Merge branch 'for-linus' of git://linux-nfs.org/~bfields/linux
[linux-2.6] / net / sched / cls_flow.c
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
24 #include <net/pkt_cls.h>
25 #include <net/ip.h>
26 #include <net/route.h>
27 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
28 #include <net/netfilter/nf_conntrack.h>
29 #endif
30
31 struct flow_head {
32         struct list_head        filters;
33 };
34
35 struct flow_filter {
36         struct list_head        list;
37         struct tcf_exts         exts;
38         struct tcf_ematch_tree  ematches;
39         u32                     handle;
40
41         u32                     nkeys;
42         u32                     keymask;
43         u32                     mode;
44         u32                     mask;
45         u32                     xor;
46         u32                     rshift;
47         u32                     addend;
48         u32                     divisor;
49         u32                     baseclass;
50 };
51
52 static u32 flow_hashrnd __read_mostly;
53 static int flow_hashrnd_initted __read_mostly;
54
55 static const struct tcf_ext_map flow_ext_map = {
56         .action = TCA_FLOW_ACT,
57         .police = TCA_FLOW_POLICE,
58 };
59
60 static inline u32 addr_fold(void *addr)
61 {
62         unsigned long a = (unsigned long)addr;
63
64         return (a & 0xFFFFFFFF) ^ (BITS_PER_LONG > 32 ? a >> 32 : 0);
65 }
66
67 static u32 flow_get_src(const struct sk_buff *skb)
68 {
69         switch (skb->protocol) {
70         case __constant_htons(ETH_P_IP):
71                 return ntohl(ip_hdr(skb)->saddr);
72         case __constant_htons(ETH_P_IPV6):
73                 return ntohl(ipv6_hdr(skb)->saddr.s6_addr32[3]);
74         default:
75                 return addr_fold(skb->sk);
76         }
77 }
78
79 static u32 flow_get_dst(const struct sk_buff *skb)
80 {
81         switch (skb->protocol) {
82         case __constant_htons(ETH_P_IP):
83                 return ntohl(ip_hdr(skb)->daddr);
84         case __constant_htons(ETH_P_IPV6):
85                 return ntohl(ipv6_hdr(skb)->daddr.s6_addr32[3]);
86         default:
87                 return addr_fold(skb->dst) ^ (__force u16)skb->protocol;
88         }
89 }
90
91 static u32 flow_get_proto(const struct sk_buff *skb)
92 {
93         switch (skb->protocol) {
94         case __constant_htons(ETH_P_IP):
95                 return ip_hdr(skb)->protocol;
96         case __constant_htons(ETH_P_IPV6):
97                 return ipv6_hdr(skb)->nexthdr;
98         default:
99                 return 0;
100         }
101 }
102
103 static int has_ports(u8 protocol)
104 {
105         switch (protocol) {
106         case IPPROTO_TCP:
107         case IPPROTO_UDP:
108         case IPPROTO_UDPLITE:
109         case IPPROTO_SCTP:
110         case IPPROTO_DCCP:
111         case IPPROTO_ESP:
112                 return 1;
113         default:
114                 return 0;
115         }
116 }
117
118 static u32 flow_get_proto_src(const struct sk_buff *skb)
119 {
120         u32 res = 0;
121
122         switch (skb->protocol) {
123         case __constant_htons(ETH_P_IP): {
124                 struct iphdr *iph = ip_hdr(skb);
125
126                 if (!(iph->frag_off&htons(IP_MF|IP_OFFSET)) &&
127                     has_ports(iph->protocol))
128                         res = ntohs(*(__be16 *)((void *)iph + iph->ihl * 4));
129                 break;
130         }
131         case __constant_htons(ETH_P_IPV6): {
132                 struct ipv6hdr *iph = ipv6_hdr(skb);
133
134                 if (has_ports(iph->nexthdr))
135                         res = ntohs(*(__be16 *)&iph[1]);
136                 break;
137         }
138         default:
139                 res = addr_fold(skb->sk);
140         }
141
142         return res;
143 }
144
145 static u32 flow_get_proto_dst(const struct sk_buff *skb)
146 {
147         u32 res = 0;
148
149         switch (skb->protocol) {
150         case __constant_htons(ETH_P_IP): {
151                 struct iphdr *iph = ip_hdr(skb);
152
153                 if (!(iph->frag_off&htons(IP_MF|IP_OFFSET)) &&
