?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
[cris-mirror.git] / net / sched / cls_u32.c
blobed8b6a24b9e9325cc99f17e6ed00ead73fe0171e
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 <linux/netdevice.h>
44 #include <linux/hash.h>
45 #include <net/netlink.h>
46 #include <net/act_api.h>
47 #include <net/pkt_cls.h>
48 #include <linux/idr.h>
49
50 struct tc_u_knode {
51 struct tc_u_knode __rcu *next;
52 u32 handle;
53 struct tc_u_hnode __rcu *ht_up;
54 struct tcf_exts exts;
55 #ifdef CONFIG_NET_CLS_IND
56 int ifindex;
57 #endif
58 u8 fshift;
59 struct tcf_result res;
60 struct tc_u_hnode __rcu *ht_down;
61 #ifdef CONFIG_CLS_U32_PERF
62 struct tc_u32_pcnt __percpu *pf;
63 #endif
64 u32 flags;
65 #ifdef CONFIG_CLS_U32_MARK
66 u32 val;
67 u32 mask;
68 u32 __percpu *pcpu_success;
69 #endif
70 struct tcf_proto *tp;
71 union {
72 struct work_struct work;
73 struct rcu_head rcu;
74 };
75 /* The 'sel' field MUST be the last field in structure to allow for
76 * tc_u32_keys allocated at end of structure.
77 */
78 struct tc_u32_sel sel;
79 };
80
81 struct tc_u_hnode {
82 struct tc_u_hnode __rcu *next;
83 u32 handle;
84 u32 prio;
85 struct tc_u_common *tp_c;
86 int refcnt;
87 unsigned int divisor;
88 struct idr handle_idr;
89 struct rcu_head rcu;
90 u32 flags;
91 /* The 'ht' field MUST be the last field in structure to allow for
92 * more entries allocated at end of structure.
93 */
94 struct tc_u_knode __rcu *ht[1];
95 };
96
97 struct tc_u_common {
98 struct tc_u_hnode __rcu *hlist;
99 void *ptr;
100 int refcnt;
101 struct idr handle_idr;
102 struct hlist_node hnode;
103 struct rcu_head rcu;
104 };
105
106 static inline unsigned int u32_hash_fold(__be32 key,
107 const struct tc_u32_sel *sel,
108 u8 fshift)
109 {
110 unsigned int h = ntohl(key & sel->hmask) >> fshift;
111
112 return h;
113 }
114
115 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp,
116 struct tcf_result *res)
117 {
118 struct {
119 struct tc_u_knode *knode;
120 unsigned int off;
121 } stack[TC_U32_MAXDEPTH];
122
123 struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
124 unsigned int off = skb_network_offset(skb);
125 struct tc_u_knode *n;
126 int sdepth = 0;
127 int off2 = 0;
128 int sel = 0;
129 #ifdef CONFIG_CLS_U32_PERF
130 int j;
131 #endif
132 int i, r;
133
134 next_ht:
135 n = rcu_dereference_bh(ht->ht[sel]);
136
137 next_knode:
138 if (n) {
139 struct tc_u32_key *key = n->sel.keys;
140
141 #ifdef CONFIG_CLS_U32_PERF
142 __this_cpu_inc(n->pf->rcnt);
143 j = 0;
144 #endif
145
146 if (tc_skip_sw(n->flags)) {
147 n = rcu_dereference_bh(n->next);
148 goto next_knode;
149 }
150
151 #ifdef CONFIG_CLS_U32_MARK
152 if ((skb->mark & n->mask) != n->val) {
153 n = rcu_dereference_bh(n->next);
154 goto next_knode;
155 } else {
156 __this_cpu_inc(*n->pcpu_success);
157 }
158 #endif
159
160 for (i = n->sel.nkeys; i > 0; i--, key++) {
161 int toff = off + key->off + (off2 & key->offmask);
162 __be32 *data, hdata;
163
164 if (skb_headroom(skb) + toff > INT_MAX)
165 goto out;
166
167 data = skb_header_pointer(skb, toff, 4, &hdata);
168 if (!data)
169 goto out;
170 if ((*data ^ key->val) & key->mask) {
171 n = rcu_dereference_bh(n->next);
172 goto next_knode;
173 }
174 #ifdef CONFIG_CLS_U32_PERF
175 __this_cpu_inc(n->pf->kcnts[j]);
176 j++;
177 #endif
178 }
179
180 ht = rcu_dereference_bh(n->ht_down);
181 if (!ht) {
182 check_terminal:
183 if (n->sel.flags & TC_U32_TERMINAL) {
184
185 *res = n->res;
186 #ifdef CONFIG_NET_CLS_IND
187 if (!tcf_match_indev(skb, n->ifindex)) {
