projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next
[deliverable/linux.git] / net / sched / cls_u32.c
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 tc_u_knode __rcu *ht[1];
82 struct rcu_head rcu;
83 };
84
85 struct tc_u_common {
86 struct tc_u_hnode __rcu *hlist;
87 struct Qdisc *q;
88 int refcnt;
89 u32 hgenerator;
90 struct rcu_head rcu;
91 };
92
93 static inline unsigned int u32_hash_fold(__be32 key,
94 const struct tc_u32_sel *sel,
95 u8 fshift)
96 {
97 unsigned int h = ntohl(key & sel->hmask) >> fshift;
98
99 return h;
100 }
101
102 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp, struct tcf_result *res)
103 {
104 struct {
105 struct tc_u_knode *knode;
106 unsigned int off;
107 } stack[TC_U32_MAXDEPTH];
108
109 struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
110 unsigned int off = skb_network_offset(skb);
111 struct tc_u_knode *n;
112 int sdepth = 0;
113 int off2 = 0;
114 int sel = 0;
115 #ifdef CONFIG_CLS_U32_PERF
116 int j;
117 #endif
118 int i, r;
119
120 next_ht:
121 n = rcu_dereference_bh(ht->ht[sel]);
122
123 next_knode:
124 if (n) {
125 struct tc_u32_key *key = n->sel.keys;
126
127 #ifdef CONFIG_CLS_U32_PERF
128 __this_cpu_inc(n->pf->rcnt);
129 j = 0;
130 #endif
131
132 #ifdef CONFIG_CLS_U32_MARK
133 if ((skb->mark & n->mask) != n->val) {
134 n = rcu_dereference_bh(n->next);
135 goto next_knode;
136 } else {
137 __this_cpu_inc(*n->pcpu_success);
138 }
139 #endif
140
141 for (i = n->sel.nkeys; i > 0; i--, key++) {
142 int toff = off + key->off + (off2 & key->offmask);
143 __be32 *data, hdata;
144
145 if (skb_headroom(skb) + toff > INT_MAX)
146 goto out;
147
148 data = skb_header_pointer(skb, toff, 4, &hdata);
149 if (!data)
150 goto out;
151 if ((*data ^ key->val) & key->mask) {
152 n = rcu_dereference_bh(n->next);
153 goto next_knode;
154 }
155 #ifdef CONFIG_CLS_U32_PERF
156 __this_cpu_inc(n->pf->kcnts[j]);
157 j++;
158 #endif
159 }
160
161 ht = rcu_dereference_bh(n->ht_down);
162 if (!ht) {
163 check_terminal:
164 if (n->sel.flags & TC_U32_TERMINAL) {
165
166 *res = n->res;
167 #ifdef CONFIG_NET_CLS_IND
168 if (!tcf_match_indev(skb, n->ifindex)) {
169 n = rcu_dereference_bh(n->next);
170 goto next_knode;
171 }
172 #endif
173 #ifdef CONFIG_CLS_U32_PERF
174 __this_cpu_inc(n->pf->rhit);
175 #endif
176 r = tcf_exts_exec(skb, &n->exts, res);
177 if (r < 0) {
178 n = rcu_dereference_bh(n->next);
179 goto next_knode;
180 }
181
182 return r;
183 }
184 n = rcu_dereference_bh(n->next);
185 goto next_knode;
186 }
187
188 /* PUSH */
189 if (sdepth >= TC_U32_MAXDEPTH)
190 goto deadloop;
191 stack[sdepth].knode = n;
192 stack[sdepth].off = off;
193 sdepth++;
194
195 ht = rcu_dereference_bh(n->ht_down);
196 sel = 0;
197 if (ht->divisor) {
198 __be32 *data, hdata;
199
200 data = skb_header_pointer(skb, off + n->sel.hoff, 4,
201 &hdata);
202 if (!data)
203 goto out;
204 sel = ht->divisor & u32_hash_fold(*data, &n->sel,
205 n->fshift);
206 }
207 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
208 goto next_ht;
209
210 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
211 off2 = n->sel.off + 3;
212 if (n->sel.flags & TC_U32_VAROFFSET) {
213 __be16 *data, hdata;
214
215 data = skb_header_pointer(skb,
216 off + n->sel.offoff,
217 2, &hdata);
218 if (!data)
219 goto out;
220 off2 += ntohs(n->sel.offmask & *data) >>
221 n->sel.offshift;
222 }
223 off2 &= ~3;
224 }
225 if (n->sel.flags & TC_U32_EAT) {
226 off += off2;
227 off2 = 0;
228 }
229
230 if (off < skb->len)
231 goto next_ht;
232 }
233
234 /* POP */
