Merge branch 'regmap-4.21' into regmap-5.0
[sfrench/cifs-2.6.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 <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         unsigned int            in_hw_count;
66 #ifdef CONFIG_CLS_U32_MARK
67         u32                     val;
68         u32                     mask;
69         u32 __percpu            *pcpu_success;
70 #endif
71         struct rcu_work         rwork;
72         /* The 'sel' field MUST be the last field in structure to allow for
73          * tc_u32_keys allocated at end of structure.
74          */
75         struct tc_u32_sel       sel;
76 };
77
78 struct tc_u_hnode {
79         struct tc_u_hnode __rcu *next;
80         u32                     handle;
81         u32                     prio;
82         int                     refcnt;
83         unsigned int            divisor;
84         struct idr              handle_idr;
85         bool                    is_root;
86         struct rcu_head         rcu;
87         u32                     flags;
88         /* The 'ht' field MUST be the last field in structure to allow for
89          * more entries allocated at end of structure.
90          */
91         struct tc_u_knode __rcu *ht[1];
92 };
93
94 struct tc_u_common {
95         struct tc_u_hnode __rcu *hlist;
96         void                    *ptr;
97         int                     refcnt;
98         struct idr              handle_idr;
99         struct hlist_node       hnode;
100         long                    knodes;
101 };
102
103 static inline unsigned int u32_hash_fold(__be32 key,
104                                          const struct tc_u32_sel *sel,
105                                          u8 fshift)
106 {
107         unsigned int h = ntohl(key & sel->hmask) >> fshift;
108
109         return h;
110 }
111
112 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp,
113                         struct tcf_result *res)
114 {
115         struct {
116                 struct tc_u_knode *knode;
117                 unsigned int      off;
118         } stack[TC_U32_MAXDEPTH];
119
120         struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
121         unsigned int off = skb_network_offset(skb);
122         struct tc_u_knode *n;
123         int sdepth = 0;
124         int off2 = 0;
125         int sel = 0;
126 #ifdef CONFIG_CLS_U32_PERF
127         int j;
128 #endif
129         int i, r;
130
131 next_ht:
132         n = rcu_dereference_bh(ht->ht[sel]);
133
134 next_knode:
135         if (n) {
136                 struct tc_u32_key *key = n->sel.keys;
137
138 #ifdef CONFIG_CLS_U32_PERF
139                 __this_cpu_inc(n->pf->rcnt);
140                 j = 0;
141 #endif
142
143                 if (tc_skip_sw(n->flags)) {
144                         n = rcu_dereference_bh(n->next);
145                         goto next_knode;
146                 }
147
148 #ifdef CONFIG_CLS_U32_MARK
149                 if ((skb->mark & n->mask) != n->val) {
150                         n = rcu_dereference_bh(n->next);
151                         goto next_knode;
152                 } else {
153                         __this_cpu_inc(*n->pcpu_success);
154                 }
155 #endif
156
157                 for (i = n->sel.nkeys; i > 0; i--, key++) {
158                         int toff = off + key->off + (off2 & key->offmask);
159                         __be32 *data, hdata;
160
161                         if (skb_headroom(skb) + toff > INT_MAX)
162                                 goto out;
163
164                         data = skb_header_pointer(skb, toff, 4, &hdata);
165                         if (!data)
166                                 goto out;
167                         if ((*data ^ key->val) & key->mask) {
168                                 n = rcu_dereference_bh(n->next);
169                                 goto next_knode;
170                         }
171 #ifdef CONFIG_CLS_U32_PERF
172                         __this_cpu_inc(n->pf->kcnts[j]);
173                         j++;
174 #endif
175                 }
176
177                 ht = rcu_dereference_bh(n->ht_down);
178                 if (!ht) {
179 check_terminal:
180                         if (n->sel.flags & TC_U32_TERMINAL) {
181
182                                 *res = n->res;
183 #ifdef CONFIG_NET_CLS_IND
184                                 if (!tcf_match_indev(skb, n->ifindex)) {
185                                         n = rcu_dereference_bh(n->next);
186                                         goto next_knode;
187                                 }
188 #endif
189 #ifdef CONFIG_CLS_U32_PERF
190                                 __this_cpu_inc(n->pf->rhit);
191 #endif
192                                 r = tcf_exts_exec(skb, &n->exts, res);
193                                 if (r < 0) {
194                                         n = rcu_dereference_bh(n->next);
195                                         goto next_knode;
196                                 }
197
198                                 return r;
199                         }
200                         n = rcu_dereference_bh(n->next);
201                         goto next_knode;
202                 }
203
204                 /* PUSH */
205                 if (sdepth >= TC_U32_MAXDEPTH)
206                         goto deadloop;
207                 stack[sdepth].knode = n;
208                 stack[sdepth].off = off;
209                 sdepth++;
210
211                 ht = rcu_dereference_bh(n->ht_down);
212                 sel = 0;
213                 if (ht->divisor) {
214                         __be32 *data, hdata;
215
216                         data = skb_header_pointer(skb, off + n->sel.hoff, 4,
217                                                   &hdata);
218                         if (!data)
219                                 goto out;
220                         sel = ht->divisor & u32_hash_fold(*data, &n->sel,
221                                                           n->fshift);
222                 }
223                 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
224                         goto next_ht;
225
226                 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
227                         off2 = n->sel.off + 3;
228                         if (n->sel.flags & TC_U32_VAROFFSET) {
229                                 __be16 *data, hdata;
230
