1
2
3
4
5
6
7
8
9
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 struct rcu_head rcu;
38};
39
40struct flow_filter {
41 struct list_head list;
42 struct tcf_exts exts;
43 struct tcf_ematch_tree ematches;
44 struct tcf_proto *tp;
45 struct timer_list perturb_timer;
46 u32 perturb_period;
47 u32 handle;
48
49 u32 nkeys;
50 u32 keymask;
51 u32 mode;
52 u32 mask;
53 u32 xor;
54 u32 rshift;
55 u32 addend;
56 u32 divisor;
57 u32 baseclass;
58 u32 hashrnd;
59 struct rcu_head rcu;
60};
61
62static inline u32 addr_fold(void *addr)
63{
64 unsigned long a = (unsigned long)addr;
65
66 return (a & 0xFFFFFFFF) ^ (BITS_PER_LONG > 32 ? a >> 32 : 0);
67}
68
69static u32 flow_get_src(const struct sk_buff *skb, const struct flow_keys *flow)
70{
71 if (flow->src)
72 return ntohl(flow->src);
73 return addr_fold(skb->sk);
74}
75
76static u32 flow_get_dst(const struct sk_buff *skb, const struct flow_keys *flow)
77{
78 if (flow->dst)
79 return ntohl(flow->dst);
80 return addr_fold(skb_dst(skb)) ^ (__force u16) tc_skb_protocol(skb);
81}
82
83static u32 flow_get_proto(const struct sk_buff *skb, const struct flow_keys *flow)
84{
85 return flow->ip_proto;
86}
87
88static u32 flow_get_proto_src(const struct sk_buff *skb, const struct flow_keys *flow)
89{
90 if (flow->ports)
91 return ntohs(flow->port16[0]);
92
93 return addr_fold(skb->sk);
94}
95
96static u32 flow_get_proto_dst(const struct sk_buff *skb, const struct flow_keys *flow)
97{
98 if (flow->ports)
99 return ntohs(flow->port16[1]);
100
101 return addr_fold(skb_dst(skb)) ^ (__force u16) tc_skb_protocol(skb);
102}
103
104static u32 flow_get_iif(const struct sk_buff *skb)
105{
106 return skb->skb_iif;
107}
108
109static u32 flow_get_priority(const struct sk_buff *skb)
110{
111 return skb->priority;
112}
113
114static u32 flow_get_mark(const struct sk_buff *skb)
115{
116 return skb->mark;
117}
118
119static u32 flow_get_nfct(const struct sk_buff *skb)
120{
121#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
122 return addr_fold(skb->nfct);
123#else
124 return 0;
125#endif
126}
127
128#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
129#define CTTUPLE(skb, member) \
130({ \
131 enum ip_conntrack_info ctinfo; \
132 const struct nf_conn *ct = nf_ct_get(skb, &ctinfo); \
133 if (ct == NULL) \
134 goto fallback; \
135 ct->tuplehash[CTINFO2DIR(ctinfo)].tuple.member; \
136})
137#else
138#define CTTUPLE(skb, member) \
139({ \
140 goto fallback; \
141 0; \
142})
143#endif
144
145static u32 flow_get_nfct_src(const struct sk_buff *skb, const struct flow_keys *flow)
146{
147 switch (tc_skb_protocol(skb)) {
148 case htons(ETH_P_IP):
149 return ntohl(CTTUPLE(skb, src.u3.ip));
150 case htons(ETH_P_IPV6):
151 return ntohl(CTTUPLE(skb, src.u3.ip6[3]));
152 }
153fallback:
154 return flow_get_src(skb, flow);
155}
156
157static u32 flow_get_nfct_dst(const struct sk_buff *skb, const struct flow_keys *flow)