235 if (sdepth--) {
236 n = stack[sdepth].knode;
237 ht = rcu_dereference_bh(n->ht_up);
238 off = stack[sdepth].off;
239 goto check_terminal;
240 }
241 out:
242 return -1;
243
244 deadloop:
245 net_warn_ratelimited("cls_u32: dead loop\n");
246 return -1;
247 }
248
249 static struct tc_u_hnode *
250 u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
251 {
252 struct tc_u_hnode *ht;
253
254 for (ht = rtnl_dereference(tp_c->hlist);
255 ht;
256 ht = rtnl_dereference(ht->next))
257 if (ht->handle == handle)
258 break;
259
260 return ht;
261 }
262
263 static struct tc_u_knode *
264 u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
265 {
266 unsigned int sel;
267 struct tc_u_knode *n = NULL;
268
269 sel = TC_U32_HASH(handle);
270 if (sel > ht->divisor)
271 goto out;
272
273 for (n = rtnl_dereference(ht->ht[sel]);
274 n;
275 n = rtnl_dereference(n->next))
276 if (n->handle == handle)
277 break;
278 out:
279 return n;
280 }
281
282
283 static unsigned long u32_get(struct tcf_proto *tp, u32 handle)
284 {
285 struct tc_u_hnode *ht;
286 struct tc_u_common *tp_c = tp->data;
287
288 if (TC_U32_HTID(handle) == TC_U32_ROOT)
289 ht = rtnl_dereference(tp->root);
290 else
291 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
292
293 if (!ht)
294 return 0;
295
296 if (TC_U32_KEY(handle) == 0)
297 return (unsigned long)ht;
298
299 return (unsigned long)u32_lookup_key(ht, handle);
300 }
301
302 static u32 gen_new_htid(struct tc_u_common *tp_c)
303 {
304 int i = 0x800;
305
306 /* hgenerator only used inside rtnl lock it is safe to increment
307 * without read _copy_ update semantics
308 */
309 do {
310 if (++tp_c->hgenerator == 0x7FF)
311 tp_c->hgenerator = 1;
312 } while (--i > 0 && u32_lookup_ht(tp_c, (tp_c->hgenerator|0x800)<<20));
313
314 return i > 0 ? (tp_c->hgenerator|0x800)<<20 : 0;
315 }
316
317 static int u32_init(struct tcf_proto *tp)
318 {
319 struct tc_u_hnode *root_ht;
320 struct tc_u_common *tp_c;
321
322 tp_c = tp->q->u32_node;
323
324 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
325 if (root_ht == NULL)
326 return -ENOBUFS;
327
328 root_ht->divisor = 0;
329 root_ht->refcnt++;
330 root_ht->handle = tp_c ? gen_new_htid(tp_c) : 0x80000000;
331 root_ht->prio = tp->prio;
332
333 if (tp_c == NULL) {
334 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
335 if (tp_c == NULL) {
336 kfree(root_ht);
337 return -ENOBUFS;
338 }
339 tp_c->q = tp->q;
340 tp->q->u32_node = tp_c;
341 }
342
343 tp_c->refcnt++;
344 RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
345 rcu_assign_pointer(tp_c->hlist, root_ht);
346 root_ht->tp_c = tp_c;
347
348 rcu_assign_pointer(tp->root, root_ht);
349 tp->data = tp_c;
350 return 0;
351 }
352
353 static int u32_destroy_key(struct tcf_proto *tp,
354 struct tc_u_knode *n,
355 bool free_pf)
356 {
357 tcf_exts_destroy(&n->exts);
358 if (n->ht_down)
359 n->ht_down->refcnt--;
360 #ifdef CONFIG_CLS_U32_PERF
361 if (free_pf)
362 free_percpu(n->pf);
363 #endif
364 #ifdef CONFIG_CLS_U32_MARK
365 if (free_pf)
366 free_percpu(n->pcpu_success);
367 #endif
368 kfree(n);
369 return 0;
370 }
371
372 /* u32_delete_key_rcu should be called when free'ing a copied
373 * version of a tc_u_knode obtained from u32_init_knode(). When
374 * copies are obtained from u32_init_knode() the statistics are
375 * shared between the old and new copies to allow readers to
376 * continue to update the statistics during the copy. To support
377 * this the u32_delete_key_rcu variant does not free the percpu
378 * statistics.
