thp: change pmd_trans_huge_lock() interface to return ptl
[linux/fpc-iii.git] / net / sched / cls_rsvp.h
blobf9c9fc075fe65402c531ada58be9e4b916d282cd
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 __rcu *ht[256];
74 struct rcu_head rcu;
77 struct rsvp_session {
78 struct rsvp_session __rcu *next;
79 __be32 dst[RSVP_DST_LEN];
80 struct tc_rsvp_gpi dpi;
81 u8 protocol;
82 u8 tunnelid;
83 /* 16 (src,sport) hash slots, and one wildcard source slot */
84 struct rsvp_filter __rcu *ht[16 + 1];
85 struct rcu_head rcu;
89 struct rsvp_filter {
90 struct rsvp_filter __rcu *next;
91 __be32 src[RSVP_DST_LEN];
92 struct tc_rsvp_gpi spi;
93 u8 tunnelhdr;
95 struct tcf_result res;
96 struct tcf_exts exts;
98 u32 handle;
99 struct rsvp_session *sess;
100 struct rcu_head rcu;
103 static inline unsigned int hash_dst(__be32 *dst, u8 protocol, u8 tunnelid)
105 unsigned int h = (__force __u32)dst[RSVP_DST_LEN - 1];
107 h ^= h>>16;
108 h ^= h>>8;
109 return (h ^ protocol ^ tunnelid) & 0xFF;
112 static inline unsigned int hash_src(__be32 *src)
114 unsigned int h = (__force __u32)src[RSVP_DST_LEN-1];
116 h ^= h>>16;
117 h ^= h>>8;
118 h ^= h>>4;
119 return h & 0xF;
122 #define RSVP_APPLY_RESULT() \
124 int r = tcf_exts_exec(skb, &f->exts, res); \
125 if (r < 0) \
126 continue; \
127 else if (r > 0) \
128 return r; \
131 static int rsvp_classify(struct sk_buff *skb, const struct tcf_proto *tp,
132 struct tcf_result *res)
134 struct rsvp_head *head = rcu_dereference_bh(tp->root);
135 struct rsvp_session *s;
136 struct rsvp_filter *f;
137 unsigned int h1, h2;
138 __be32 *dst, *src;
139 u8 protocol;
140 u8 tunnelid = 0;
141 u8 *xprt;
142 #if RSVP_DST_LEN == 4
143 struct ipv6hdr *nhptr;
145 if (!pskb_network_may_pull(skb, sizeof(*nhptr)))
146 return -1;
147 nhptr = ipv6_hdr(skb);
148 #else
149 struct iphdr *nhptr;
151 if (!pskb_network_may_pull(skb, sizeof(*nhptr)))
152 return -1;
153 nhptr = ip_hdr(skb);
154 #endif
156 restart:
158 #if RSVP_DST_LEN == 4
159 src = &nhptr->saddr.s6_addr32[0];
160 dst = &nhptr->daddr.s6_addr32[0];
161 protocol = nhptr->nexthdr;
162 xprt = ((u8 *)nhptr) + sizeof(struct ipv6hdr);
163 #else
164 src = &nhptr->saddr;
165 dst = &nhptr->daddr;
166 protocol = nhptr->protocol;
167 xprt = ((u8 *)nhptr) + (nhptr->ihl<<2);
168 if (ip_is_fragment(nhptr))
169 return -1;
170 #endif
172 h1 = hash_dst(dst, protocol, tunnelid);
173 h2 = hash_src(src);
175 for (s = rcu_dereference_bh(head->ht[h1]); s;
176 s = rcu_dereference_bh(s->next)) {
177 if (dst[RSVP_DST_LEN-1] == s->dst[RSVP_DST_LEN - 1] &&
178 protocol == s->protocol &&
179 !(s->dpi.mask &
180 (*(u32 *)(xprt + s->dpi.offset) ^ s->dpi.key)) &&
181 #if RSVP_DST_LEN == 4
182 dst[0] == s->dst[0] &&
183 dst[1] == s->dst[1] &&
184 dst[2] == s->dst[2] &&
185 #endif
186 tunnelid == s->tunnelid) {
188 for (f = rcu_dereference_bh(s->ht[h2]); f;
189 f = rcu_dereference_bh(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 = rcu_dereference_bh(s->ht[16]); f;
214 f = rcu_dereference_bh(f->next)) {
215 *res = f->res;
216 RSVP_APPLY_RESULT();
217 goto matched;
219 return -1;
222 return -1;
225 static void rsvp_replace(struct tcf_proto *tp, struct rsvp_filter *n, u32 h)
227 struct rsvp_head *head = rtnl_dereference(tp->root);
228 struct rsvp_session *s;
229 struct rsvp_filter __rcu **ins;
230 struct rsvp_filter *pins;
231 unsigned int h1 = h & 0xFF;
232 unsigned int h2 = (h >> 8) & 0xFF;
234 for (s = rtnl_dereference(head->ht[h1]); s;
235 s = rtnl_dereference(s->next)) {
236 for (ins = &s->ht[h2], pins = rtnl_dereference(*ins); ;
237 ins = &pins->next, pins = rtnl_dereference(*ins)) {
238 if (pins->handle == h) {
239 RCU_INIT_POINTER(n->next, pins->next);
240 rcu_assign_pointer(*ins, n);
241 return;
246 /* Something went wrong if we are trying to replace a non-existant
247 * node. Mind as well halt instead of silently failing.
