projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
net: sched: add cls_u32 offload hooks for netdevs
[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 #include <linux/netdevice.h>
47
48 struct tc_u_knode {
49 struct tc_u_knode __rcu *next;
50 u32 handle;
51 struct tc_u_hnode __rcu *ht_up;
52 struct tcf_exts exts;
53 #ifdef CONFIG_NET_CLS_IND
54 int ifindex;
55 #endif
56 u8 fshift;
57 struct tcf_result res;
58 struct tc_u_hnode __rcu *ht_down;
59 #ifdef CONFIG_CLS_U32_PERF
60 struct tc_u32_pcnt __percpu *pf;
61 #endif
62 #ifdef CONFIG_CLS_U32_MARK
63 u32 val;
64 u32 mask;
65 u32 __percpu *pcpu_success;
66 #endif
67 struct tcf_proto *tp;
68 struct rcu_head rcu;
69 /* The 'sel' field MUST be the last field in structure to allow for
70 * tc_u32_keys allocated at end of structure.
71 */
72 struct tc_u32_sel sel;
73 };
74
75 struct tc_u_hnode {
76 struct tc_u_hnode __rcu *next;
77 u32 handle;
78 u32 prio;
79 struct tc_u_common *tp_c;
80 int refcnt;
81 unsigned int divisor;
82 struct rcu_head rcu;
83 /* The 'ht' field MUST be the last field in structure to allow for
84 * more entries allocated at end of structure.
85 */
86 struct tc_u_knode __rcu *ht[1];
87 };
88
89 struct tc_u_common {
90 struct tc_u_hnode __rcu *hlist;
91 struct Qdisc *q;
92 int refcnt;
93 u32 hgenerator;
94 struct rcu_head rcu;
95 };
96
97 static inline unsigned int u32_hash_fold(__be32 key,
98 const struct tc_u32_sel *sel,
99 u8 fshift)
100 {
101 unsigned int h = ntohl(key & sel->hmask) >> fshift;
102
103 return h;
104 }
105
106 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp, struct tcf_result *res)
107 {
108 struct {
109 struct tc_u_knode *knode;
110 unsigned int off;
111 } stack[TC_U32_MAXDEPTH];
112
113 struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
114 unsigned int off = skb_network_offset(skb);
115 struct tc_u_knode *n;
116 int sdepth = 0;
117 int off2 = 0;
118 int sel = 0;
119 #ifdef CONFIG_CLS_U32_PERF
120 int j;
121 #endif
122 int i, r;
123
124 next_ht:
125 n = rcu_dereference_bh(ht->ht[sel]);
126
127 next_knode:
128 if (n) {
129 struct tc_u32_key *key = n->sel.keys;
130
131 #ifdef CONFIG_CLS_U32_PERF
132 __this_cpu_inc(n->pf->rcnt);
133 j = 0;
134 #endif
135
136 #ifdef CONFIG_CLS_U32_MARK
137 if ((skb->mark & n->mask) != n->val) {
138 n = rcu_dereference_bh(n->next);
139 goto next_knode;
140 } else {
141 __this_cpu_inc(*n->pcpu_success);
142 }
143 #endif
144
145 for (i = n->sel.nkeys; i > 0; i--, key++) {
146 int toff = off + key->off + (off2 & key->offmask);
147 __be32 *data, hdata;
148
149 if (skb_headroom(skb) + toff > INT_MAX)
150 goto out;
151
152 data = skb_header_pointer(skb, toff, 4, &hdata);
153 if (!data)
154 goto out;
155 if ((*data ^ key->val) & key->mask) {
156 n = rcu_dereference_bh(n->next);
157 goto next_knode;
158 }
159 #ifdef CONFIG_CLS_U32_PERF
160 __this_cpu_inc(n->pf->kcnts[j]);
161 j++;
162 #endif
163 }
164
165 ht = rcu_dereference_bh(n->ht_down);
166 if (!ht) {
167 check_terminal:
168 if (n->sel.flags & TC_U32_TERMINAL) {
169
170 *res = n->res;
171 #ifdef CONFIG_NET_CLS_IND
172 if (!tcf_match_indev(skb, n->ifindex)) {
173 n = rcu_dereference_bh(n->next);
174 goto next_knode;
175 }
176 #endif
177 #ifdef CONFIG_CLS_U32_PERF
178 __this_cpu_inc(n->pf->rhit);
179 #endif
180 r = tcf_exts_exec(skb, &n->exts, res);
181 if (r < 0) {
182 n = rcu_dereference_bh(n->next);
183 goto next_knode;
184 }
185
186 return r;
187 }
188 n = rcu_dereference_bh(n->next);
189 goto next_knode;
190 }
191
192 /* PUSH */
193 if (sdepth >= TC_U32_MAXDEPTH)
194 goto deadloop;
195 stack[sdepth].knode = n;