379 */
380 static void u32_delete_key_rcu(struct rcu_head *rcu)
381 {
382 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
383
384 u32_destroy_key(key->tp, key, false);
385 }
386
387 /* u32_delete_key_freepf_rcu is the rcu callback variant
388 * that free's the entire structure including the statistics
389 * percpu variables. Only use this if the key is not a copy
390 * returned by u32_init_knode(). See u32_delete_key_rcu()
391 * for the variant that should be used with keys return from
392 * u32_init_knode()
393 */
394 static void u32_delete_key_freepf_rcu(struct rcu_head *rcu)
395 {
396 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
397
398 u32_destroy_key(key->tp, key, true);
399 }
400
401 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
402 {
403 struct tc_u_knode __rcu **kp;
404 struct tc_u_knode *pkp;
405 struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
406
407 if (ht) {
408 kp = &ht->ht[TC_U32_HASH(key->handle)];
409 for (pkp = rtnl_dereference(*kp); pkp;
410 kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
411 if (pkp == key) {
412 RCU_INIT_POINTER(*kp, key->next);
413
414 tcf_unbind_filter(tp, &key->res);
415 call_rcu(&key->rcu, u32_delete_key_freepf_rcu);
416 return 0;
417 }
418 }
419 }
420 WARN_ON(1);
421 return 0;
422 }
423
424 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
425 {
426 struct tc_u_knode *n;
427 unsigned int h;
428
429 for (h = 0; h <= ht->divisor; h++) {
430 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
431 RCU_INIT_POINTER(ht->ht[h],
432 rtnl_dereference(n->next));
433 tcf_unbind_filter(tp, &n->res);
434 call_rcu(&n->rcu, u32_delete_key_freepf_rcu);
435 }
436 }
437 }
438
439 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
440 {
441 struct tc_u_common *tp_c = tp->data;
442 struct tc_u_hnode __rcu **hn;
443 struct tc_u_hnode *phn;
444
445 WARN_ON(ht->refcnt);
446
447 u32_clear_hnode(tp, ht);
448
449 hn = &tp_c->hlist;
450 for (phn = rtnl_dereference(*hn);
451 phn;
452 hn = &phn->next, phn = rtnl_dereference(*hn)) {
453 if (phn == ht) {
454 RCU_INIT_POINTER(*hn, ht->next);
455 kfree_rcu(ht, rcu);
456 return 0;
457 }
458 }
459
460 return -ENOENT;
461 }
462
463 static void u32_destroy(struct tcf_proto *tp)
464 {
465 struct tc_u_common *tp_c = tp->data;
466 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
467
468 WARN_ON(root_ht == NULL);
469
470 if (root_ht && --root_ht->refcnt == 0)
471 u32_destroy_hnode(tp, root_ht);
472
473 if (--tp_c->refcnt == 0) {
474 struct tc_u_hnode *ht;
475
476 tp->q->u32_node = NULL;
477
478 for (ht = rtnl_dereference(tp_c->hlist);
479 ht;
480 ht = rtnl_dereference(ht->next)) {
481 ht->refcnt--;
482 u32_clear_hnode(tp, ht);
483 }
484
485 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
486 RCU_INIT_POINTER(tp_c->hlist, ht->next);
487 kfree_rcu(ht, rcu);
488 }
489
490 kfree(tp_c);
491 }
492
493 tp->data = NULL;
494 }
495
496 static int u32_delete(struct tcf_proto *tp, unsigned long arg)
497 {
498 struct tc_u_hnode *ht = (struct tc_u_hnode *)arg;
499 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
500
501 if (ht == NULL)
502 return 0;
503
504 if (TC_U32_KEY(ht->handle))
505 return u32_delete_key(tp, (struct tc_u_knode *)ht);