249 BUG_ON(1);
252 static unsigned long rsvp_get(struct tcf_proto *tp, u32 handle)
254 struct rsvp_head *head = rtnl_dereference(tp->root);
255 struct rsvp_session *s;
256 struct rsvp_filter *f;
257 unsigned int h1 = handle & 0xFF;
258 unsigned int h2 = (handle >> 8) & 0xFF;
260 if (h2 > 16)
261 return 0;
263 for (s = rtnl_dereference(head->ht[h1]); s;
264 s = rtnl_dereference(s->next)) {
265 for (f = rtnl_dereference(s->ht[h2]); f;
266 f = rtnl_dereference(f->next)) {
267 if (f->handle == handle)
268 return (unsigned long)f;
271 return 0;
274 static int rsvp_init(struct tcf_proto *tp)
276 struct rsvp_head *data;
278 data = kzalloc(sizeof(struct rsvp_head), GFP_KERNEL);
279 if (data) {
280 rcu_assign_pointer(tp->root, data);
281 return 0;
283 return -ENOBUFS;
286 static void rsvp_delete_filter_rcu(struct rcu_head *head)
288 struct rsvp_filter *f = container_of(head, struct rsvp_filter, rcu);
290 tcf_exts_destroy(&f->exts);
291 kfree(f);
294 static void rsvp_delete_filter(struct tcf_proto *tp, struct rsvp_filter *f)
296 tcf_unbind_filter(tp, &f->res);
297 /* all classifiers are required to call tcf_exts_destroy() after rcu
298 * grace period, since converted-to-rcu actions are relying on that
299 * in cleanup() callback
301 call_rcu(&f->rcu, rsvp_delete_filter_rcu);
304 static bool rsvp_destroy(struct tcf_proto *tp, bool force)
306 struct rsvp_head *data = rtnl_dereference(tp->root);
307 int h1, h2;
309 if (data == NULL)
310 return true;
312 if (!force) {
313 for (h1 = 0; h1 < 256; h1++) {
314 if (rcu_access_pointer(data->ht[h1]))
315 return false;
319 RCU_INIT_POINTER(tp->root, NULL);
321 for (h1 = 0; h1 < 256; h1++) {
322 struct rsvp_session *s;
324 while ((s = rtnl_dereference(data->ht[h1])) != NULL) {
325 RCU_INIT_POINTER(data->ht[h1], s->next);
327 for (h2 = 0; h2 <= 16; h2++) {
328 struct rsvp_filter *f;
330 while ((f = rtnl_dereference(s->ht[h2])) != NULL) {
331 rcu_assign_pointer(s->ht[h2], f->next);
332 rsvp_delete_filter(tp, f);
335 kfree_rcu(s, rcu);
338 kfree_rcu(data, rcu);
339 return true;
342 static int rsvp_delete(struct tcf_proto *tp, unsigned long arg)
344 struct rsvp_head *head = rtnl_dereference(tp->root);
345 struct rsvp_filter *nfp, *f = (struct rsvp_filter *)arg;
346 struct rsvp_filter __rcu **fp;
347 unsigned int h = f->handle;
348 struct rsvp_session __rcu **sp;
349 struct rsvp_session *nsp, *s = f->sess;
350 int i;
352 fp = &s->ht[(h >> 8) & 0xFF];
353 for (nfp = rtnl_dereference(*fp); nfp;
354 fp = &nfp->next, nfp = rtnl_dereference(*fp)) {
355 if (nfp == f) {
356 RCU_INIT_POINTER(*fp, f->next);
357 rsvp_delete_filter(tp, f);
359 /* Strip tree */
361 for (i = 0; i <= 16; i++)
362 if (s->ht[i])
363 return 0;
365 /* OK, session has no flows */
366 sp = &head->ht[h & 0xFF];
367 for (nsp = rtnl_dereference(*sp); nsp;
368 sp = &nsp->next, nsp = rtnl_dereference(*sp)) {
