Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6/kmemtrace.git] / net / sched / cls_rsvp.h
blob22f9ede70e8fb47dbf02b8900c1f1a146c5caa60
1 /*
2 * net/sched/cls_rsvp.h Template file for RSVPv[46] classifiers.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
13 Comparing to general packet classification problem,
14 RSVP needs only sevaral relatively simple rules:
16 * (dst, protocol) are always specified,
17 so that we are able to hash them.
18 * src may be exact, or may be wildcard, so that
19 we can keep a hash table plus one wildcard entry.
20 * source port (or flow label) is important only if src is given.
22 IMPLEMENTATION.
24 We use a two level hash table: The top level is keyed by
25 destination address and protocol ID, every bucket contains a list
26 of "rsvp sessions", identified by destination address, protocol and
27 DPI(="Destination Port ID"): triple (key, mask, offset).
29 Every bucket has a smaller hash table keyed by source address
30 (cf. RSVP flowspec) and one wildcard entry for wildcard reservations.
31 Every bucket is again a list of "RSVP flows", selected by
32 source address and SPI(="Source Port ID" here rather than
33 "security parameter index"): triple (key, mask, offset).
36 NOTE 1. All the packets with IPv6 extension headers (but AH and ESP)
37 and all fragmented packets go to the best-effort traffic class.
40 NOTE 2. Two "port id"'s seems to be redundant, rfc2207 requires
41 only one "Generalized Port Identifier". So that for classic
42 ah, esp (and udp,tcp) both *pi should coincide or one of them
43 should be wildcard.
45 At first sight, this redundancy is just a waste of CPU
46 resources. But DPI and SPI add the possibility to assign different
47 priorities to GPIs. Look also at note 4 about tunnels below.
50 NOTE 3. One complication is the case of tunneled packets.
51 We implement it as following: if the first lookup
52 matches a special session with "tunnelhdr" value not zero,
53 flowid doesn't contain the true flow ID, but the tunnel ID (1...255).
54 In this case, we pull tunnelhdr bytes and restart lookup
55 with tunnel ID added to the list of keys. Simple and stupid 8)8)
56 It's enough for PIMREG and IPIP.
59 NOTE 4. Two GPIs make it possible to parse even GRE packets.
60 F.e. DPI can select ETH_P_IP (and necessary flags to make
61 tunnelhdr correct) in GRE protocol field and SPI matches
62 GRE key. Is it not nice? 8)8)
65 Well, as result, despite its simplicity, we get a pretty
66 powerful classification engine. */
69 struct rsvp_head
71 u32 tmap[256/32];
72 u32 hgenerator;
73 u8 tgenerator;
74 struct rsvp_session *ht[256];
77 struct rsvp_session
79 struct rsvp_session *next;
80 __be32 dst[RSVP_DST_LEN];
81 struct tc_rsvp_gpi dpi;
82 u8 protocol;
83 u8 tunnelid;
84 /* 16 (src,sport) hash slots, and one wildcard source slot */
85 struct rsvp_filter *ht[16+1];
89 struct rsvp_filter
91 struct rsvp_filter *next;
92 __be32 src[RSVP_DST_LEN];
93 struct tc_rsvp_gpi spi;
94 u8 tunnelhdr;
96 struct tcf_result res;
97 struct tcf_exts exts;
99 u32 handle;
