?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
phy: mdio-octeon: Use devm_mdiobus_alloc_size()
[linux-2.6/btrfs-unstable.git] / net / sched / cls_u32.c
blob4fbb67430ce4821fd66576132afec2d7a301b379
1 /*
2 * net/sched/cls_u32.c Ugly (or Universal) 32bit key Packet Classifier.
3 *
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.
8 *
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 *
11 * The filters are packed to hash tables of key nodes
12 * with a set of 32bit key/mask pairs at every node.
13 * Nodes reference next level hash tables etc.
14 *
15 * This scheme is the best universal classifier I managed to
16 * invent; it is not super-fast, but it is not slow (provided you
17 * program it correctly), and general enough. And its relative
18 * speed grows as the number of rules becomes larger.
19 *
20 * It seems that it represents the best middle point between
21 * speed and manageability both by human and by machine.
22 *
23 * It is especially useful for link sharing combined with QoS;
24 * pure RSVP doesn't need such a general approach and can use
25 * much simpler (and faster) schemes, sort of cls_rsvp.c.
26 *
27 * JHS: We should remove the CONFIG_NET_CLS_IND from here
28 * eventually when the meta match extension is made available
29 *
30 * nfmark match added by Catalin(ux aka Dino) BOIE <catab at umbrella.ro>
31 */
32
33 #include <linux/module.h>
34 #include <linux/slab.h>
35 #include <linux/types.h>
36 #include <linux/kernel.h>
37 #include <linux/string.h>
38 #include <linux/errno.h>
39 #include <linux/percpu.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/skbuff.h>
42 #include <linux/bitmap.h>
43 #include <net/netlink.h>
44 #include <net/act_api.h>
45 #include <net/pkt_cls.h>
46
47 struct tc_u_knode {
48 struct tc_u_knode __rcu *next;
49 u32 handle;
50 struct tc_u_hnode __rcu *ht_up;
51 struct tcf_exts exts;
52 #ifdef CONFIG_NET_CLS_IND
53 int ifindex;
54 #endif
55 u8 fshift;
56 struct tcf_result res;
57 struct tc_u_hnode __rcu *ht_down;
58 #ifdef CONFIG_CLS_U32_PERF
59 struct tc_u32_pcnt __percpu *pf;
60 #endif
61 #ifdef CONFIG_CLS_U32_MARK
62 u32 val;
63 u32 mask;
64 u32 __percpu *pcpu_success;
65 #endif
66 struct tcf_proto *tp;
67 struct rcu_head rcu;
68 /* The 'sel' field MUST be the last field in structure to allow for
69 * tc_u32_keys allocated at end of structure.
70 */
71 struct tc_u32_sel sel;
72 };
73
74 struct tc_u_hnode {
75 struct tc_u_hnode __rcu *next;
76 u32 handle;
77 u32 prio;
78 struct tc_u_common *tp_c;
79 int refcnt;
80 unsigned int divisor;
81 struct rcu_head rcu;
82 /* The 'ht' field MUST be the last field in structure to allow for
83 * more entries allocated at end of structure.
