sched: make early bootup sched_clock() use safer
[wrt350n-kernel.git] / net / sched / cls_rsvp.h
blob7034ea4530e578c0e59da24ece7dc1790125d4cb
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 const struct nla_policy rsvp_policy[TCA_RSVP_MAX + 1] = {
401 [TCA_RSVP_CLASSID] = { .type = NLA_U32 },
402 [TCA_RSVP_DST] = { .type = NLA_BINARY,
403 .len = RSVP_DST_LEN * sizeof(u32) },
404 [TCA_RSVP_SRC] = { .type = NLA_BINARY,
405 .len = RSVP_DST_LEN * sizeof(u32) },
406 [TCA_RSVP_PINFO] = { .len = sizeof(struct tc_rsvp_pinfo) },
409 static int rsvp_change(struct tcf_proto *tp, unsigned long base,
410 u32 handle,
411 struct nlattr **tca,
412 unsigned long *arg)
414 struct rsvp_head *data = tp->root;
415 struct rsvp_filter *f, **fp;
416 struct rsvp_session *s, **sp;
417 struct tc_rsvp_pinfo *pinfo = NULL;
418 struct nlattr *opt = tca[TCA_OPTIONS-1];
419 struct nlattr *tb[TCA_RSVP_MAX + 1];
420 struct tcf_exts e;
421 unsigned h1, h2;
422 __be32 *dst;
423 int err;
425 if (opt == NULL)
426 return handle ? -EINVAL : 0;
428 err = nla_parse_nested(tb, TCA_RSVP_MAX, opt, rsvp_policy);
429 if (err < 0)
430 return err;
432 err = tcf_exts_validate(tp, tb, tca[TCA_RATE-1], &e, &rsvp_ext_map);
433 if (err < 0)
434 return err;
436 if ((f = (struct rsvp_filter*)*arg) != NULL) {
437 /* Node exists: adjust only classid */
439 if (f->handle != handle && handle)
440 goto errout2;
441 if (tb[TCA_RSVP_CLASSID-1]) {
442 f->res.classid = nla_get_u32(tb[TCA_RSVP_CLASSID-1]);
443 tcf_bind_filter(tp, &f->res, base);
446 tcf_exts_change(tp, &f->exts, &e);
447 return 0;
450 /* Now more serious part... */
451 err = -EINVAL;
452 if (handle)
453 goto errout2;
454 if (tb[TCA_RSVP_DST-1] == NULL)
455 goto errout2;
457 err = -ENOBUFS;
458 f = kzalloc(sizeof(struct rsvp_filter), GFP_KERNEL);
459 if (f == NULL)
460 goto errout2;
462 h2 = 16;
463 if (tb[TCA_RSVP_SRC-1]) {
464 memcpy(f->src, nla_data(tb[TCA_RSVP_SRC-1]), sizeof(f->src));
465 h2 = hash_src(f->src);
467 if (tb[TCA_RSVP_PINFO-1]) {
468 pinfo = nla_data(tb[TCA_RSVP_PINFO-1]);
469 f->spi = pinfo->spi;
470 f->tunnelhdr = pinfo->tunnelhdr;
472 if (tb[TCA_RSVP_CLASSID-1])
473 f->res.classid = nla_get_u32(tb[TCA_RSVP_CLASSID-1]);
475 dst = nla_data(tb[TCA_RSVP_DST-1]);
476 h1 = hash_dst(dst, pinfo ? pinfo->protocol : 0, pinfo ? pinfo->tunnelid : 0);
478 err = -ENOMEM;
479 if ((f->handle = gen_handle(tp, h1 | (h2<<8))) == 0)
480 goto errout;
482 if (f->tunnelhdr) {
483 err = -EINVAL;
484 if (f->res.classid > 255)
485 goto errout;
487 err = -ENOMEM;
488 if (f->res.classid == 0 &&
489 (f->res.classid = gen_tunnel(data)) == 0)
490 goto errout;
493 for (sp = &data->ht[h1]; (s=*sp) != NULL; sp = &s->next) {
494 if (dst[RSVP_DST_LEN-1] == s->dst[RSVP_DST_LEN-1] &&
495 pinfo && pinfo->protocol == s->protocol &&
496 memcmp(&pinfo->dpi, &s->dpi, sizeof(s->dpi)) == 0
497 #if RSVP_DST_LEN == 4
498 && dst[0] == s->dst[0]
499 && dst[1] == s->dst[1]
500 && dst[2] == s->dst[2]
501 #endif
502 && pinfo->tunnelid == s->tunnelid) {
504 insert:
505 /* OK, we found appropriate session */
507 fp = &s->ht[h2];
509 f->sess = s;
510 if (f->tunnelhdr == 0)
511 tcf_bind_filter(tp, &f->res, base);
