?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 3.18.26
[linux/fpc-iii.git] / net / sched / cls_u32.c
blobb78d81f5ffda5a7993f7fba73fd33f5d5fe0d9bd
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 void u32_put(struct tcf_proto *tp, unsigned long f)
306 {
307 }
308
309 static u32 gen_new_htid(struct tc_u_common *tp_c)
310 {
311 int i = 0x800;
312
313 /* hgenerator only used inside rtnl lock it is safe to increment
314 * without read _copy_ update semantics
315 */
316 do {
317 if (++tp_c->hgenerator == 0x7FF)
318 tp_c->hgenerator = 1;
319 } while (--i > 0 && u32_lookup_ht(tp_c, (tp_c->hgenerator|0x800)<<20));
320
321 return i > 0 ? (tp_c->hgenerator|0x800)<<20 : 0;
322 }
323
324 static int u32_init(struct tcf_proto *tp)
325 {
326 struct tc_u_hnode *root_ht;
327 struct tc_u_common *tp_c;
328
329 tp_c = tp->q->u32_node;
330
331 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
332 if (root_ht == NULL)
333 return -ENOBUFS;
334
335 root_ht->divisor = 0;
336 root_ht->refcnt++;
337 root_ht->handle = tp_c ? gen_new_htid(tp_c) : 0x80000000;
338 root_ht->prio = tp->prio;
339
340 if (tp_c == NULL) {
341 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
342 if (tp_c == NULL) {
343 kfree(root_ht);
344 return -ENOBUFS;
345 }
346 tp_c->q = tp->q;
347 tp->q->u32_node = tp_c;
348 }
349
350 tp_c->refcnt++;
351 RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
352 rcu_assign_pointer(tp_c->hlist, root_ht);
353 root_ht->tp_c = tp_c;
354
355 rcu_assign_pointer(tp->root, root_ht);
356 tp->data = tp_c;
357 return 0;
358 }
359
360 static int u32_destroy_key(struct tcf_proto *tp,
361 struct tc_u_knode *n,
362 bool free_pf)
363 {
364 tcf_exts_destroy(&n->exts);
365 if (n->ht_down)
366 n->ht_down->refcnt--;
367 #ifdef CONFIG_CLS_U32_PERF
368 if (free_pf)
369 free_percpu(n->pf);
370 #endif
371 #ifdef CONFIG_CLS_U32_MARK
372 if (free_pf)
373 free_percpu(n->pcpu_success);
374 #endif
375 kfree(n);
376 return 0;
377 }
378
379 /* u32_delete_key_rcu should be called when free'ing a copied
380 * version of a tc_u_knode obtained from u32_init_knode(). When
381 * copies are obtained from u32_init_knode() the statistics are
382 * shared between the old and new copies to allow readers to
383 * continue to update the statistics during the copy. To support
384 * this the u32_delete_key_rcu variant does not free the percpu
385 * statistics.