506
507 if (root_ht == ht)
508 return -EINVAL;
509
510 if (ht->refcnt == 1) {
511 ht->refcnt--;
512 u32_destroy_hnode(tp, ht);
513 } else {
514 return -EBUSY;
515 }
516
517 return 0;
518 }
519
520 #define NR_U32_NODE (1<<12)
521 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 handle)
522 {
523 struct tc_u_knode *n;
524 unsigned long i;
525 unsigned long *bitmap = kzalloc(BITS_TO_LONGS(NR_U32_NODE) * sizeof(unsigned long),
526 GFP_KERNEL);
527 if (!bitmap)
528 return handle | 0xFFF;
529
530 for (n = rtnl_dereference(ht->ht[TC_U32_HASH(handle)]);
531 n;
532 n = rtnl_dereference(n->next))
533 set_bit(TC_U32_NODE(n->handle), bitmap);
534
535 i = find_next_zero_bit(bitmap, NR_U32_NODE, 0x800);
536 if (i >= NR_U32_NODE)
537 i = find_next_zero_bit(bitmap, NR_U32_NODE, 1);
538
539 kfree(bitmap);
540 return handle | (i >= NR_U32_NODE ? 0xFFF : i);
541 }
542
543 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
544 [TCA_U32_CLASSID] = { .type = NLA_U32 },
545 [TCA_U32_HASH] = { .type = NLA_U32 },
546 [TCA_U32_LINK] = { .type = NLA_U32 },
547 [TCA_U32_DIVISOR] = { .type = NLA_U32 },
548 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) },
549 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ },
550 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) },
551 };
552
553 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
554 unsigned long base, struct tc_u_hnode *ht,
555 struct tc_u_knode *n, struct nlattr **tb,
556 struct nlattr *est, bool ovr)
557 {
558 int err;
559 struct tcf_exts e;
560
561 tcf_exts_init(&e, TCA_U32_ACT, TCA_U32_POLICE);
562 err = tcf_exts_validate(net, tp, tb, est, &e, ovr);
563 if (err < 0)
564 return err;
565
566 err = -EINVAL;
567 if (tb[TCA_U32_LINK]) {
568 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
569 struct tc_u_hnode *ht_down = NULL, *ht_old;
570
571 if (TC_U32_KEY(handle))
572 goto errout;
573
574 if (handle) {
575 ht_down = u32_lookup_ht(ht->tp_c, handle);
576
577 if (ht_down == NULL)
578 goto errout;
579 ht_down->refcnt++;
580 }
581
582 ht_old = rtnl_dereference(n->ht_down);
583 rcu_assign_pointer(n->ht_down, ht_down);
584
585 if (ht_old)
586 ht_old->refcnt--;
587 }
588 if (tb[TCA_U32_CLASSID]) {
589 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
590 tcf_bind_filter(tp, &n->res, base);
591 }
592
593 #ifdef CONFIG_NET_CLS_IND
594 if (tb[TCA_U32_INDEV]) {
595 int ret;
596 ret = tcf_change_indev(net, tb[TCA_U32_INDEV]);
597 if (ret < 0)
598 goto errout;
599 n->ifindex = ret;
600 }
601 #endif
602 tcf_exts_change(tp, &n->exts, &e);
603
604 return 0;
605 errout:
606 tcf_exts_destroy(&e);
607 return err;
608 }
609
610 static void u32_replace_knode(struct tcf_proto *tp,
611 struct tc_u_common *tp_c,
612 struct tc_u_knode *n)
613 {
614 struct tc_u_knode __rcu **ins;
615 struct tc_u_knode *pins;
616 struct tc_u_hnode *ht;
617
618 if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
619 ht = rtnl_dereference(tp->root);
620 else
621 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
622
623 ins = &ht->ht[TC_U32_HASH(n->handle)];
624
625 /* The node must always exist for it to be replaced if this is not the
626 * case then something went very wrong elsewhere.