188 n = rcu_dereference_bh(n->next);
189 goto next_knode;
190 }
191 #endif
192 #ifdef CONFIG_CLS_U32_PERF
193 __this_cpu_inc(n->pf->rhit);
194 #endif
195 r = tcf_exts_exec(skb, &n->exts, res);
196 if (r < 0) {
197 n = rcu_dereference_bh(n->next);
198 goto next_knode;
199 }
200
201 return r;
202 }
203 n = rcu_dereference_bh(n->next);
204 goto next_knode;
205 }
206
207 /* PUSH */
208 if (sdepth >= TC_U32_MAXDEPTH)
209 goto deadloop;
210 stack[sdepth].knode = n;
211 stack[sdepth].off = off;
212 sdepth++;
213
214 ht = rcu_dereference_bh(n->ht_down);
215 sel = 0;
216 if (ht->divisor) {
217 __be32 *data, hdata;
218
219 data = skb_header_pointer(skb, off + n->sel.hoff, 4,
220 &hdata);
221 if (!data)
222 goto out;
223 sel = ht->divisor & u32_hash_fold(*data, &n->sel,
224 n->fshift);
225 }
226 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
227 goto next_ht;
228
229 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
230 off2 = n->sel.off + 3;
231 if (n->sel.flags & TC_U32_VAROFFSET) {
232 __be16 *data, hdata;
233
234 data = skb_header_pointer(skb,
235 off + n->sel.offoff,
236 2, &hdata);
237 if (!data)
238 goto out;
239 off2 += ntohs(n->sel.offmask & *data) >>
240 n->sel.offshift;
241 }
242 off2 &= ~3;
243 }
244 if (n->sel.flags & TC_U32_EAT) {
245 off += off2;
246 off2 = 0;
247 }
248
249 if (off < skb->len)
250 goto next_ht;
251 }
252
253 /* POP */
254 if (sdepth--) {
255 n = stack[sdepth].knode;
256 ht = rcu_dereference_bh(n->ht_up);
257 off = stack[sdepth].off;
258 goto check_terminal;
259 }
260 out:
261 return -1;
262
263 deadloop:
264 net_warn_ratelimited("cls_u32: dead loop\n");
265 return -1;
266 }
267
268 static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
269 {
270 struct tc_u_hnode *ht;
271
272 for (ht = rtnl_dereference(tp_c->hlist);
273 ht;
274 ht = rtnl_dereference(ht->next))
275 if (ht->handle == handle)
276 break;
277
278 return ht;
279 }
280
281 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
282 {
283 unsigned int sel;
284 struct tc_u_knode *n = NULL;
285
286 sel = TC_U32_HASH(handle);
287 if (sel > ht->divisor)
288 goto out;
289
290 for (n = rtnl_dereference(ht->ht[sel]);
291 n;
292 n = rtnl_dereference(n->next))
293 if (n->handle == handle)
294 break;
295 out:
296 return n;
297 }
298
299
300 static void *u32_get(struct tcf_proto *tp, u32 handle)
301 {
302 struct tc_u_hnode *ht;
303 struct tc_u_common *tp_c = tp->data;
304
305 if (TC_U32_HTID(handle) == TC_U32_ROOT)
306 ht = rtnl_dereference(tp->root);
307 else
308 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
309
310 if (!ht)
311 return NULL;
312
313 if (TC_U32_KEY(handle) == 0)
314 return ht;
315
316 return u32_lookup_key(ht, handle);
317 }
318
319 /* Protected by rtnl lock */
320 static u32 gen_new_htid(struct tc_u_common *tp_c, struct tc_u_hnode *ptr)
321 {
322 int id = idr_alloc_cyclic(&tp_c->handle_idr, ptr, 1, 0x7FF, GFP_KERNEL);
323 if (id < 0)
324 return 0;
325 return (id | 0x800U) << 20;
326 }
327
328 static struct hlist_head *tc_u_common_hash;
329
330 #define U32_HASH_SHIFT 10
331 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
332
333 static void *tc_u_common_ptr(const struct tcf_proto *tp)
334 {
335 struct tcf_block *block = tp->chain->block;
336
337 /* The block sharing is currently supported only
338 * for classless qdiscs. In that case we use block
339 * for tc_u_common identification. In case the
340 * block is not shared, block->q is a valid pointer
341 * and we can use that. That works for classful qdiscs.