196 stack[sdepth].off = off;
197 sdepth++;
198
199 ht = rcu_dereference_bh(n->ht_down);
200 sel = 0;
201 if (ht->divisor) {
202 __be32 *data, hdata;
203
204 data = skb_header_pointer(skb, off + n->sel.hoff, 4,
205 &hdata);
206 if (!data)
207 goto out;
208 sel = ht->divisor & u32_hash_fold(*data, &n->sel,
209 n->fshift);
210 }
211 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
212 goto next_ht;
213
214 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
215 off2 = n->sel.off + 3;
216 if (n->sel.flags & TC_U32_VAROFFSET) {
217 __be16 *data, hdata;
218
219 data = skb_header_pointer(skb,
220 off + n->sel.offoff,
221 2, &hdata);
222 if (!data)
223 goto out;
224 off2 += ntohs(n->sel.offmask & *data) >>
225 n->sel.offshift;
226 }
227 off2 &= ~3;
228 }
229 if (n->sel.flags & TC_U32_EAT) {
230 off += off2;
231 off2 = 0;
232 }
233
234 if (off < skb->len)
235 goto next_ht;
236 }
237
238 /* POP */
239 if (sdepth--) {
240 n = stack[sdepth].knode;
241 ht = rcu_dereference_bh(n->ht_up);
242 off = stack[sdepth].off;
243 goto check_terminal;
244 }
245 out:
246 return -1;
247
248 deadloop:
249 net_warn_ratelimited("cls_u32: dead loop\n");
250 return -1;
251 }
252
253 static struct tc_u_hnode *
254 u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
255 {
256 struct tc_u_hnode *ht;
257
258 for (ht = rtnl_dereference(tp_c->hlist);
259 ht;
260 ht = rtnl_dereference(ht->next))
261 if (ht->handle == handle)
262 break;
263
264 return ht;
265 }
266
267 static struct tc_u_knode *
268 u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
269 {
270 unsigned int sel;
271 struct tc_u_knode *n = NULL;
272
273 sel = TC_U32_HASH(handle);
274 if (sel > ht->divisor)
275 goto out;
276
277 for (n = rtnl_dereference(ht->ht[sel]);
278 n;
279 n = rtnl_dereference(n->next))
280 if (n->handle == handle)
281 break;
282 out:
283 return n;
284 }
285
286
287 static unsigned long u32_get(struct tcf_proto *tp, u32 handle)
288 {
289 struct tc_u_hnode *ht;
290 struct tc_u_common *tp_c = tp->data;
291
292 if (TC_U32_HTID(handle) == TC_U32_ROOT)
293 ht = rtnl_dereference(tp->root);
294 else
295 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
296
297 if (!ht)
298 return 0;
299
300 if (TC_U32_KEY(handle) == 0)
301 return (unsigned long)ht;
302
303 return (unsigned long)u32_lookup_key(ht, handle);
304 }
305
306 static u32 gen_new_htid(struct tc_u_common *tp_c)
307 {
308 int i = 0x800;
309
310 /* hgenerator only used inside rtnl lock it is safe to increment
311 * without read _copy_ update semantics
312 */
313 do {
314 if (++tp_c->hgenerator == 0x7FF)
315 tp_c->hgenerator = 1;
316 } while (--i > 0 && u32_lookup_ht(tp_c, (tp_c->hgenerator|0x800)<<20));
317
318 return i > 0 ? (tp_c->hgenerator|0x800)<<20 : 0;
319 }
320
321 static int u32_init(struct tcf_proto *tp)
322 {
323 struct tc_u_hnode *root_ht;
324 struct tc_u_common *tp_c;
325
326 tp_c = tp->q->u32_node;
327
328 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
329 if (root_ht == NULL)
330 return -ENOBUFS;
331
332 root_ht->divisor = 0;
333 root_ht->refcnt++;
334 root_ht->handle = tp_c ? gen_new_htid(tp_c) : 0x80000000;
335 root_ht->prio = tp->prio;
336
337 if (tp_c == NULL) {
338 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
339 if (tp_c == NULL) {
340 kfree(root_ht);
341 return -ENOBUFS;
342 }
343 tp_c->q = tp->q;
344 tp->q->u32_node = tp_c;
345 }
346
347 tp_c->refcnt++;
348 RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
349 rcu_assign_pointer(tp_c->hlist, root_ht);
350 root_ht->tp_c = tp_c;
351
352 rcu_assign_pointer(tp->root, root_ht);
353 tp->data = tp_c;
354 return 0;
355 }
356
