Btrfs: fix direct-io vs nodatacow
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / sched / cls_rsvp.h
blobb01427924f8112dd0b923287976f285fd7674fbf
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 {
70 u32 tmap[256/32];
71 u32 hgenerator;
72 u8 tgenerator;
73 struct rsvp_session *ht[256];
76 struct rsvp_session {
77 struct rsvp_session *next;
78 __be32 dst[RSVP_DST_LEN];
79 struct tc_rsvp_gpi dpi;
80 u8 protocol;
81 u8 tunnelid;
82 /* 16 (src,sport) hash slots, and one wildcard source slot */
83 struct rsvp_filter *ht[16 + 1];
87 struct rsvp_filter {
88 struct rsvp_filter *next;
89 __be32 src[RSVP_DST_LEN];
90 struct tc_rsvp_gpi spi;
91 u8 tunnelhdr;
93 struct tcf_result res;
94 struct tcf_exts exts;
96 u32 handle;
97 struct rsvp_session *sess;
100 static inline unsigned int hash_dst(__be32 *dst, u8 protocol, u8 tunnelid)
102 unsigned int h = (__force __u32)dst[RSVP_DST_LEN - 1];
104 h ^= h>>16;
105 h ^= h>>8;
106 return (h ^ protocol ^ tunnelid) & 0xFF;
109 static inline unsigned int hash_src(__be32 *src)
111 unsigned int h = (__force __u32)src[RSVP_DST_LEN-1];
113 h ^= h>>16;
114 h ^= h>>8;
115 h ^= h>>4;
116 return h & 0xF;
119 static struct tcf_ext_map rsvp_ext_map = {
120 .police = TCA_RSVP_POLICE,
121 .action = TCA_RSVP_ACT
124 #define RSVP_APPLY_RESULT() \
126 int r = tcf_exts_exec(skb, &f->exts, res); \
127 if (r < 0) \
128 continue; \
129 else if (r > 0) \
130 return r; \
133 static int rsvp_classify(struct sk_buff *skb, const struct tcf_proto *tp,
134 struct tcf_result *res)
136 struct rsvp_session **sht = ((struct rsvp_head *)tp->root)->ht;
137 struct rsvp_session *s;
138 struct rsvp_filter *f;
139 unsigned int h1, h2;
140 __be32 *dst, *src;
141 u8 protocol;
142 u8 tunnelid = 0;
143 u8 *xprt;
144 #if RSVP_DST_LEN == 4
145 struct ipv6hdr *nhptr;
147 if (!pskb_network_may_pull(skb, sizeof(*nhptr)))
148 return -1;
149 nhptr = ipv6_hdr(skb);
150 #else
151 struct iphdr *nhptr;
153 if (!pskb_network_may_pull(skb, sizeof(*nhptr)))
154 return -1;
155 nhptr = ip_hdr(skb);
156 #endif
158 restart:
160 #if RSVP_DST_LEN == 4
161 src = &nhptr->saddr.s6_addr32[0];
162 dst = &nhptr->daddr.s6_addr32[0];
163 protocol = nhptr->nexthdr;
164 xprt = ((u8 *)nhptr) + sizeof(struct ipv6hdr);
165 #else
166 src = &nhptr->saddr;
167 dst = &nhptr->daddr;
168 protocol = nhptr->protocol;
169 xprt = ((u8 *)nhptr) + (nhptr->ihl<<2);
170 if (ip_is_fragment(nhptr))
171 return -1;
172 #endif
174 h1 = hash_dst(dst, protocol, tunnelid);
175 h2 = hash_src(src);
177 for (s = sht[h1]; s; s = s->next) {
178 if (dst[RSVP_DST_LEN-1] == s->dst[RSVP_DST_LEN - 1] &&
179 protocol == s->protocol &&
180 !(s->dpi.mask &
181 (*(u32 *)(xprt + s->dpi.offset) ^ s->dpi.key)) &&
182 #if RSVP_DST_LEN == 4
183 dst[0] == s->dst[0] &&
184 dst[1] == s->dst[1] &&
185 dst[2] == s->dst[2] &&
186 #endif
187 tunnelid == s->tunnelid) {
189 for (f = s->ht[h2]; f; f = f->next) {
190 if (src[RSVP_DST_LEN-1] == f->src[RSVP_DST_LEN - 1] &&
191 !(f->spi.mask & (*(u32 *)(xprt + f->spi.offset) ^ f->spi.key))
192 #if RSVP_DST_LEN == 4
194 src[0] == f->src[0] &&
195 src[1] == f->src[1] &&
196 src[2] == f->src[2]
197 #endif
199 *res = f->res;
200 RSVP_APPLY_RESULT();
202 matched:
203 if (f->tunnelhdr == 0)
204 return 0;
206 tunnelid = f->res.classid;