342 */
343 if (tcf_block_shared(block))
344 return block;
345 else
346 return block->q;
347 }
348
349 static unsigned int tc_u_hash(const struct tcf_proto *tp)
350 {
351 return hash_ptr(tc_u_common_ptr(tp), U32_HASH_SHIFT);
352 }
353
354 static struct tc_u_common *tc_u_common_find(const struct tcf_proto *tp)
355 {
356 struct tc_u_common *tc;
357 unsigned int h;
358
359 h = tc_u_hash(tp);
360 hlist_for_each_entry(tc, &tc_u_common_hash[h], hnode) {
361 if (tc->ptr == tc_u_common_ptr(tp))
362 return tc;
363 }
364 return NULL;
365 }
366
367 static int u32_init(struct tcf_proto *tp)
368 {
369 struct tc_u_hnode *root_ht;
370 struct tc_u_common *tp_c;
371 unsigned int h;
372
373 tp_c = tc_u_common_find(tp);
374
375 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
376 if (root_ht == NULL)
377 return -ENOBUFS;
378
379 root_ht->refcnt++;
380 root_ht->handle = tp_c ? gen_new_htid(tp_c, root_ht) : 0x80000000;
381 root_ht->prio = tp->prio;
382 idr_init(&root_ht->handle_idr);
383
384 if (tp_c == NULL) {
385 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
386 if (tp_c == NULL) {
387 kfree(root_ht);
388 return -ENOBUFS;
389 }
390 tp_c->ptr = tc_u_common_ptr(tp);
391 INIT_HLIST_NODE(&tp_c->hnode);
392 idr_init(&tp_c->handle_idr);
393
394 h = tc_u_hash(tp);
395 hlist_add_head(&tp_c->hnode, &tc_u_common_hash[h]);
396 }
397
398 tp_c->refcnt++;
399 RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
400 rcu_assign_pointer(tp_c->hlist, root_ht);
401 root_ht->tp_c = tp_c;
402
403 rcu_assign_pointer(tp->root, root_ht);
404 tp->data = tp_c;
405 return 0;
406 }
407
408 static int u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n,
409 bool free_pf)
410 {
411 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
412
413 tcf_exts_destroy(&n->exts);
414 tcf_exts_put_net(&n->exts);
415 if (ht && --ht->refcnt == 0)
416 kfree(ht);
417 #ifdef CONFIG_CLS_U32_PERF
418 if (free_pf)
419 free_percpu(n->pf);
420 #endif
421 #ifdef CONFIG_CLS_U32_MARK
422 if (free_pf)
423 free_percpu(n->pcpu_success);
424 #endif
425 kfree(n);
426 return 0;
427 }
428
429 /* u32_delete_key_rcu should be called when free'ing a copied
430 * version of a tc_u_knode obtained from u32_init_knode(). When
431 * copies are obtained from u32_init_knode() the statistics are
432 * shared between the old and new copies to allow readers to
433 * continue to update the statistics during the copy. To support
434 * this the u32_delete_key_rcu variant does not free the percpu
435 * statistics.
436 */
437 static void u32_delete_key_work(struct work_struct *work)
438 {
439 struct tc_u_knode *key = container_of(work, struct tc_u_knode, work);
440
441 rtnl_lock();
442 u32_destroy_key(key->tp, key, false);
443 rtnl_unlock();
444 }
445
446 static void u32_delete_key_rcu(struct rcu_head *rcu)
447 {
448 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
449
450 INIT_WORK(&key->work, u32_delete_key_work);
451 tcf_queue_work(&key->work);
452 }
453
454 /* u32_delete_key_freepf_rcu is the rcu callback variant
455 * that free's the entire structure including the statistics
456 * percpu variables. Only use this if the key is not a copy
457 * returned by u32_init_knode(). See u32_delete_key_rcu()
458 * for the variant that should be used with keys return from
459 * u32_init_knode()
460 */
461 static void u32_delete_key_freepf_work(struct work_struct *work)
462 {
463 struct tc_u_knode *key = container_of(work, struct tc_u_knode, work);
464
465 rtnl_lock();
466 u32_destroy_key(key->tp, key, true);
467 rtnl_unlock();
468 }
469
470 static void u32_delete_key_freepf_rcu(struct rcu_head *rcu)
471 {
472 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
473
474 INIT_WORK(&key->work, u32_delete_key_freepf_work);
475 tcf_queue_work(&key->work);
476 }
477
478 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
479 {
480 struct tc_u_knode __rcu **kp;
481 struct tc_u_knode *pkp;
482 struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
483
484 if (ht) {
485 kp = &ht->ht[TC_U32_HASH(key->handle)];
486 for (pkp = rtnl_dereference(*kp); pkp;
487 kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
488 if (pkp == key) {
489 RCU_INIT_POINTER(*kp, key->next);
490
491 tcf_unbind_filter(tp, &key->res);
492 tcf_exts_get_net(&key->exts);
493 call_rcu(&key->rcu, u32_delete_key_freepf_rcu);
494 return 0;
495 }
496 }
497 }
498 WARN_ON(1);
499 return 0;
500 }
501
502 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
503 struct netlink_ext_ack *extack)
504 {
505 struct tcf_block *block = tp->chain->block;
506 struct tc_cls_u32_offload cls_u32 = {};
507
508 tc_cls_common_offload_init(&cls_u32.common, tp, h->flags, extack);
509 cls_u32.command = TC_CLSU32_DELETE_HNODE;
510 cls_u32.hnode.divisor = h->divisor;
511 cls_u32.hnode.handle = h->handle;
512 cls_u32.hnode.prio = h->prio;
513
514 tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false);
515 }
516
517 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
