int flow_delete(struct tcf_proto *tp, void *arg, bool *last,
579 struct netlink_ext_ack *extack)
580{
581 struct flow_head *head = rtnl_dereference(tp->root);
582 struct flow_filter *f = arg;
583
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;
589}
590
591static int flow_init(struct tcf_proto *tp)
592{
593 struct flow_head *head;
594
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;
601}
602
603static void flow_destroy(struct tcf_proto *tp, struct netlink_ext_ack *extack)
604{
605 struct flow_head *head = rtnl_dereference(tp->root);
606 struct flow_filter *f, *next;
607
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);
614 }
615 kfree_rcu(head, rcu);
616}
617
618static void *flow_get(struct tcf_proto *tp, u32 handle)
619{
620 struct flow_head *head = rtnl_dereference(tp->root);
621 struct flow_filter *f;
622
623 list_for_each_entry(f, &head->filters, list)
624 if (f->handle == handle)
625 return f;
626 return NULL;
627}
628
629static int flow_dump(struct net *net, struct tcf_proto *tp, void *fh,
630 struct sk_buff *skb, struct tcmsg *t)
631{
632 struct flow_filter *f = fh;
633 struct nlattr *nest;
634
635 if (f == NULL)
636 return skb->len;
637
638 t->tcm_handle = f->handle;
639
640 nest = nla_nest_start(skb, TCA_OPTIONS);
641 if (nest == NULL)
642 goto nla_put_failure;
643
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;
647
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;
652 }
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;
659
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;
666
667 if (f->perturb_period &&
668 nla_put_u32(skb, TCA_FLOW_PERTURB, f->perturb_period / HZ))
669 goto nla_put_failure;
670
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);
679
680 if (tcf_exts_dump_stats(skb, &f->exts) < 0)
681 goto nla_put_failure;
682
683 return skb->len;
684
685nla_put_failure:
686 nla_nest_cancel(skb, nest);
687 return -1;
688}
689
690static void flow_walk(struct tcf_proto *tp, struct tcf_walker *arg)
691{
692 struct flow_head *head = rtnl_dereference(tp->root);
693 struct flow_filter *f;
694
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;
701 }
702skip:
703 arg->count++;
704 }
705}
706
707static 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,
718};
719
720static int __init cls_flow_init(void)
721{
722 return register_tcf_proto_ops(&cls_flow_ops);
723}
724
725static void __exit cls_flow_exit(void)
726{
727 unregister_tcf_proto_ops(&cls_flow_ops);
728}
729
730module_init(cls_flow_init);
731module_exit(cls_flow_exit);
732
733MODULE_LICENSE("GPL");
734MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
735MODULE_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 const struct tcf_ext_map flow_ext_map = {
60 .action = TCA_FLOW_ACT,
61 .police = TCA_FLOW_POLICE,
62};
63
64static inline u32 addr_fold(void *addr)
65{
66 unsigned long a = (unsigned long)addr;
67
68 return (a & 0xFFFFFFFF) ^ (BITS_PER_LONG > 32 ? a >> 32 : 0);
69}
70
71static u32 flow_get_src(const struct sk_buff *skb, const struct flow_keys *flow)
72{
73 if (flow->src)
74 return ntohl(flow->src);
75 return addr_fold(skb->sk);
76}
77
78static u32 flow_get_dst(const struct sk_buff *skb, const struct flow_keys *flow)
79{
80 if (flow->dst)
81 return ntohl(flow->dst);
82 return addr_fold(skb_dst(skb)) ^ (__force u16)skb->protocol;
83}
84
85static u32 flow_get_proto(const struct sk_buff *skb, const struct flow_keys *flow)
86{
87 return flow->ip_proto;
88}
89