231                                 data = skb_header_pointer(skb,
232                                                           off + n->sel.offoff,
233                                                           2, &hdata);
234                                 if (!data)
235                                         goto out;
236                                 off2 += ntohs(n->sel.offmask & *data) >>
237                                         n->sel.offshift;
238                         }
239                         off2 &= ~3;
240                 }
241                 if (n->sel.flags & TC_U32_EAT) {
242                         off += off2;
243                         off2 = 0;
244                 }
245
246                 if (off < skb->len)
247                         goto next_ht;
248         }
249
250         /* POP */
251         if (sdepth--) {
252                 n = stack[sdepth].knode;
253                 ht = rcu_dereference_bh(n->ht_up);
254                 off = stack[sdepth].off;
255                 goto check_terminal;
256         }
257 out:
258         return -1;
259
260 deadloop:
261         net_warn_ratelimited("cls_u32: dead loop\n");
262         return -1;
263 }
264
265 static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
266 {
267         struct tc_u_hnode *ht;
268
269         for (ht = rtnl_dereference(tp_c->hlist);
270              ht;
271              ht = rtnl_dereference(ht->next))
272                 if (ht->handle == handle)
273                         break;
274
275         return ht;
276 }
277
278 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
279 {
280         unsigned int sel;
281         struct tc_u_knode *n = NULL;
282
283         sel = TC_U32_HASH(handle);
284         if (sel > ht->divisor)
285                 goto out;
286
287         for (n = rtnl_dereference(ht->ht[sel]);
288              n;
289              n = rtnl_dereference(n->next))
290                 if (n->handle == handle)
291                         break;
292 out:
293         return n;
294 }
295
296
297 static void *u32_get(struct tcf_proto *tp, u32 handle)
298 {
299         struct tc_u_hnode *ht;
300         struct tc_u_common *tp_c = tp->data;
301
302         if (TC_U32_HTID(handle) == TC_U32_ROOT)
303                 ht = rtnl_dereference(tp->root);
304         else
305                 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
306
307         if (!ht)
308                 return NULL;
309
310         if (TC_U32_KEY(handle) == 0)
311                 return ht;
312
313         return u32_lookup_key(ht, handle);
314 }
315
316 /* Protected by rtnl lock */
317 static u32 gen_new_htid(struct tc_u_common *tp_c, struct tc_u_hnode *ptr)
318 {
319         int id = idr_alloc_cyclic(&tp_c->handle_idr, ptr, 1, 0x7FF, GFP_KERNEL);
320         if (id < 0)
321                 return 0;
322         return (id | 0x800U) << 20;
323 }
324
325 static struct hlist_head *tc_u_common_hash;
326
327 #define U32_HASH_SHIFT 10
328 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
329
330 static void *tc_u_common_ptr(const struct tcf_proto *tp)
331 {
332         struct tcf_block *block = tp->chain->block;
333
334         /* The block sharing is currently supported only
335          * for classless qdiscs. In that case we use block
336          * for tc_u_common identification. In case the
337          * block is not shared, block->q is a valid pointer
338          * and we can use that. That works for classful qdiscs.
339          */
340         if (tcf_block_shared(block))
341                 return block;
342         else
343                 return block->q;
344 }
345
346 static struct hlist_head *tc_u_hash(void *key)
347 {
348         return tc_u_common_hash + hash_ptr(key, U32_HASH_SHIFT);
349 }
350
351 static struct tc_u_common *tc_u_common_find(void *key)
352 {
353         struct tc_u_common *tc;
354         hlist_for_each_entry(tc, tc_u_hash(key), hnode) {
355                 if (tc->ptr == key)
356                         return tc;
357         }
358         return NULL;
359 }
360
361 static int u32_init(struct tcf_proto *tp)
362 {
363         struct tc_u_hnode *root_ht;
364         void *key = tc_u_common_ptr(tp);
365         struct tc_u_common *tp_c = tc_u_common_find(key);
366
367         root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
368         if (root_ht == NULL)
369                 return -ENOBUFS;
370
371         root_ht->refcnt++;
372         root_ht->handle = tp_c ? gen_new_htid(tp_c, root_ht) : 0x80000000;
373         root_ht->prio = tp->prio;
374         root_ht->is_root = true;
375         idr_init(&root_ht->handle_idr);
376
377         if (tp_c == NULL) {
378                 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
379                 if (tp_c == NULL) {
380                         kfree(root_ht);
381                         return -ENOBUFS;
382                 }
383                 tp_c->ptr = key;
384                 INIT_HLIST_NODE(&tp_c->hnode);
385                 idr_init(&tp_c->handle_idr);
386
387                 hlist_add_head(&tp_c->hnode, tc_u_hash(key));
388         }
389
390         tp_c->refcnt++;
391         RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
392         rcu_assign_pointer(tp_c->hlist, root_ht);
393
394         root_ht->refcnt++;
395         rcu_assign_pointer(tp->root, root_ht);
396         tp->data = tp_c;
397         return 0;
398 }
399
400 static int u32_destroy_key(struct tc_u_knode *n, bool free_pf)
401 {
402         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
403
404         tcf_exts_destroy(&n->exts);
405         tcf_exts_put_net(&n->exts);
406         if (ht && --ht->refcnt == 0)
407                 kfree(ht);
408 #ifdef CONFIG_CLS_U32_PERF
409         if (free_pf)
410                 free_percpu(n->pf);
411 #endif
412 #ifdef CONFIG_CLS_U32_MARK
413         if (free_pf)
414                 free_percpu(n->pcpu_success);
415 #endif
416         kfree(n);
417         return 0;
418 }
419
420 /* u32_delete_key_rcu should be called when free'ing a copied
421  * version of a tc_u_knode obtained from u32_init_knode(). When
422  * copies are obtained from u32_init_knode() the statistics are
423  * shared between the old and new copies to allow readers to
424  * continue to update the statistics during the copy. To support
425  * this the u32_delete_key_rcu variant does not free the percpu
426  * statistics.