84 */
85 struct tc_u_knode __rcu *ht[1];
86 };
87
88 struct tc_u_common {
89 struct tc_u_hnode __rcu *hlist;
90 struct Qdisc *q;
91 int refcnt;
92 u32 hgenerator;
93 struct rcu_head rcu;
94 };
95
96 static inline unsigned int u32_hash_fold(__be32 key,
97 const struct tc_u32_sel *sel,
98 u8 fshift)
99 {
100 unsigned int h = ntohl(key & sel->hmask) >> fshift;
101
102 return h;
103 }
104
105 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp, struct tcf_result *res)
106 {
107 struct {
108 struct tc_u_knode *knode;
109 unsigned int off;
110 } stack[TC_U32_MAXDEPTH];
111
112 struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
113 unsigned int off = skb_network_offset(skb);
114 struct tc_u_knode *n;
115 int sdepth = 0;
116 int off2 = 0;
117 int sel = 0;
118 #ifdef CONFIG_CLS_U32_PERF
119 int j;
120 #endif
121 int i, r;
122
123 next_ht:
124 n = rcu_dereference_bh(ht->ht[sel]);
125
126 next_knode:
127 if (n) {
128 struct tc_u32_key *key = n->sel.keys;
129
130 #ifdef CONFIG_CLS_U32_PERF
131 __this_cpu_inc(n->pf->rcnt);
132 j = 0;
133 #endif
134
135 #ifdef CONFIG_CLS_U32_MARK
136 if ((skb->mark & n->mask) != n->val) {
137 n = rcu_dereference_bh(n->next);
138 goto next_knode;
139 } else {
140 __this_cpu_inc(*n->pcpu_success);
141 }
142 #endif
143
144 for (i = n->sel.nkeys; i > 0; i--, key++) {
145 int toff = off + key->off + (off2 & key->offmask);
146 __be32 *data, hdata;
147
148 if (skb_headroom(skb) + toff > INT_MAX)
149 goto out;
150
151 data = skb_header_pointer(skb, toff, 4, &hdata);
152 if (!data)
153 goto out;
154 if ((*data ^ key->val) & key->mask) {
155 n = rcu_dereference_bh(n->next);
156 goto next_knode;
157 }
158 #ifdef CONFIG_CLS_U32_PERF
159 __this_cpu_inc(n->pf->kcnts[j]);
160 j++;
161 #endif
162 }
163
164 ht = rcu_dereference_bh(n->ht_down);
165 if (!ht) {
166 check_terminal:
167 if (n->sel.flags & TC_U32_TERMINAL) {
168
169 *res = n->res;
170 #ifdef CONFIG_NET_CLS_IND
171 if (!tcf_match_indev(skb, n->ifindex)) {
172 n = rcu_dereference_bh(n->next);
173 goto next_knode;
174 }
175 #endif
176 #ifdef CONFIG_CLS_U32_PERF
177 __this_cpu_inc(n->pf->rhit);
178 #endif
179 r = tcf_exts_exec(skb, &n->exts, res);
180 if (r < 0) {
181 n = rcu_dereference_bh(n->next);
182 goto next_knode;
183 }
184
185 return r;
186 }
187 n = rcu_dereference_bh(n->next);
188 goto next_knode;
189 }
190
191 /* PUSH */
192 if (sdepth >= TC_U32_MAXDEPTH)
193 goto deadloop;
194 stack[sdepth].knode = n;
195 stack[sdepth].off = off;
196 sdepth++;
197
198 ht = rcu_dereference_bh(n->ht_down);
199 sel = 0;
200 if (ht->divisor) {
201 __be32 *data, hdata;
202
203 data = skb_header_pointer(skb, off + n->sel.hoff, 4,
204 &hdata);
205 if (!data)
206 goto out;
207 sel = ht->divisor & u32_hash_fold(*data, &n->sel,
208 n->fshift);
209 }
210 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
211 goto next_ht;
212
213 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
214 off2 = n->sel.off + 3;
215 if (n->sel.flags & TC_U32_VAROFFSET) {
216 __be16 *data, hdata;
217
218 data = skb_header_pointer(skb,
219 off + n->sel.offoff,
220 2, &hdata);
221 if (!data)
222 goto out;
223 off2 += ntohs(n->sel.offmask & *data) >>
224 n->sel.offshift;
225 }
226 off2 &= ~3;
227 }
228 if (n->sel.flags & TC_U32_EAT) {
229 off += off2;
230 off2 = 0;
231 }
232
233 if (off < skb->len)
234 goto next_ht;
235 }
236
237 /* POP */
238 if (sdepth--) {
239 n = stack[sdepth].knode;
240 ht = rcu_dereference_bh(n->ht_up);