518 u32 flags, struct netlink_ext_ack *extack)
519 {
520 struct tcf_block *block = tp->chain->block;
521 struct tc_cls_u32_offload cls_u32 = {};
522 bool skip_sw = tc_skip_sw(flags);
523 bool offloaded = false;
524 int err;
525
526 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
527 cls_u32.command = TC_CLSU32_NEW_HNODE;
528 cls_u32.hnode.divisor = h->divisor;
529 cls_u32.hnode.handle = h->handle;
530 cls_u32.hnode.prio = h->prio;
531
532 err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw);
533 if (err < 0) {
534 u32_clear_hw_hnode(tp, h, NULL);
535 return err;
536 } else if (err > 0) {
537 offloaded = true;
538 }
539
540 if (skip_sw && !offloaded)
541 return -EINVAL;
542
543 return 0;
544 }
545
546 static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
547 struct netlink_ext_ack *extack)
548 {
549 struct tcf_block *block = tp->chain->block;
550 struct tc_cls_u32_offload cls_u32 = {};
551
552 tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
553 cls_u32.command = TC_CLSU32_DELETE_KNODE;
554 cls_u32.knode.handle = n->handle;
555
556 tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false);
557 tcf_block_offload_dec(block, &n->flags);
558 }
559
560 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
561 u32 flags, struct netlink_ext_ack *extack)
562 {
563 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
564 struct tcf_block *block = tp->chain->block;
565 struct tc_cls_u32_offload cls_u32 = {};
566 bool skip_sw = tc_skip_sw(flags);
567 int err;
568
569 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
570 cls_u32.command = TC_CLSU32_REPLACE_KNODE;
571 cls_u32.knode.handle = n->handle;
572 cls_u32.knode.fshift = n->fshift;
573 #ifdef CONFIG_CLS_U32_MARK
574 cls_u32.knode.val = n->val;
575 cls_u32.knode.mask = n->mask;
576 #else
577 cls_u32.knode.val = 0;
578 cls_u32.knode.mask = 0;
579 #endif
580 cls_u32.knode.sel = &n->sel;
581 cls_u32.knode.exts = &n->exts;
582 if (n->ht_down)
583 cls_u32.knode.link_handle = ht->handle;
584
585 err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw);
586 if (err < 0) {
587 u32_remove_hw_knode(tp, n, NULL);
588 return err;
589 } else if (err > 0) {
590 tcf_block_offload_inc(block, &n->flags);
591 }
592
593 if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
594 return -EINVAL;
595
596 return 0;
597 }
598
599 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
600 struct netlink_ext_ack *extack)
601 {
602 struct tc_u_knode *n;
603 unsigned int h;
604
605 for (h = 0; h <= ht->divisor; h++) {
606 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
607 RCU_INIT_POINTER(ht->ht[h],
608 rtnl_dereference(n->next));
609 tcf_unbind_filter(tp, &n->res);
610 u32_remove_hw_knode(tp, n, extack);
611 idr_remove(&ht->handle_idr, n->handle);
612 if (tcf_exts_get_net(&n->exts))
613 call_rcu(&n->rcu, u32_delete_key_freepf_rcu);
614 else
615 u32_destroy_key(n->tp, n, true);
616 }
617 }
618 }
619
620 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
621 struct netlink_ext_ack *extack)
622 {
623 struct tc_u_common *tp_c = tp->data;
624 struct tc_u_hnode __rcu **hn;
625 struct tc_u_hnode *phn;
626
627 WARN_ON(ht->refcnt);
628
629 u32_clear_hnode(tp, ht, extack);
630
631 hn = &tp_c->hlist;
632 for (phn = rtnl_dereference(*hn);
633 phn;
634 hn = &phn->next, phn = rtnl_dereference(*hn)) {
635 if (phn == ht) {
636 u32_clear_hw_hnode(tp, ht, extack);
637 idr_destroy(&ht->handle_idr);
638 idr_remove(&tp_c->handle_idr, ht->handle);
639 RCU_INIT_POINTER(*hn, ht->next);
640 kfree_rcu(ht, rcu);
641 return 0;
642 }
643 }
644
645 return -ENOENT;
646 }
647
648 static bool ht_empty(struct tc_u_hnode *ht)
649 {
650 unsigned int h;
651
652 for (h = 0; h <= ht->divisor; h++)
653 if (rcu_access_pointer(ht->ht[h]))
654 return false;
655
656 return true;
657 }
658
659 static void u32_destroy(struct tcf_proto *tp, struct netlink_ext_ack *extack)
660 {
661 struct tc_u_common *tp_c = tp->data;
662 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
663
664 WARN_ON(root_ht == NULL);
665
666 if (root_ht && --root_ht->refcnt == 0)
667 u32_destroy_hnode(tp, root_ht, extack);
668
669 if (--tp_c->refcnt == 0) {
670 struct tc_u_hnode *ht;
671
672 hlist_del(&tp_c->hnode);
673
674 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
675 u32_clear_hnode(tp, ht, extack);
676 RCU_INIT_POINTER(tp_c->hlist, ht->next);
677
678 /* u32_destroy_key() will later free ht for us, if it's
679 * still referenced by some knode
680 */
681 if (--ht->refcnt == 0)
682 kfree_rcu(ht, rcu);
683 }
684
685 idr_destroy(&tp_c->handle_idr);
686 kfree(tp_c);
687 }
688
689 tp->data = NULL;
690 }
691
692 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last,
693 struct netlink_ext_ack *extack)
694 {
695 struct tc_u_hnode *ht = arg;
696 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
697 struct tc_u_common *tp_c = tp->data;
698 int ret = 0;
699
700 if (ht == NULL)
701 goto out;
702
703 if (TC_U32_KEY(ht->handle)) {
704 u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack);