158{
159 switch (tc_skb_protocol(skb)) {
160 case htons(ETH_P_IP):
161 return ntohl(CTTUPLE(skb, dst.u3.ip));
162 case htons(ETH_P_IPV6):
163 return ntohl(CTTUPLE(skb, dst.u3.ip6[3]));
164 }
165fallback:
166 return flow_get_dst(skb, flow);
167}
168
169static u32 flow_get_nfct_proto_src(const struct sk_buff *skb, const struct flow_keys *flow)
170{
171 return ntohs(CTTUPLE(skb, src.u.all));
172fallback:
173 return flow_get_proto_src(skb, flow);
174}
175
176static u32 flow_get_nfct_proto_dst(const struct sk_buff *skb, const struct flow_keys *flow)
177{
178 return ntohs(CTTUPLE(skb, dst.u.all));
179fallback:
180 return flow_get_proto_dst(skb, flow);
181}
182
183static u32 flow_get_rtclassid(const struct sk_buff *skb)
184{
185#ifdef CONFIG_IP_ROUTE_CLASSID
186 if (skb_dst(skb))
187 return skb_dst(skb)->tclassid;
188#endif
189 return 0;
190}
191
192static u32 flow_get_skuid(const struct sk_buff *skb)
193{
194 if (skb->sk && skb->sk->sk_socket && skb->sk->sk_socket->file) {
195 kuid_t skuid = skb->sk->sk_socket->file->f_cred->fsuid;
196 return from_kuid(&init_user_ns, skuid);
197 }
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 kgid_t skgid = skb->sk->sk_socket->file->f_cred->fsgid;
205 return from_kgid(&init_user_ns, skgid);
206 }
207 return 0;
208}
209
210static u32 flow_get_vlan_tag(const struct sk_buff *skb)
211{
212 u16 uninitialized_var(tag);
213
214 if (vlan_get_tag(skb, &tag) < 0)
215 return 0;
216 return tag & VLAN_VID_MASK;
217}
218
219static u32 flow_get_rxhash(struct sk_buff *skb)
220{
221 return skb_get_hash(skb);
222}
223
224static u32 flow_key_get(struct sk_buff *skb, int key, struct flow_keys *flow)
225{
226 switch (key) {
227 case FLOW_KEY_SRC:
228 return flow_get_src(skb, flow);
229 case FLOW_KEY_DST:
230 return flow_get_dst(skb, flow);
231 case FLOW_KEY_PROTO:
232 return flow_get_proto(skb, flow);
233 case FLOW_KEY_PROTO_SRC:
234 return flow_get_proto_src(skb, flow);
235 case FLOW_KEY_PROTO_DST:
236 return flow_get_proto_dst(skb, flow);
237 case FLOW_KEY_IIF:
238 return flow_get_iif(skb);
239 case FLOW_KEY_PRIORITY:
240 return flow_get_priority(skb);
241 case FLOW_KEY_MARK:
242 return flow_get_mark(skb);
243 case FLOW_KEY_NFCT:
244 return flow_get_nfct(skb);
245 case FLOW_KEY_NFCT_SRC:
246 return flow_get_nfct_src(skb, flow);
247 case FLOW_KEY_NFCT_DST:
248 return flow_get_nfct_dst(skb, flow);
249 case FLOW_KEY_NFCT_PROTO_SRC:
250 return flow_get_nfct_proto_src(skb, flow);
251 case FLOW_KEY_NFCT_PROTO_DST:
252 return flow_get_nfct_proto_dst(skb, flow);
253 case FLOW_KEY_RTCLASSID:
254 return flow_get_rtclassid(skb);
255 case FLOW_KEY_SKUID:
256 return flow_get_skuid(skb);
257 case FLOW_KEY_SKGID:
258 return flow_get_skgid(skb);
259 case FLOW_KEY_VLAN_TAG:
260 return flow_get_vlan_tag(skb);
261 case FLOW_KEY_RXHASH:
262 return flow_get_rxhash(skb);