154                     has_ports(iph->protocol))
155                         res = ntohs(*(__be16 *)((void *)iph + iph->ihl * 4 + 2));
156                 break;
157         }
158         case __constant_htons(ETH_P_IPV6): {
159                 struct ipv6hdr *iph = ipv6_hdr(skb);
160
161                 if (has_ports(iph->nexthdr))
162                         res = ntohs(*(__be16 *)((void *)&iph[1] + 2));
163                 break;
164         }
165         default:
166                 res = addr_fold(skb->dst) ^ (__force u16)skb->protocol;
167         }
168
169         return res;
170 }
171
172 static u32 flow_get_iif(const struct sk_buff *skb)
173 {
174         return skb->iif;
175 }
176
177 static u32 flow_get_priority(const struct sk_buff *skb)
178 {
179         return skb->priority;
180 }
181
182 static u32 flow_get_mark(const struct sk_buff *skb)
183 {
184         return skb->mark;
185 }
186
187 static u32 flow_get_nfct(const struct sk_buff *skb)
188 {
189 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
190         return addr_fold(skb->nfct);
191 #else
192         return 0;
193 #endif
194 }
195
196 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
197 #define CTTUPLE(skb, member)                                            \
198 ({                                                                      \
199         enum ip_conntrack_info ctinfo;                                  \
200         struct nf_conn *ct = nf_ct_get(skb, &ctinfo);                   \
201         if (ct == NULL)                                                 \
202                 goto fallback;                                          \
203         ct->tuplehash[CTINFO2DIR(ctinfo)].tuple.member;                 \
204 })
205 #else
206 #define CTTUPLE(skb, member)                                            \
207 ({                                                                      \
208         goto fallback;                                                  \
209         0;                                                              \
210 })
211 #endif
212
213 static u32 flow_get_nfct_src(const struct sk_buff *skb)
214 {
215         switch (skb->protocol) {
216         case __constant_htons(ETH_P_IP):
217                 return ntohl(CTTUPLE(skb, src.u3.ip));
218         case __constant_htons(ETH_P_IPV6):
219                 return ntohl(CTTUPLE(skb, src.u3.ip6[3]));
220         }
221 fallback:
222         return flow_get_src(skb);
223 }
224
225 static u32 flow_get_nfct_dst(const struct sk_buff *skb)
226 {
227         switch (skb->protocol) {
228         case __constant_htons(ETH_P_IP):
229                 return ntohl(CTTUPLE(skb, dst.u3.ip));
230         case __constant_htons(ETH_P_IPV6):
231                 return ntohl(CTTUPLE(skb, dst.u3.ip6[3]));
232         }
233 fallback:
234         return flow_get_dst(skb);
235 }
236
237 static u32 flow_get_nfct_proto_src(const struct sk_buff *skb)
238 {
239         return ntohs(CTTUPLE(skb, src.u.all));
240 fallback:
241         return flow_get_proto_src(skb);
242 }
243
244 static u32 flow_get_nfct_proto_dst(const struct sk_buff *skb)
245 {
246         return ntohs(CTTUPLE(skb, dst.u.all));
247 fallback:
248         return flow_get_proto_dst(skb);
249 }
250
251 static u32 flow_get_rtclassid(const struct sk_buff *skb)
252 {
253 #ifdef CONFIG_NET_CLS_ROUTE
254         if (skb->dst)
255                 return skb->dst->tclassid;
256 #endif
257         return 0;
258 }
259
260 static u32 flow_get_skuid(const struct sk_buff *skb)
261 {
262         if (skb->sk && skb->sk->sk_socket && skb->sk->sk_socket->file)
263                 return skb->sk->sk_socket->file->f_uid;
264         return 0;
265 }
266
267 static u32 flow_get_skgid(const struct sk_buff *skb)
268 {
269         if (skb->sk && skb->sk->sk_socket && skb->sk->sk_socket->file)
270                 return skb->sk->sk_socket->file->f_gid;
271         return 0;
272 }
273
274 static u32 flow_get_vlan_tag(const struct sk_buff *skb)
275 {
276         u16 uninitialized_var(tag);
277
278         if (vlan_get_tag(skb, &tag) < 0)