386 */
387 static void u32_delete_key_rcu(struct rcu_head *rcu)
388 {
389 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
390
391 u32_destroy_key(key->tp, key, false);
392 }
393
394 /* u32_delete_key_freepf_rcu is the rcu callback variant
395 * that free's the entire structure including the statistics
396 * percpu variables. Only use this if the key is not a copy
397 * returned by u32_init_knode(). See u32_delete_key_rcu()
398 * for the variant that should be used with keys return from
399 * u32_init_knode()
400 */
401 static void u32_delete_key_freepf_rcu(struct rcu_head *rcu)
402 {
403 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
404
405 u32_destroy_key(key->tp, key, true);
406 }
407
408 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
409 {
410 struct tc_u_knode __rcu **kp;
411 struct tc_u_knode *pkp;
412 struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
413
414 if (ht) {
415 kp = &ht->ht[TC_U32_HASH(key->handle)];
416 for (pkp = rtnl_dereference(*kp); pkp;
417 kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
418 if (pkp == key) {
419 RCU_INIT_POINTER(*kp, key->next);
420
421 tcf_unbind_filter(tp, &key->res);
422 call_rcu(&key->rcu, u32_delete_key_freepf_rcu);
423 return 0;
424 }
425 }
426 }
427 WARN_ON(1);
428 return 0;
429 }
430
431 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
432 {
433 struct tc_u_knode *n;
434 unsigned int h;
435
436 for (h = 0; h <= ht->divisor; h++) {
437 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
438 RCU_INIT_POINTER(ht->ht[h],
439 rtnl_dereference(n->next));
440 tcf_unbind_filter(tp, &n->res);
441 call_rcu(&n->rcu, u32_delete_key_freepf_rcu);
442 }
443 }
444 }
445
446 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
447 {
448 struct tc_u_common *tp_c = tp->data;
449 struct tc_u_hnode __rcu **hn;
450 struct tc_u_hnode *phn;
451
452 WARN_ON(ht->refcnt);
453
454 u32_clear_hnode(tp, ht);
455
456 hn = &tp_c->hlist;
457 for (phn = rtnl_dereference(*hn);
458 phn;
459 hn = &phn->next, phn = rtnl_dereference(*hn)) {
460 if (phn == ht) {
461 RCU_INIT_POINTER(*hn, ht->next);
462 kfree_rcu(ht, rcu);
463 return 0;
464 }
465 }
466
467 return -ENOENT;
468 }
469
470 static void u32_destroy(struct tcf_proto *tp)
471 {
472 struct tc_u_common *tp_c = tp->data;
473 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
474
475 WARN_ON(root_ht == NULL);
476
477 if (root_ht && --root_ht->refcnt == 0)
478 u32_destroy_hnode(tp, root_ht);
479
480 if (--tp_c->refcnt == 0) {
481 struct tc_u_hnode *ht;
482
483 tp->q->u32_node = NULL;
484
485 for (ht = rtnl_dereference(tp_c->hlist);
486 ht;
487 ht = rtnl_dereference(ht->next)) {
488 ht->refcnt--;
489 u32_clear_hnode(tp, ht);
490 }
491
492 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
493 RCU_INIT_POINTER(tp_c->hlist, ht->next);
494 kfree_rcu(ht, rcu);
495 }
496
497 kfree(tp_c);
498 }
499
500 tp->data = NULL;
501 }
502
503 static int u32_delete(struct tcf_proto *tp, unsigned long arg)
504 {
505 struct tc_u_hnode *ht = (struct tc_u_hnode *)arg;
506 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
507
508 if (ht == NULL)
509 return 0;
510
511 if (TC_U32_KEY(ht->handle))
512 return u32_delete_key(tp, (struct tc_u_knode *)ht);
513
514 if (root_ht == ht)
515 return -EINVAL;
516
517 if (ht->refcnt == 1) {
518 ht->refcnt--;
519 u32_destroy_hnode(tp, ht);
520 } else {
521 return -EBUSY;
522 }
523
524 return 0;
525 }
526
527 #define NR_U32_NODE (1<<12)
528 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 handle)
529 {
530 struct tc_u_knode *n;
531 unsigned long i;
532 unsigned long *bitmap = kzalloc(BITS_TO_LONGS(NR_U32_NODE) * sizeof(unsigned long),
533 GFP_KERNEL);
534 if (!bitmap)
535 return handle | 0xFFF;
536
537 for (n = rtnl_dereference(ht->ht[TC_U32_HASH(handle)]);
538 n;