90static u32 flow_get_proto_src(const struct sk_buff *skb, const struct flow_keys *flow)
91{
92 if (flow->ports)
93 return ntohs(flow->port16[0]);
94
95 return addr_fold(skb->sk);
96}
97
98static u32 flow_get_proto_dst(const struct sk_buff *skb, const struct flow_keys *flow)
99{
100 if (flow->ports)
101 return ntohs(flow->port16[1]);
102
103 return addr_fold(skb_dst(skb)) ^ (__force u16)skb->protocol;
104}
105
106static u32 flow_get_iif(const struct sk_buff *skb)
107{
108 return skb->skb_iif;
109}
110
111static u32 flow_get_priority(const struct sk_buff *skb)
112{
113 return skb->priority;
114}
115
116static u32 flow_get_mark(const struct sk_buff *skb)
117{
118 return skb->mark;
119}
120
121static u32 flow_get_nfct(const struct sk_buff *skb)
122{
123#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
124 return addr_fold(skb->nfct);
125#else
126 return 0;
127#endif
128}
129
130#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
131#define CTTUPLE(skb, member) \
132({ \
133 enum ip_conntrack_info ctinfo; \
134 const struct nf_conn *ct = nf_ct_get(skb, &ctinfo); \
135 if (ct == NULL) \
136 goto fallback; \
137 ct->tuplehash[CTINFO2DIR(ctinfo)].tuple.member; \
138})
139#else
140#define CTTUPLE(skb, member) \
141({ \
142 goto fallback; \
143 0; \
144})
145#endif
146
147static u32 flow_get_nfct_src(const struct sk_buff *skb, const struct flow_keys *flow)
148{
149 switch (skb->protocol) {
150 case htons(ETH_P_IP):
151 return ntohl(CTTUPLE(skb, src.u3.ip));
152 case htons(ETH_P_IPV6):
153 return ntohl(CTTUPLE(skb, src.u3.ip6[3]));
154 }
155fallback:
156 return flow_get_src(skb, flow);
157}
158
159static u32 flow_get_nfct_dst(const struct sk_buff *skb, const struct flow_keys *flow)
160{
161 switch (skb->protocol) {
162 case htons(ETH_P_IP):
163 return ntohl(CTTUPLE(skb, dst.u3.ip));
164 case htons(ETH_P_IPV6):
165 return ntohl(CTTUPLE(skb, dst.u3.ip6[3]));
166 }
167fallback:
168 return flow_get_dst(skb, flow);
169}
170
171static u32 flow_get_nfct_proto_src(const struct sk_buff *skb, const struct flow_keys *flow)
172{
173 return ntohs(CTTUPLE(skb, src.u.all));
174fallback:
175 return flow_get_proto_src(skb, flow);
176}
177
178static u32 flow_get_nfct_proto_dst(const struct sk_buff *skb, const struct flow_keys *flow)
179{
180 return ntohs(CTTUPLE(skb, dst.u.all));
181fallback:
182 return flow_get_proto_dst(skb, flow);
183}
184
185static u32 flow_get_rtclassid(const struct sk_buff *skb)
186{
187#ifdef CONFIG_IP_ROUTE_CLASSID
188 if (skb_dst(skb))
189 return skb_dst(skb)->tclassid;
190#endif
191 return 0;
192}
193
194static u32 flow_get_skuid(const struct sk_buff *skb)
195{
196 if (skb->sk && skb->sk->sk_socket && skb->sk->sk_socket->file)
197 return skb->sk->sk_socket->file->f_cred->fsuid;
198 return 0;
199}
200
201static u32 flow_get_skgid(const struct sk_buff *skb)
202{
203 if (skb->sk && skb->sk->sk_socket && skb->sk->sk_socket->file)
204 return skb->sk->sk_socket->file->f_cred->fsgid;
205 return 0;
206}
207
208static u32 flow_get_vlan_tag(const struct sk_buff *skb)
209{
210 u16 uninitialized_var(tag);
211
212 if (vlan_get_tag(skb, &tag) < 0)
213 return 0;
214 return tag & VLAN_VID_MASK;
215}
216
217static u32 flow_get_rxhash(struct sk_buff *skb)
218{
219 return skb_get_rxhash(skb);
220}
221
222static u32 flow_key_get(struct sk_buff *skb, int key, struct flow_keys *flow)
223{
224 switch (key) {
225 case FLOW_KEY_SRC:
226 return flow_get_src(skb, flow);
227 case FLOW_KEY_DST:
228 return flow_get_dst(skb, flow);
229 case FLOW_KEY_PROTO:
230 return flow_get_proto(skb, flow);
231 case FLOW_KEY_PROTO_SRC:
232 return flow_get_proto_src(skb, flow);
233 case FLOW_KEY_PROTO_DST:
234 return flow_get_proto_dst(skb, flow);
235 case FLOW_KEY_IIF:
236 return flow_get_iif(skb);
237 case FLOW_KEY_PRIORITY:
238 return flow_get_priority(skb);
239 case FLOW_KEY_MARK:
240 return flow_get_mark(skb);
241 case FLOW_KEY_NFCT:
242 return flow_get_nfct(skb);
243 case FLOW_KEY_NFCT_SRC:
244 return flow_get_nfct_src(skb, flow);
245 case FLOW_KEY_NFCT_DST:
246 return flow_get_nfct_dst(skb, flow);
247 case FLOW_KEY_NFCT_PROTO_SRC:
248 return flow_get_nfct_proto_src(skb, flow);
249 case FLOW_KEY_NFCT_PROTO_DST:
250 return flow_get_nfct_proto_dst(skb, flow);
251 case FLOW_KEY_RTCLASSID:
252 return flow_get_rtclassid(skb);
253 case FLOW_KEY_SKUID:
254 return flow_get_skuid(skb);
255 case FLOW_KEY_SKGID:
256 return flow_get_skgid(skb);
257 case FLOW_KEY_VLAN_TAG:
258 return flow_get_vlan_tag(skb);
259 case FLOW_KEY_RXHASH:
260 return flow_get_rxhash(skb);
261 default:
262 WARN_ON(1);
263 return 0;
264 }
265}
266
267#define FLOW_KEYS_NEEDED ((1 << FLOW_KEY_SRC) | \
268 (1 << FLOW_KEY_DST) | \
269 (1 << FLOW_KEY_PROTO) | \
270 (1 << FLOW_KEY_PROTO_SRC) | \
271 (1 << FLOW_KEY_PROTO_DST) | \
272 (1 << FLOW_KEY_NFCT_SRC) | \
273 (1 << FLOW_KEY_NFCT_DST) | \
274 (1 << FLOW_KEY_NFCT_PROTO_SRC) | \
275 (1 << FLOW_KEY_NFCT_PROTO_DST))
276
277static int flow_classify(struct sk_buff *skb, const struct tcf_proto *tp,
278 struct tcf_result *res)
279{
280 struct flow_head *head = tp->root;
281 struct flow_filter *f;
282 u32 keymask;
283 u32 classid;
284 unsigned int n, key;
285 int r;
286
287 list_for_each_entry(f, &head->filters, list) {
288 u32 keys[FLOW_KEY_MAX + 1];
289 struct flow_keys flow_keys;
290
291 if (!tcf_em_tree_match(skb, &f->ematches, NULL))
292 continue;
293
294 keymask = f->keymask;
295 if (keymask & FLOW_KEYS_NEEDED)
296 skb_flow_dissect(skb, &flow_keys);
297
298 for (n = 0; n < f->nkeys; n++) {
299 key = ffs(keymask) - 1;
300 keymask &= ~(1 << key);
301 keys[n] = flow_key_get(skb, key, &flow_keys);
302 }
303
304 if (f->mode == FLOW_MODE_HASH)
305 classid = jhash2(keys, f->nkeys, f->hashrnd);
306 else {
307 classid = keys[0];
308 classid = (classid & f->mask) ^ f->xor;
309 classid = (classid >> f->rshift) + f->addend;
310 }
311
312 if (f->divisor)
313 classid %= f->divisor;
314
315 res->class = 0;
316 res->classid = TC_H_MAKE(f->baseclass, f->baseclass + classid);
317
318 r = tcf_exts_exec(skb, &f->exts, res);
319 if (r < 0)
320 continue;
321 return r;
322 }
323 return -1;
324}
325
326static void flow_perturbation(unsigned long arg)
327{
328 struct flow_filter *f = (struct flow_filter *)arg;
329
330 get_random_bytes(&f->hashrnd, 4);
331 if (f->perturb_period)
332 mod_timer(&f->perturb_timer, jiffies + f->perturb_period);
333}
334
335static const struct nla_policy flow_policy[TCA_FLOW_MAX + 1] = {
336 [TCA_FLOW_KEYS] = { .type = NLA_U32 },
337 [TCA_FLOW_MODE] = { .type = NLA_U32 },
338 [TCA_FLOW_BASECLASS] = { .type = NLA_U32 },
339 [TCA_FLOW_RSHIFT] = { .type = NLA_U32 },
340 [TCA_FLOW_ADDEND] = { .type = NLA_U32 },
341 [TCA_FLOW_MASK] = { .type = NLA_U32 },
342 [TCA_FLOW_XOR] = { .type = NLA_U32 },
343 [TCA_FLOW_DIVISOR] = { .type = NLA_U32 },
344 [TCA_FLOW_ACT] = { .type = NLA_NESTED },
345 [TCA_FLOW_POLICE] = { .type = NLA_NESTED },
346 [TCA_FLOW_EMATCHES] = { .type = NLA_NESTED },
347 [TCA_FLOW_PERTURB] = { .type = NLA_U32 },
348};
349