427  */
428 static void u32_delete_key_work(struct work_struct *work)
429 {
430         struct tc_u_knode *key = container_of(to_rcu_work(work),
431                                               struct tc_u_knode,
432                                               rwork);
433         rtnl_lock();
434         u32_destroy_key(key, false);
435         rtnl_unlock();
436 }
437
438 /* u32_delete_key_freepf_rcu is the rcu callback variant
439  * that free's the entire structure including the statistics
440  * percpu variables. Only use this if the key is not a copy
441  * returned by u32_init_knode(). See u32_delete_key_rcu()
442  * for the variant that should be used with keys return from
443  * u32_init_knode()
444  */
445 static void u32_delete_key_freepf_work(struct work_struct *work)
446 {
447         struct tc_u_knode *key = container_of(to_rcu_work(work),
448                                               struct tc_u_knode,
449                                               rwork);
450         rtnl_lock();
451         u32_destroy_key(key, true);
452         rtnl_unlock();
453 }
454
455 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
456 {
457         struct tc_u_common *tp_c = tp->data;
458         struct tc_u_knode __rcu **kp;
459         struct tc_u_knode *pkp;
460         struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
461
462         if (ht) {
463                 kp = &ht->ht[TC_U32_HASH(key->handle)];
464                 for (pkp = rtnl_dereference(*kp); pkp;
465                      kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
466                         if (pkp == key) {
467                                 RCU_INIT_POINTER(*kp, key->next);
468                                 tp_c->knodes--;
469
470                                 tcf_unbind_filter(tp, &key->res);
471                                 idr_remove(&ht->handle_idr, key->handle);
472                                 tcf_exts_get_net(&key->exts);
473                                 tcf_queue_work(&key->rwork, u32_delete_key_freepf_work);
474                                 return 0;
475                         }
476                 }
477         }
478         WARN_ON(1);
479         return 0;
480 }
481
482 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
483                                struct netlink_ext_ack *extack)
484 {
485         struct tcf_block *block = tp->chain->block;
486         struct tc_cls_u32_offload cls_u32 = {};
487
488         tc_cls_common_offload_init(&cls_u32.common, tp, h->flags, extack);
489         cls_u32.command = TC_CLSU32_DELETE_HNODE;
490         cls_u32.hnode.divisor = h->divisor;
491         cls_u32.hnode.handle = h->handle;
492         cls_u32.hnode.prio = h->prio;
493
494         tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, false);
495 }
496
497 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
498                                 u32 flags, struct netlink_ext_ack *extack)
499 {
500         struct tcf_block *block = tp->chain->block;
501         struct tc_cls_u32_offload cls_u32 = {};
502         bool skip_sw = tc_skip_sw(flags);
503         bool offloaded = false;
504         int err;
505
506         tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
507         cls_u32.command = TC_CLSU32_NEW_HNODE;
508         cls_u32.hnode.divisor = h->divisor;
509         cls_u32.hnode.handle = h->handle;
510         cls_u32.hnode.prio = h->prio;
511
512         err = tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, skip_sw);
513         if (err < 0) {
514                 u32_clear_hw_hnode(tp, h, NULL);
515                 return err;
516         } else if (err > 0) {
517                 offloaded = true;
518         }
519
520         if (skip_sw && !offloaded)
521                 return -EINVAL;
522
523         return 0;
524 }
525
526 static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
527                                 struct netlink_ext_ack *extack)
528 {
529         struct tcf_block *block = tp->chain->block;
530         struct tc_cls_u32_offload cls_u32 = {};
531
532         tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
533         cls_u32.command = TC_CLSU32_DELETE_KNODE;
534         cls_u32.knode.handle = n->handle;
535
536         tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, false);
537         tcf_block_offload_dec(block, &n->flags);
538 }
539
540 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
541                                 u32 flags, struct netlink_ext_ack *extack)
542 {
543         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
544         struct tcf_block *block = tp->chain->block;
545         struct tc_cls_u32_offload cls_u32 = {};
546         bool skip_sw = tc_skip_sw(flags);
547         int err;
548
549         tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
550         cls_u32.command = TC_CLSU32_REPLACE_KNODE;
551         cls_u32.knode.handle = n->handle;
552         cls_u32.knode.fshift = n->fshift;
553 #ifdef CONFIG_CLS_U32_MARK
554         cls_u32.knode.val = n->val;
555         cls_u32.knode.mask = n->mask;
556 #else
557         cls_u32.knode.val = 0;
558         cls_u32.knode.mask = 0;
559 #endif
560         cls_u32.knode.sel = &n->sel;
561         cls_u32.knode.res = &n->res;
562         cls_u32.knode.exts = &n->exts;
563         if (n->ht_down)
564                 cls_u32.knode.link_handle = ht->handle;
565
566         err = tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, skip_sw);
567         if (err < 0) {
568                 u32_remove_hw_knode(tp, n, NULL);
569                 return err;
570         } else if (err > 0) {
571                 n->in_hw_count = err;
572                 tcf_block_offload_inc(block, &n->flags);
573         }
574
575         if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
576                 return -EINVAL;
577
578         return 0;
579 }
580
581 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
582                             struct netlink_ext_ack *extack)
583 {
584         struct tc_u_common *tp_c = tp->data;
585         struct tc_u_knode *n;
586         unsigned int h;
587
588         for (h = 0; h <= ht->divisor; h++) {
589                 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
590                         RCU_INIT_POINTER(ht->ht[h],
591                                          rtnl_dereference(n->next));
592                         tp_c->knodes--;
593                         tcf_unbind_filter(tp, &n->res);
594                         u32_remove_hw_knode(tp, n, extack);
595                         idr_remove(&ht->handle_idr, n->handle);
596                         if (tcf_exts_get_net(&n->exts))
597                                 tcf_queue_work(&n->rwork, u32_delete_key_freepf_work);
598                         else
599                                 u32_destroy_key(n, true);
600                 }
601         }
602 }
603
604 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