705 ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
706 goto out;
707 }
708
709 if (root_ht == ht) {
710 NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node");
711 return -EINVAL;
712 }
713
714 if (ht->refcnt == 1) {
715 ht->refcnt--;
716 u32_destroy_hnode(tp, ht, extack);
717 } else {
718 NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter");
719 return -EBUSY;
720 }
721
722 out:
723 *last = true;
724 if (root_ht) {
725 if (root_ht->refcnt > 1) {
726 *last = false;
727 goto ret;
728 }
729 if (root_ht->refcnt == 1) {
730 if (!ht_empty(root_ht)) {
731 *last = false;
732 goto ret;
733 }
734 }
735 }
736
737 if (tp_c->refcnt > 1) {
738 *last = false;
739 goto ret;
740 }
741
742 if (tp_c->refcnt == 1) {
743 struct tc_u_hnode *ht;
744
745 for (ht = rtnl_dereference(tp_c->hlist);
746 ht;
747 ht = rtnl_dereference(ht->next))
748 if (!ht_empty(ht)) {
749 *last = false;
750 break;
751 }
752 }
753
754 ret:
755 return ret;
756 }
757
758 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
759 {
760 u32 index = htid | 0x800;
761 u32 max = htid | 0xFFF;
762
763 if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max, GFP_KERNEL)) {
764 index = htid + 1;
765 if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max,
766 GFP_KERNEL))
767 index = max;
768 }
769
770 return index;
771 }
772
773 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
774 [TCA_U32_CLASSID] = { .type = NLA_U32 },
775 [TCA_U32_HASH] = { .type = NLA_U32 },
776 [TCA_U32_LINK] = { .type = NLA_U32 },
777 [TCA_U32_DIVISOR] = { .type = NLA_U32 },
778 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) },
779 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ },
780 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) },
781 [TCA_U32_FLAGS] = { .type = NLA_U32 },
782 };
783
784 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
785 unsigned long base, struct tc_u_hnode *ht,
786 struct tc_u_knode *n, struct nlattr **tb,
787 struct nlattr *est, bool ovr,
788 struct netlink_ext_ack *extack)
789 {
790 int err;
791
792 err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr, extack);
793 if (err < 0)
794 return err;
795
796 if (tb[TCA_U32_LINK]) {
797 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
798 struct tc_u_hnode *ht_down = NULL, *ht_old;
799
800 if (TC_U32_KEY(handle)) {
801 NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table");
802 return -EINVAL;
803 }
804
805 if (handle) {
806 ht_down = u32_lookup_ht(ht->tp_c, handle);
807
808 if (!ht_down) {
809 NL_SET_ERR_MSG_MOD(extack, "Link hash table not found");
810 return -EINVAL;
811 }
812 ht_down->refcnt++;
813 }
814
815 ht_old = rtnl_dereference(n->ht_down);
816 rcu_assign_pointer(n->ht_down, ht_down);
817
818 if (ht_old)
819 ht_old->refcnt--;
820 }
821 if (tb[TCA_U32_CLASSID]) {
822 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
823 tcf_bind_filter(tp, &n->res, base);
824 }
825
826 #ifdef CONFIG_NET_CLS_IND
827 if (tb[TCA_U32_INDEV]) {
828 int ret;
829 ret = tcf_change_indev(net, tb[TCA_U32_INDEV], extack);
830 if (ret < 0)
831 return -EINVAL;
832 n->ifindex = ret;
833 }
834 #endif
835 return 0;
836 }
837
838 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
839 struct tc_u_knode *n)
840 {
841 struct tc_u_knode __rcu **ins;
842 struct tc_u_knode *pins;
843 struct tc_u_hnode *ht;
844
845 if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
846 ht = rtnl_dereference(tp->root);
847 else
848 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
849
850 ins = &ht->ht[TC_U32_HASH(n->handle)];
851
852 /* The node must always exist for it to be replaced if this is not the
853 * case then something went very wrong elsewhere.
854 */
855 for (pins = rtnl_dereference(*ins); ;
856 ins = &pins->next, pins = rtnl_dereference(*ins))
857 if (pins->handle == n->handle)
858 break;
859
860 idr_replace(&ht->handle_idr, n, n->handle);
861 RCU_INIT_POINTER(n->next, pins->next);
862 rcu_assign_pointer(*ins, n);
863 }
864
865 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp,
866 struct tc_u_knode *n)
867 {
868 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
869 struct tc_u32_sel *s = &n->sel;
870 struct tc_u_knode *new;
871
872 new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key),
873 GFP_KERNEL);
874
875 if (!new)
876 return NULL;
877
878 RCU_INIT_POINTER(new->next, n->next);
879 new->handle = n->handle;
880 RCU_INIT_POINTER(new->ht_up, n->ht_up);
881
882 #ifdef CONFIG_NET_CLS_IND
883 new->ifindex = n->ifindex;
884 #endif
885 new->fshift = n->fshift;
886 new->res = n->res;
887 new->flags = n->flags;
888 RCU_INIT_POINTER(new->ht_down, ht);
889
890 /* bump reference count as long as we hold pointer to structure */
891 if (ht)
892 ht->refcnt++;
893
894 #ifdef CONFIG_CLS_U32_PERF
895 /* Statistics may be incremented by readers during update
896 * so we must keep them in tact. When the node is later destroyed
897 * a special destroy call must be made to not free the pf memory.