357 static int u32_destroy_key(struct tcf_proto *tp,
358 struct tc_u_knode *n,
359 bool free_pf)
360 {
361 tcf_exts_destroy(&n->exts);
362 if (n->ht_down)
363 n->ht_down->refcnt--;
364 #ifdef CONFIG_CLS_U32_PERF
365 if (free_pf)
366 free_percpu(n->pf);
367 #endif
368 #ifdef CONFIG_CLS_U32_MARK
369 if (free_pf)
370 free_percpu(n->pcpu_success);
371 #endif
372 kfree(n);
373 return 0;
374 }
375
376 /* u32_delete_key_rcu should be called when free'ing a copied
377 * version of a tc_u_knode obtained from u32_init_knode(). When
378 * copies are obtained from u32_init_knode() the statistics are
379 * shared between the old and new copies to allow readers to
380 * continue to update the statistics during the copy. To support
381 * this the u32_delete_key_rcu variant does not free the percpu
382 * statistics.
383 */
384 static void u32_delete_key_rcu(struct rcu_head *rcu)
385 {
386 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
387
388 u32_destroy_key(key->tp, key, false);
389 }
390
391 /* u32_delete_key_freepf_rcu is the rcu callback variant
392 * that free's the entire structure including the statistics
393 * percpu variables. Only use this if the key is not a copy
394 * returned by u32_init_knode(). See u32_delete_key_rcu()
395 * for the variant that should be used with keys return from
396 * u32_init_knode()
397 */
398 static void u32_delete_key_freepf_rcu(struct rcu_head *rcu)
399 {
400 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
401
402 u32_destroy_key(key->tp, key, true);
403 }
404
405 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
406 {
407 struct tc_u_knode __rcu **kp;
408 struct tc_u_knode *pkp;
409 struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
410
411 if (ht) {
412 kp = &ht->ht[TC_U32_HASH(key->handle)];
413 for (pkp = rtnl_dereference(*kp); pkp;
414 kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
415 if (pkp == key) {
416 RCU_INIT_POINTER(*kp, key->next);
417
418 tcf_unbind_filter(tp, &key->res);
419 call_rcu(&key->rcu, u32_delete_key_freepf_rcu);
420 return 0;
421 }
422 }
423 }
424 WARN_ON(1);
425 return 0;
426 }
427
428 static void u32_remove_hw_knode(struct tcf_proto *tp, u32 handle)
429 {
430 struct net_device *dev = tp->q->dev_queue->dev;
431 struct tc_cls_u32_offload u32_offload = {0};
432 struct tc_to_netdev offload;
433
434 offload.type = TC_SETUP_CLSU32;
435 offload.cls_u32 = &u32_offload;
436
437 if (dev->netdev_ops->ndo_setup_tc) {
438 offload.cls_u32->command = TC_CLSU32_DELETE_KNODE;
439 offload.cls_u32->knode.handle = handle;
440 dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle,
441 tp->protocol, &offload);
442 }
443 }
444
445 static void u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h)
446 {
447 struct net_device *dev = tp->q->dev_queue->dev;
448 struct tc_cls_u32_offload u32_offload = {0};
449 struct tc_to_netdev offload;
450
451 offload.type = TC_SETUP_CLSU32;
452 offload.cls_u32 = &u32_offload;
453
454 if (dev->netdev_ops->ndo_setup_tc) {
455 offload.cls_u32->command = TC_CLSU32_NEW_HNODE;
456 offload.cls_u32->hnode.divisor = h->divisor;
457 offload.cls_u32->hnode.handle = h->handle;
458 offload.cls_u32->hnode.prio = h->prio;
459
460 dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle,
461 tp->protocol, &offload);
462 }
463 }
464
465 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h)
466 {
467 struct net_device *dev = tp->q->dev_queue->dev;
468 struct tc_cls_u32_offload u32_offload = {0};
469 struct tc_to_netdev offload;
470
471 offload.type = TC_SETUP_CLSU32;
472 offload.cls_u32 = &u32_offload;
473
474 if (dev->netdev_ops->ndo_setup_tc) {