539 n = rtnl_dereference(n->next))
540 set_bit(TC_U32_NODE(n->handle), bitmap);
541
542 i = find_next_zero_bit(bitmap, NR_U32_NODE, 0x800);
543 if (i >= NR_U32_NODE)
544 i = find_next_zero_bit(bitmap, NR_U32_NODE, 1);
545
546 kfree(bitmap);
547 return handle | (i >= NR_U32_NODE ? 0xFFF : i);
548 }
549
550 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
551 [TCA_U32_CLASSID] = { .type = NLA_U32 },
552 [TCA_U32_HASH] = { .type = NLA_U32 },
553 [TCA_U32_LINK] = { .type = NLA_U32 },
554 [TCA_U32_DIVISOR] = { .type = NLA_U32 },
555 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) },
556 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ },
557 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) },
558 };
559
560 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
561 unsigned long base, struct tc_u_hnode *ht,
562 struct tc_u_knode *n, struct nlattr **tb,
563 struct nlattr *est, bool ovr)
564 {
565 int err;
566 struct tcf_exts e;
567
568 tcf_exts_init(&e, TCA_U32_ACT, TCA_U32_POLICE);
569 err = tcf_exts_validate(net, tp, tb, est, &e, ovr);
570 if (err < 0)
571 return err;
572
573 err = -EINVAL;
574 if (tb[TCA_U32_LINK]) {
575 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
576 struct tc_u_hnode *ht_down = NULL, *ht_old;
577
578 if (TC_U32_KEY(handle))
579 goto errout;
580
581 if (handle) {
582 ht_down = u32_lookup_ht(ht->tp_c, handle);
583
584 if (ht_down == NULL)
585 goto errout;
586 ht_down->refcnt++;
587 }
588
589 ht_old = rtnl_dereference(n->ht_down);
590 rcu_assign_pointer(n->ht_down, ht_down);
591
592 if (ht_old)
593 ht_old->refcnt--;
594 }
595 if (tb[TCA_U32_CLASSID]) {
596 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
597 tcf_bind_filter(tp, &n->res, base);
598 }
599
600 #ifdef CONFIG_NET_CLS_IND
601 if (tb[TCA_U32_INDEV]) {
602 int ret;
603 ret = tcf_change_indev(net, tb[TCA_U32_INDEV]);
604 if (ret < 0)
605 goto errout;
606 n->ifindex = ret;
607 }
608 #endif
609 tcf_exts_change(tp, &n->exts, &e);
610
611 return 0;
612 errout:
613 tcf_exts_destroy(&e);
614 return err;
615 }
616
617 static void u32_replace_knode(struct tcf_proto *tp,
618 struct tc_u_common *tp_c,
619 struct tc_u_knode *n)
620 {
621 struct tc_u_knode __rcu **ins;
622 struct tc_u_knode *pins;
623 struct tc_u_hnode *ht;
624
625 if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
626 ht = rtnl_dereference(tp->root);
627 else
628 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
629
630 ins = &ht->ht[TC_U32_HASH(n->handle)];
631
632 /* The node must always exist for it to be replaced if this is not the
633 * case then something went very wrong elsewhere.
634 */
635 for (pins = rtnl_dereference(*ins); ;
636 ins = &pins->next, pins = rtnl_dereference(*ins))
637 if (pins->handle == n->handle)
638 break;
639
640 RCU_INIT_POINTER(n->next, pins->next);
641 rcu_assign_pointer(*ins, n);
642 }
643
644 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp,
645 struct tc_u_knode *n)
646 {
647 struct tc_u_knode *new;
648 struct tc_u32_sel *s = &n->sel;
649
650 new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key),
651 GFP_KERNEL);
652
653 if (!new)
654 return NULL;
655
656 RCU_INIT_POINTER(new->next, n->next);
657 new->handle = n->handle;
658 RCU_INIT_POINTER(new->ht_up, n->ht_up);
659
660 #ifdef CONFIG_NET_CLS_IND
661 new->ifindex = n->ifindex;
662 #endif
663 new->fshift = n->fshift;
664 new->res = n->res;
665 RCU_INIT_POINTER(new->ht_down, n->ht_down);
666
667 /* bump reference count as long as we hold pointer to structure */
668 if (new->ht_down)
669 new->ht_down->refcnt++;
670
671 #ifdef CONFIG_CLS_U32_PERF
672 /* Statistics may be incremented by readers during update
673 * so we must keep them in tact. When the node is later destroyed
674 * a special destroy call must be made to not free the pf memory.