350static int flow_change(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
391 err = tcf_exts_validate(tp, tb, tca[TCA_RATE], &e, &flow_ext_map);
392 if (err < 0)
393 return err;
394
395 err = tcf_em_tree_validate(tp, tb[TCA_FLOW_EMATCHES], &t);
396 if (err < 0)
397 goto err1;
398
399 f = (struct flow_filter *)*arg;
400 if (f != NULL) {
401 err = -EINVAL;
402 if (f->handle != handle && handle)
403 goto err2;
404
405 mode = f->mode;
406 if (tb[TCA_FLOW_MODE])
407 mode = nla_get_u32(tb[TCA_FLOW_MODE]);
408 if (mode != FLOW_MODE_HASH && nkeys > 1)
409 goto err2;
410
411 if (mode == FLOW_MODE_HASH)
412 perturb_period = f->perturb_period;
413 if (tb[TCA_FLOW_PERTURB]) {
414 if (mode != FLOW_MODE_HASH)
415 goto err2;
416 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
417 }
418 } else {
419 err = -EINVAL;
420 if (!handle)
421 goto err2;
422 if (!tb[TCA_FLOW_KEYS])
423 goto err2;
424
425 mode = FLOW_MODE_MAP;
426 if (tb[TCA_FLOW_MODE])
427 mode = nla_get_u32(tb[TCA_FLOW_MODE]);
428 if (mode != FLOW_MODE_HASH && nkeys > 1)
429 goto err2;
430
431 if (tb[TCA_FLOW_PERTURB]) {
432 if (mode != FLOW_MODE_HASH)
433 goto err2;
434 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
435 }
436
437 if (TC_H_MAJ(baseclass) == 0)
438 baseclass = TC_H_MAKE(tp->q->handle, baseclass);
439 if (TC_H_MIN(baseclass) == 0)
440 baseclass = TC_H_MAKE(baseclass, 1);
441
442 err = -ENOBUFS;
443 f = kzalloc(sizeof(*f), GFP_KERNEL);
444 if (f == NULL)
445 goto err2;
446
447 f->handle = handle;
448 f->mask = ~0U;
449
450 get_random_bytes(&f->hashrnd, 4);
451 f->perturb_timer.function = flow_perturbation;
452 f->perturb_timer.data = (unsigned long)f;
453 init_timer_deferrable(&f->perturb_timer);
454 }
455
456 tcf_exts_change(tp, &f->exts, &e);
457 tcf_em_tree_change(tp, &f->ematches, &t);
458
459 tcf_tree_lock(tp);
460
461 if (tb[TCA_FLOW_KEYS]) {
462 f->keymask = keymask;
463 f->nkeys = nkeys;
464 }
465
466 f->mode = mode;
467
468 if (tb[TCA_FLOW_MASK])
469 f->mask = nla_get_u32(tb[TCA_FLOW_MASK]);
470 if (tb[TCA_FLOW_XOR])
471 f->xor = nla_get_u32(tb[TCA_FLOW_XOR]);
472 if (tb[TCA_FLOW_RSHIFT])
473 f->rshift = nla_get_u32(tb[TCA_FLOW_RSHIFT]);
474 if (tb[TCA_FLOW_ADDEND])
475 f->addend = nla_get_u32(tb[TCA_FLOW_ADDEND]);
476
477 if (tb[TCA_FLOW_DIVISOR])
478 f->divisor = nla_get_u32(tb[TCA_FLOW_DIVISOR]);
479 if (baseclass)
480 f->baseclass = baseclass;
481
482 f->perturb_period = perturb_period;
483 del_timer(&f->perturb_timer);
484 if (perturb_period)
485 mod_timer(&f->perturb_timer, jiffies + perturb_period);
486
487 if (*arg == 0)
488 list_add_tail(&f->list, &head->filters);
489
490 tcf_tree_unlock(tp);
491
492 *arg = (unsigned long)f;
493 return 0;
494
495err2:
496 tcf_em_tree_destroy(tp, &t);
497err1:
498 tcf_exts_destroy(tp, &e);
499 return err;
500}
501
502static void flow_destroy_filter(struct tcf_proto *tp, struct flow_filter *f)
503{
504 del_timer_sync(&f->perturb_timer);
505 tcf_exts_destroy(tp, &f->exts);
506 tcf_em_tree_destroy(tp, &f->ematches);
507 kfree(f);
508}
509
510static int flow_delete(struct tcf_proto *tp, unsigned long arg)
511{
512 struct flow_filter *f = (struct flow_filter *)arg;
513
514 tcf_tree_lock(tp);
515 list_del(&f->list);
516 tcf_tree_unlock(tp);
517 flow_destroy_filter(tp, f);
518 return 0;
519}
520
521static int flow_init(struct tcf_proto *tp)
522{
523 struct flow_head *head;
524
525 head = kzalloc(sizeof(*head), GFP_KERNEL);
526 if (head == NULL)
527 return -ENOBUFS;
528 INIT_LIST_HEAD(&head->filters);