207 nhptr = (void *)(xprt + f->tunnelhdr - sizeof(*nhptr));
208 goto restart;
212 /* And wildcard bucket... */
213 for (f = s->ht[16]; f; f = f->next) {
214 *res = f->res;
215 RSVP_APPLY_RESULT();
216 goto matched;
218 return -1;
221 return -1;
224 static unsigned long rsvp_get(struct tcf_proto *tp, u32 handle)
226 struct rsvp_session **sht = ((struct rsvp_head *)tp->root)->ht;
227 struct rsvp_session *s;
228 struct rsvp_filter *f;
229 unsigned int h1 = handle & 0xFF;
230 unsigned int h2 = (handle >> 8) & 0xFF;
232 if (h2 > 16)
233 return 0;
235 for (s = sht[h1]; s; s = s->next) {
236 for (f = s->ht[h2]; f; f = f->next) {
237 if (f->handle == handle)
238 return (unsigned long)f;
241 return 0;
244 static void rsvp_put(struct tcf_proto *tp, unsigned long f)
248 static int rsvp_init(struct tcf_proto *tp)
250 struct rsvp_head *data;
252 data = kzalloc(sizeof(struct rsvp_head), GFP_KERNEL);
253 if (data) {
254 tp->root = data;
255 return 0;
257 return -ENOBUFS;
260 static void
261 rsvp_delete_filter(struct tcf_proto *tp, struct rsvp_filter *f)
263 tcf_unbind_filter(tp, &f->res);
264 tcf_exts_destroy(tp, &f->exts);
265 kfree(f);
268 static void rsvp_destroy(struct tcf_proto *tp)
270 struct rsvp_head *data = xchg(&tp->root, NULL);
271 struct rsvp_session **sht;
272 int h1, h2;
274 if (data == NULL)
275 return;
277 sht = data->ht;
279 for (h1 = 0; h1 < 256; h1++) {
280 struct rsvp_session *s;
282 while ((s = sht[h1]) != NULL) {
283 sht[h1] = s->next;
285 for (h2 = 0; h2 <= 16; h2++) {
286 struct rsvp_filter *f;
288 while ((f = s->ht[h2]) != NULL) {
289 s->ht[h2] = f->next;
290 rsvp_delete_filter(tp, f);
293 kfree(s);
296 kfree(data);
299 static int rsvp_delete(struct tcf_proto *tp, unsigned long arg)
301 struct rsvp_filter **fp, *f = (struct rsvp_filter *)arg;
302 unsigned int h = f->handle;
303 struct rsvp_session **sp;
304 struct rsvp_session *s = f->sess;
305 int i;
307 for (fp = &s->ht[(h >> 8) & 0xFF]; *fp; fp = &(*fp)->next) {
308 if (*fp == f) {
309 tcf_tree_lock(tp);
310 *fp = f->next;
311 tcf_tree_unlock(tp);
312 rsvp_delete_filter(tp, f);
314 /* Strip tree */
316 for (i = 0; i <= 16; i++)
317 if (s->ht[i])
318 return 0;
320 /* OK, session has no flows */
321 for (sp = &((struct rsvp_head *)tp->root)->ht[h & 0xFF];
322 *sp; sp = &(*sp)->next) {
323 if (*sp == s) {
324 tcf_tree_lock(tp);
325 *sp = s->next;
326 tcf_tree_unlock(tp);
328 kfree(s);
329 return 0;
333 return 0;
336 return 0;
339 static unsigned int gen_handle(struct tcf_proto *tp, unsigned salt)
341 struct rsvp_head *data = tp->root;
342 int i = 0xFFFF;
344 while (i-- > 0) {
345 u32 h;
347 if ((data->hgenerator += 0x10000) == 0)
348 data->hgenerator = 0x10000;
349 h = data->hgenerator|salt;
350 if (rsvp_get(tp, h) == 0)
351 return h;
353 return 0;
356 static int tunnel_bts(struct rsvp_head *data)
358 int n = data->tgenerator >> 5;
359 u32 b = 1 << (data->tgenerator & 0x1F);
361 if (data->tmap[n] & b)
362 return 0;
363 data->tmap[n] |= b;
364 return 1;
367 static void tunnel_recycle(struct rsvp_head *data)
369 struct rsvp_session **sht = data->ht;
370 u32 tmap[256/32];
371 int h1, h2;
373 memset(tmap, 0, sizeof(tmap));
375 for (h1 = 0; h1 < 256; h1++) {
376 struct rsvp_session *s;
377 for (s = sht[h1]; s; s = s->next) {
378 for (h2 = 0; h2 <= 16; h2++) {
379 struct rsvp_filter *f;