475 offload.cls_u32->command = TC_CLSU32_DELETE_HNODE;
476 offload.cls_u32->hnode.divisor = h->divisor;
477 offload.cls_u32->hnode.handle = h->handle;
478 offload.cls_u32->hnode.prio = h->prio;
479
480 dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle,
481 tp->protocol, &offload);
482 }
483 }
484
485 static void u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n)
486 {
487 struct net_device *dev = tp->q->dev_queue->dev;
488 struct tc_cls_u32_offload u32_offload = {0};
489 struct tc_to_netdev offload;
490
491 offload.type = TC_SETUP_CLSU32;
492 offload.cls_u32 = &u32_offload;
493
494 if (dev->netdev_ops->ndo_setup_tc) {
495 offload.cls_u32->command = TC_CLSU32_REPLACE_KNODE;
496 offload.cls_u32->knode.handle = n->handle;
497 offload.cls_u32->knode.fshift = n->fshift;
498 #ifdef CONFIG_CLS_U32_MARK
499 offload.cls_u32->knode.val = n->val;
500 offload.cls_u32->knode.mask = n->mask;
501 #else
502 offload.cls_u32->knode.val = 0;
503 offload.cls_u32->knode.mask = 0;
504 #endif
505 offload.cls_u32->knode.sel = &n->sel;
506 offload.cls_u32->knode.exts = &n->exts;
507 if (n->ht_down)
508 offload.cls_u32->knode.link_handle = n->ht_down->handle;
509
510 dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle,
511 tp->protocol, &offload);
512 }
513 }
514
515 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
516 {
517 struct tc_u_knode *n;
518 unsigned int h;
519
520 for (h = 0; h <= ht->divisor; h++) {
521 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
522 RCU_INIT_POINTER(ht->ht[h],
523 rtnl_dereference(n->next));
524 tcf_unbind_filter(tp, &n->res);
525 u32_remove_hw_knode(tp, n->handle);
526 call_rcu(&n->rcu, u32_delete_key_freepf_rcu);
527 }
528 }
529 }
530
531 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
532 {
533 struct tc_u_common *tp_c = tp->data;
534 struct tc_u_hnode __rcu **hn;
535 struct tc_u_hnode *phn;
536
537 WARN_ON(ht->refcnt);
538
539 u32_clear_hnode(tp, ht);
540
541 hn = &tp_c->hlist;
542 for (phn = rtnl_dereference(*hn);
543 phn;
544 hn = &phn->next, phn = rtnl_dereference(*hn)) {
545 if (phn == ht) {
546 u32_clear_hw_hnode(tp, ht);
547 RCU_INIT_POINTER(*hn, ht->next);
548 kfree_rcu(ht, rcu);
549 return 0;
550 }
551 }
552
553 return -ENOENT;
554 }
555
556 static bool ht_empty(struct tc_u_hnode *ht)
557 {
558 unsigned int h;
559
560 for (h = 0; h <= ht->divisor; h++)
561 if (rcu_access_pointer(ht->ht[h]))
562 return false;
563
564 return true;
565 }
566
567 static bool u32_destroy(struct tcf_proto *tp, bool force)
568 {
569 struct tc_u_common *tp_c = tp->data;
570 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
571
572 WARN_ON(root_ht == NULL);
573
574 if (!force) {
575 if (root_ht) {
576 if (root_ht->refcnt > 1)
577 return false;
578 if (root_ht->refcnt == 1) {
579 if (!ht_empty(root_ht))
580 return false;
581 }
582 }
583
584 if (tp_c->refcnt > 1)
585 return false;
586
587 if (tp_c->refcnt == 1) {
588 struct tc_u_hnode *ht;
589
590 for (ht = rtnl_dereference(tp_c->hlist);
591 ht;
592 ht = rtnl_dereference(ht->next))
593 if (!ht_empty(ht))
594 return false;
595 }
596 }
597
598 if (root_ht && --root_ht->refcnt == 0)
599 u32_destroy_hnode(tp, root_ht);
600
601 if (--tp_c->refcnt == 0) {
602 struct tc_u_hnode *ht;
603
604 tp->q->u32_node = NULL;
605
606 for (ht = rtnl_dereference(tp_c->hlist);
607 ht;
608 ht = rtnl_dereference(ht->next)) {
609 ht->refcnt--;
610 u32_clear_hnode(tp, ht);
611 }
612
613 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
614 RCU_INIT_POINTER(tp_c->hlist, ht->next);
615 kfree_rcu(ht, rcu);
616 }
617
618 kfree(tp_c);
619 }
620