369 if (nsp == s) {
370 RCU_INIT_POINTER(*sp, s->next);
371 kfree_rcu(s, rcu);
372 return 0;
376 return 0;
379 return 0;
382 static unsigned int gen_handle(struct tcf_proto *tp, unsigned salt)
384 struct rsvp_head *data = rtnl_dereference(tp->root);
385 int i = 0xFFFF;
387 while (i-- > 0) {
388 u32 h;
390 if ((data->hgenerator += 0x10000) == 0)
391 data->hgenerator = 0x10000;
392 h = data->hgenerator|salt;
393 if (rsvp_get(tp, h) == 0)
394 return h;
396 return 0;
399 static int tunnel_bts(struct rsvp_head *data)
401 int n = data->tgenerator >> 5;
402 u32 b = 1 << (data->tgenerator & 0x1F);
404 if (data->tmap[n] & b)
405 return 0;
406 data->tmap[n] |= b;
407 return 1;
410 static void tunnel_recycle(struct rsvp_head *data)
412 struct rsvp_session __rcu **sht = data->ht;
413 u32 tmap[256/32];
414 int h1, h2;
416 memset(tmap, 0, sizeof(tmap));
418 for (h1 = 0; h1 < 256; h1++) {
419 struct rsvp_session *s;
420 for (s = rtnl_dereference(sht[h1]); s;
421 s = rtnl_dereference(s->next)) {
422 for (h2 = 0; h2 <= 16; h2++) {
423 struct rsvp_filter *f;
425 for (f = rtnl_dereference(s->ht[h2]); f;
426 f = rtnl_dereference(f->next)) {
427 if (f->tunnelhdr == 0)
428 continue;
429 data->tgenerator = f->res.classid;
430 tunnel_bts(data);
436 memcpy(data->tmap, tmap, sizeof(tmap));
439 static u32 gen_tunnel(struct rsvp_head *data)
441 int i, k;
443 for (k = 0; k < 2; k++) {
444 for (i = 255; i > 0; i--) {
445 if (++data->tgenerator == 0)
446 data->tgenerator = 1;
447 if (tunnel_bts(data))
448 return data->tgenerator;
450 tunnel_recycle(data);
452 return 0;
455 static const struct nla_policy rsvp_policy[TCA_RSVP_MAX + 1] = {
456 [TCA_RSVP_CLASSID] = { .type = NLA_U32 },
457 [TCA_RSVP_DST] = { .type = NLA_BINARY,
458 .len = RSVP_DST_LEN * sizeof(u32) },
459 [TCA_RSVP_SRC] = { .type = NLA_BINARY,
460 .len = RSVP_DST_LEN * sizeof(u32) },
461 [TCA_RSVP_PINFO] = { .len = sizeof(struct tc_rsvp_pinfo) },
464 static int rsvp_change(struct net *net, struct sk_buff *in_skb,
465 struct tcf_proto *tp, unsigned long base,
466 u32 handle,
467 struct nlattr **tca,
468 unsigned long *arg, bool ovr)
470 struct rsvp_head *data = rtnl_dereference(tp->root);
471 struct rsvp_filter *f, *nfp;
472 struct rsvp_filter __rcu **fp;
473 struct rsvp_session *nsp, *s;
474 struct rsvp_session __rcu **sp;
475 struct tc_rsvp_pinfo *pinfo = NULL;
476 struct nlattr *opt = tca[TCA_OPTIONS];
477 struct nlattr *tb[TCA_RSVP_MAX + 1];
478 struct tcf_exts e;
479 unsigned int h1, h2;
480 __be32 *dst;
481 int err;
483 if (opt == NULL)
484 return handle ? -EINVAL : 0;
486 err = nla_parse_nested(tb, TCA_RSVP_MAX, opt, rsvp_policy);
487 if (err < 0)
488 return err;
490 tcf_exts_init(&e, TCA_RSVP_ACT, TCA_RSVP_POLICE);
491 err = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &e, ovr);
492 if (err < 0)
493 return err;