675 */
676 new->pf = n->pf;
677 #endif
678
679 #ifdef CONFIG_CLS_U32_MARK
680 new->val = n->val;
681 new->mask = n->mask;
682 /* Similarly success statistics must be moved as pointers */
683 new->pcpu_success = n->pcpu_success;
684 #endif
685 new->tp = tp;
686 memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
687
688 tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE);
689
690 return new;
691 }
692
693 static int u32_change(struct net *net, struct sk_buff *in_skb,
694 struct tcf_proto *tp, unsigned long base, u32 handle,
695 struct nlattr **tca,
696 unsigned long *arg, bool ovr)
697 {
698 struct tc_u_common *tp_c = tp->data;
699 struct tc_u_hnode *ht;
700 struct tc_u_knode *n;
701 struct tc_u32_sel *s;
702 struct nlattr *opt = tca[TCA_OPTIONS];
703 struct nlattr *tb[TCA_U32_MAX + 1];
704 u32 htid;
705 int err;
706 #ifdef CONFIG_CLS_U32_PERF
707 size_t size;
708 #endif
709
710 if (opt == NULL)
711 return handle ? -EINVAL : 0;
712
713 err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy);
714 if (err < 0)
715 return err;
716
717 n = (struct tc_u_knode *)*arg;
718 if (n) {
719 struct tc_u_knode *new;
720
721 if (TC_U32_KEY(n->handle) == 0)
722 return -EINVAL;
723
724 new = u32_init_knode(tp, n);
725 if (!new)
726 return -ENOMEM;
727
728 err = u32_set_parms(net, tp, base,
729 rtnl_dereference(n->ht_up), new, tb,
730 tca[TCA_RATE], ovr);
731
732 if (err) {
733 u32_destroy_key(tp, new, false);
734 return err;
735 }
736
737 u32_replace_knode(tp, tp_c, new);
738 tcf_unbind_filter(tp, &n->res);
739 call_rcu(&n->rcu, u32_delete_key_rcu);
740 return 0;
741 }
742
743 if (tb[TCA_U32_DIVISOR]) {
744 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
745
746 if (--divisor > 0x100)
747 return -EINVAL;
748 if (TC_U32_KEY(handle))
749 return -EINVAL;
750 if (handle == 0) {
751 handle = gen_new_htid(tp->data);
752 if (handle == 0)
753 return -ENOMEM;
754 }
755 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
756 if (ht == NULL)
757 return -ENOBUFS;
758 ht->tp_c = tp_c;
759 ht->refcnt = 1;
760 ht->divisor = divisor;
761 ht->handle = handle;
762 ht->prio = tp->prio;
763 RCU_INIT_POINTER(ht->next, tp_c->hlist);
764 rcu_assign_pointer(tp_c->hlist, ht);
765 *arg = (unsigned long)ht;
766 return 0;
767 }
768
769 if (tb[TCA_U32_HASH]) {
770 htid = nla_get_u32(tb[TCA_U32_HASH]);
771 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
772 ht = rtnl_dereference(tp->root);
773 htid = ht->handle;
774 } else {
775 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
776 if (ht == NULL)
777 return -EINVAL;
778 }
779 } else {
780 ht = rtnl_dereference(tp->root);
781 htid = ht->handle;
782 }
783
784 if (ht->divisor < TC_U32_HASH(htid))
785 return -EINVAL;
786
787 if (handle) {
788 if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid))
789 return -EINVAL;
790 handle = htid | TC_U32_NODE(handle);
791 } else
792 handle = gen_new_kid(ht, htid);
793
794 if (tb[TCA_U32_SEL] == NULL)
795 return -EINVAL;
796
797 s = nla_data(tb[TCA_U32_SEL]);
798
799 n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL);
800 if (n == NULL)
801 return -ENOBUFS;
802