279                 return 0;
280         return tag & VLAN_VID_MASK;
281 }
282
283 static u32 flow_key_get(const struct sk_buff *skb, int key)
284 {
285         switch (key) {
286         case FLOW_KEY_SRC:
287                 return flow_get_src(skb);
288         case FLOW_KEY_DST:
289                 return flow_get_dst(skb);
290         case FLOW_KEY_PROTO:
291                 return flow_get_proto(skb);
292         case FLOW_KEY_PROTO_SRC:
293                 return flow_get_proto_src(skb);
294         case FLOW_KEY_PROTO_DST:
295                 return flow_get_proto_dst(skb);
296         case FLOW_KEY_IIF:
297                 return flow_get_iif(skb);
298         case FLOW_KEY_PRIORITY:
299                 return flow_get_priority(skb);
300         case FLOW_KEY_MARK:
301                 return flow_get_mark(skb);
302         case FLOW_KEY_NFCT:
303                 return flow_get_nfct(skb);
304         case FLOW_KEY_NFCT_SRC:
305                 return flow_get_nfct_src(skb);
306         case FLOW_KEY_NFCT_DST:
307                 return flow_get_nfct_dst(skb);
308         case FLOW_KEY_NFCT_PROTO_SRC:
309                 return flow_get_nfct_proto_src(skb);
310         case FLOW_KEY_NFCT_PROTO_DST:
311                 return flow_get_nfct_proto_dst(skb);
312         case FLOW_KEY_RTCLASSID:
313                 return flow_get_rtclassid(skb);
314         case FLOW_KEY_SKUID:
315                 return flow_get_skuid(skb);
316         case FLOW_KEY_SKGID:
317                 return flow_get_skgid(skb);
318         case FLOW_KEY_VLAN_TAG:
319                 return flow_get_vlan_tag(skb);
320         default:
321                 WARN_ON(1);
322                 return 0;
323         }
324 }
325
326 static int flow_classify(struct sk_buff *skb, struct tcf_proto *tp,
327                          struct tcf_result *res)
328 {
329         struct flow_head *head = tp->root;
330         struct flow_filter *f;
331         u32 keymask;
332         u32 classid;
333         unsigned int n, key;
334         int r;
335
336         list_for_each_entry(f, &head->filters, list) {
337                 u32 keys[f->nkeys];
338
339                 if (!tcf_em_tree_match(skb, &f->ematches, NULL))
340                         continue;
341
342                 keymask = f->keymask;
343
344                 for (n = 0; n < f->nkeys; n++) {
345                         key = ffs(keymask) - 1;
346                         keymask &= ~(1 << key);
347                         keys[n] = flow_key_get(skb, key);
348                 }
349
350                 if (f->mode == FLOW_MODE_HASH)
351                         classid = jhash2(keys, f->nkeys, flow_hashrnd);
352                 else {
353                         classid = keys[0];
354                         classid = (classid & f->mask) ^ f->xor;
355                         classid = (classid >> f->rshift) + f->addend;
356                 }
357
358                 if (f->divisor)
359                         classid %= f->divisor;
360
361                 res->class   = 0;
362                 res->classid = TC_H_MAKE(f->baseclass, f->baseclass + classid);
363
364                 r = tcf_exts_exec(skb, &f->exts, res);
365                 if (r < 0)
366                         continue;
367                 return r;
368         }
369         return -1;
370 }
371
372 static const struct nla_policy flow_policy[TCA_FLOW_MAX + 1] = {
373         [TCA_FLOW_KEYS]         = { .type = NLA_U32 },
374         [TCA_FLOW_MODE]         = { .type = NLA_U32 },
375         [TCA_FLOW_BASECLASS]    = { .type = NLA_U32 },
376         [TCA_FLOW_RSHIFT]       = { .type = NLA_U32 },
377         [TCA_FLOW_ADDEND]       = { .type = NLA_U32 },
378         [TCA_FLOW_MASK]         = { .type = NLA_U32 },
379         [TCA_FLOW_XOR]          = { .type = NLA_U32 },
380         [TCA_FLOW_DIVISOR]      = { .type = NLA_U32 },
381         [TCA_FLOW_ACT]          = { .type = NLA_NESTED },
382         [TCA_FLOW_POLICE]       = { .type = NLA_NESTED },
383         [TCA_FLOW_EMATCHES]     = { .type = NLA_NESTED },
384 };
385