529 tp->root = head;
530 return 0;
531}
532
533static void flow_destroy(struct tcf_proto *tp)
534{
535 struct flow_head *head = tp->root;
536 struct flow_filter *f, *next;
537
538 list_for_each_entry_safe(f, next, &head->filters, list) {
539 list_del(&f->list);
540 flow_destroy_filter(tp, f);
541 }
542 kfree(head);
543}
544
545static unsigned long flow_get(struct tcf_proto *tp, u32 handle)
546{
547 struct flow_head *head = tp->root;
548 struct flow_filter *f;
549
550 list_for_each_entry(f, &head->filters, list)
551 if (f->handle == handle)
552 return (unsigned long)f;
553 return 0;
554}
555
556static void flow_put(struct tcf_proto *tp, unsigned long f)
557{
558}
559
560static int flow_dump(struct tcf_proto *tp, unsigned long fh,
561 struct sk_buff *skb, struct tcmsg *t)
562{
563 struct flow_filter *f = (struct flow_filter *)fh;
564 struct nlattr *nest;
565
566 if (f == NULL)
567 return skb->len;
568
569 t->tcm_handle = f->handle;
570
571 nest = nla_nest_start(skb, TCA_OPTIONS);
572 if (nest == NULL)
573 goto nla_put_failure;
574
575 if (nla_put_u32(skb, TCA_FLOW_KEYS, f->keymask) ||
576 nla_put_u32(skb, TCA_FLOW_MODE, f->mode))
577 goto nla_put_failure;
578
579 if (f->mask != ~0 || f->xor != 0) {
580 if (nla_put_u32(skb, TCA_FLOW_MASK, f->mask) ||
581 nla_put_u32(skb, TCA_FLOW_XOR, f->xor))
582 goto nla_put_failure;
583 }
584 if (f->rshift &&
585 nla_put_u32(skb, TCA_FLOW_RSHIFT, f->rshift))
586 goto nla_put_failure;
587 if (f->addend &&
588 nla_put_u32(skb, TCA_FLOW_ADDEND, f->addend))
589 goto nla_put_failure;
590
591 if (f->divisor &&
592 nla_put_u32(skb, TCA_FLOW_DIVISOR, f->divisor))
593 goto nla_put_failure;
594 if (f->baseclass &&
595 nla_put_u32(skb, TCA_FLOW_BASECLASS, f->baseclass))
596 goto nla_put_failure;
597
598 if (f->perturb_period &&
599 nla_put_u32(skb, TCA_FLOW_PERTURB, f->perturb_period / HZ))
600 goto nla_put_failure;
601
602 if (tcf_exts_dump(skb, &f->exts, &flow_ext_map) < 0)
603 goto nla_put_failure;
604#ifdef CONFIG_NET_EMATCH
605 if (f->ematches.hdr.nmatches &&
606 tcf_em_tree_dump(skb, &f->ematches, TCA_FLOW_EMATCHES) < 0)
607 goto nla_put_failure;
608#endif
609 nla_nest_end(skb, nest);
610
611 if (tcf_exts_dump_stats(skb, &f->exts, &flow_ext_map) < 0)
612 goto nla_put_failure;
613
614 return skb->len;
615
616nla_put_failure:
617 nlmsg_trim(skb, nest);
618 return -1;
619}
620
621static void flow_walk(struct tcf_proto *tp, struct tcf_walker *arg)
622{
623 struct flow_head *head = tp->root;
624 struct flow_filter *f;
625
626 list_for_each_entry(f, &head->filters, list) {
627 if (arg->count < arg->skip)
628 goto skip;
629 if (arg->fn(tp, (unsigned long)f, arg) < 0) {
630 arg->stop = 1;
631 break;
632 }
633skip:
634 arg->count++;
635 }
636}
637
638static struct tcf_proto_ops cls_flow_ops __read_mostly = {
639 .kind = "flow",
640 .classify = flow_classify,
641 .init = flow_init,
642 .destroy = flow_destroy,
643 .change = flow_change,
644 .delete = flow_delete,
645 .get = flow_get,
646 .put = flow_put,
647 .dump = flow_dump,
648 .walk = flow_walk,
649 .owner = THIS_MODULE,
650};
651
652static int __init cls_flow_init(void)
653{
654 return register_tcf_proto_ops(&cls_flow_ops);
655}
656
657static void __exit cls_flow_exit(void)
658{
659 unregister_tcf_proto_ops(&cls_flow_ops);
660}
661
662module_init(cls_flow_init);
663module_exit(cls_flow_exit);
664
665MODULE_LICENSE("GPL");
666MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
667MODULE_DESCRIPTION("TC flow classifier");