263 default:
264 WARN_ON(1);
265 return 0;
266 }
267}
268
269#define FLOW_KEYS_NEEDED ((1 << FLOW_KEY_SRC) | \
270 (1 << FLOW_KEY_DST) | \
271 (1 << FLOW_KEY_PROTO) | \
272 (1 << FLOW_KEY_PROTO_SRC) | \
273 (1 << FLOW_KEY_PROTO_DST) | \
274 (1 << FLOW_KEY_NFCT_SRC) | \
275 (1 << FLOW_KEY_NFCT_DST) | \
276 (1 << FLOW_KEY_NFCT_PROTO_SRC) | \
277 (1 << FLOW_KEY_NFCT_PROTO_DST))
278
279static int flow_classify(struct sk_buff *skb, const struct tcf_proto *tp,
280 struct tcf_result *res)
281{
282 struct flow_head *head = rcu_dereference_bh(tp->root);
283 struct flow_filter *f;
284 u32 keymask;
285 u32 classid;
286 unsigned int n, key;
287 int r;
288
289 list_for_each_entry_rcu(f, &head->filters, list) {
290 u32 keys[FLOW_KEY_MAX + 1];
291 struct flow_keys flow_keys;
292
293 if (!tcf_em_tree_match(skb, &f->ematches, NULL))
294 continue;
295
296 keymask = f->keymask;
297 if (keymask & FLOW_KEYS_NEEDED)
298 skb_flow_dissect(skb, &flow_keys);
299
300 for (n = 0; n < f->nkeys; n++) {
301 key = ffs(keymask) - 1;
302 keymask &= ~(1 << key);
303 keys[n] = flow_key_get(skb, key, &flow_keys);
304 }
305
306 if (f->mode == FLOW_MODE_HASH)
307 classid = jhash2(keys, f->nkeys, f->hashrnd);
308 else {
309 classid = keys[0];
310 classid = (classid & f->mask) ^ f->xor;
311 classid = (classid >> f->rshift) + f->addend;
312 }
313
314 if (f->divisor)
315 classid %= f->divisor;
316
317 res->class = 0;
318 res->classid = TC_H_MAKE(f->baseclass, f->baseclass + classid);
319
320 r = tcf_exts_exec(skb, &f->exts, res);
321 if (r < 0)
322 continue;
323 return r;
324 }
325 return -1;
326}
327
328static void flow_perturbation(unsigned long arg)
329{
330 struct flow_filter *f = (struct flow_filter *)arg;
331
332 get_random_bytes(&f->hashrnd, 4);
333 if (f->perturb_period)
334 mod_timer(&f->perturb_timer, jiffies + f->perturb_period);
335}
336
337static const struct nla_policy flow_policy[TCA_FLOW_MAX + 1] = {
338 [TCA_FLOW_KEYS] = { .type = NLA_U32 },
339 [TCA_FLOW_MODE] = { .type = NLA_U32 },
340 [TCA_FLOW_BASECLASS] = { .type = NLA_U32 },
341 [TCA_FLOW_RSHIFT] = { .type = NLA_U32 },
342 [TCA_FLOW_ADDEND] = { .type = NLA_U32 },
343 [TCA_FLOW_MASK] = { .type = NLA_U32 },
344 [TCA_FLOW_XOR] = { .type = NLA_U32 },
345 [TCA_FLOW_DIVISOR] = { .type = NLA_U32 },
346 [TCA_FLOW_ACT] = { .type = NLA_NESTED },
347 [TCA_FLOW_POLICE] = { .type = NLA_NESTED },
348 [TCA_FLOW_EMATCHES] = { .type = NLA_NESTED },
349 [TCA_FLOW_PERTURB] = { .type = NLA_U32 },
350};
351
352static void flow_destroy_filter(struct rcu_head *head)
353{
354 struct flow_filter *f = container_of(head, struct flow_filter, rcu);
355
356 del_timer_sync(&f->perturb_timer);
357 tcf_exts_destroy(&f->exts);
358 tcf_em_tree_destroy(&f->ematches);
359 kfree(f);
360}
361
362static int flow_change(struct net *net, struct sk_buff *in_skb,
363 struct tcf_proto *tp, unsigned long base,