627 */
628 for (pins = rtnl_dereference(*ins); ;
629 ins = &pins->next, pins = rtnl_dereference(*ins))
630 if (pins->handle == n->handle)
631 break;
632
633 RCU_INIT_POINTER(n->next, pins->next);
634 rcu_assign_pointer(*ins, n);
635 }
636
637 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp,
638 struct tc_u_knode *n)
639 {
640 struct tc_u_knode *new;
641 struct tc_u32_sel *s = &n->sel;
642
643 new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key),
644 GFP_KERNEL);
645
646 if (!new)
647 return NULL;
648
649 RCU_INIT_POINTER(new->next, n->next);
650 new->handle = n->handle;
651 RCU_INIT_POINTER(new->ht_up, n->ht_up);
652
653 #ifdef CONFIG_NET_CLS_IND
654 new->ifindex = n->ifindex;
655 #endif
656 new->fshift = n->fshift;
657 new->res = n->res;
658 RCU_INIT_POINTER(new->ht_down, n->ht_down);
659
660 /* bump reference count as long as we hold pointer to structure */
661 if (new->ht_down)
662 new->ht_down->refcnt++;
663
664 #ifdef CONFIG_CLS_U32_PERF
665 /* Statistics may be incremented by readers during update
666 * so we must keep them in tact. When the node is later destroyed
667 * a special destroy call must be made to not free the pf memory.
668 */
669 new->pf = n->pf;
670 #endif
671
672 #ifdef CONFIG_CLS_U32_MARK
673 new->val = n->val;
674 new->mask = n->mask;
675 /* Similarly success statistics must be moved as pointers */
676 new->pcpu_success = n->pcpu_success;
677 #endif
678 new->tp = tp;
679 memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
680
681 tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE);
682
683 return new;
684 }
685
686 static int u32_change(struct net *net, struct sk_buff *in_skb,
687 struct tcf_proto *tp, unsigned long base, u32 handle,
688 struct nlattr **tca,
689 unsigned long *arg, bool ovr)
690 {
691 struct tc_u_common *tp_c = tp->data;
692 struct tc_u_hnode *ht;
693 struct tc_u_knode *n;
694 struct tc_u32_sel *s;
695 struct nlattr *opt = tca[TCA_OPTIONS];
696 struct nlattr *tb[TCA_U32_MAX + 1];
697 u32 htid;
698 int err;
699 #ifdef CONFIG_CLS_U32_PERF
700 size_t size;
701 #endif
702
703 if (opt == NULL)
704 return handle ? -EINVAL : 0;
705
706 err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy);
707 if (err < 0)
708 return err;
709
710 n = (struct tc_u_knode *)*arg;
711 if (n) {
712 struct tc_u_knode *new;
713
714 if (TC_U32_KEY(n->handle) == 0)
715 return -EINVAL;
716
717 new = u32_init_knode(tp, n);
718 if (!new)
719 return -ENOMEM;
720
721 err = u32_set_parms(net, tp, base,
722 rtnl_dereference(n->ht_up), new, tb,
723 tca[TCA_RATE], ovr);
724
725 if (err) {
726 u32_destroy_key(tp, new, false);
727 return err;
728 }
729
730 u32_replace_knode(tp, tp_c, new);
731 tcf_unbind_filter(tp, &n->res);
732 call_rcu(&n->rcu, u32_delete_key_rcu);
733 return 0;
734 }
735
736 if (tb[TCA_U32_DIVISOR]) {
737 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
738
739 if (--divisor > 0x100)
740 return -EINVAL;
741 if (TC_U32_KEY(handle))
742 return -EINVAL;
743 if (handle == 0) {
744 handle = gen_new_htid(tp->data);
745 if (handle == 0)
746 return -ENOMEM;
747 }
748 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
749 if (ht == NULL)
750 return -ENOBUFS;
751 ht->tp_c = tp_c;
752 ht->refcnt = 1;
753 ht->divisor = divisor;
754 ht->handle = handle;
755 ht->prio = tp->prio;
756 RCU_INIT_POINTER(ht->next, tp_c->hlist);
757 rcu_assign_pointer(tp_c->hlist, ht);
758 *arg = (unsigned long)ht;
759 return 0;
760 }
761
762 if (tb[TCA_U32_HASH]) {
763 htid = nla_get_u32(tb[TCA_U32_HASH]);