241 off = stack[sdepth].off;
242 goto check_terminal;
243 }
244 out:
245 return -1;
246
247 deadloop:
248 net_warn_ratelimited("cls_u32: dead loop\n");
249 return -1;
250 }
251
252 static struct tc_u_hnode *
253 u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
254 {
255 struct tc_u_hnode *ht;
256
257 for (ht = rtnl_dereference(tp_c->hlist);
258 ht;
259 ht = rtnl_dereference(ht->next))
260 if (ht->handle == handle)
261 break;
262
263 return ht;
264 }
265
266 static struct tc_u_knode *
267 u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
268 {
269 unsigned int sel;
270 struct tc_u_knode *n = NULL;
271
272 sel = TC_U32_HASH(handle);
273 if (sel > ht->divisor)
274 goto out;
275
276 for (n = rtnl_dereference(ht->ht[sel]);
277 n;
278 n = rtnl_dereference(n->next))
279 if (n->handle == handle)
280 break;
281 out:
282 return n;
283 }
284
285
286 static unsigned long u32_get(struct tcf_proto *tp, u32 handle)
287 {
288 struct tc_u_hnode *ht;
289 struct tc_u_common *tp_c = tp->data;
290
291 if (TC_U32_HTID(handle) == TC_U32_ROOT)
292 ht = rtnl_dereference(tp->root);
293 else
294 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
295
296 if (!ht)
297 return 0;
298
299 if (TC_U32_KEY(handle) == 0)
300 return (unsigned long)ht;
301
302 return (unsigned long)u32_lookup_key(ht, handle);
303 }
304
305 static u32 gen_new_htid(struct tc_u_common *tp_c)
306 {
307 int i = 0x800;
308
309 /* hgenerator only used inside rtnl lock it is safe to increment
310 * without read _copy_ update semantics
311 */
312 do {
313 if (++tp_c->hgenerator == 0x7FF)
314 tp_c->hgenerator = 1;
315 } while (--i > 0 && u32_lookup_ht(tp_c, (tp_c->hgenerator|0x800)<<20));
316
317 return i > 0 ? (tp_c->hgenerator|0x800)<<20 : 0;
318 }
319
320 static int u32_init(struct tcf_proto *tp)
321 {
322 struct tc_u_hnode *root_ht;
323 struct tc_u_common *tp_c;
324
325 tp_c = tp->q->u32_node;
326
327 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
328 if (root_ht == NULL)
329 return -ENOBUFS;
330
331 root_ht->divisor = 0;
332 root_ht->refcnt++;
333 root_ht->handle = tp_c ? gen_new_htid(tp_c) : 0x80000000;
334 root_ht->prio = tp->prio;
335
336 if (tp_c == NULL) {
337 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
338 if (tp_c == NULL) {
339 kfree(root_ht);
340 return -ENOBUFS;
341 }
342 tp_c->q = tp->q;
343 tp->q->u32_node = tp_c;
344 }
345
346 tp_c->refcnt++;
347 RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
348 rcu_assign_pointer(tp_c->hlist, root_ht);
349 root_ht->tp_c = tp_c;
350
351 rcu_assign_pointer(tp->root, root_ht);
352 tp->data = tp_c;
353 return 0;
354 }
355
356 static int u32_destroy_key(struct tcf_proto *tp,
357 struct tc_u_knode *n,
358 bool free_pf)
359 {
360 tcf_exts_destroy(&n->exts);
361 if (n->ht_down)
362 n->ht_down->refcnt--;
363 #ifdef CONFIG_CLS_U32_PERF
364 if (free_pf)
365 free_percpu(n->pf);
366 #endif
367 #ifdef CONFIG_CLS_U32_MARK
368 if (free_pf)
369 free_percpu(n->pcpu_success);
370 #endif
371 kfree(n);
372 return 0;
373 }
374
375 /* u32_delete_key_rcu should be called when free'ing a copied
376 * version of a tc_u_knode obtained from u32_init_knode(). When
377 * copies are obtained from u32_init_knode() the statistics are
378 * shared between the old and new copies to allow readers to
379 * continue to update the statistics during the copy. To support
380 * this the u32_delete_key_rcu variant does not free the percpu
381 * statistics.