364 u32 handle, struct nlattr **tca,
365 unsigned long *arg, bool ovr)
366{
367 struct flow_head *head = rtnl_dereference(tp->root);
368 struct flow_filter *fold, *fnew;
369 struct nlattr *opt = tca[TCA_OPTIONS];
370 struct nlattr *tb[TCA_FLOW_MAX + 1];
371 struct tcf_exts e;
372 struct tcf_ematch_tree t;
373 unsigned int nkeys = 0;
374 unsigned int perturb_period = 0;
375 u32 baseclass = 0;
376 u32 keymask = 0;
377 u32 mode;
378 int err;
379
380 if (opt == NULL)
381 return -EINVAL;
382
383 err = nla_parse_nested(tb, TCA_FLOW_MAX, opt, flow_policy);
384 if (err < 0)
385 return err;
386
387 if (tb[TCA_FLOW_BASECLASS]) {
388 baseclass = nla_get_u32(tb[TCA_FLOW_BASECLASS]);
389 if (TC_H_MIN(baseclass) == 0)
390 return -EINVAL;
391 }
392
393 if (tb[TCA_FLOW_KEYS]) {
394 keymask = nla_get_u32(tb[TCA_FLOW_KEYS]);
395
396 nkeys = hweight32(keymask);
397 if (nkeys == 0)
398 return -EINVAL;
399
400 if (fls(keymask) - 1 > FLOW_KEY_MAX)
401 return -EOPNOTSUPP;
402
403 if ((keymask & (FLOW_KEY_SKUID|FLOW_KEY_SKGID)) &&
404 sk_user_ns(NETLINK_CB(in_skb).sk) != &init_user_ns)
405 return -EOPNOTSUPP;
406 }
407
408 tcf_exts_init(&e, TCA_FLOW_ACT, TCA_FLOW_POLICE);
409 err = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &e, ovr);
410 if (err < 0)
411 return err;
412
413 err = tcf_em_tree_validate(tp, tb[TCA_FLOW_EMATCHES], &t);
414 if (err < 0)
415 goto err1;
416
417 err = -ENOBUFS;
418 fnew = kzalloc(sizeof(*fnew), GFP_KERNEL);
419 if (!fnew)
420 goto err2;
421
422 fold = (struct flow_filter *)*arg;
423 if (fold) {
424 err = -EINVAL;
425 if (fold->handle != handle && handle)
426 goto err2;
427
428
429 fnew->tp = fold->tp;
430 fnew->handle = fold->handle;
431 fnew->nkeys = fold->nkeys;
432 fnew->keymask = fold->keymask;
433 fnew->mode = fold->mode;
434 fnew->mask = fold->mask;
435 fnew->xor = fold->xor;
436 fnew->rshift = fold->rshift;
437 fnew->addend = fold->addend;
438 fnew->divisor = fold->divisor;
439 fnew->baseclass = fold->baseclass;
440 fnew->hashrnd = fold->hashrnd;
441
442 mode = fold->mode;
443 if (tb[TCA_FLOW_MODE])
444 mode = nla_get_u32(tb[TCA_FLOW_MODE]);
445 if (mode != FLOW_MODE_HASH && nkeys > 1)
446 goto err2;
447
448 if (mode == FLOW_MODE_HASH)
449 perturb_period = fold->perturb_period;
450 if (tb[TCA_FLOW_PERTURB]) {
451 if (mode != FLOW_MODE_HASH)
452 goto err2;
453 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
454 }
455 } else {
456 err = -EINVAL;
457 if (!handle)
458 goto err2;
459 if (!tb[TCA_FLOW_KEYS])
460 goto err2;
461
462 mode = FLOW_MODE_MAP;
463 if (tb[TCA_FLOW_MODE])
464 mode = nla_get_u32(tb[TCA_FLOW_MODE]);
465 if (mode != FLOW_MODE_HASH && nkeys > 1)
466 goto err2;
467
468 if (tb[TCA_FLOW_PERTURB]) {
469 if (mode != FLOW_MODE_HASH)
470 goto err2;
471 perturb_period = nla_get_u32(tb[TCA_FLOW_PERTURB]) * HZ;
472 }
473
474 if (TC_H_MAJ(baseclass) == 0)