898 */
899 new->pf = n->pf;
900 #endif
901
902 #ifdef CONFIG_CLS_U32_MARK
903 new->val = n->val;
904 new->mask = n->mask;
905 /* Similarly success statistics must be moved as pointers */
906 new->pcpu_success = n->pcpu_success;
907 #endif
908 new->tp = tp;
909 memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
910
911 if (tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE)) {
912 kfree(new);
913 return NULL;
914 }
915
916 return new;
917 }
918
919 static int u32_change(struct net *net, struct sk_buff *in_skb,
920 struct tcf_proto *tp, unsigned long base, u32 handle,
921 struct nlattr **tca, void **arg, bool ovr,
922 struct netlink_ext_ack *extack)
923 {
924 struct tc_u_common *tp_c = tp->data;
925 struct tc_u_hnode *ht;
926 struct tc_u_knode *n;
927 struct tc_u32_sel *s;
928 struct nlattr *opt = tca[TCA_OPTIONS];
929 struct nlattr *tb[TCA_U32_MAX + 1];
930 u32 htid, flags = 0;
931 int err;
932 #ifdef CONFIG_CLS_U32_PERF
933 size_t size;
934 #endif
935
936 if (!opt) {
937 if (handle) {
938 NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options");
939 return -EINVAL;
940 } else {
941 return 0;
942 }
943 }
944
945 err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy, extack);
946 if (err < 0)
947 return err;
948
949 if (tb[TCA_U32_FLAGS]) {
950 flags = nla_get_u32(tb[TCA_U32_FLAGS]);
951 if (!tc_flags_valid(flags)) {
952 NL_SET_ERR_MSG_MOD(extack, "Invalid filter flags");
953 return -EINVAL;
954 }
955 }
956
957 n = *arg;
958 if (n) {
959 struct tc_u_knode *new;
960
961 if (TC_U32_KEY(n->handle) == 0) {
962 NL_SET_ERR_MSG_MOD(extack, "Key node id cannot be zero");
963 return -EINVAL;
964 }
965
966 if ((n->flags ^ flags) &
967 ~(TCA_CLS_FLAGS_IN_HW | TCA_CLS_FLAGS_NOT_IN_HW)) {
968 NL_SET_ERR_MSG_MOD(extack, "Key node flags do not match passed flags");
969 return -EINVAL;
970 }
971
972 new = u32_init_knode(tp, n);
973 if (!new)
974 return -ENOMEM;
975
976 err = u32_set_parms(net, tp, base,
977 rtnl_dereference(n->ht_up), new, tb,
978 tca[TCA_RATE], ovr, extack);
979
980 if (err) {
981 u32_destroy_key(tp, new, false);
982 return err;
983 }
984
985 err = u32_replace_hw_knode(tp, new, flags, extack);
986 if (err) {
987 u32_destroy_key(tp, new, false);
988 return err;
989 }
990
991 if (!tc_in_hw(new->flags))
992 new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
993
994 u32_replace_knode(tp, tp_c, new);
995 tcf_unbind_filter(tp, &n->res);
996 tcf_exts_get_net(&n->exts);
997 call_rcu(&n->rcu, u32_delete_key_rcu);
998 return 0;
999 }
1000
1001 if (tb[TCA_U32_DIVISOR]) {
1002 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
1003
1004 if (--divisor > 0x100) {
1005 NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets");
1006 return -EINVAL;
1007 }
1008 if (TC_U32_KEY(handle)) {
1009 NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table");
1010 return -EINVAL;
1011 }
1012 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
1013 if (ht == NULL)
1014 return -ENOBUFS;
1015 if (handle == 0) {
1016 handle = gen_new_htid(tp->data, ht);
1017 if (handle == 0) {
1018 kfree(ht);
1019 return -ENOMEM;
1020 }
1021 } else {
1022 err = idr_alloc_u32(&tp_c->handle_idr, ht, &handle,
1023 handle, GFP_KERNEL);
1024 if (err) {
1025 kfree(ht);
1026 return err;
1027 }
1028 }
1029 ht->tp_c = tp_c;
1030 ht->refcnt = 1;
1031 ht->divisor = divisor;
1032 ht->handle = handle;
1033 ht->prio = tp->prio;
1034 idr_init(&ht->handle_idr);
1035 ht->flags = flags;
1036
1037 err = u32_replace_hw_hnode(tp, ht, flags, extack);
1038 if (err) {
1039 idr_remove(&tp_c->handle_idr, handle);
1040 kfree(ht);
1041 return err;
1042 }
1043
1044 RCU_INIT_POINTER(ht->next, tp_c->hlist);
1045 rcu_assign_pointer(tp_c->hlist, ht);
1046 *arg = ht;
1047
1048 return 0;
1049 }
1050
1051 if (tb[TCA_U32_HASH]) {
1052 htid = nla_get_u32(tb[TCA_U32_HASH]);
1053 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
1054 ht = rtnl_dereference(tp->root);
1055 htid = ht->handle;
1056 } else {
1057 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
1058 if (!ht) {
1059 NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found");
1060 return -EINVAL;
1061 }
1062 }
1063 } else {