605                              struct netlink_ext_ack *extack)
606 {
607         struct tc_u_common *tp_c = tp->data;
608         struct tc_u_hnode __rcu **hn;
609         struct tc_u_hnode *phn;
610
611         WARN_ON(--ht->refcnt);
612
613         u32_clear_hnode(tp, ht, extack);
614
615         hn = &tp_c->hlist;
616         for (phn = rtnl_dereference(*hn);
617              phn;
618              hn = &phn->next, phn = rtnl_dereference(*hn)) {
619                 if (phn == ht) {
620                         u32_clear_hw_hnode(tp, ht, extack);
621                         idr_destroy(&ht->handle_idr);
622                         idr_remove(&tp_c->handle_idr, ht->handle);
623                         RCU_INIT_POINTER(*hn, ht->next);
624                         kfree_rcu(ht, rcu);
625                         return 0;
626                 }
627         }
628
629         return -ENOENT;
630 }
631
632 static void u32_destroy(struct tcf_proto *tp, struct netlink_ext_ack *extack)
633 {
634         struct tc_u_common *tp_c = tp->data;
635         struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
636
637         WARN_ON(root_ht == NULL);
638
639         if (root_ht && --root_ht->refcnt == 1)
640                 u32_destroy_hnode(tp, root_ht, extack);
641
642         if (--tp_c->refcnt == 0) {
643                 struct tc_u_hnode *ht;
644
645                 hlist_del(&tp_c->hnode);
646
647                 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
648                         u32_clear_hnode(tp, ht, extack);
649                         RCU_INIT_POINTER(tp_c->hlist, ht->next);
650
651                         /* u32_destroy_key() will later free ht for us, if it's
652                          * still referenced by some knode
653                          */
654                         if (--ht->refcnt == 0)
655                                 kfree_rcu(ht, rcu);
656                 }
657
658                 idr_destroy(&tp_c->handle_idr);
659                 kfree(tp_c);
660         }
661
662         tp->data = NULL;
663 }
664
665 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last,
666                       struct netlink_ext_ack *extack)
667 {
668         struct tc_u_hnode *ht = arg;
669         struct tc_u_common *tp_c = tp->data;
670         int ret = 0;
671
672         if (TC_U32_KEY(ht->handle)) {
673                 u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack);
674                 ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
675                 goto out;
676         }
677
678         if (ht->is_root) {
679                 NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node");
680                 return -EINVAL;
681         }
682
683         if (ht->refcnt == 1) {
684                 u32_destroy_hnode(tp, ht, extack);
685         } else {
686                 NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter");
687                 return -EBUSY;
688         }
689
690 out:
691         *last = tp_c->refcnt == 1 && tp_c->knodes == 0;
692         return ret;
693 }
694
695 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
696 {
697         u32 index = htid | 0x800;
698         u32 max = htid | 0xFFF;
699
700         if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max, GFP_KERNEL)) {
701                 index = htid + 1;
702                 if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max,
703                                  GFP_KERNEL))
704                         index = max;
705         }
706
707         return index;
708 }
709
710 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
711         [TCA_U32_CLASSID]       = { .type = NLA_U32 },
712         [TCA_U32_HASH]          = { .type = NLA_U32 },
713         [TCA_U32_LINK]          = { .type = NLA_U32 },
714         [TCA_U32_DIVISOR]       = { .type = NLA_U32 },
715         [TCA_U32_SEL]           = { .len = sizeof(struct tc_u32_sel) },
716         [TCA_U32_INDEV]         = { .type = NLA_STRING, .len = IFNAMSIZ },
717         [TCA_U32_MARK]          = { .len = sizeof(struct tc_u32_mark) },
718         [TCA_U32_FLAGS]         = { .type = NLA_U32 },
719 };
720
721 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
722                          unsigned long base,
723                          struct tc_u_knode *n, struct nlattr **tb,
724                          struct nlattr *est, bool ovr,
725                          struct netlink_ext_ack *extack)
726 {
727         int err;
728
729         err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr, extack);
730         if (err < 0)
731                 return err;
732
733         if (tb[TCA_U32_LINK]) {
734                 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
735                 struct tc_u_hnode *ht_down = NULL, *ht_old;
736
737                 if (TC_U32_KEY(handle)) {
738                         NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table");
739                         return -EINVAL;
740                 }
741
742                 if (handle) {
743                         ht_down = u32_lookup_ht(tp->data, handle);
744
745                         if (!ht_down) {
746                                 NL_SET_ERR_MSG_MOD(extack, "Link hash table not found");
747                                 return -EINVAL;
748                         }
749                         if (ht_down->is_root) {
750                                 NL_SET_ERR_MSG_MOD(extack, "Not linking to root node");
751                                 return -EINVAL;
752                         }
753                         ht_down->refcnt++;
754                 }
755
756                 ht_old = rtnl_dereference(n->ht_down);
757                 rcu_assign_pointer(n->ht_down, ht_down);
758
759                 if (ht_old)
760                         ht_old->refcnt--;
761         }
762         if (tb[TCA_U32_CLASSID]) {
763                 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
764                 tcf_bind_filter(tp, &n->res, base);
765         }
766
767 #ifdef CONFIG_NET_CLS_IND
768         if (tb[TCA_U32_INDEV]) {
769                 int ret;
770                 ret = tcf_change_indev(net, tb[TCA_U32_INDEV], extack);
771                 if (ret < 0)
772                         return -EINVAL;
773                 n->ifindex = ret;
774         }
775 #endif
776         return 0;
777 }
778
779 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
780                               struct tc_u_knode *n)
781 {
782         struct tc_u_knode __rcu **ins;
783         struct tc_u_knode *pins;
784         struct tc_u_hnode *ht;
785
786         if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
787                 ht = rtnl_dereference(tp->root);
788         else
789                 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
790
791         ins = &ht->ht[TC_U32_HASH(n->handle)];
792
793         /* The node must always exist for it to be replaced if this is not the
794          * case then something went very wrong elsewhere.