513 tcf_exts_change(tp, &f->exts, &e);
515 for (fp = &s->ht[h2]; *fp; fp = &(*fp)->next)
516 if (((*fp)->spi.mask&f->spi.mask) != f->spi.mask)
517 break;
518 f->next = *fp;
519 wmb();
520 *fp = f;
522 *arg = (unsigned long)f;
523 return 0;
527 /* No session found. Create new one. */
529 err = -ENOBUFS;
530 s = kzalloc(sizeof(struct rsvp_session), GFP_KERNEL);
531 if (s == NULL)
532 goto errout;
533 memcpy(s->dst, dst, sizeof(s->dst));
535 if (pinfo) {
536 s->dpi = pinfo->dpi;
537 s->protocol = pinfo->protocol;
538 s->tunnelid = pinfo->tunnelid;
540 for (sp = &data->ht[h1]; *sp; sp = &(*sp)->next) {
541 if (((*sp)->dpi.mask&s->dpi.mask) != s->dpi.mask)
542 break;
544 s->next = *sp;
545 wmb();
546 *sp = s;
548 goto insert;
550 errout:
551 kfree(f);
552 errout2:
553 tcf_exts_destroy(tp, &e);
554 return err;
557 static void rsvp_walk(struct tcf_proto *tp, struct tcf_walker *arg)
559 struct rsvp_head *head = tp->root;
560 unsigned h, h1;
562 if (arg->stop)
563 return;
565 for (h = 0; h < 256; h++) {
566 struct rsvp_session *s;
568 for (s = head->ht[h]; s; s = s->next) {
569 for (h1 = 0; h1 <= 16; h1++) {
570 struct rsvp_filter *f;
572 for (f = s->ht[h1]; f; f = f->next) {
573 if (arg->count < arg->skip) {
574 arg->count++;
575 continue;
577 if (arg->fn(tp, (unsigned long)f, arg) < 0) {
578 arg->stop = 1;
579 return;
581 arg->count++;
588 static int rsvp_dump(struct tcf_proto *tp, unsigned long fh,
589 struct sk_buff *skb, struct tcmsg *t)
591 struct rsvp_filter *f = (struct rsvp_filter*)fh;
592 struct rsvp_session *s;
593 unsigned char *b = skb_tail_pointer(skb);
594 struct nlattr *nest;
595 struct tc_rsvp_pinfo pinfo;
597 if (f == NULL)
598 return skb->len;
599 s = f->sess;
601 t->tcm_handle = f->handle;
603 nest = nla_nest_start(skb, TCA_OPTIONS);
604 if (nest == NULL)
605 goto nla_put_failure;
607 NLA_PUT(skb, TCA_RSVP_DST, sizeof(s->dst), &s->dst);
608 pinfo.dpi = s->dpi;
609 pinfo.spi = f->spi;
610 pinfo.protocol = s->protocol;
611 pinfo.tunnelid = s->tunnelid;
612 pinfo.tunnelhdr = f->tunnelhdr;
613 pinfo.pad = 0;
614 NLA_PUT(skb, TCA_RSVP_PINFO, sizeof(pinfo), &pinfo);
615 if (f->res.classid)
616 NLA_PUT_U32(skb, TCA_RSVP_CLASSID, f->res.classid);
617 if (((f->handle>>8)&0xFF) != 16)
618 NLA_PUT(skb, TCA_RSVP_SRC, sizeof(f->src), f->src);
620 if (tcf_exts_dump(skb, &f->exts, &rsvp_ext_map) < 0)
621 goto nla_put_failure;
623 nla_nest_end(skb, nest);
625 if (tcf_exts_dump_stats(skb, &f->exts, &rsvp_ext_map) < 0)
626 goto nla_put_failure;
627 return skb->len;
629 nla_put_failure:
630 nlmsg_trim(skb, b);
631 return -1;
634 static struct tcf_proto_ops RSVP_OPS = {
635 .next = NULL,
636 .kind = RSVP_ID,
637 .classify = rsvp_classify,
638 .init = rsvp_init,
639 .destroy = rsvp_destroy,
640 .get = rsvp_get,
641 .put = rsvp_put,
642 .change = rsvp_change,
643 .delete = rsvp_delete,
644 .walk = rsvp_walk,
645 .dump = rsvp_dump,
646 .owner = THIS_MODULE,
649 static int __init init_rsvp(void)
651 return register_tcf_proto_ops(&RSVP_OPS);
654 static void __exit exit_rsvp(void)
656 unregister_tcf_proto_ops(&RSVP_OPS);
659 module_init(init_rsvp)
660 module_exit(exit_rsvp)