621 tp->data = NULL;
622 return true;
623 }
624
625 static int u32_delete(struct tcf_proto *tp, unsigned long arg)
626 {
627 struct tc_u_hnode *ht = (struct tc_u_hnode *)arg;
628 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
629
630 if (ht == NULL)
631 return 0;
632
633 if (TC_U32_KEY(ht->handle)) {
634 u32_remove_hw_knode(tp, ht->handle);
635 return u32_delete_key(tp, (struct tc_u_knode *)ht);
636 }
637
638 if (root_ht == ht)
639 return -EINVAL;
640
641 if (ht->refcnt == 1) {
642 ht->refcnt--;
643 u32_destroy_hnode(tp, ht);
644 } else {
645 return -EBUSY;
646 }
647
648 return 0;
649 }
650
651 #define NR_U32_NODE (1<<12)
652 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 handle)
653 {
654 struct tc_u_knode *n;
655 unsigned long i;
656 unsigned long *bitmap = kzalloc(BITS_TO_LONGS(NR_U32_NODE) * sizeof(unsigned long),
657 GFP_KERNEL);
658 if (!bitmap)
659 return handle | 0xFFF;
660
661 for (n = rtnl_dereference(ht->ht[TC_U32_HASH(handle)]);
662 n;
663 n = rtnl_dereference(n->next))
664 set_bit(TC_U32_NODE(n->handle), bitmap);
665
666 i = find_next_zero_bit(bitmap, NR_U32_NODE, 0x800);
667 if (i >= NR_U32_NODE)
668 i = find_next_zero_bit(bitmap, NR_U32_NODE, 1);
669
670 kfree(bitmap);
671 return handle | (i >= NR_U32_NODE ? 0xFFF : i);
672 }
673
674 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
675 [TCA_U32_CLASSID] = { .type = NLA_U32 },
676 [TCA_U32_HASH] = { .type = NLA_U32 },
677 [TCA_U32_LINK] = { .type = NLA_U32 },
678 [TCA_U32_DIVISOR] = { .type = NLA_U32 },
679 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) },
680 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ },
681 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) },
682 };
683
684 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
685 unsigned long base, struct tc_u_hnode *ht,
686 struct tc_u_knode *n, struct nlattr **tb,
687 struct nlattr *est, bool ovr)
688 {
689 int err;
690 struct tcf_exts e;
691
692 tcf_exts_init(&e, TCA_U32_ACT, TCA_U32_POLICE);
693 err = tcf_exts_validate(net, tp, tb, est, &e, ovr);
694 if (err < 0)
695 return err;
696
697 err = -EINVAL;
698 if (tb[TCA_U32_LINK]) {
699 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
700 struct tc_u_hnode *ht_down = NULL, *ht_old;
701
702 if (TC_U32_KEY(handle))
703 goto errout;
704
705 if (handle) {
706 ht_down = u32_lookup_ht(ht->tp_c, handle);
707
708 if (ht_down == NULL)
709 goto errout;
710 ht_down->refcnt++;
711 }
712
713 ht_old = rtnl_dereference(n->ht_down);
714 rcu_assign_pointer(n->ht_down, ht_down);
715
716 if (ht_old)
717 ht_old->refcnt--;
718 }
719 if (tb[TCA_U32_CLASSID]) {
720 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
721 tcf_bind_filter(tp, &n->res, base);
722 }
723
724 #ifdef CONFIG_NET_CLS_IND
725 if (tb[TCA_U32_INDEV]) {
726 int ret;
727 ret = tcf_change_indev(net, tb[TCA_U32_INDEV]);
728 if (ret < 0)
729 goto errout;
730 n->ifindex = ret;
731 }
732 #endif
733 tcf_exts_change(tp, &n->exts, &e);
734
735 return 0;
736 errout:
737 tcf_exts_destroy(&e);
738 return err;
739 }
740
741 static void u32_replace_knode(struct tcf_proto *tp,
742 struct tc_u_common *tp_c,
743 struct tc_u_knode *n)
744 {
745 struct tc_u_knode __rcu **ins;
746 struct tc_u_knode *pins;
747 struct tc_u_hnode *ht;
748
749 if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
750 ht = rtnl_dereference(tp->root);
751 else
752 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
753
754 ins = &ht->ht[TC_U32_HASH(n->handle)];
755
756 /* The node must always exist for it to be replaced if this is not the
757 * case then something went very wrong elsewhere.