386 static int flow_change(struct tcf_proto *tp, unsigned long base,
387                        u32 handle, struct nlattr **tca,
388                        unsigned long *arg)
389 {
390         struct flow_head *head = tp->root;
391         struct flow_filter *f;
392         struct nlattr *opt = tca[TCA_OPTIONS];
393         struct nlattr *tb[TCA_FLOW_MAX + 1];
394         struct tcf_exts e;
395         struct tcf_ematch_tree t;
396         unsigned int nkeys = 0;
397         u32 baseclass = 0;
398         u32 keymask = 0;
399         u32 mode;
400         int err;
401
402         if (opt == NULL)
403                 return -EINVAL;
404
405         err = nla_parse_nested(tb, TCA_FLOW_MAX, opt, flow_policy);
406         if (err < 0)
407                 return err;
408
409         if (tb[TCA_FLOW_BASECLASS]) {
410                 baseclass = nla_get_u32(tb[TCA_FLOW_BASECLASS]);
411                 if (TC_H_MIN(baseclass) == 0)
412                         return -EINVAL;
413         }
414
415         if (tb[TCA_FLOW_KEYS]) {
416                 keymask = nla_get_u32(tb[TCA_FLOW_KEYS]);
417
418                 nkeys = hweight32(keymask);
419                 if (nkeys == 0)
420                         return -EINVAL;
421
422                 if (fls(keymask) - 1 > FLOW_KEY_MAX)
423                         return -EOPNOTSUPP;
424         }
425
426         err = tcf_exts_validate(tp, tb, tca[TCA_RATE], &e, &flow_ext_map);
427         if (err < 0)
428                 return err;
429
430         err = tcf_em_tree_validate(tp, tb[TCA_FLOW_EMATCHES], &t);
431         if (err < 0)
432                 goto err1;
433
434         f = (struct flow_filter *)*arg;
435         if (f != NULL) {
436                 err = -EINVAL;
437                 if (f->handle != handle && handle)
438                         goto err2;
439
440                 mode = f->mode;
441                 if (tb[TCA_FLOW_MODE])
442                         mode = nla_get_u32(tb[TCA_FLOW_MODE]);
443                 if (mode != FLOW_MODE_HASH && nkeys > 1)
444                         goto err2;
445         } else {
446                 err = -EINVAL;
447                 if (!handle)
448                         goto err2;
449                 if (!tb[TCA_FLOW_KEYS])
450                         goto err2;
451
452                 mode = FLOW_MODE_MAP;
453                 if (tb[TCA_FLOW_MODE])
454                         mode = nla_get_u32(tb[TCA_FLOW_MODE]);
455                 if (mode != FLOW_MODE_HASH && nkeys > 1)
456                         goto err2;
457
458                 if (TC_H_MAJ(baseclass) == 0)
459                         baseclass = TC_H_MAKE(tp->q->handle, baseclass);
460                 if (TC_H_MIN(baseclass) == 0)
461                         baseclass = TC_H_MAKE(baseclass, 1);
462
463                 err = -ENOBUFS;
464                 f = kzalloc(sizeof(*f), GFP_KERNEL);
465                 if (f == NULL)
466                         goto err2;
467
468                 f->handle = handle;
469                 f->mask   = ~0U;
470         }
471
472         tcf_exts_change(tp, &f->exts, &e);
473         tcf_em_tree_change(tp, &f->ematches, &t);
474
475         tcf_tree_lock(tp);
476
477         if (tb[TCA_FLOW_KEYS]) {
478                 f->keymask = keymask;
479                 f->nkeys   = nkeys;
480         }
481
482         f->mode = mode;
483
484         if (tb[TCA_FLOW_MASK])
485                 f->mask = nla_get_u32(tb[TCA_FLOW_MASK]);
486         if (tb[TCA_FLOW_XOR])
487                 f->xor = nla_get_u32(tb[TCA_FLOW_XOR]);
488         if (tb[TCA_FLOW_RSHIFT])
489                 f->rshift = nla_get_u32(tb[TCA_FLOW_RSHIFT]);
490         if (tb[TCA_FLOW_ADDEND])
491                 f->addend = nla_get_u32(tb[TCA_FLOW_ADDEND]);
492
493         if (tb[TCA_FLOW_DIVISOR])
494                 f->divisor = nla_get_u32(tb[TCA_FLOW_DIVISOR]);
495         if (baseclass)
496                 f->baseclass = baseclass;
497
498         if (*arg == 0)
499                 list_add_tail(&f->list, &head->filters);
500
501         tcf_tree_unlock(tp);
502