475 baseclass = TC_H_MAKE(tp->q->handle, baseclass);
476 if (TC_H_MIN(baseclass) == 0)
477 baseclass = TC_H_MAKE(baseclass, 1);
478
479 fnew->handle = handle;
480 fnew->mask = ~0U;
481 fnew->tp = tp;
482 get_random_bytes(&fnew->hashrnd, 4);
483 tcf_exts_init(&fnew->exts, TCA_FLOW_ACT, TCA_FLOW_POLICE);
484 }
485
486 fnew->perturb_timer.function = flow_perturbation;
487 fnew->perturb_timer.data = (unsigned long)fnew;
488 init_timer_deferrable(&fnew->perturb_timer);
489
490 tcf_exts_change(tp, &fnew->exts, &e);
491 tcf_em_tree_change(tp, &fnew->ematches, &t);
492
493 netif_keep_dst(qdisc_dev(tp->q));
494
495 if (tb[TCA_FLOW_KEYS]) {
496 fnew->keymask = keymask;
497 fnew->nkeys = nkeys;
498 }
499
500 fnew->mode = mode;
501
502 if (tb[TCA_FLOW_MASK])
503 fnew->mask = nla_get_u32(tb[TCA_FLOW_MASK]);
504 if (tb[TCA_FLOW_XOR])
505 fnew->xor = nla_get_u32(tb[TCA_FLOW_XOR]);
506 if (tb[TCA_FLOW_RSHIFT])
507 fnew->rshift = nla_get_u32(tb[TCA_FLOW_RSHIFT]);
508 if (tb[TCA_FLOW_ADDEND])
509 fnew->addend = nla_get_u32(tb[TCA_FLOW_ADDEND]);
510
511 if (tb[TCA_FLOW_DIVISOR])
512 fnew->divisor = nla_get_u32(tb[TCA_FLOW_DIVISOR]);
513 if (baseclass)
514 fnew->baseclass = baseclass;
515
516 fnew->perturb_period = perturb_period;
517 if (perturb_period)
518 mod_timer(&fnew->perturb_timer, jiffies + perturb_period);
519
520 if (*arg == 0)
521 list_add_tail_rcu(&fnew->list, &head->filters);
522 else
523 list_replace_rcu(&fnew->list, &fold->list);
524
525 *arg = (unsigned long)fnew;
526
527 if (fold)
528 call_rcu(&fold->rcu, flow_destroy_filter);
529 return 0;
530
531err2:
532 tcf_em_tree_destroy(&t);
533 kfree(fnew);
534err1:
535 tcf_exts_destroy(&e);
536 return err;
537}
538
539static int flow_delete(struct tcf_proto *tp, unsigned long arg)
540{
541 struct flow_filter *f = (struct flow_filter *)arg;
542
543 list_del_rcu(&f->list);
544 call_rcu(&f->rcu, flow_destroy_filter);
545 return 0;
546}
547
548static int flow_init(struct tcf_proto *tp)
549{
550 struct flow_head *head;
551
552 head = kzalloc(sizeof(*head), GFP_KERNEL);
553 if (head == NULL)
554 return -ENOBUFS;
555 INIT_LIST_HEAD(&head->filters);
556 rcu_assign_pointer(tp->root, head);
557 return 0;
558}
559
560static bool flow_destroy(struct tcf_proto *tp, bool force)
561{
562 struct flow_head *head = rtnl_dereference(tp->root);
563 struct flow_filter *f, *next;
564
565 if (!force && !list_empty(&head->filters))
566 return false;
567
568 list_for_each_entry_safe(f, next, &head->filters, list) {
569 list_del_rcu(&f->list);
570 call_rcu(&f->rcu, flow_destroy_filter);
571 }
572 RCU_INIT_POINTER(tp->root, NULL);
573 kfree_rcu(head, rcu);
574 return true;
575}
576
577static unsigned long flow_get(struct tcf_proto *tp, u32 handle)
578{
579 struct flow_head *head = rtnl_dereference(tp->root);
580 struct flow_filter *f;
581
582 list_for_each_entry(f, &head->filters, list)