795          */
796         for (pins = rtnl_dereference(*ins); ;
797              ins = &pins->next, pins = rtnl_dereference(*ins))
798                 if (pins->handle == n->handle)
799                         break;
800
801         idr_replace(&ht->handle_idr, n, n->handle);
802         RCU_INIT_POINTER(n->next, pins->next);
803         rcu_assign_pointer(*ins, n);
804 }
805
806 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp,
807                                          struct tc_u_knode *n)
808 {
809         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
810         struct tc_u32_sel *s = &n->sel;
811         struct tc_u_knode *new;
812
813         new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key),
814                       GFP_KERNEL);
815
816         if (!new)
817                 return NULL;
818
819         RCU_INIT_POINTER(new->next, n->next);
820         new->handle = n->handle;
821         RCU_INIT_POINTER(new->ht_up, n->ht_up);
822
823 #ifdef CONFIG_NET_CLS_IND
824         new->ifindex = n->ifindex;
825 #endif
826         new->fshift = n->fshift;
827         new->res = n->res;
828         new->flags = n->flags;
829         RCU_INIT_POINTER(new->ht_down, ht);
830
831         /* bump reference count as long as we hold pointer to structure */
832         if (ht)
833                 ht->refcnt++;
834
835 #ifdef CONFIG_CLS_U32_PERF
836         /* Statistics may be incremented by readers during update
837          * so we must keep them in tact. When the node is later destroyed
838          * a special destroy call must be made to not free the pf memory.
839          */
840         new->pf = n->pf;
841 #endif
842
843 #ifdef CONFIG_CLS_U32_MARK
844         new->val = n->val;
845         new->mask = n->mask;
846         /* Similarly success statistics must be moved as pointers */
847         new->pcpu_success = n->pcpu_success;
848 #endif
849         memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
850
851         if (tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE)) {
852                 kfree(new);
853                 return NULL;
854         }
855
856         return new;
857 }
858
859 static int u32_change(struct net *net, struct sk_buff *in_skb,
860                       struct tcf_proto *tp, unsigned long base, u32 handle,
861                       struct nlattr **tca, void **arg, bool ovr,
862                       struct netlink_ext_ack *extack)
863 {
864         struct tc_u_common *tp_c = tp->data;
865         struct tc_u_hnode *ht;
866         struct tc_u_knode *n;
867         struct tc_u32_sel *s;
868         struct nlattr *opt = tca[TCA_OPTIONS];
869         struct nlattr *tb[TCA_U32_MAX + 1];
870         u32 htid, flags = 0;
871         size_t sel_size;
872         int err;
873 #ifdef CONFIG_CLS_U32_PERF
874         size_t size;
875 #endif
876
877         if (!opt) {
878                 if (handle) {
879                         NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options");
880                         return -EINVAL;
881                 } else {
882                         return 0;
883                 }
884         }
885
886         err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy, extack);
887         if (err < 0)
888                 return err;
889
890         if (tb[TCA_U32_FLAGS]) {
891                 flags = nla_get_u32(tb[TCA_U32_FLAGS]);
892                 if (!tc_flags_valid(flags)) {
893                         NL_SET_ERR_MSG_MOD(extack, "Invalid filter flags");
894                         return -EINVAL;
895                 }
896         }
897
898         n = *arg;
899         if (n) {
900                 struct tc_u_knode *new;
901
902                 if (TC_U32_KEY(n->handle) == 0) {
903                         NL_SET_ERR_MSG_MOD(extack, "Key node id cannot be zero");
904                         return -EINVAL;
905                 }
906
907                 if ((n->flags ^ flags) &
908                     ~(TCA_CLS_FLAGS_IN_HW | TCA_CLS_FLAGS_NOT_IN_HW)) {
909                         NL_SET_ERR_MSG_MOD(extack, "Key node flags do not match passed flags");
910                         return -EINVAL;
911                 }
912
913                 new = u32_init_knode(tp, n);
914                 if (!new)
915                         return -ENOMEM;
916
917                 err = u32_set_parms(net, tp, base, new, tb,
918                                     tca[TCA_RATE], ovr, extack);
919
920                 if (err) {
921                         u32_destroy_key(new, false);
922                         return err;
923                 }
924
925                 err = u32_replace_hw_knode(tp, new, flags, extack);
926                 if (err) {
927                         u32_destroy_key(new, false);
928                         return err;
929                 }
930
931                 if (!tc_in_hw(new->flags))
932                         new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
933
934                 u32_replace_knode(tp, tp_c, new);
935                 tcf_unbind_filter(tp, &n->res);
936                 tcf_exts_get_net(&n->exts);
937                 tcf_queue_work(&n->rwork, u32_delete_key_work);
938                 return 0;
939         }
940
941         if (tb[TCA_U32_DIVISOR]) {
942                 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
943
944                 if (!is_power_of_2(divisor)) {
945                         NL_SET_ERR_MSG_MOD(extack, "Divisor is not a power of 2");
946                         return -EINVAL;
947                 }
948                 if (divisor-- > 0x100) {
949                         NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets");