382 */
383 static void u32_delete_key_rcu(struct rcu_head *rcu)
384 {
385 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
386
387 u32_destroy_key(key->tp, key, false);
388 }
389
390 /* u32_delete_key_freepf_rcu is the rcu callback variant
391 * that free's the entire structure including the statistics
392 * percpu variables. Only use this if the key is not a copy
393 * returned by u32_init_knode(). See u32_delete_key_rcu()
394 * for the variant that should be used with keys return from
395 * u32_init_knode()
396 */
397 static void u32_delete_key_freepf_rcu(struct rcu_head *rcu)
398 {
399 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
400
401 u32_destroy_key(key->tp, key, true);
402 }
403
404 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
405 {
406 struct tc_u_knode __rcu **kp;
407 struct tc_u_knode *pkp;
408 struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
409
410 if (ht) {
411 kp = &ht->ht[TC_U32_HASH(key->handle)];
412 for (pkp = rtnl_dereference(*kp); pkp;
413 kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
414 if (pkp == key) {
415 RCU_INIT_POINTER(*kp, key->next);
416
417 tcf_unbind_filter(tp, &key->res);
418 call_rcu(&key->rcu, u32_delete_key_freepf_rcu);
419 return 0;
420 }
421 }
422 }
423 WARN_ON(1);
424 return 0;
425 }
426
427 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
428 {
429 struct tc_u_knode *n;
430 unsigned int h;
431
432 for (h = 0; h <= ht->divisor; h++) {
433 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
434 RCU_INIT_POINTER(ht->ht[h],
435 rtnl_dereference(n->next));
436 tcf_unbind_filter(tp, &n->res);
437 call_rcu(&n->rcu, u32_delete_key_freepf_rcu);
438 }
439 }
440 }
441
442 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
443 {
444 struct tc_u_common *tp_c = tp->data;
445 struct tc_u_hnode __rcu **hn;
446 struct tc_u_hnode *phn;
447
448 WARN_ON(ht->refcnt);
449
450 u32_clear_hnode(tp, ht);
451
452 hn = &tp_c->hlist;
453 for (phn = rtnl_dereference(*hn);
454 phn;
455 hn = &phn->next, phn = rtnl_dereference(*hn)) {
456 if (phn == ht) {
457 RCU_INIT_POINTER(*hn, ht->next);
458 kfree_rcu(ht, rcu);
459 return 0;
460 }
461 }
462
463 return -ENOENT;
464 }
465
466 static bool ht_empty(struct tc_u_hnode *ht)
467 {
468 unsigned int h;
469
470 for (h = 0; h <= ht->divisor; h++)
471 if (rcu_access_pointer(ht->ht[h]))
472 return false;
473
474 return true;
475 }
476
477 static bool u32_destroy(struct tcf_proto *tp, bool force)
478 {
479 struct tc_u_common *tp_c = tp->data;
480 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
481
482 WARN_ON(root_ht == NULL);
483
484 if (!force) {
485 if (root_ht) {
486 if (root_ht->refcnt > 1)
487 return false;
488 if (root_ht->refcnt == 1) {
489 if (!ht_empty(root_ht))
490 return false;
491 }
492 }
493
494 if (tp_c->refcnt > 1)
495 return false;
496
497 if (tp_c->refcnt == 1) {
498 struct tc_u_hnode *ht;
499
500 for (ht = rtnl_dereference(tp_c->hlist);
501 ht;
502 ht = rtnl_dereference(ht->next))
503 if (!ht_empty(ht))
504 return false;
505 }
506 }
507
508 if (root_ht && --root_ht->refcnt == 0)
509 u32_destroy_hnode(tp, root_ht);
510
511 if (--tp_c->refcnt == 0) {
512 struct tc_u_hnode *ht;
513
514 tp->q->u32_node = NULL;
515
516 for (ht = rtnl_dereference(tp_c->hlist);
517 ht;
518 ht = rtnl_dereference(ht->next)) {
519 ht->refcnt--;
520 u32_clear_hnode(tp, ht);
521 }
522
523 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
524 RCU_INIT_POINTER(tp_c->hlist, ht->next);
525 kfree_rcu(ht, rcu);
526 }
527
528 kfree(tp_c);
529 }
530
531 tp->data = NULL;