503         *arg = (unsigned long)f;
504         return 0;
505
506 err2:
507         tcf_em_tree_destroy(tp, &t);
508 err1:
509         tcf_exts_destroy(tp, &e);
510         return err;
511 }
512
513 static void flow_destroy_filter(struct tcf_proto *tp, struct flow_filter *f)
514 {
515         tcf_exts_destroy(tp, &f->exts);
516         tcf_em_tree_destroy(tp, &f->ematches);
517         kfree(f);
518 }
519
520 static int flow_delete(struct tcf_proto *tp, unsigned long arg)
521 {
522         struct flow_filter *f = (struct flow_filter *)arg;
523
524         tcf_tree_lock(tp);
525         list_del(&f->list);
526         tcf_tree_unlock(tp);
527         flow_destroy_filter(tp, f);
528         return 0;
529 }
530
531 static int flow_init(struct tcf_proto *tp)
532 {
533         struct flow_head *head;
534
535         if (!flow_hashrnd_initted) {
536                 get_random_bytes(&flow_hashrnd, 4);
537                 flow_hashrnd_initted = 1;
538         }
539
540         head = kzalloc(sizeof(*head), GFP_KERNEL);
541         if (head == NULL)
542                 return -ENOBUFS;
543         INIT_LIST_HEAD(&head->filters);
544         tp->root = head;
545         return 0;
546 }
547
548 static void flow_destroy(struct tcf_proto *tp)
549 {
550         struct flow_head *head = tp->root;
551         struct flow_filter *f, *next;
552
553         list_for_each_entry_safe(f, next, &head->filters, list) {
554                 list_del(&f->list);
555                 flow_destroy_filter(tp, f);
556         }
557         kfree(head);
558 }
559
560 static unsigned long flow_get(struct tcf_proto *tp, u32 handle)
561 {
562         struct flow_head *head = tp->root;
563         struct flow_filter *f;
564
565         list_for_each_entry(f, &head->filters, list)
566                 if (f->handle == handle)
567                         return (unsigned long)f;
568         return 0;
569 }
570
571 static void flow_put(struct tcf_proto *tp, unsigned long f)
572 {
573         return;
574 }
575
576 static int flow_dump(struct tcf_proto *tp, unsigned long fh,
577                      struct sk_buff *skb, struct tcmsg *t)
578 {
579         struct flow_filter *f = (struct flow_filter *)fh;
580         struct nlattr *nest;
581
582         if (f == NULL)
583                 return skb->len;
584
585         t->tcm_handle = f->handle;
586
587         nest = nla_nest_start(skb, TCA_OPTIONS);
588         if (nest == NULL)
589                 goto nla_put_failure;
590
591         NLA_PUT_U32(skb, TCA_FLOW_KEYS, f->keymask);
592         NLA_PUT_U32(skb, TCA_FLOW_MODE, f->mode);
593
594         if (f->mask != ~0 || f->xor != 0) {
595                 NLA_PUT_U32(skb, TCA_FLOW_MASK, f->mask);
596                 NLA_PUT_U32(skb, TCA_FLOW_XOR, f->xor);
597         }
598         if (f->rshift)
599                 NLA_PUT_U32(skb, TCA_FLOW_RSHIFT, f->rshift);
600         if (f->addend)
601                 NLA_PUT_U32(skb, TCA_FLOW_ADDEND, f->addend);
602
603         if (f->divisor)
604                 NLA_PUT_U32(skb, TCA_FLOW_DIVISOR, f->divisor);
605         if (f->baseclass)
606                 NLA_PUT_U32(skb, TCA_FLOW_BASECLASS, f->baseclass);
607
608         if (tcf_exts_dump(skb, &f->exts, &flow_ext_map) < 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, &flow_ext_map) < 0)
618                 goto nla_put_failure;
619
620         return skb->len;
621
622 nla_put_failure:
623         nlmsg_trim(skb, nest);
624         return -1;
625 }
626
627 static 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                 }
639 skip:
640                 arg->count++;
641         }
642 }
643
644 static 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
658 static int __init cls_flow_init(void)
659 {
660         return register_tcf_proto_ops(&cls_flow_ops);
661 }
662
663 static void __exit cls_flow_exit(void)
664 {
665         unregister_tcf_proto_ops(&cls_flow_ops);
666 }
667
668 module_init(cls_flow_init);
669 module_exit(cls_flow_exit);
670
671 MODULE_LICENSE("GPL");
672 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
673 MODULE_DESCRIPTION("TC flow classifier");