758 */
759 for (pins = rtnl_dereference(*ins); ;
760 ins = &pins->next, pins = rtnl_dereference(*ins))
761 if (pins->handle == n->handle)
762 break;
763
764 RCU_INIT_POINTER(n->next, pins->next);
765 rcu_assign_pointer(*ins, n);
766 }
767
768 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp,
769 struct tc_u_knode *n)
770 {
771 struct tc_u_knode *new;
772 struct tc_u32_sel *s = &n->sel;
773
774 new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key),
775 GFP_KERNEL);
776
777 if (!new)
778 return NULL;
779
780 RCU_INIT_POINTER(new->next, n->next);
781 new->handle = n->handle;
782 RCU_INIT_POINTER(new->ht_up, n->ht_up);
783
784 #ifdef CONFIG_NET_CLS_IND
785 new->ifindex = n->ifindex;
786 #endif
787 new->fshift = n->fshift;
788 new->res = n->res;
789 RCU_INIT_POINTER(new->ht_down, n->ht_down);
790
791 /* bump reference count as long as we hold pointer to structure */
792 if (new->ht_down)
793 new->ht_down->refcnt++;
794
795 #ifdef CONFIG_CLS_U32_PERF
796 /* Statistics may be incremented by readers during update
797 * so we must keep them in tact. When the node is later destroyed
798 * a special destroy call must be made to not free the pf memory.
799 */
800 new->pf = n->pf;
801 #endif
802
803 #ifdef CONFIG_CLS_U32_MARK
804 new->val = n->val;
805 new->mask = n->mask;
806 /* Similarly success statistics must be moved as pointers */
807 new->pcpu_success = n->pcpu_success;
808 #endif
809 new->tp = tp;
810 memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
811
812 tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE);
813
814 return new;
815 }
816
817 static int u32_change(struct net *net, struct sk_buff *in_skb,
818 struct tcf_proto *tp, unsigned long base, u32 handle,
819 struct nlattr **tca,
820 unsigned long *arg, bool ovr)
821 {
822 struct tc_u_common *tp_c = tp->data;
823 struct tc_u_hnode *ht;
824 struct tc_u_knode *n;
825 struct tc_u32_sel *s;
826 struct nlattr *opt = tca[TCA_OPTIONS];
827 struct nlattr *tb[TCA_U32_MAX + 1];
828 u32 htid;
829 int err;
830 #ifdef CONFIG_CLS_U32_PERF
831 size_t size;
832 #endif
833
834 if (opt == NULL)
835 return handle ? -EINVAL : 0;
836
837 err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy);
838 if (err < 0)
839 return err;
840
841 n = (struct tc_u_knode *)*arg;
842 if (n) {
843 struct tc_u_knode *new;
844
845 if (TC_U32_KEY(n->handle) == 0)
846 return -EINVAL;
847
848 new = u32_init_knode(tp, n);
849 if (!new)
850 return -ENOMEM;
851
852 err = u32_set_parms(net, tp, base,
853 rtnl_dereference(n->ht_up), new, tb,
854 tca[TCA_RATE], ovr);
855
856 if (err) {
857 u32_destroy_key(tp, new, false);
858 return err;
859 }
860
861 u32_replace_knode(tp, tp_c, new);
862 tcf_unbind_filter(tp, &n->res);
863 call_rcu(&n->rcu, u32_delete_key_rcu);
864 u32_replace_hw_knode(tp, new);
865 return 0;
866 }
867
868 if (tb[TCA_U32_DIVISOR]) {
869 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
870
871 if (--divisor > 0x100)
872 return -EINVAL;
873 if (TC_U32_KEY(handle))
874 return -EINVAL;
875 if (handle == 0) {
876 handle = gen_new_htid(tp->data);
877 if (handle == 0)
878 return -ENOMEM;
879 }
880 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
881 if (ht == NULL)
882 return -ENOBUFS;
883 ht->tp_c = tp_c;
884 ht->refcnt = 1;
885 ht->divisor = divisor;
886 ht->handle = handle;
887 ht->prio = tp->prio;
888 RCU_INIT_POINTER(ht->next, tp_c->hlist);
889 rcu_assign_pointer(tp_c->hlist, ht);
890 *arg = (unsigned long)ht;
891
892 u32_replace_hw_hnode(tp, ht);
893 return 0;
894 }
895
896 if (tb[TCA_U32_HASH]) {
897 htid = nla_get_u32(tb[TCA_U32_HASH]);