381 for (f = s->ht[h2]; f; f = f->next) {
382 if (f->tunnelhdr == 0)
383 continue;
384 data->tgenerator = f->res.classid;
385 tunnel_bts(data);
391 memcpy(data->tmap, tmap, sizeof(tmap));
394 static u32 gen_tunnel(struct rsvp_head *data)
396 int i, k;
398 for (k = 0; k < 2; k++) {
399 for (i = 255; i > 0; i--) {
400 if (++data->tgenerator == 0)
401 data->tgenerator = 1;
402 if (tunnel_bts(data))
403 return data->tgenerator;
405 tunnel_recycle(data);
407 return 0;
410 static const struct nla_policy rsvp_policy[TCA_RSVP_MAX + 1] = {
411 [TCA_RSVP_CLASSID] = { .type = NLA_U32 },
412 [TCA_RSVP_DST] = { .type = NLA_BINARY,
413 .len = RSVP_DST_LEN * sizeof(u32) },
414 [TCA_RSVP_SRC] = { .type = NLA_BINARY,
415 .len = RSVP_DST_LEN * sizeof(u32) },
416 [TCA_RSVP_PINFO] = { .len = sizeof(struct tc_rsvp_pinfo) },
419 static int rsvp_change(struct tcf_proto *tp, unsigned long base,
420 u32 handle,
421 struct nlattr **tca,
422 unsigned long *arg)
424 struct rsvp_head *data = tp->root;
425 struct rsvp_filter *f, **fp;
426 struct rsvp_session *s, **sp;
427 struct tc_rsvp_pinfo *pinfo = NULL;
428 struct nlattr *opt = tca[TCA_OPTIONS];
429 struct nlattr *tb[TCA_RSVP_MAX + 1];
430 struct tcf_exts e;
431 unsigned int h1, h2;
432 __be32 *dst;
433 int err;
435 if (opt == NULL)
436 return handle ? -EINVAL : 0;
438 err = nla_parse_nested(tb, TCA_RSVP_MAX, opt, rsvp_policy);
439 if (err < 0)
440 return err;
442 err = tcf_exts_validate(tp, tb, tca[TCA_RATE], &e, &rsvp_ext_map);
443 if (err < 0)
444 return err;
446 f = (struct rsvp_filter *)*arg;
447 if (f) {
448 /* Node exists: adjust only classid */
450 if (f->handle != handle && handle)
451 goto errout2;
452 if (tb[TCA_RSVP_CLASSID]) {
453 f->res.classid = nla_get_u32(tb[TCA_RSVP_CLASSID]);
454 tcf_bind_filter(tp, &f->res, base);
457 tcf_exts_change(tp, &f->exts, &e);
458 return 0;
461 /* Now more serious part... */
462 err = -EINVAL;
463 if (handle)
464 goto errout2;
465 if (tb[TCA_RSVP_DST] == NULL)
466 goto errout2;
468 err = -ENOBUFS;
469 f = kzalloc(sizeof(struct rsvp_filter), GFP_KERNEL);
470 if (f == NULL)
471 goto errout2;
473 h2 = 16;
474 if (tb[TCA_RSVP_SRC]) {
475 memcpy(f->src, nla_data(tb[TCA_RSVP_SRC]), sizeof(f->src));
476 h2 = hash_src(f->src);
478 if (tb[TCA_RSVP_PINFO]) {
479 pinfo = nla_data(tb[TCA_RSVP_PINFO]);
480 f->spi = pinfo->spi;
481 f->tunnelhdr = pinfo->tunnelhdr;
483 if (tb[TCA_RSVP_CLASSID])
484 f->res.classid = nla_get_u32(tb[TCA_RSVP_CLASSID]);
486 dst = nla_data(tb[TCA_RSVP_DST]);
487 h1 = hash_dst(dst, pinfo ? pinfo->protocol : 0, pinfo ? pinfo->tunnelid : 0);
489 err = -ENOMEM;
490 if ((f->handle = gen_handle(tp, h1 | (h2<<8))) == 0)
491 goto errout;
493 if (f->tunnelhdr) {
494 err = -EINVAL;
495 if (f->res.classid > 255)
496 goto errout;
498 err = -ENOMEM;
499 if (f->res.classid == 0 &&
500 (f->res.classid = gen_tunnel(data)) == 0)
501 goto errout;
504 for (sp = &data->ht[h1]; (s = *sp) != NULL; sp = &s->next) {
505 if (dst[RSVP_DST_LEN-1] == s->dst[RSVP_DST_LEN-1] &&
506 pinfo && pinfo->protocol == s->protocol &&
507 memcmp(&pinfo->dpi, &s->dpi, sizeof(s->dpi)) == 0 &&
508 #if RSVP_DST_LEN == 4
509 dst[0] == s->dst[0] &&
510 dst[1] == s->dst[1] &&
511 dst[2] == s->dst[2] &&
512 #endif
513 pinfo->tunnelid == s->tunnelid) {
515 insert:
516 /* OK, we found appropriate session */
518 fp = &s->ht[h2];
520 f->sess = s;
521 if (f->tunnelhdr == 0)
522 tcf_bind_filter(tp, &f->res, base);
524 tcf_exts_change(tp, &f->exts, &e);
526 for (fp = &s->ht[h2]; *fp; fp = &(*fp)->next)
527 if (((*fp)->spi.mask & f->spi.mask) != f->spi.mask)
528 break;
529 f->next = *fp;
530 wmb();
531 *fp = f;
533 *arg = (unsigned long)f;
534 return 0;
538 /* No session found. Create new one. */
540 err = -ENOBUFS;
541 s = kzalloc(sizeof(struct rsvp_session), GFP_KERNEL);
542 if (s == NULL)
543 goto errout;
544 memcpy(s->dst, dst, sizeof(s->dst));
546 if (pinfo) {
547 s->dpi = pinfo->dpi;
548 s->protocol = pinfo->protocol;
549 s->tunnelid = pinfo->tunnelid;
551 for (sp = &data->ht[h1]; *sp; sp = &(*sp)->next) {
552 if (((*sp)->dpi.mask&s->dpi.mask) != s->dpi.mask)
553 break;
555 s->next = *sp;
556 wmb();
557 *sp = s;
559 goto insert;
561 errout:
562 kfree(f);
563 errout2:
564 tcf_exts_destroy(tp, &e);
565 return err;
568 static void rsvp_walk(struct tcf_proto *tp, struct tcf_walker *arg)
570 struct rsvp_head *head = tp->root;
571 unsigned int h, h1;
573 if (arg->stop)
574 return;
576 for (h = 0; h < 256; h++) {
577 struct rsvp_session *s;
579 for (s = head->ht[h]; s; s = s->next) {
580 for (h1 = 0; h1 <= 16; h1++) {
581 struct rsvp_filter *f;
583 for (f = s->ht[h1]; f; f = f->next) {
584 if (arg->count < arg->skip) {
585 arg->count++;
586 continue;
588 if (arg->fn(tp, (unsigned long)f, arg) < 0) {
589 arg->stop = 1;
590 return;
592 arg->count++;
599 static int rsvp_dump(struct tcf_proto *tp, unsigned long fh,
600 struct sk_buff *skb, struct tcmsg *t)
602 struct rsvp_filter *f = (struct rsvp_filter *)fh;
603 struct rsvp_session *s;
604 unsigned char *b = skb_tail_pointer(skb);
605 struct nlattr *nest;
606 struct tc_rsvp_pinfo pinfo;
608 if (f == NULL)
609 return skb->len;
610 s = f->sess;
612 t->tcm_handle = f->handle;
614 nest = nla_nest_start(skb, TCA_OPTIONS);
615 if (nest == NULL)
616 goto nla_put_failure;
618 NLA_PUT(skb, TCA_RSVP_DST, sizeof(s->dst), &s->dst);
619 pinfo.dpi = s->dpi;
620 pinfo.spi = f->spi;
621 pinfo.protocol = s->protocol;
622 pinfo.tunnelid = s->tunnelid;
623 pinfo.tunnelhdr = f->tunnelhdr;
624 pinfo.pad = 0;
625 NLA_PUT(skb, TCA_RSVP_PINFO, sizeof(pinfo), &pinfo);
626 if (f->res.classid)
627 NLA_PUT_U32(skb, TCA_RSVP_CLASSID, f->res.classid);
628 if (((f->handle >> 8) & 0xFF) != 16)
629 NLA_PUT(skb, TCA_RSVP_SRC, sizeof(f->src), f->src);
631 if (tcf_exts_dump(skb, &f->exts, &rsvp_ext_map) < 0)
632 goto nla_put_failure;
634 nla_nest_end(skb, nest);
636 if (tcf_exts_dump_stats(skb, &f->exts, &rsvp_ext_map) < 0)
637 goto nla_put_failure;
638 return skb->len;
640 nla_put_failure:
641 nlmsg_trim(skb, b);
642 return -1;
645 static struct tcf_proto_ops RSVP_OPS __read_mostly = {
646 .kind = RSVP_ID,
647 .classify = rsvp_classify,
648 .init = rsvp_init,
649 .destroy = rsvp_destroy,
650 .get = rsvp_get,
651 .put = rsvp_put,
652 .change = rsvp_change,
653 .delete = rsvp_delete,
654 .walk = rsvp_walk,
655 .dump = rsvp_dump,
656 .owner = THIS_MODULE,
659 static int __init init_rsvp(void)
661 return register_tcf_proto_ops(&RSVP_OPS);
664 static void __exit exit_rsvp(void)
666 unregister_tcf_proto_ops(&RSVP_OPS);
669 module_init(init_rsvp)
670 module_exit(exit_rsvp)