764 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
765 ht = rtnl_dereference(tp->root);
766 htid = ht->handle;
767 } else {
768 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
769 if (ht == NULL)
770 return -EINVAL;
771 }
772 } else {
773 ht = rtnl_dereference(tp->root);
774 htid = ht->handle;
775 }
776
777 if (ht->divisor < TC_U32_HASH(htid))
778 return -EINVAL;
779
780 if (handle) {
781 if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid))
782 return -EINVAL;
783 handle = htid | TC_U32_NODE(handle);
784 } else
785 handle = gen_new_kid(ht, htid);
786
787 if (tb[TCA_U32_SEL] == NULL)
788 return -EINVAL;
789
790 s = nla_data(tb[TCA_U32_SEL]);
791
792 n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL);
793 if (n == NULL)
794 return -ENOBUFS;
795
796 #ifdef CONFIG_CLS_U32_PERF
797 size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64);
798 n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt));
799 if (!n->pf) {
800 kfree(n);
801 return -ENOBUFS;
802 }
803 #endif
804
805 memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
806 RCU_INIT_POINTER(n->ht_up, ht);
807 n->handle = handle;
808 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
809 tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE);
810 n->tp = tp;
811
812 #ifdef CONFIG_CLS_U32_MARK
813 n->pcpu_success = alloc_percpu(u32);
814 if (!n->pcpu_success) {
815 err = -ENOMEM;
816 goto errout;
817 }
818
819 if (tb[TCA_U32_MARK]) {
820 struct tc_u32_mark *mark;
821
822 mark = nla_data(tb[TCA_U32_MARK]);
823 n->val = mark->val;
824 n->mask = mark->mask;
825 }
826 #endif
827
828 err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr);
829 if (err == 0) {
830 struct tc_u_knode __rcu **ins;
831 struct tc_u_knode *pins;
832
833 ins = &ht->ht[TC_U32_HASH(handle)];
834 for (pins = rtnl_dereference(*ins); pins;
835 ins = &pins->next, pins = rtnl_dereference(*ins))
836 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
837 break;
838
839 RCU_INIT_POINTER(n->next, pins);
840 rcu_assign_pointer(*ins, n);
841
842 *arg = (unsigned long)n;
843 return 0;
844 }
845
846 #ifdef CONFIG_CLS_U32_MARK
847 free_percpu(n->pcpu_success);
848 errout:
849 #endif
850
851 #ifdef CONFIG_CLS_U32_PERF
852 free_percpu(n->pf);
853 #endif
854 kfree(n);
855 return err;
856 }
857
858 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg)
859 {
860 struct tc_u_common *tp_c = tp->data;
861 struct tc_u_hnode *ht;
862 struct tc_u_knode *n;
863 unsigned int h;
864
865 if (arg->stop)
866 return;
867
868 for (ht = rtnl_dereference(tp_c->hlist);
869 ht;
870 ht = rtnl_dereference(ht->next)) {
871 if (ht->prio != tp->prio)
872 continue;
873 if (arg->count >= arg->skip) {
874 if (arg->fn(tp, (unsigned long)ht, arg) < 0) {
875 arg->stop = 1;
876 return;
877 }
878 }
879 arg->count++;
880 for (h = 0; h <= ht->divisor; h++) {
881 for (n = rtnl_dereference(ht->ht[h]);
882 n;
883 n = rtnl_dereference(n->next)) {
884 if (arg->count < arg->skip) {
885 arg->count++;
886 continue;
887 }
888 if (arg->fn(tp, (unsigned long)n, arg) < 0) {
889 arg->stop = 1;
890 return;
891 }
892 arg->count++;
893 }
894 }
895 }
896 }
897
898 static int u32_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
899 struct sk_buff *skb, struct tcmsg *t)
900 {
901 struct tc_u_knode *n = (struct tc_u_knode *)fh;
902 struct tc_u_hnode *ht_up, *ht_down;
903 struct nlattr *nest;
904
905 if (n == NULL)
906 return skb->len;
907
908 t->tcm_handle = n->handle;
909
910 nest = nla_nest_start(skb, TCA_OPTIONS);
911 if (nest == NULL)