803 #ifdef CONFIG_CLS_U32_PERF
804 size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64);
805 n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt));
806 if (!n->pf) {
807 kfree(n);
808 return -ENOBUFS;
809 }
810 #endif
811
812 memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
813 RCU_INIT_POINTER(n->ht_up, ht);
814 n->handle = handle;
815 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
816 tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE);
817 n->tp = tp;
818
819 #ifdef CONFIG_CLS_U32_MARK
820 n->pcpu_success = alloc_percpu(u32);
821 if (!n->pcpu_success) {
822 err = -ENOMEM;
823 goto errout;
824 }
825
826 if (tb[TCA_U32_MARK]) {
827 struct tc_u32_mark *mark;
828
829 mark = nla_data(tb[TCA_U32_MARK]);
830 n->val = mark->val;
831 n->mask = mark->mask;
832 }
833 #endif
834
835 err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr);
836 if (err == 0) {
837 struct tc_u_knode __rcu **ins;
838 struct tc_u_knode *pins;
839
840 ins = &ht->ht[TC_U32_HASH(handle)];
841 for (pins = rtnl_dereference(*ins); pins;
842 ins = &pins->next, pins = rtnl_dereference(*ins))
843 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
844 break;
845
846 RCU_INIT_POINTER(n->next, pins);
847 rcu_assign_pointer(*ins, n);
848
849 *arg = (unsigned long)n;
850 return 0;
851 }
852
853 #ifdef CONFIG_CLS_U32_MARK
854 free_percpu(n->pcpu_success);
855 errout:
856 #endif
857
858 #ifdef CONFIG_CLS_U32_PERF
859 free_percpu(n->pf);
860 #endif
861 kfree(n);
862 return err;
863 }
864
865 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg)
866 {
867 struct tc_u_common *tp_c = tp->data;
868 struct tc_u_hnode *ht;
869 struct tc_u_knode *n;
870 unsigned int h;
871
872 if (arg->stop)
873 return;
874
875 for (ht = rtnl_dereference(tp_c->hlist);
876 ht;
877 ht = rtnl_dereference(ht->next)) {
878 if (ht->prio != tp->prio)
879 continue;
880 if (arg->count >= arg->skip) {
881 if (arg->fn(tp, (unsigned long)ht, arg) < 0) {
882 arg->stop = 1;
883 return;
884 }
885 }
886 arg->count++;
887 for (h = 0; h <= ht->divisor; h++) {
888 for (n = rtnl_dereference(ht->ht[h]);
889 n;
890 n = rtnl_dereference(n->next)) {
891 if (arg->count < arg->skip) {
892 arg->count++;
893 continue;
894 }
895 if (arg->fn(tp, (unsigned long)n, arg) < 0) {
896 arg->stop = 1;
897 return;
898 }
899 arg->count++;
900 }
901 }
902 }
903 }
904
905 static int u32_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
906 struct sk_buff *skb, struct tcmsg *t)
907 {
908 struct tc_u_knode *n = (struct tc_u_knode *)fh;
909 struct tc_u_hnode *ht_up, *ht_down;
910 struct nlattr *nest;
911
912 if (n == NULL)
913 return skb->len;
914
915 t->tcm_handle = n->handle;
916
917 nest = nla_nest_start(skb, TCA_OPTIONS);
918 if (nest == NULL)
919 goto nla_put_failure;
920
921 if (TC_U32_KEY(n->handle) == 0) {
922 struct tc_u_hnode *ht = (struct tc_u_hnode *)fh;
923 u32 divisor = ht->divisor + 1;
924
925 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
926 goto nla_put_failure;
927 } else {
928 #ifdef CONFIG_CLS_U32_PERF
929 struct tc_u32_pcnt *gpf;
930 int cpu;
931 #endif
932
933 if (nla_put(skb, TCA_U32_SEL,