1064 ht = rtnl_dereference(tp->root);
1065 htid = ht->handle;
1066 }
1067
1068 if (ht->divisor < TC_U32_HASH(htid)) {
1069 NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value");
1070 return -EINVAL;
1071 }
1072
1073 if (handle) {
1074 if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) {
1075 NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch");
1076 return -EINVAL;
1077 }
1078 handle = htid | TC_U32_NODE(handle);
1079 err = idr_alloc_u32(&ht->handle_idr, NULL, &handle, handle,
1080 GFP_KERNEL);
1081 if (err)
1082 return err;
1083 } else
1084 handle = gen_new_kid(ht, htid);
1085
1086 if (tb[TCA_U32_SEL] == NULL) {
1087 NL_SET_ERR_MSG_MOD(extack, "Selector not specified");
1088 err = -EINVAL;
1089 goto erridr;
1090 }
1091
1092 s = nla_data(tb[TCA_U32_SEL]);
1093
1094 n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL);
1095 if (n == NULL) {
1096 err = -ENOBUFS;
1097 goto erridr;
1098 }
1099
1100 #ifdef CONFIG_CLS_U32_PERF
1101 size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64);
1102 n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt));
1103 if (!n->pf) {
1104 err = -ENOBUFS;
1105 goto errfree;
1106 }
1107 #endif
1108
1109 memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
1110 RCU_INIT_POINTER(n->ht_up, ht);
1111 n->handle = handle;
1112 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1113 n->flags = flags;
1114 n->tp = tp;
1115
1116 err = tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE);
1117 if (err < 0)
1118 goto errout;
1119
1120 #ifdef CONFIG_CLS_U32_MARK
1121 n->pcpu_success = alloc_percpu(u32);
1122 if (!n->pcpu_success) {
1123 err = -ENOMEM;
1124 goto errout;
1125 }
1126
1127 if (tb[TCA_U32_MARK]) {
1128 struct tc_u32_mark *mark;
1129
1130 mark = nla_data(tb[TCA_U32_MARK]);
1131 n->val = mark->val;
1132 n->mask = mark->mask;
1133 }
1134 #endif
1135
1136 err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr,
1137 extack);
1138 if (err == 0) {
1139 struct tc_u_knode __rcu **ins;
1140 struct tc_u_knode *pins;
1141
1142 err = u32_replace_hw_knode(tp, n, flags, extack);
1143 if (err)
1144 goto errhw;
1145
1146 if (!tc_in_hw(n->flags))
1147 n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1148
1149 ins = &ht->ht[TC_U32_HASH(handle)];
1150 for (pins = rtnl_dereference(*ins); pins;
1151 ins = &pins->next, pins = rtnl_dereference(*ins))
1152 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1153 break;
1154
1155 RCU_INIT_POINTER(n->next, pins);
1156 rcu_assign_pointer(*ins, n);
1157 *arg = n;
1158 return 0;
1159 }
1160
1161 errhw:
1162 #ifdef CONFIG_CLS_U32_MARK
1163 free_percpu(n->pcpu_success);
1164 #endif
1165
1166 errout:
1167 tcf_exts_destroy(&n->exts);
1168 #ifdef CONFIG_CLS_U32_PERF
1169 errfree:
1170 free_percpu(n->pf);
1171 #endif
1172 kfree(n);
1173 erridr:
1174 idr_remove(&ht->handle_idr, handle);
1175 return err;
1176 }
1177
1178 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg)
1179 {
1180 struct tc_u_common *tp_c = tp->data;
1181 struct tc_u_hnode *ht;
1182 struct tc_u_knode *n;
1183 unsigned int h;
1184
1185 if (arg->stop)
1186 return;
1187
1188 for (ht = rtnl_dereference(tp_c->hlist);
1189 ht;
1190 ht = rtnl_dereference(ht->next)) {
1191 if (ht->prio != tp->prio)
1192 continue;
1193 if (arg->count >= arg->skip) {
1194 if (arg->fn(tp, ht, arg) < 0) {
1195 arg->stop = 1;
1196 return;
1197 }
1198 }
1199 arg->count++;
1200 for (h = 0; h <= ht->divisor; h++) {
1201 for (n = rtnl_dereference(ht->ht[h]);
1202 n;
1203 n = rtnl_dereference(n->next)) {
1204 if (arg->count < arg->skip) {
1205 arg->count++;
1206 continue;
1207 }
1208 if (arg->fn(tp, n, arg) < 0) {
1209 arg->stop = 1;
1210 return;
1211 }
1212 arg->count++;
1213 }
1214 }
1215 }
1216 }
1217
1218 static void u32_bind_class(void *fh, u32 classid, unsigned long cl)
1219 {
1220 struct tc_u_knode *n = fh;
1221
1222 if (n && n->res.classid == classid)
1223 n->res.class = cl;
1224 }
1225
1226 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1227 struct sk_buff *skb, struct tcmsg *t)
1228 {
1229 struct tc_u_knode *n = fh;
1230 struct tc_u_hnode *ht_up, *ht_down;
1231 struct nlattr *nest;
1232
1233 if (n == NULL)
1234 return skb->len;