495 f = (struct rsvp_filter *)*arg;
496 if (f) {
497 /* Node exists: adjust only classid */
498 struct rsvp_filter *n;
500 if (f->handle != handle && handle)
501 goto errout2;
503 n = kmemdup(f, sizeof(*f), GFP_KERNEL);
504 if (!n) {
505 err = -ENOMEM;
506 goto errout2;
509 tcf_exts_init(&n->exts, TCA_RSVP_ACT, TCA_RSVP_POLICE);
511 if (tb[TCA_RSVP_CLASSID]) {
512 n->res.classid = nla_get_u32(tb[TCA_RSVP_CLASSID]);
513 tcf_bind_filter(tp, &n->res, base);
516 tcf_exts_change(tp, &n->exts, &e);
517 rsvp_replace(tp, n, handle);
518 return 0;
521 /* Now more serious part... */
522 err = -EINVAL;
523 if (handle)
524 goto errout2;
525 if (tb[TCA_RSVP_DST] == NULL)
526 goto errout2;
528 err = -ENOBUFS;
529 f = kzalloc(sizeof(struct rsvp_filter), GFP_KERNEL);
530 if (f == NULL)
531 goto errout2;
533 tcf_exts_init(&f->exts, TCA_RSVP_ACT, TCA_RSVP_POLICE);
534 h2 = 16;
535 if (tb[TCA_RSVP_SRC]) {
536 memcpy(f->src, nla_data(tb[TCA_RSVP_SRC]), sizeof(f->src));
537 h2 = hash_src(f->src);
539 if (tb[TCA_RSVP_PINFO]) {
540 pinfo = nla_data(tb[TCA_RSVP_PINFO]);
541 f->spi = pinfo->spi;
542 f->tunnelhdr = pinfo->tunnelhdr;
544 if (tb[TCA_RSVP_CLASSID])
545 f->res.classid = nla_get_u32(tb[TCA_RSVP_CLASSID]);
547 dst = nla_data(tb[TCA_RSVP_DST]);
548 h1 = hash_dst(dst, pinfo ? pinfo->protocol : 0, pinfo ? pinfo->tunnelid : 0);
550 err = -ENOMEM;
551 if ((f->handle = gen_handle(tp, h1 | (h2<<8))) == 0)
552 goto errout;
554 if (f->tunnelhdr) {
555 err = -EINVAL;
556 if (f->res.classid > 255)
557 goto errout;
559 err = -ENOMEM;
560 if (f->res.classid == 0 &&
561 (f->res.classid = gen_tunnel(data)) == 0)
562 goto errout;
565 for (sp = &data->ht[h1];
566 (s = rtnl_dereference(*sp)) != NULL;
567 sp = &s->next) {
568 if (dst[RSVP_DST_LEN-1] == s->dst[RSVP_DST_LEN-1] &&
569 pinfo && pinfo->protocol == s->protocol &&
570 memcmp(&pinfo->dpi, &s->dpi, sizeof(s->dpi)) == 0 &&
571 #if RSVP_DST_LEN == 4
572 dst[0] == s->dst[0] &&
573 dst[1] == s->dst[1] &&
574 dst[2] == s->dst[2] &&
575 #endif
576 pinfo->tunnelid == s->tunnelid) {
578 insert:
579 /* OK, we found appropriate session */
581 fp = &s->ht[h2];
583 f->sess = s;
584 if (f->tunnelhdr == 0)
585 tcf_bind_filter(tp, &f->res, base);
587 tcf_exts_change(tp, &f->exts, &e);
589 fp = &s->ht[h2];
590 for (nfp = rtnl_dereference(*fp); nfp;
591 fp = &nfp->next, nfp = rtnl_dereference(*fp)) {
592 __u32 mask = nfp->spi.mask & f->spi.mask;
594 if (mask != f->spi.mask)
595 break;
597 RCU_INIT_POINTER(f->next, nfp);
598 rcu_assign_pointer(*fp, f);
600 *arg = (unsigned long)f;
601 return 0;
605 /* No session found. Create new one. */
607 err = -ENOBUFS;
608 s = kzalloc(sizeof(struct rsvp_session), GFP_KERNEL);
609 if (s == NULL)
610 goto errout;
611 memcpy(s->dst, dst, sizeof(s->dst));
613 if (pinfo) {
614 s->dpi = pinfo->dpi;
615 s->protocol = pinfo->protocol;