898 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
899 ht = rtnl_dereference(tp->root);
900 htid = ht->handle;
901 } else {
902 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
903 if (ht == NULL)
904 return -EINVAL;
905 }
906 } else {
907 ht = rtnl_dereference(tp->root);
908 htid = ht->handle;
909 }
910
911 if (ht->divisor < TC_U32_HASH(htid))
912 return -EINVAL;
913
914 if (handle) {
915 if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid))
916 return -EINVAL;
917 handle = htid | TC_U32_NODE(handle);
918 } else
919 handle = gen_new_kid(ht, htid);
920
921 if (tb[TCA_U32_SEL] == NULL)
922 return -EINVAL;
923
924 s = nla_data(tb[TCA_U32_SEL]);
925
926 n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL);
927 if (n == NULL)
928 return -ENOBUFS;
929
930 #ifdef CONFIG_CLS_U32_PERF
931 size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64);
932 n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt));
933 if (!n->pf) {
934 kfree(n);
935 return -ENOBUFS;
936 }
937 #endif
938
939 memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
940 RCU_INIT_POINTER(n->ht_up, ht);
941 n->handle = handle;
942 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
943 tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE);
944 n->tp = tp;
945
946 #ifdef CONFIG_CLS_U32_MARK
947 n->pcpu_success = alloc_percpu(u32);
948 if (!n->pcpu_success) {
949 err = -ENOMEM;
950 goto errout;
951 }
952
953 if (tb[TCA_U32_MARK]) {
954 struct tc_u32_mark *mark;
955
956 mark = nla_data(tb[TCA_U32_MARK]);
957 n->val = mark->val;
958 n->mask = mark->mask;
959 }
960 #endif
961
962 err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr);
963 if (err == 0) {
964 struct tc_u_knode __rcu **ins;
965 struct tc_u_knode *pins;
966
967 ins = &ht->ht[TC_U32_HASH(handle)];
968 for (pins = rtnl_dereference(*ins); pins;
969 ins = &pins->next, pins = rtnl_dereference(*ins))
970 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
971 break;
972
973 RCU_INIT_POINTER(n->next, pins);
974 rcu_assign_pointer(*ins, n);
975 u32_replace_hw_knode(tp, n);
976 *arg = (unsigned long)n;
977 return 0;
978 }
979
980 #ifdef CONFIG_CLS_U32_MARK
981 free_percpu(n->pcpu_success);
982 errout:
983 #endif
984
985 #ifdef CONFIG_CLS_U32_PERF
986 free_percpu(n->pf);
987 #endif
988 kfree(n);
989 return err;
990 }
991
992 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg)
993 {
994 struct tc_u_common *tp_c = tp->data;
995 struct tc_u_hnode *ht;
996 struct tc_u_knode *n;
997 unsigned int h;
998
999 if (arg->stop)
1000 return;
1001
1002 for (ht = rtnl_dereference(tp_c->hlist);
1003 ht;
1004 ht = rtnl_dereference(ht->next)) {
1005 if (ht->prio != tp->prio)
1006 continue;
1007 if (arg->count >= arg->skip) {
1008 if (arg->fn(tp, (unsigned long)ht, arg) < 0) {
1009 arg->stop = 1;
1010 return;
1011 }
1012 }
1013 arg->count++;
1014 for (h = 0; h <= ht->divisor; h++) {
1015 for (n = rtnl_dereference(ht->ht[h]);
1016 n;
1017 n = rtnl_dereference(n->next)) {
1018 if (arg->count < arg->skip) {
1019 arg->count++;
1020 continue;
1021 }
1022 if (arg->fn(tp, (unsigned long)n, arg) < 0) {
1023 arg->stop = 1;
1024 return;
1025 }
1026 arg->count++;
1027 }
1028 }
1029 }
1030 }
1031
1032 static int u32_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
1033 struct sk_buff *skb, struct tcmsg *t)
1034 {
1035 struct tc_u_knode *n = (struct tc_u_knode *)fh;
1036 struct tc_u_hnode *ht_up, *ht_down;
1037 struct nlattr *nest;
1038
1039 if (n == NULL)
1040 return skb->len;
1041
1042 t->tcm_handle = n->handle;
1043
1044 nest = nla_nest_start(skb, TCA_OPTIONS);
1045 if (nest == NULL)