912 goto nla_put_failure;
913
914 if (TC_U32_KEY(n->handle) == 0) {
915 struct tc_u_hnode *ht = (struct tc_u_hnode *)fh;
916 u32 divisor = ht->divisor + 1;
917
918 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
919 goto nla_put_failure;
920 } else {
921 #ifdef CONFIG_CLS_U32_PERF
922 struct tc_u32_pcnt *gpf;
923 int cpu;
924 #endif
925
926 if (nla_put(skb, TCA_U32_SEL,
927 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
928 &n->sel))
929 goto nla_put_failure;
930
931 ht_up = rtnl_dereference(n->ht_up);
932 if (ht_up) {
933 u32 htid = n->handle & 0xFFFFF000;
934 if (nla_put_u32(skb, TCA_U32_HASH, htid))
935 goto nla_put_failure;
936 }
937 if (n->res.classid &&
938 nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
939 goto nla_put_failure;
940
941 ht_down = rtnl_dereference(n->ht_down);
942 if (ht_down &&
943 nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
944 goto nla_put_failure;
945
946 #ifdef CONFIG_CLS_U32_MARK
947 if ((n->val || n->mask)) {
948 struct tc_u32_mark mark = {.val = n->val,
949 .mask = n->mask,
950 .success = 0};
951 int cpum;
952
953 for_each_possible_cpu(cpum) {
954 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
955
956 mark.success += cnt;
957 }
958
959 if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
960 goto nla_put_failure;
961 }
962 #endif
963
964 if (tcf_exts_dump(skb, &n->exts) < 0)
965 goto nla_put_failure;
966
967 #ifdef CONFIG_NET_CLS_IND
968 if (n->ifindex) {
969 struct net_device *dev;
970 dev = __dev_get_by_index(net, n->ifindex);
971 if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
972 goto nla_put_failure;
973 }
974 #endif
975 #ifdef CONFIG_CLS_U32_PERF
976 gpf = kzalloc(sizeof(struct tc_u32_pcnt) +
977 n->sel.nkeys * sizeof(u64),
978 GFP_KERNEL);
979 if (!gpf)
980 goto nla_put_failure;
981
982 for_each_possible_cpu(cpu) {
983 int i;
984 struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
985
986 gpf->rcnt += pf->rcnt;
987 gpf->rhit += pf->rhit;
988 for (i = 0; i < n->sel.nkeys; i++)
989 gpf->kcnts[i] += pf->kcnts[i];
990 }
991
992 if (nla_put(skb, TCA_U32_PCNT,
993 sizeof(struct tc_u32_pcnt) + n->sel.nkeys*sizeof(u64),
994 gpf)) {
995 kfree(gpf);
996 goto nla_put_failure;
997 }
998 kfree(gpf);
999 #endif
1000 }
1001
1002 nla_nest_end(skb, nest);
1003
1004 if (TC_U32_KEY(n->handle))
1005 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1006 goto nla_put_failure;
1007 return skb->len;
1008
1009 nla_put_failure:
1010 nla_nest_cancel(skb, nest);
1011 return -1;
1012 }
1013
1014 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1015 .kind = "u32",
1016 .classify = u32_classify,
1017 .init = u32_init,
1018 .destroy = u32_destroy,
1019 .get = u32_get,
1020 .change = u32_change,
1021 .delete = u32_delete,
1022 .walk = u32_walk,
1023 .dump = u32_dump,
1024 .owner = THIS_MODULE,
1025 };
1026
1027 static int __init init_u32(void)
1028 {
1029 pr_info("u32 classifier\n");
1030 #ifdef CONFIG_CLS_U32_PERF
1031 pr_info(" Performance counters on\n");
1032 #endif
1033 #ifdef CONFIG_NET_CLS_IND
1034 pr_info(" input device check on\n");
1035 #endif
1036 #ifdef CONFIG_NET_CLS_ACT
1037 pr_info(" Actions configured\n");
1038 #endif
1039 return register_tcf_proto_ops(&cls_u32_ops);
1040 }
1041
1042 static void __exit exit_u32(void)
1043 {
1044 unregister_tcf_proto_ops(&cls_u32_ops);
1045 }
1046
1047 module_init(init_u32)
1048 module_exit(exit_u32)
1049 MODULE_LICENSE("GPL");
Linux kernel - Deliverables
This page took 0.060179 seconds and 6 git commands to generate.