934 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
935 &n->sel))
936 goto nla_put_failure;
937
938 ht_up = rtnl_dereference(n->ht_up);
939 if (ht_up) {
940 u32 htid = n->handle & 0xFFFFF000;
941 if (nla_put_u32(skb, TCA_U32_HASH, htid))
942 goto nla_put_failure;
943 }
944 if (n->res.classid &&
945 nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
946 goto nla_put_failure;
947
948 ht_down = rtnl_dereference(n->ht_down);
949 if (ht_down &&
950 nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
951 goto nla_put_failure;
952
953 #ifdef CONFIG_CLS_U32_MARK
954 if ((n->val || n->mask)) {
955 struct tc_u32_mark mark = {.val = n->val,
956 .mask = n->mask,
957 .success = 0};
958 int cpum;
959
960 for_each_possible_cpu(cpum) {
961 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
962
963 mark.success += cnt;
964 }
965
966 if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
967 goto nla_put_failure;
968 }
969 #endif
970
971 if (tcf_exts_dump(skb, &n->exts) < 0)
972 goto nla_put_failure;
973
974 #ifdef CONFIG_NET_CLS_IND
975 if (n->ifindex) {
976 struct net_device *dev;
977 dev = __dev_get_by_index(net, n->ifindex);
978 if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
979 goto nla_put_failure;
980 }
981 #endif
982 #ifdef CONFIG_CLS_U32_PERF
983 gpf = kzalloc(sizeof(struct tc_u32_pcnt) +
984 n->sel.nkeys * sizeof(u64),
985 GFP_KERNEL);
986 if (!gpf)
987 goto nla_put_failure;
988
989 for_each_possible_cpu(cpu) {
990 int i;
991 struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
992
993 gpf->rcnt += pf->rcnt;
994 gpf->rhit += pf->rhit;
995 for (i = 0; i < n->sel.nkeys; i++)
996 gpf->kcnts[i] += pf->kcnts[i];
997 }
998
999 if (nla_put(skb, TCA_U32_PCNT,
1000 sizeof(struct tc_u32_pcnt) + n->sel.nkeys*sizeof(u64),
1001 gpf)) {
1002 kfree(gpf);
1003 goto nla_put_failure;
1004 }
1005 kfree(gpf);
1006 #endif
1007 }
1008
1009 nla_nest_end(skb, nest);
1010
1011 if (TC_U32_KEY(n->handle))
1012 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1013 goto nla_put_failure;
1014 return skb->len;
1015
1016 nla_put_failure:
1017 nla_nest_cancel(skb, nest);
1018 return -1;
1019 }
1020
1021 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1022 .kind = "u32",
1023 .classify = u32_classify,
1024 .init = u32_init,
1025 .destroy = u32_destroy,
1026 .get = u32_get,
1027 .put = u32_put,
1028 .change = u32_change,
1029 .delete = u32_delete,
1030 .walk = u32_walk,
1031 .dump = u32_dump,
1032 .owner = THIS_MODULE,
1033 };
1034
1035 static int __init init_u32(void)
1036 {
1037 pr_info("u32 classifier\n");
1038 #ifdef CONFIG_CLS_U32_PERF
1039 pr_info(" Performance counters on\n");
1040 #endif
1041 #ifdef CONFIG_NET_CLS_IND
1042 pr_info(" input device check on\n");
1043 #endif
1044 #ifdef CONFIG_NET_CLS_ACT
1045 pr_info(" Actions configured\n");
1046 #endif
1047 return register_tcf_proto_ops(&cls_u32_ops);
1048 }
1049
1050 static void __exit exit_u32(void)
1051 {
1052 unregister_tcf_proto_ops(&cls_u32_ops);
1053 }
1054
1055 module_init(init_u32)
1056 module_exit(exit_u32)
1057 MODULE_LICENSE("GPL");
Linux with added FPC-III support