950                         return -EINVAL;
951                 }
952                 if (TC_U32_KEY(handle)) {
953                         NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table");
954                         return -EINVAL;
955                 }
956                 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
957                 if (ht == NULL)
958                         return -ENOBUFS;
959                 if (handle == 0) {
960                         handle = gen_new_htid(tp->data, ht);
961                         if (handle == 0) {
962                                 kfree(ht);
963                                 return -ENOMEM;
964                         }
965                 } else {
966                         err = idr_alloc_u32(&tp_c->handle_idr, ht, &handle,
967                                             handle, GFP_KERNEL);
968                         if (err) {
969                                 kfree(ht);
970                                 return err;
971                         }
972                 }
973                 ht->refcnt = 1;
974                 ht->divisor = divisor;
975                 ht->handle = handle;
976                 ht->prio = tp->prio;
977                 idr_init(&ht->handle_idr);
978                 ht->flags = flags;
979
980                 err = u32_replace_hw_hnode(tp, ht, flags, extack);
981                 if (err) {
982                         idr_remove(&tp_c->handle_idr, handle);
983                         kfree(ht);
984                         return err;
985                 }
986
987                 RCU_INIT_POINTER(ht->next, tp_c->hlist);
988                 rcu_assign_pointer(tp_c->hlist, ht);
989                 *arg = ht;
990
991                 return 0;
992         }
993
994         if (tb[TCA_U32_HASH]) {
995                 htid = nla_get_u32(tb[TCA_U32_HASH]);
996                 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
997                         ht = rtnl_dereference(tp->root);
998                         htid = ht->handle;
999                 } else {
1000                         ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
1001                         if (!ht) {
1002                                 NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found");
1003                                 return -EINVAL;
1004                         }
1005                 }
1006         } else {
1007                 ht = rtnl_dereference(tp->root);
1008                 htid = ht->handle;
1009         }
1010
1011         if (ht->divisor < TC_U32_HASH(htid)) {
1012                 NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value");
1013                 return -EINVAL;
1014         }
1015
1016         if (handle) {
1017                 if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) {
1018                         NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch");
1019                         return -EINVAL;
1020                 }
1021                 handle = htid | TC_U32_NODE(handle);
1022                 err = idr_alloc_u32(&ht->handle_idr, NULL, &handle, handle,
1023                                     GFP_KERNEL);
1024                 if (err)
1025                         return err;
1026         } else
1027                 handle = gen_new_kid(ht, htid);
1028
1029         if (tb[TCA_U32_SEL] == NULL) {
1030                 NL_SET_ERR_MSG_MOD(extack, "Selector not specified");
1031                 err = -EINVAL;
1032                 goto erridr;
1033         }
1034
1035         s = nla_data(tb[TCA_U32_SEL]);
1036         sel_size = struct_size(s, keys, s->nkeys);
1037         if (nla_len(tb[TCA_U32_SEL]) < sel_size) {
1038                 err = -EINVAL;
1039                 goto erridr;
1040         }
1041
1042         n = kzalloc(offsetof(typeof(*n), sel) + sel_size, GFP_KERNEL);
1043         if (n == NULL) {
1044                 err = -ENOBUFS;
1045                 goto erridr;
1046         }
1047
1048 #ifdef CONFIG_CLS_U32_PERF
1049         size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64);
1050         n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt));
1051         if (!n->pf) {
1052                 err = -ENOBUFS;
1053                 goto errfree;
1054         }
1055 #endif
1056
1057         memcpy(&n->sel, s, sel_size);
1058         RCU_INIT_POINTER(n->ht_up, ht);
1059         n->handle = handle;
1060         n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1061         n->flags = flags;
1062
1063         err = tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE);
1064         if (err < 0)
1065                 goto errout;
1066
1067 #ifdef CONFIG_CLS_U32_MARK
1068         n->pcpu_success = alloc_percpu(u32);
1069         if (!n->pcpu_success) {
1070                 err = -ENOMEM;
1071                 goto errout;
1072         }
1073
1074         if (tb[TCA_U32_MARK]) {
1075                 struct tc_u32_mark *mark;
1076
1077                 mark = nla_data(tb[TCA_U32_MARK]);
1078                 n->val = mark->val;
1079                 n->mask = mark->mask;
1080         }
1081 #endif
1082
1083         err = u32_set_parms(net, tp, base, n, tb, tca[TCA_RATE], ovr,
1084                             extack);
1085         if (err == 0) {
1086                 struct tc_u_knode __rcu **ins;
1087                 struct tc_u_knode *pins;
1088
1089                 err = u32_replace_hw_knode(tp, n, flags, extack);
1090                 if (err)
1091                         goto errhw;
1092
1093                 if (!tc_in_hw(n->flags))
1094                         n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1095