1235
1236 t->tcm_handle = n->handle;
1237
1238 nest = nla_nest_start(skb, TCA_OPTIONS);
1239 if (nest == NULL)
1240 goto nla_put_failure;
1241
1242 if (TC_U32_KEY(n->handle) == 0) {
1243 struct tc_u_hnode *ht = fh;
1244 u32 divisor = ht->divisor + 1;
1245
1246 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1247 goto nla_put_failure;
1248 } else {
1249 #ifdef CONFIG_CLS_U32_PERF
1250 struct tc_u32_pcnt *gpf;
1251 int cpu;
1252 #endif
1253
1254 if (nla_put(skb, TCA_U32_SEL,
1255 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
1256 &n->sel))
1257 goto nla_put_failure;
1258
1259 ht_up = rtnl_dereference(n->ht_up);
1260 if (ht_up) {
1261 u32 htid = n->handle & 0xFFFFF000;
1262 if (nla_put_u32(skb, TCA_U32_HASH, htid))
1263 goto nla_put_failure;
1264 }
1265 if (n->res.classid &&
1266 nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1267 goto nla_put_failure;
1268
1269 ht_down = rtnl_dereference(n->ht_down);
1270 if (ht_down &&
1271 nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1272 goto nla_put_failure;
1273
1274 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1275 goto nla_put_failure;
1276
1277 #ifdef CONFIG_CLS_U32_MARK
1278 if ((n->val || n->mask)) {
1279 struct tc_u32_mark mark = {.val = n->val,
1280 .mask = n->mask,
1281 .success = 0};
1282 int cpum;
1283
1284 for_each_possible_cpu(cpum) {
1285 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1286
1287 mark.success += cnt;
1288 }
1289
1290 if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1291 goto nla_put_failure;
1292 }
1293 #endif
1294
1295 if (tcf_exts_dump(skb, &n->exts) < 0)
1296 goto nla_put_failure;
1297
1298 #ifdef CONFIG_NET_CLS_IND
1299 if (n->ifindex) {
1300 struct net_device *dev;
1301 dev = __dev_get_by_index(net, n->ifindex);
1302 if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1303 goto nla_put_failure;
1304 }
1305 #endif
1306 #ifdef CONFIG_CLS_U32_PERF
1307 gpf = kzalloc(sizeof(struct tc_u32_pcnt) +
1308 n->sel.nkeys * sizeof(u64),
1309 GFP_KERNEL);
1310 if (!gpf)
1311 goto nla_put_failure;
1312
1313 for_each_possible_cpu(cpu) {
1314 int i;
1315 struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1316
1317 gpf->rcnt += pf->rcnt;
1318 gpf->rhit += pf->rhit;
1319 for (i = 0; i < n->sel.nkeys; i++)
1320 gpf->kcnts[i] += pf->kcnts[i];
1321 }
1322
1323 if (nla_put_64bit(skb, TCA_U32_PCNT,
1324 sizeof(struct tc_u32_pcnt) +
1325 n->sel.nkeys * sizeof(u64),
1326 gpf, TCA_U32_PAD)) {
1327 kfree(gpf);
1328 goto nla_put_failure;
1329 }
1330 kfree(gpf);
1331 #endif
1332 }
1333
1334 nla_nest_end(skb, nest);
1335
1336 if (TC_U32_KEY(n->handle))
1337 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1338 goto nla_put_failure;
1339 return skb->len;
1340
1341 nla_put_failure:
1342 nla_nest_cancel(skb, nest);
1343 return -1;
1344 }
1345
1346 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1347 .kind = "u32",
1348 .classify = u32_classify,
1349 .init = u32_init,
1350 .destroy = u32_destroy,
1351 .get = u32_get,
1352 .change = u32_change,
1353 .delete = u32_delete,
1354 .walk = u32_walk,
1355 .dump = u32_dump,
1356 .bind_class = u32_bind_class,
1357 .owner = THIS_MODULE,
1358 };
1359
1360 static int __init init_u32(void)
1361 {
1362 int i, ret;
1363
1364 pr_info("u32 classifier\n");
1365 #ifdef CONFIG_CLS_U32_PERF
1366 pr_info(" Performance counters on\n");
1367 #endif
1368 #ifdef CONFIG_NET_CLS_IND
1369 pr_info(" input device check on\n");
1370 #endif
1371 #ifdef CONFIG_NET_CLS_ACT
1372 pr_info(" Actions configured\n");
1373 #endif
1374 tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1375 sizeof(struct hlist_head),
1376 GFP_KERNEL);
1377 if (!tc_u_common_hash)
1378 return -ENOMEM;
1379
1380 for (i = 0; i < U32_HASH_SIZE; i++)
1381 INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1382
1383 ret = register_tcf_proto_ops(&cls_u32_ops);
1384 if (ret)
1385 kvfree(tc_u_common_hash);
1386 return ret;
1387 }
1388
1389 static void __exit exit_u32(void)
1390 {
1391 unregister_tcf_proto_ops(&cls_u32_ops);
1392 kvfree(tc_u_common_hash);
1393 }
1394
1395 module_init(init_u32)
1396 module_exit(exit_u32)
1397 MODULE_LICENSE("GPL");
CRIS linux kernel tree