1046 goto nla_put_failure;
1047
1048 if (TC_U32_KEY(n->handle) == 0) {
1049 struct tc_u_hnode *ht = (struct tc_u_hnode *)fh;
1050 u32 divisor = ht->divisor + 1;
1051
1052 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1053 goto nla_put_failure;
1054 } else {
1055 #ifdef CONFIG_CLS_U32_PERF
1056 struct tc_u32_pcnt *gpf;
1057 int cpu;
1058 #endif
1059
1060 if (nla_put(skb, TCA_U32_SEL,
1061 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
1062 &n->sel))
1063 goto nla_put_failure;
1064
1065 ht_up = rtnl_dereference(n->ht_up);
1066 if (ht_up) {
1067 u32 htid = n->handle & 0xFFFFF000;
1068 if (nla_put_u32(skb, TCA_U32_HASH, htid))
1069 goto nla_put_failure;
1070 }
1071 if (n->res.classid &&
1072 nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1073 goto nla_put_failure;
1074
1075 ht_down = rtnl_dereference(n->ht_down);
1076 if (ht_down &&
1077 nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1078 goto nla_put_failure;
1079
1080 #ifdef CONFIG_CLS_U32_MARK
1081 if ((n->val || n->mask)) {
1082 struct tc_u32_mark mark = {.val = n->val,
1083 .mask = n->mask,
1084 .success = 0};
1085 int cpum;
1086
1087 for_each_possible_cpu(cpum) {
1088 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1089
1090 mark.success += cnt;
1091 }
1092
1093 if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1094 goto nla_put_failure;
1095 }
1096 #endif
1097
1098 if (tcf_exts_dump(skb, &n->exts) < 0)
1099 goto nla_put_failure;
1100
1101 #ifdef CONFIG_NET_CLS_IND
1102 if (n->ifindex) {
1103 struct net_device *dev;
1104 dev = __dev_get_by_index(net, n->ifindex);
1105 if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1106 goto nla_put_failure;
1107 }
1108 #endif
1109 #ifdef CONFIG_CLS_U32_PERF
1110 gpf = kzalloc(sizeof(struct tc_u32_pcnt) +
1111 n->sel.nkeys * sizeof(u64),
1112 GFP_KERNEL);
1113 if (!gpf)
1114 goto nla_put_failure;
1115
1116 for_each_possible_cpu(cpu) {
1117 int i;
1118 struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1119
1120 gpf->rcnt += pf->rcnt;
1121 gpf->rhit += pf->rhit;
1122 for (i = 0; i < n->sel.nkeys; i++)
1123 gpf->kcnts[i] += pf->kcnts[i];
1124 }
1125
1126 if (nla_put(skb, TCA_U32_PCNT,
1127 sizeof(struct tc_u32_pcnt) + n->sel.nkeys*sizeof(u64),
1128 gpf)) {
1129 kfree(gpf);
1130 goto nla_put_failure;
1131 }
1132 kfree(gpf);
1133 #endif
1134 }
1135
1136 nla_nest_end(skb, nest);
1137
1138 if (TC_U32_KEY(n->handle))
1139 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1140 goto nla_put_failure;
1141 return skb->len;
1142
1143 nla_put_failure:
1144 nla_nest_cancel(skb, nest);
1145 return -1;
1146 }
1147
1148 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1149 .kind = "u32",
1150 .classify = u32_classify,
1151 .init = u32_init,
1152 .destroy = u32_destroy,
1153 .get = u32_get,
1154 .change = u32_change,
1155 .delete = u32_delete,
1156 .walk = u32_walk,
1157 .dump = u32_dump,
1158 .owner = THIS_MODULE,
1159 };
1160
1161 static int __init init_u32(void)
1162 {
1163 pr_info("u32 classifier\n");
1164 #ifdef CONFIG_CLS_U32_PERF
1165 pr_info(" Performance counters on\n");
1166 #endif
1167 #ifdef CONFIG_NET_CLS_IND
1168 pr_info(" input device check on\n");
1169 #endif
1170 #ifdef CONFIG_NET_CLS_ACT
1171 pr_info(" Actions configured\n");
1172 #endif
1173 return register_tcf_proto_ops(&cls_u32_ops);
1174 }
1175
1176 static void __exit exit_u32(void)
1177 {
1178 unregister_tcf_proto_ops(&cls_u32_ops);
1179 }
1180
1181 module_init(init_u32)
1182 module_exit(exit_u32)
1183 MODULE_LICENSE("GPL");
Linux kernel - Deliverables
This page took 0.053864 seconds and 5 git commands to generate.