1096                 ins = &ht->ht[TC_U32_HASH(handle)];
1097                 for (pins = rtnl_dereference(*ins); pins;
1098                      ins = &pins->next, pins = rtnl_dereference(*ins))
1099                         if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1100                                 break;
1101
1102                 RCU_INIT_POINTER(n->next, pins);
1103                 rcu_assign_pointer(*ins, n);
1104                 tp_c->knodes++;
1105                 *arg = n;
1106                 return 0;
1107         }
1108
1109 errhw:
1110 #ifdef CONFIG_CLS_U32_MARK
1111         free_percpu(n->pcpu_success);
1112 #endif
1113
1114 errout:
1115         tcf_exts_destroy(&n->exts);
1116 #ifdef CONFIG_CLS_U32_PERF
1117 errfree:
1118         free_percpu(n->pf);
1119 #endif
1120         kfree(n);
1121 erridr:
1122         idr_remove(&ht->handle_idr, handle);
1123         return err;
1124 }
1125
1126 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg)
1127 {
1128         struct tc_u_common *tp_c = tp->data;
1129         struct tc_u_hnode *ht;
1130         struct tc_u_knode *n;
1131         unsigned int h;
1132
1133         if (arg->stop)
1134                 return;
1135
1136         for (ht = rtnl_dereference(tp_c->hlist);
1137              ht;
1138              ht = rtnl_dereference(ht->next)) {
1139                 if (ht->prio != tp->prio)
1140                         continue;
1141                 if (arg->count >= arg->skip) {
1142                         if (arg->fn(tp, ht, arg) < 0) {
1143                                 arg->stop = 1;
1144                                 return;
1145                         }
1146                 }
1147                 arg->count++;
1148                 for (h = 0; h <= ht->divisor; h++) {
1149                         for (n = rtnl_dereference(ht->ht[h]);
1150                              n;
1151                              n = rtnl_dereference(n->next)) {
1152                                 if (arg->count < arg->skip) {
1153                                         arg->count++;
1154                                         continue;
1155                                 }
1156                                 if (arg->fn(tp, n, arg) < 0) {
1157                                         arg->stop = 1;
1158                                         return;
1159                                 }
1160                                 arg->count++;
1161                         }
1162                 }
1163         }
1164 }
1165
1166 static int u32_reoffload_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
1167                                bool add, tc_setup_cb_t *cb, void *cb_priv,
1168                                struct netlink_ext_ack *extack)
1169 {
1170         struct tc_cls_u32_offload cls_u32 = {};
1171         int err;
1172
1173         tc_cls_common_offload_init(&cls_u32.common, tp, ht->flags, extack);
1174         cls_u32.command = add ? TC_CLSU32_NEW_HNODE : TC_CLSU32_DELETE_HNODE;
1175         cls_u32.hnode.divisor = ht->divisor;
1176         cls_u32.hnode.handle = ht->handle;
1177         cls_u32.hnode.prio = ht->prio;
1178
1179         err = cb(TC_SETUP_CLSU32, &cls_u32, cb_priv);
1180         if (err && add && tc_skip_sw(ht->flags))
1181                 return err;
1182
1183         return 0;
1184 }
1185
1186 static int u32_reoffload_knode(struct tcf_proto *tp, struct tc_u_knode *n,
1187                                bool add, tc_setup_cb_t *cb, void *cb_priv,
1188                                struct netlink_ext_ack *extack)
1189 {
1190         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
1191         struct tcf_block *block = tp->chain->block;
1192         struct tc_cls_u32_offload cls_u32 = {};
1193         int err;
1194
1195         tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
1196         cls_u32.command = add ?
1197                 TC_CLSU32_REPLACE_KNODE : TC_CLSU32_DELETE_KNODE;
1198         cls_u32.knode.handle = n->handle;
1199
1200         if (add) {
1201                 cls_u32.knode.fshift = n->fshift;
1202 #ifdef CONFIG_CLS_U32_MARK
1203                 cls_u32.knode.val = n->val;
1204                 cls_u32.knode.mask = n->mask;
1205 #else
1206                 cls_u32.knode.val = 0;
1207                 cls_u32.knode.mask = 0;
1208 #endif
1209                 cls_u32.knode.sel = &n->sel;
1210                 cls_u32.knode.res = &n->res;
1211                 cls_u32.knode.exts = &n->exts;
1212                 if (n->ht_down)
1213                         cls_u32.knode.link_handle = ht->handle;
1214         }
1215
1216         err = cb(TC_SETUP_CLSU32, &cls_u32, cb_priv);
1217         if (err) {
1218                 if (add && tc_skip_sw(n->flags))
1219                         return err;
1220                 return 0;
1221         }
1222
1223         tc_cls_offload_cnt_update(block, &n->in_hw_count, &n->flags, add);
1224
1225         return 0;
1226 }
1227
1228 static int u32_reoffload(struct tcf_proto *tp, bool add, tc_setup_cb_t *cb,
1229                          void *cb_priv, struct netlink_ext_ack *extack)
1230 {
1231         struct tc_u_common *tp_c = tp->data;
1232         struct tc_u_hnode *ht;
1233         struct tc_u_knode *n;
1234         unsigned int h;
1235         int err;
1236
1237         for (ht = rtnl_dereference(tp_c->hlist);
1238              ht;
1239              ht = rtnl_dereference(ht->next)) {
1240                 if (ht->prio != tp->prio)
1241                         continue;
1242
1243                 /* When adding filters to a new dev, try to offload the
1244                  * hashtable first. When removing, do the filters before the
1245                  * hashtable.