583 if (f->handle == handle)
584 return (unsigned long)f;
585 return 0;
586}
587
588static int flow_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
589 struct sk_buff *skb, struct tcmsg *t)
590{
591 struct flow_filter *f = (struct flow_filter *)fh;
592 struct nlattr *nest;
593
594 if (f == NULL)
595 return skb->len;
596
597 t->tcm_handle = f->handle;
598
599 nest = nla_nest_start(skb, TCA_OPTIONS);
600 if (nest == NULL)
601 goto nla_put_failure;
602
603 if (nla_put_u32(skb, TCA_FLOW_KEYS, f->keymask) ||
604 nla_put_u32(skb, TCA_FLOW_MODE, f->mode))
605 goto nla_put_failure;
606
607 if (f->mask != ~0 || f->xor != 0) {
608 if (nla_put_u32(skb, TCA_FLOW_MASK, f->mask) ||
609 nla_put_u32(skb, TCA_FLOW_XOR, f->xor))
610 goto nla_put_failure;
611 }
612 if (f->rshift &&
613 nla_put_u32(skb, TCA_FLOW_RSHIFT, f->rshift))
614 goto nla_put_failure;
615 if (f->addend &&
616 nla_put_u32(skb, TCA_FLOW_ADDEND, f->addend))
617 goto nla_put_failure;
618
619 if (f->divisor &&
620 nla_put_u32(skb, TCA_FLOW_DIVISOR, f->divisor))
621 goto nla_put_failure;
622 if (f->baseclass &&
623 nla_put_u32(skb, TCA_FLOW_BASECLASS, f->baseclass))
624 goto nla_put_failure;
625
626 if (f->perturb_period &&
627 nla_put_u32(skb, TCA_FLOW_PERTURB, f->perturb_period / HZ))
628 goto nla_put_failure;
629
630 if (tcf_exts_dump(skb, &f->exts) < 0)
631 goto nla_put_failure;
632#ifdef CONFIG_NET_EMATCH
633 if (f->ematches.hdr.nmatches &&
634 tcf_em_tree_dump(skb, &f->ematches, TCA_FLOW_EMATCHES) < 0)
635 goto nla_put_failure;
636#endif
637 nla_nest_end(skb, nest);
638
639 if (tcf_exts_dump_stats(skb, &f->exts) < 0)
640 goto nla_put_failure;
641
642 return skb->len;
643
644nla_put_failure:
645 nla_nest_cancel(skb, nest);
646 return -1;
647}
648
649static void flow_walk(struct tcf_proto *tp, struct tcf_walker *arg)
650{
651 struct flow_head *head = rtnl_dereference(tp->root);
652 struct flow_filter *f;
653
654 list_for_each_entry(f, &head->filters, list) {
655 if (arg->count < arg->skip)
656 goto skip;
657 if (arg->fn(tp, (unsigned long)f, arg) < 0) {
658 arg->stop = 1;
659 break;
660 }
661skip:
662 arg->count++;
663 }
664}
665
666static struct tcf_proto_ops cls_flow_ops __read_mostly = {
667 .kind = "flow",
668 .classify = flow_classify,
669 .init = flow_init,
670 .destroy = flow_destroy,
671 .change = flow_change,
672 .delete = flow_delete,
673 .get = flow_get,
674 .dump = flow_dump,
675 .walk = flow_walk,
676 .owner = THIS_MODULE,
677};
678
679static int __init cls_flow_init(void)
680{
681 return register_tcf_proto_ops(&cls_flow_ops);
682}
683
684static void __exit cls_flow_exit(void)
685{
686 unregister_tcf_proto_ops(&cls_flow_ops);
687}
688
689module_init(cls_flow_init);
690module_exit(cls_flow_exit);
691
692MODULE_LICENSE("GPL");
693MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
694MODULE_DESCRIPTION("TC flow classifier");
695