532 return true;
533 }
534
535 static int u32_delete(struct tcf_proto *tp, unsigned long arg)
536 {
537 struct tc_u_hnode *ht = (struct tc_u_hnode *)arg;
538 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
539
540 if (ht == NULL)
541 return 0;
542
543 if (TC_U32_KEY(ht->handle))
544 return u32_delete_key(tp, (struct tc_u_knode *)ht);
545
546 if (root_ht == ht)
547 return -EINVAL;
548
549 if (ht->refcnt == 1) {
550 ht->refcnt--;
551 u32_destroy_hnode(tp, ht);
552 } else {
553 return -EBUSY;
554 }
555
556 return 0;
557 }
558
559 #define NR_U32_NODE (1<<12)
560 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 handle)
561 {
562 struct tc_u_knode *n;
563 unsigned long i;
564 unsigned long *bitmap = kzalloc(BITS_TO_LONGS(NR_U32_NODE) * sizeof(unsigned long),
565 GFP_KERNEL);
566 if (!bitmap)
567 return handle | 0xFFF;
568
569 for (n = rtnl_dereference(ht->ht[TC_U32_HASH(handle)]);
570 n;
571 n = rtnl_dereference(n->next))
572 set_bit(TC_U32_NODE(n->handle), bitmap);
573
574 i = find_next_zero_bit(bitmap, NR_U32_NODE, 0x800);
575 if (i >= NR_U32_NODE)
576 i = find_next_zero_bit(bitmap, NR_U32_NODE, 1);
577
578 kfree(bitmap);
579 return handle | (i >= NR_U32_NODE ? 0xFFF : i);
580 }
581
582 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
583 [TCA_U32_CLASSID] = { .type = NLA_U32 },
584 [TCA_U32_HASH] = { .type = NLA_U32 },
585 [TCA_U32_LINK] = { .type = NLA_U32 },
586 [TCA_U32_DIVISOR] = { .type = NLA_U32 },
587 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) },
588 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ },
589 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) },
590 };
591
592 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
593 unsigned long base, struct tc_u_hnode *ht,
594 struct tc_u_knode *n, struct nlattr **tb,
595 struct nlattr *est, bool ovr)
596 {
597 int err;
598 struct tcf_exts e;
599
600 tcf_exts_init(&e, TCA_U32_ACT, TCA_U32_POLICE);
601 err = tcf_exts_validate(net, tp, tb, est, &e, ovr);
602 if (err < 0)
603 return err;
604
605 err = -EINVAL;
606 if (tb[TCA_U32_LINK]) {
607 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
608 struct tc_u_hnode *ht_down = NULL, *ht_old;
609
610 if (TC_U32_KEY(handle))
611 goto errout;
612
613 if (handle) {
614 ht_down = u32_lookup_ht(ht->tp_c, handle);
615
616 if (ht_down == NULL)
617 goto errout;
618 ht_down->refcnt++;
619 }
620
621 ht_old = rtnl_dereference(n->ht_down);
622 rcu_assign_pointer(n->ht_down, ht_down);
623
624 if (ht_old)
625 ht_old->refcnt--;
626 }
627 if (tb[TCA_U32_CLASSID]) {
628 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
629 tcf_bind_filter(tp, &n->res, base);
630 }
631
632 #ifdef CONFIG_NET_CLS_IND
633 if (tb[TCA_U32_INDEV]) {
634 int ret;
635 ret = tcf_change_indev(net, tb[TCA_U32_INDEV]);
636 if (ret < 0)
637 goto errout;
638 n->ifindex = ret;
639 }
640 #endif
641 tcf_exts_change(tp, &n->exts, &e);
642
643 return 0;
644 errout:
645 tcf_exts_destroy(&e);
646 return err;
647 }
648
649 static void u32_replace_knode(struct tcf_proto *tp,
650 struct tc_u_common *tp_c,
651 struct tc_u_knode *n)
652 {
653 struct tc_u_knode __rcu **ins;
654 struct tc_u_knode *pins;
655 struct tc_u_hnode *ht;
656
657 if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
658 ht = rtnl_dereference(tp->root);
659 else
660 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
661
662 ins = &ht->ht[TC_U32_HASH(n->handle)];
663
664 /* The node must always exist for it to be replaced if this is not the
665 * case then something went very wrong elsewhere.