1246                  */
1247                 if (add && !tc_skip_hw(ht->flags)) {
1248                         err = u32_reoffload_hnode(tp, ht, add, cb, cb_priv,
1249                                                   extack);
1250                         if (err)
1251                                 return err;
1252                 }
1253
1254                 for (h = 0; h <= ht->divisor; h++) {
1255                         for (n = rtnl_dereference(ht->ht[h]);
1256                              n;
1257                              n = rtnl_dereference(n->next)) {
1258                                 if (tc_skip_hw(n->flags))
1259                                         continue;
1260
1261                                 err = u32_reoffload_knode(tp, n, add, cb,
1262                                                           cb_priv, extack);
1263                                 if (err)
1264                                         return err;
1265                         }
1266                 }
1267
1268                 if (!add && !tc_skip_hw(ht->flags))
1269                         u32_reoffload_hnode(tp, ht, add, cb, cb_priv, extack);
1270         }
1271
1272         return 0;
1273 }
1274
1275 static void u32_bind_class(void *fh, u32 classid, unsigned long cl)
1276 {
1277         struct tc_u_knode *n = fh;
1278
1279         if (n && n->res.classid == classid)
1280                 n->res.class = cl;
1281 }
1282
1283 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1284                     struct sk_buff *skb, struct tcmsg *t)
1285 {
1286         struct tc_u_knode *n = fh;
1287         struct tc_u_hnode *ht_up, *ht_down;
1288         struct nlattr *nest;
1289
1290         if (n == NULL)
1291                 return skb->len;
1292
1293         t->tcm_handle = n->handle;
1294
1295         nest = nla_nest_start(skb, TCA_OPTIONS);
1296         if (nest == NULL)
1297                 goto nla_put_failure;
1298
1299         if (TC_U32_KEY(n->handle) == 0) {
1300                 struct tc_u_hnode *ht = fh;
1301                 u32 divisor = ht->divisor + 1;
1302
1303                 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1304                         goto nla_put_failure;
1305         } else {
1306 #ifdef CONFIG_CLS_U32_PERF
1307                 struct tc_u32_pcnt *gpf;
1308                 int cpu;
1309 #endif
1310
1311                 if (nla_put(skb, TCA_U32_SEL,
1312                             sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
1313                             &n->sel))
1314                         goto nla_put_failure;
1315
1316                 ht_up = rtnl_dereference(n->ht_up);
1317                 if (ht_up) {
1318                         u32 htid = n->handle & 0xFFFFF000;
1319                         if (nla_put_u32(skb, TCA_U32_HASH, htid))
1320                                 goto nla_put_failure;
1321                 }
1322                 if (n->res.classid &&
1323                     nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1324                         goto nla_put_failure;
1325
1326                 ht_down = rtnl_dereference(n->ht_down);
1327                 if (ht_down &&
1328                     nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1329                         goto nla_put_failure;
1330
1331                 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1332                         goto nla_put_failure;
1333
1334 #ifdef CONFIG_CLS_U32_MARK
1335                 if ((n->val || n->mask)) {
1336                         struct tc_u32_mark mark = {.val = n->val,
1337                                                    .mask = n->mask,
1338                                                    .success = 0};
1339                         int cpum;
1340
1341                         for_each_possible_cpu(cpum) {
1342                                 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1343
1344                                 mark.success += cnt;
1345                         }
1346
1347                         if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1348                                 goto nla_put_failure;
1349                 }
1350 #endif
1351
1352                 if (tcf_exts_dump(skb, &n->exts) < 0)
1353                         goto nla_put_failure;
1354
1355 #ifdef CONFIG_NET_CLS_IND
1356                 if (n->ifindex) {
1357                         struct net_device *dev;
1358                         dev = __dev_get_by_index(net, n->ifindex);
1359                         if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1360                                 goto nla_put_failure;
1361                 }
1362 #endif
1363 #ifdef CONFIG_CLS_U32_PERF
1364                 gpf = kzalloc(sizeof(struct tc_u32_pcnt) +
1365                               n->sel.nkeys * sizeof(u64),
1366                               GFP_KERNEL);
1367                 if (!gpf)
1368                         goto nla_put_failure;
1369
1370                 for_each_possible_cpu(cpu) {
1371                         int i;
1372                         struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1373
1374                         gpf->rcnt += pf->rcnt;
1375                         gpf->rhit += pf->rhit;
1376                         for (i = 0; i < n->sel.nkeys; i++)
1377                                 gpf->kcnts[i] += pf->kcnts[i];
1378                 }
1379
1380                 if (nla_put_64bit(skb, TCA_U32_PCNT,
1381                                   sizeof(struct tc_u32_pcnt) +
1382                                   n->sel.nkeys * sizeof(u64),
1383                                   gpf, TCA_U32_PAD)) {
1384                         kfree(gpf);
1385                         goto nla_put_failure;
1386                 }
1387                 kfree(gpf);
1388 #endif
1389         }
1390
1391         nla_nest_end(skb, nest);
1392
1393         if (TC_U32_KEY(n->handle))
1394                 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1395                         goto nla_put_failure;
1396         return skb->len;
1397
1398 nla_put_failure:
1399         nla_nest_cancel(skb, nest);
1400         return -1;
1401 }
1402
1403 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1404         .kind           =       "u32",
1405         .classify       =       u32_classify,
1406         .init           =       u32_init,
1407         .destroy        =       u32_destroy,
1408         .get            =       u32_get,
1409         .change         =       u32_change,
1410         .delete         =       u32_delete,
1411         .walk           =       u32_walk,
1412         .reoffload      =       u32_reoffload,
1413         .dump           =       u32_dump,
1414         .bind_class     =       u32_bind_class,
1415         .owner          =       THIS_MODULE,
1416 };
1417
1418 static int __init init_u32(void)
1419 {
1420         int i, ret;
1421
1422         pr_info("u32 classifier\n");
1423 #ifdef CONFIG_CLS_U32_PERF
1424         pr_info("    Performance counters on\n");
1425 #endif
1426 #ifdef CONFIG_NET_CLS_IND
1427         pr_info("    input device check on\n");
1428 #endif
1429 #ifdef CONFIG_NET_CLS_ACT
1430         pr_info("    Actions configured\n");
1431 #endif
1432         tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1433                                           sizeof(struct hlist_head),
1434                                           GFP_KERNEL);
1435         if (!tc_u_common_hash)
1436                 return -ENOMEM;
1437
1438         for (i = 0; i < U32_HASH_SIZE; i++)
1439                 INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1440
1441         ret = register_tcf_proto_ops(&cls_u32_ops);
1442         if (ret)
1443                 kvfree(tc_u_common_hash);
1444         return ret;
1445 }
1446
1447 static void __exit exit_u32(void)
1448 {
1449         unregister_tcf_proto_ops(&cls_u32_ops);
1450         kvfree(tc_u_common_hash);
1451 }
1452
1453 module_init(init_u32)
1454 module_exit(exit_u32)
1455 MODULE_LICENSE("GPL");