100 struct rsvp_session *sess;
103 static __inline__ unsigned hash_dst(__be32 *dst, u8 protocol, u8 tunnelid)
105 unsigned h = (__force __u32)dst[RSVP_DST_LEN-1];
106 h ^= h>>16;
107 h ^= h>>8;
108 return (h ^ protocol ^ tunnelid) & 0xFF;
111 static __inline__ unsigned hash_src(__be32 *src)
113 unsigned h = (__force __u32)src[RSVP_DST_LEN-1];
114 h ^= h>>16;
115 h ^= h>>8;
116 h ^= h>>4;
117 return h & 0xF;
120 static struct tcf_ext_map rsvp_ext_map = {
121 .police = TCA_RSVP_POLICE,
122 .action = TCA_RSVP_ACT
125 #define RSVP_APPLY_RESULT() \
127 int r = tcf_exts_exec(skb, &f->exts, res); \
128 if (r < 0) \
129 continue; \
130 else if (r > 0) \
131 return r; \
134 static int rsvp_classify(struct sk_buff *skb, struct tcf_proto *tp,
135 struct tcf_result *res)
137 struct rsvp_session **sht = ((struct rsvp_head*)tp->root)->ht;
138 struct rsvp_session *s;
139 struct rsvp_filter *f;
140 unsigned h1, h2;
141 __be32 *dst, *src;
142 u8 protocol;
143 u8 tunnelid = 0;
144 u8 *xprt;
145 #if RSVP_DST_LEN == 4
146 struct ipv6hdr *nhptr = ipv6_hdr(skb);
147 #else
148 struct iphdr *nhptr = ip_hdr(skb);
149 #endif
151 restart:
153 #if RSVP_DST_LEN == 4
154 src = &nhptr->saddr.s6_addr32[0];
155 dst = &nhptr->daddr.s6_addr32[0];
156 protocol = nhptr->nexthdr;
157 xprt = ((u8*)nhptr) + sizeof(struct ipv6hdr);
158 #else
159 src = &nhptr->saddr;
160 dst = &nhptr->daddr;
161 protocol = nhptr->protocol;
162 xprt = ((u8*)nhptr) + (nhptr->ihl<<2);
163 if (nhptr->frag_off & htons(IP_MF|IP_OFFSET))
164 return -1;
165 #endif
167 h1 = hash_dst(dst, protocol, tunnelid);
168 h2 = hash_src(src);
170 for (s = sht[h1]; s; s = s->next) {
171 if (dst[RSVP_DST_LEN-1] == s->dst[RSVP_DST_LEN-1] &&
172 protocol == s->protocol &&
173 !(s->dpi.mask & (*(u32*)(xprt+s->dpi.offset)^s->dpi.key))
174 #if RSVP_DST_LEN == 4
175 && dst[0] == s->dst[0]
176 && dst[1] == s->dst[1]
177 && dst[2] == s->dst[2]
178 #endif
179 && tunnelid == s->tunnelid) {
181 for (f = s->ht[h2]; f; f = f->next) {
182 if (src[RSVP_DST_LEN-1] == f->src[RSVP_DST_LEN-1] &&
183 !(f->spi.mask & (*(u32*)(xprt+f->spi.offset)^f->spi.key))
184 #if RSVP_DST_LEN == 4
185 && src[0] == f->src[0]
186 && src[1] == f->src[1]
187 && src[2] == f->src[2]
188 #endif
190 *res = f->res;
191 RSVP_APPLY_RESULT();
193 matched:
194 if (f->tunnelhdr == 0)
195 return 0;
197 tunnelid = f->res.classid;
198 nhptr = (void*)(xprt + f->tunnelhdr - sizeof(*nhptr));
199 goto restart;
203 /* And wildcard bucket... */
204 for (f = s->ht[16]; f; f = f->next) {
205 *res = f->res;
206 RSVP_APPLY_RESULT();
207 goto matched;
209 return -1;
212 return -1;
215 static unsigned long rsvp_get(struct tcf_proto *tp, u32 handle)
217 struct rsvp_session **sht = ((struct rsvp_head*)tp->root)->ht;
218 struct rsvp_session *s;
219 struct rsvp_filter *f;
220 unsigned h1 = handle&0xFF;
221 unsigned h2 = (handle>>8)&0xFF;
223 if (h2 > 16)
224 return 0;
226 for (s = sht[h1]; s; s = s->next) {
227 for (f = s->ht[h2]; f; f = f->next) {
228 if (f->handle == handle)
229 return (unsigned long)f;
232 return 0;