666 */
667 for (pins = rtnl_dereference(*ins); ;
668 ins = &pins->next, pins = rtnl_dereference(*ins))
669 if (pins->handle == n->handle)
670 break;
671
672 RCU_INIT_POINTER(n->next, pins->next);
673 rcu_assign_pointer(*ins, n);
674 }
675
676 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp,
677 struct tc_u_knode *n)
678 {
679 struct tc_u_knode *new;
680 struct tc_u32_sel *s = &n->sel;
681
682 new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key),
683 GFP_KERNEL);
684
685 if (!new)
686 return NULL;
687
688 RCU_INIT_POINTER(new->next, n->next);
689 new->handle = n->handle;
690 RCU_INIT_POINTER(new->ht_up, n->ht_up);
691
692 #ifdef CONFIG_NET_CLS_IND
693 new->ifindex = n->ifindex;
694 #endif
695 new->fshift = n->fshift;
696 new->res = n->res;
697 RCU_INIT_POINTER(new->ht_down, n->ht_down);
698
699 /* bump reference count as long as we hold pointer to structure */
700 if (new->ht_down)
701 new->ht_down->refcnt++;
702
703 #ifdef CONFIG_CLS_U32_PERF
704 /* Statistics may be incremented by readers during update
705 * so we must keep them in tact. When the node is later destroyed
706 * a special destroy call must be made to not free the pf memory.
707 */
708 new->pf = n->pf;
709 #endif
710
711 #ifdef CONFIG_CLS_U32_MARK
712 new->val = n->val;
713 new->mask = n->mask;
714 /* Similarly success statistics must be moved as pointers */
715 new->pcpu_success = n->pcpu_success;
716 #endif
717 new->tp = tp;
718 memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
719
720 tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE);
721
722 return new;
723 }
724
725 static int u32_change(struct net *net, struct sk_buff *in_skb,
726 struct tcf_proto *tp, unsigned long base, u32 handle,
727 struct nlattr **tca,
728 unsigned long *arg, bool ovr)
729 {
730 struct tc_u_common *tp_c = tp->data;
731 struct tc_u_hnode *ht;
732 struct tc_u_knode *n;
733 struct tc_u32_sel *s;
734 struct nlattr *opt = tca[TCA_OPTIONS];
735 struct nlattr *tb[TCA_U32_MAX + 1];
736 u32 htid;
737 int err;
738 #ifdef CONFIG_CLS_U32_PERF
739 size_t size;
740 #endif
741
742 if (opt == NULL)
743 return handle ? -EINVAL : 0;
744
745 err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy);
746 if (err < 0)
747 return err;
748
749 n = (struct tc_u_knode *)*arg;
750 if (n) {
751 struct tc_u_knode *new;
752
753 if (TC_U32_KEY(n->handle) == 0)
754 return -EINVAL;
755
756 new = u32_init_knode(tp, n);
757 if (!new)
758 return -ENOMEM;
759
760 err = u32_set_parms(net, tp, base,
761 rtnl_dereference(n->ht_up), new, tb,
762 tca[TCA_RATE], ovr);
763
764 if (err) {
765 u32_destroy_key(tp, new, false);
766 return err;
767 }
768
769 u32_replace_knode(tp, tp_c, new);
770 tcf_unbind_filter(tp, &n->res);
771 call_rcu(&n->rcu, u32_delete_key_rcu);
772 return 0;
773 }
774
775 if (tb[TCA_U32_DIVISOR]) {
776 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
777
778 if (--divisor > 0x100)
779 return -EINVAL;
780 if (TC_U32_KEY(handle))
781 return -EINVAL;
782 if (handle == 0) {
783 handle = gen_new_htid(tp->data);
784 if (handle == 0)
785 return -ENOMEM;
786 }
787 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
788 if (ht == NULL)
789 return -ENOBUFS;
790 ht->tp_c = tp_c;
791 ht->refcnt = 1;
792 ht->divisor = divisor;
793 ht->handle = handle;
794 ht->prio = tp->prio;
795 RCU_INIT_POINTER(ht->next, tp_c->hlist);
796 rcu_assign_pointer(tp_c->hlist, ht);
797 *arg = (unsigned long)ht;
798 return 0;
799 }
800
801 if (tb[TCA_U32_HASH]) {
802 htid = nla_get_u32(tb[TCA_U32_HASH]);