616 s->tunnelid = pinfo->tunnelid;
618 sp = &data->ht[h1];
619 for (nsp = rtnl_dereference(*sp); nsp;
620 sp = &nsp->next, nsp = rtnl_dereference(*sp)) {
621 if ((nsp->dpi.mask & s->dpi.mask) != s->dpi.mask)
622 break;
624 RCU_INIT_POINTER(s->next, nsp);
625 rcu_assign_pointer(*sp, s);
627 goto insert;
629 errout:
630 kfree(f);
631 errout2:
632 tcf_exts_destroy(&e);
633 return err;
636 static void rsvp_walk(struct tcf_proto *tp, struct tcf_walker *arg)
638 struct rsvp_head *head = rtnl_dereference(tp->root);
639 unsigned int h, h1;
641 if (arg->stop)
642 return;
644 for (h = 0; h < 256; h++) {
645 struct rsvp_session *s;
647 for (s = rtnl_dereference(head->ht[h]); s;
648 s = rtnl_dereference(s->next)) {
649 for (h1 = 0; h1 <= 16; h1++) {
650 struct rsvp_filter *f;
652 for (f = rtnl_dereference(s->ht[h1]); f;
653 f = rtnl_dereference(f->next)) {
654 if (arg->count < arg->skip) {
655 arg->count++;
656 continue;
658 if (arg->fn(tp, (unsigned long)f, arg) < 0) {
659 arg->stop = 1;
660 return;
662 arg->count++;
669 static int rsvp_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
670 struct sk_buff *skb, struct tcmsg *t)
672 struct rsvp_filter *f = (struct rsvp_filter *)fh;
673 struct rsvp_session *s;
674 struct nlattr *nest;
675 struct tc_rsvp_pinfo pinfo;
677 if (f == NULL)
678 return skb->len;
679 s = f->sess;
681 t->tcm_handle = f->handle;
683 nest = nla_nest_start(skb, TCA_OPTIONS);
684 if (nest == NULL)
685 goto nla_put_failure;
687 if (nla_put(skb, TCA_RSVP_DST, sizeof(s->dst), &s->dst))
688 goto nla_put_failure;
689 pinfo.dpi = s->dpi;
690 pinfo.spi = f->spi;
691 pinfo.protocol = s->protocol;
692 pinfo.tunnelid = s->tunnelid;
693 pinfo.tunnelhdr = f->tunnelhdr;
694 pinfo.pad = 0;
695 if (nla_put(skb, TCA_RSVP_PINFO, sizeof(pinfo), &pinfo))
696 goto nla_put_failure;
697 if (f->res.classid &&
698 nla_put_u32(skb, TCA_RSVP_CLASSID, f->res.classid))
699 goto nla_put_failure;
700 if (((f->handle >> 8) & 0xFF) != 16 &&
701 nla_put(skb, TCA_RSVP_SRC, sizeof(f->src), f->src))
702 goto nla_put_failure;
704 if (tcf_exts_dump(skb, &f->exts) < 0)
705 goto nla_put_failure;
707 nla_nest_end(skb, nest);
709 if (tcf_exts_dump_stats(skb, &f->exts) < 0)
710 goto nla_put_failure;
711 return skb->len;
713 nla_put_failure:
714 nla_nest_cancel(skb, nest);
715 return -1;
718 static struct tcf_proto_ops RSVP_OPS __read_mostly = {
719 .kind = RSVP_ID,
720 .classify = rsvp_classify,
721 .init = rsvp_init,
722 .destroy = rsvp_destroy,
723 .get = rsvp_get,
724 .change = rsvp_change,
725 .delete = rsvp_delete,
726 .walk = rsvp_walk,
727 .dump = rsvp_dump,
728 .owner = THIS_MODULE,
731 static int __init init_rsvp(void)
733 return register_tcf_proto_ops(&RSVP_OPS);
736 static void __exit exit_rsvp(void)
738 unregister_tcf_proto_ops(&RSVP_OPS);
741 module_init(init_rsvp)
742 module_exit(exit_rsvp)