235 static void rsvp_put(struct tcf_proto *tp, unsigned long f)
239 static int rsvp_init(struct tcf_proto *tp)
241 struct rsvp_head *data;
243 data = kzalloc(sizeof(struct rsvp_head), GFP_KERNEL);
244 if (data) {
245 tp->root = data;
246 return 0;
248 return -ENOBUFS;
251 static inline void
252 rsvp_delete_filter(struct tcf_proto *tp, struct rsvp_filter *f)
254 tcf_unbind_filter(tp, &f->res);
255 tcf_exts_destroy(tp, &f->exts);
256 kfree(f);
259 static void rsvp_destroy(struct tcf_proto *tp)
261 struct rsvp_head *data = xchg(&tp->root, NULL);
262 struct rsvp_session **sht;
263 int h1, h2;
265 if (data == NULL)
266 return;
268 sht = data->ht;
270 for (h1=0; h1<256; h1++) {
271 struct rsvp_session *s;
273 while ((s = sht[h1]) != NULL) {
274 sht[h1] = s->next;
276 for (h2=0; h2<=16; h2++) {
277 struct rsvp_filter *f;
279 while ((f = s->ht[h2]) != NULL) {
280 s->ht[h2] = f->next;
281 rsvp_delete_filter(tp, f);
284 kfree(s);
287 kfree(data);
290 static int rsvp_delete(struct tcf_proto *tp, unsigned long arg)
292 struct rsvp_filter **fp, *f = (struct rsvp_filter*)arg;
293 unsigned h = f->handle;
294 struct rsvp_session **sp;
295 struct rsvp_session *s = f->sess;
296 int i;
298 for (fp = &s->ht[(h>>8)&0xFF]; *fp; fp = &(*fp)->next) {
299 if (*fp == f) {
300 tcf_tree_lock(tp);
301 *fp = f->next;
302 tcf_tree_unlock(tp);
303 rsvp_delete_filter(tp, f);
305 /* Strip tree */
307 for (i=0; i<=16; i++)
308 if (s->ht[i])
309 return 0;
311 /* OK, session has no flows */
312 for (sp = &((struct rsvp_head*)tp->root)->ht[h&0xFF];
313 *sp; sp = &(*sp)->next) {
314 if (*sp == s) {
315 tcf_tree_lock(tp);
316 *sp = s->next;
317 tcf_tree_unlock(tp);
319 kfree(s);
320 return 0;
324 return 0;
327 return 0;
330 static unsigned gen_handle(struct tcf_proto *tp, unsigned salt)
332 struct rsvp_head *data = tp->root;
333 int i = 0xFFFF;
335 while (i-- > 0) {
336 u32 h;
337 if ((data->hgenerator += 0x10000) == 0)
338 data->hgenerator = 0x10000;
339 h = data->hgenerator|salt;
340 if (rsvp_get(tp, h) == 0)
341 return h;
343 return 0;
346 static int tunnel_bts(struct rsvp_head *data)
348 int n = data->tgenerator>>5;
349 u32 b = 1<<(data->tgenerator&0x1F);
351 if (data->tmap[n]&b)
352 return 0;
353 data->tmap[n] |= b;
354 return 1;
357 static void tunnel_recycle(struct rsvp_head *data)
359 struct rsvp_session **sht = data->ht;
360 u32 tmap[256/32];
361 int h1, h2;
363 memset(tmap, 0, sizeof(tmap));
365 for (h1=0; h1<256; h1++) {
366 struct rsvp_session *s;
367 for (s = sht[h1]; s; s = s->next) {
368 for (h2=0; h2<=16; h2++) {
369 struct rsvp_filter *f;
371 for (f = s->ht[h2]; f; f = f->next) {
372 if (f->tunnelhdr == 0)
373 continue;
374 data->tgenerator = f->res.classid;
375 tunnel_bts(data);
381 memcpy(data->tmap, tmap, sizeof(tmap));
384 static u32 gen_tunnel(struct rsvp_head *data)
386 int i, k;
388 for (k=0; k<2; k++) {
389 for (i=255; i>0; i--) {
390 if (++data->tgenerator == 0)
391 data->tgenerator = 1;
392 if (tunnel_bts(data))
393 return data->tgenerator;
395 tunnel_recycle(data);
397 return 0;