803 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
804 ht = rtnl_dereference(tp->root);
805 htid = ht->handle;
806 } else {
807 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
808 if (ht == NULL)
809 return -EINVAL;
810 }
811 } else {
812 ht = rtnl_dereference(tp->root);
813 htid = ht->handle;
814 }
815
816 if (ht->divisor < TC_U32_HASH(htid))
817 return -EINVAL;
818
819 if (handle) {
820 if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid))
821 return -EINVAL;
822 handle = htid | TC_U32_NODE(handle);
823 } else
824 handle = gen_new_kid(ht, htid);
825
826 if (tb[TCA_U32_SEL] == NULL)
827 return -EINVAL;
828
829 s = nla_data(tb[TCA_U32_SEL]);
830
831 n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL);
832 if (n == NULL)
833 return -ENOBUFS;
834
835 #ifdef CONFIG_CLS_U32_PERF
836 size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64);
837 n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt));
838 if (!n->pf) {
839 kfree(n);
840 return -ENOBUFS;
841 }
842 #endif
843
844 memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
845 RCU_INIT_POINTER(n->ht_up, ht);
846 n->handle = handle;
847 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
848 tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE);
849 n->tp = tp;
850
851 #ifdef CONFIG_CLS_U32_MARK
852 n->pcpu_success = alloc_percpu(u32);
853 if (!n->pcpu_success) {
854 err = -ENOMEM;
855 goto errout;
856 }
857
858 if (tb[TCA_U32_MARK]) {
859 struct tc_u32_mark *mark;
860
861 mark = nla_data(tb[TCA_U32_MARK]);
862 n->val = mark->val;
863 n->mask = mark->mask;
864 }
865 #endif
866
867 err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr);
868 if (err == 0) {
869 struct tc_u_knode __rcu **ins;
870 struct tc_u_knode *pins;
871
872 ins = &ht->ht[TC_U32_HASH(handle)];
873 for (pins = rtnl_dereference(*ins); pins;
874 ins = &pins->next, pins = rtnl_dereference(*ins))
875 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
876 break;
877
878 RCU_INIT_POINTER(n->next, pins);
879 rcu_assign_pointer(*ins, n);
880
881 *arg = (unsigned long)n;
882 return 0;
883 }
884
885 #ifdef CONFIG_CLS_U32_MARK
886 free_percpu(n->pcpu_success);
887 errout:
888 #endif
889
890 #ifdef CONFIG_CLS_U32_PERF
891 free_percpu(n->pf);
892 #endif
893 kfree(n);
894 return err;
895 }
896
897 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg)
898 {
899 struct tc_u_common *tp_c = tp->data;
900 struct tc_u_hnode *ht;
901 struct tc_u_knode *n;
902 unsigned int h;
903
904 if (arg->stop)
905 return;
906
907 for (ht = rtnl_dereference(tp_c->hlist);
908 ht;
909 ht = rtnl_dereference(ht->next)) {
910 if (ht->prio != tp->prio)
911 continue;
912 if (arg->count >= arg->skip) {
913 if (arg->fn(tp, (unsigned long)ht, arg) < 0) {
914 arg->stop = 1;
915 return;
916 }
917 }
918 arg->count++;
919 for (h = 0; h <= ht->divisor; h++) {
920 for (n = rtnl_dereference(ht->ht[h]);
921 n;
922 n = rtnl_dereference(n->next)) {
923 if (arg->count < arg->skip) {
924 arg->count++;
925 continue;
926 }
927 if (arg->fn(tp, (unsigned long)n, arg) < 0) {
928 arg->stop = 1;
929 return;
930 }
931 arg->count++;
932 }
933 }
934 }
935 }
936
937 static int u32_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
938 struct sk_buff *skb, struct tcmsg *t)
939 {
940 struct tc_u_knode *n = (struct tc_u_knode *)fh;
941 struct tc_u_hnode *ht_up, *ht_down;
942 struct nlattr *nest;
943
944 if (n == NULL)
945 return skb->len;
946
947 t->tcm_handle = n->handle;
948
949 nest = nla_nest_start(skb, TCA_OPTIONS);