400 static int rsvp_change(struct tcf_proto *tp, unsigned long base,
401 u32 handle,
402 struct rtattr **tca,
403 unsigned long *arg)
405 struct rsvp_head *data = tp->root;
406 struct rsvp_filter *f, **fp;
407 struct rsvp_session *s, **sp;
408 struct tc_rsvp_pinfo *pinfo = NULL;
409 struct rtattr *opt = tca[TCA_OPTIONS-1];
410 struct rtattr *tb[TCA_RSVP_MAX];
411 struct tcf_exts e;
412 unsigned h1, h2;
413 __be32 *dst;
414 int err;
416 if (opt == NULL)
417 return handle ? -EINVAL : 0;
419 if (rtattr_parse_nested(tb, TCA_RSVP_MAX, opt) < 0)
420 return -EINVAL;
422 err = tcf_exts_validate(tp, tb, tca[TCA_RATE-1], &e, &rsvp_ext_map);
423 if (err < 0)
424 return err;
426 if ((f = (struct rsvp_filter*)*arg) != NULL) {
427 /* Node exists: adjust only classid */
429 if (f->handle != handle && handle)
430 goto errout2;
431 if (tb[TCA_RSVP_CLASSID-1]) {
432 f->res.classid = *(u32*)RTA_DATA(tb[TCA_RSVP_CLASSID-1]);
433 tcf_bind_filter(tp, &f->res, base);
436 tcf_exts_change(tp, &f->exts, &e);
437 return 0;
440 /* Now more serious part... */
441 err = -EINVAL;
442 if (handle)
443 goto errout2;
444 if (tb[TCA_RSVP_DST-1] == NULL)
445 goto errout2;
447 err = -ENOBUFS;
448 f = kzalloc(sizeof(struct rsvp_filter), GFP_KERNEL);
449 if (f == NULL)
450 goto errout2;
452 h2 = 16;
453 if (tb[TCA_RSVP_SRC-1]) {
454 err = -EINVAL;
455 if (RTA_PAYLOAD(tb[TCA_RSVP_SRC-1]) != sizeof(f->src))
456 goto errout;
457 memcpy(f->src, RTA_DATA(tb[TCA_RSVP_SRC-1]), sizeof(f->src));
458 h2 = hash_src(f->src);
460 if (tb[TCA_RSVP_PINFO-1]) {
461 err = -EINVAL;
462 if (RTA_PAYLOAD(tb[TCA_RSVP_PINFO-1]) < sizeof(struct tc_rsvp_pinfo))
463 goto errout;
464 pinfo = RTA_DATA(tb[TCA_RSVP_PINFO-1]);
465 f->spi = pinfo->spi;
466 f->tunnelhdr = pinfo->tunnelhdr;
468 if (tb[TCA_RSVP_CLASSID-1]) {
469 err = -EINVAL;
470 if (RTA_PAYLOAD(tb[TCA_RSVP_CLASSID-1]) != 4)
471 goto errout;
472 f->res.classid = *(u32*)RTA_DATA(tb[TCA_RSVP_CLASSID-1]);
475 err = -EINVAL;
476 if (RTA_PAYLOAD(tb[TCA_RSVP_DST-1]) != sizeof(f->src))
477 goto errout;
478 dst = RTA_DATA(tb[TCA_RSVP_DST-1]);
479 h1 = hash_dst(dst, pinfo ? pinfo->protocol : 0, pinfo ? pinfo->tunnelid : 0);
481 err = -ENOMEM;
482 if ((f->handle = gen_handle(tp, h1 | (h2<<8))) == 0)
483 goto errout;
485 if (f->tunnelhdr) {
486 err = -EINVAL;
487 if (f->res.classid > 255)
488 goto errout;
490 err = -ENOMEM;
491 if (f->res.classid == 0 &&
492 (f->res.classid = gen_tunnel(data)) == 0)
493 goto errout;
496 for (sp = &data->ht[h1]; (s=*sp) != NULL; sp = &s->next) {
497 if (dst[RSVP_DST_LEN-1] == s->dst[RSVP_DST_LEN-1] &&
498 pinfo && pinfo->protocol == s->protocol &&
499 memcmp(&pinfo->dpi, &s->dpi, sizeof(s->dpi)) == 0
500 #if RSVP_DST_LEN == 4
501 && dst[0] == s->dst[0]
502 && dst[1] == s->dst[1]
503 && dst[2] == s->dst[2]
504 #endif
505 && pinfo->tunnelid == s->tunnelid) {
507 insert:
508 /* OK, we found appropriate session */
510 fp = &s->ht[h2];
512 f->sess = s;
513 if (f->tunnelhdr == 0)
514 tcf_bind_filter(tp, &f->res, base);
516 tcf_exts_change(tp, &f->exts, &e);
518 for (fp = &s->ht[h2]; *fp; fp = &(*fp)->next)