950 if (nest == NULL)
951 goto nla_put_failure;
952
953 if (TC_U32_KEY(n->handle) == 0) {
954 struct tc_u_hnode *ht = (struct tc_u_hnode *)fh;
955 u32 divisor = ht->divisor + 1;
956
957 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
958 goto nla_put_failure;
959 } else {
960 #ifdef CONFIG_CLS_U32_PERF
961 struct tc_u32_pcnt *gpf;
962 int cpu;
963 #endif
964
965 if (nla_put(skb, TCA_U32_SEL,
966 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
967 &n->sel))
968 goto nla_put_failure;
969
970 ht_up = rtnl_dereference(n->ht_up);
971 if (ht_up) {
972 u32 htid = n->handle & 0xFFFFF000;
973 if (nla_put_u32(skb, TCA_U32_HASH, htid))
974 goto nla_put_failure;
975 }
976 if (n->res.classid &&
977 nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
978 goto nla_put_failure;
979
980 ht_down = rtnl_dereference(n->ht_down);
981 if (ht_down &&
982 nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
983 goto nla_put_failure;
984
985 #ifdef CONFIG_CLS_U32_MARK
986 if ((n->val || n->mask)) {
987 struct tc_u32_mark mark = {.val = n->val,
988 .mask = n->mask,
989 .success = 0};
990 int cpum;
991
992 for_each_possible_cpu(cpum) {
993 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
994
995 mark.success += cnt;
996 }
997
998 if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
999 goto nla_put_failure;
1000 }
1001 #endif
1002
1003 if (tcf_exts_dump(skb, &n->exts) < 0)
1004 goto nla_put_failure;
1005
1006 #ifdef CONFIG_NET_CLS_IND
1007 if (n->ifindex) {
1008 struct net_device *dev;
1009 dev = __dev_get_by_index(net, n->ifindex);
1010 if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1011 goto nla_put_failure;
1012 }
1013 #endif
1014 #ifdef CONFIG_CLS_U32_PERF
1015 gpf = kzalloc(sizeof(struct tc_u32_pcnt) +
1016 n->sel.nkeys * sizeof(u64),
1017 GFP_KERNEL);
1018 if (!gpf)
1019 goto nla_put_failure;
1020
1021 for_each_possible_cpu(cpu) {
1022 int i;
1023 struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1024
1025 gpf->rcnt += pf->rcnt;
1026 gpf->rhit += pf->rhit;
1027 for (i = 0; i < n->sel.nkeys; i++)
1028 gpf->kcnts[i] += pf->kcnts[i];
1029 }
1030
1031 if (nla_put(skb, TCA_U32_PCNT,
1032 sizeof(struct tc_u32_pcnt) + n->sel.nkeys*sizeof(u64),
1033 gpf)) {
1034 kfree(gpf);
1035 goto nla_put_failure;
1036 }
1037 kfree(gpf);
1038 #endif
1039 }
1040
1041 nla_nest_end(skb, nest);
1042
1043 if (TC_U32_KEY(n->handle))
1044 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1045 goto nla_put_failure;
1046 return skb->len;
1047
1048 nla_put_failure:
1049 nla_nest_cancel(skb, nest);
1050 return -1;
1051 }
1052
1053 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1054 .kind = "u32",
1055 .classify = u32_classify,
1056 .init = u32_init,
1057 .destroy = u32_destroy,
1058 .get = u32_get,
1059 .change = u32_change,
1060 .delete = u32_delete,
1061 .walk = u32_walk,
1062 .dump = u32_dump,
1063 .owner = THIS_MODULE,
1064 };
1065
1066 static int __init init_u32(void)
1067 {
1068 pr_info("u32 classifier\n");
1069 #ifdef CONFIG_CLS_U32_PERF
1070 pr_info(" Performance counters on\n");
1071 #endif
1072 #ifdef CONFIG_NET_CLS_IND
1073 pr_info(" input device check on\n");
1074 #endif
1075 #ifdef CONFIG_NET_CLS_ACT
1076 pr_info(" Actions configured\n");
1077 #endif
1078 return register_tcf_proto_ops(&cls_u32_ops);
1079 }
1080
1081 static void __exit exit_u32(void)
1082 {
1083 unregister_tcf_proto_ops(&cls_u32_ops);
1084 }
1085
1086 module_init(init_u32)
1087 module_exit(exit_u32)
1088 MODULE_LICENSE("GPL");
(deprecated) Btrfs devel tree