519 if (((*fp)->spi.mask&f->spi.mask) != f->spi.mask)
520 break;
521 f->next = *fp;
522 wmb();
523 *fp = f;
525 *arg = (unsigned long)f;
526 return 0;
530 /* No session found. Create new one. */
532 err = -ENOBUFS;
533 s = kzalloc(sizeof(struct rsvp_session), GFP_KERNEL);
534 if (s == NULL)
535 goto errout;
536 memcpy(s->dst, dst, sizeof(s->dst));
538 if (pinfo) {
539 s->dpi = pinfo->dpi;
540 s->protocol = pinfo->protocol;
541 s->tunnelid = pinfo->tunnelid;
543 for (sp = &data->ht[h1]; *sp; sp = &(*sp)->next) {
544 if (((*sp)->dpi.mask&s->dpi.mask) != s->dpi.mask)
545 break;
547 s->next = *sp;
548 wmb();
549 *sp = s;
551 goto insert;
553 errout:
554 kfree(f);
555 errout2:
556 tcf_exts_destroy(tp, &e);
557 return err;
560 static void rsvp_walk(struct tcf_proto *tp, struct tcf_walker *arg)
562 struct rsvp_head *head = tp->root;
563 unsigned h, h1;
565 if (arg->stop)
566 return;
568 for (h = 0; h < 256; h++) {
569 struct rsvp_session *s;
571 for (s = head->ht[h]; s; s = s->next) {
572 for (h1 = 0; h1 <= 16; h1++) {
573 struct rsvp_filter *f;
575 for (f = s->ht[h1]; f; f = f->next) {
576 if (arg->count < arg->skip) {
577 arg->count++;
578 continue;
580 if (arg->fn(tp, (unsigned long)f, arg) < 0) {
581 arg->stop = 1;
582 return;
584 arg->count++;
591 static int rsvp_dump(struct tcf_proto *tp, unsigned long fh,
592 struct sk_buff *skb, struct tcmsg *t)
594 struct rsvp_filter *f = (struct rsvp_filter*)fh;
595 struct rsvp_session *s;
596 unsigned char *b = skb_tail_pointer(skb);
597 struct rtattr *rta;
598 struct tc_rsvp_pinfo pinfo;
600 if (f == NULL)
601 return skb->len;
602 s = f->sess;
604 t->tcm_handle = f->handle;
607 rta = (struct rtattr*)b;
608 RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
610 RTA_PUT(skb, TCA_RSVP_DST, sizeof(s->dst), &s->dst);
611 pinfo.dpi = s->dpi;
612 pinfo.spi = f->spi;
613 pinfo.protocol = s->protocol;
614 pinfo.tunnelid = s->tunnelid;
615 pinfo.tunnelhdr = f->tunnelhdr;
616 pinfo.pad = 0;
617 RTA_PUT(skb, TCA_RSVP_PINFO, sizeof(pinfo), &pinfo);
618 if (f->res.classid)
619 RTA_PUT(skb, TCA_RSVP_CLASSID, 4, &f->res.classid);
620 if (((f->handle>>8)&0xFF) != 16)
621 RTA_PUT(skb, TCA_RSVP_SRC, sizeof(f->src), f->src);
623 if (tcf_exts_dump(skb, &f->exts, &rsvp_ext_map) < 0)
624 goto rtattr_failure;
626 rta->rta_len = skb_tail_pointer(skb) - b;
628 if (tcf_exts_dump_stats(skb, &f->exts, &rsvp_ext_map) < 0)
629 goto rtattr_failure;
630 return skb->len;
632 rtattr_failure:
633 nlmsg_trim(skb, b);
634 return -1;
637 static struct tcf_proto_ops RSVP_OPS = {
638 .next = NULL,
639 .kind = RSVP_ID,
640 .classify = rsvp_classify,
641 .init = rsvp_init,
642 .destroy = rsvp_destroy,
643 .get = rsvp_get,
644 .put = rsvp_put,
645 .change = rsvp_change,
646 .delete = rsvp_delete,
647 .walk = rsvp_walk,
648 .dump = rsvp_dump,
649 .owner = THIS_MODULE,
652 static int __init init_rsvp(void)
654 return register_tcf_proto_ops(&RSVP_OPS);
657 static void __exit exit_rsvp(void)
659 unregister_tcf_proto_ops(&RSVP_OPS);
662 module_init(init_rsvp)
663 module_exit(exit_rsvp)