Merge tag 'tty-4.15-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty
[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 #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         /* The 'ht' field MUST be the last field in structure to allow for
91          * more entries allocated at end of structure.
92          */
93         struct tc_u_knode __rcu *ht[1];
94 };
95
96 struct tc_u_common {
97         struct tc_u_hnode __rcu *hlist;
98         struct tcf_block        *block;
99         int                     refcnt;
100         struct idr              handle_idr;
101         struct hlist_node       hnode;
102         struct rcu_head         rcu;
103 };
104
105 static inline unsigned int u32_hash_fold(__be32 key,
106                                          const struct tc_u32_sel *sel,
107                                          u8 fshift)
108 {
109         unsigned int h = ntohl(key & sel->hmask) >> fshift;
110
111         return h;
112 }
113
114 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp,
115                         struct tcf_result *res)
116 {
117         struct {
118                 struct tc_u_knode *knode;
119                 unsigned int      off;
120         } stack[TC_U32_MAXDEPTH];
121
122         struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
123         unsigned int off = skb_network_offset(skb);
124         struct tc_u_knode *n;
125         int sdepth = 0;
126         int off2 = 0;
127         int sel = 0;
128 #ifdef CONFIG_CLS_U32_PERF
129         int j;
130 #endif
131         int i, r;
132
133 next_ht:
134         n = rcu_dereference_bh(ht->ht[sel]);
135
136 next_knode:
137         if (n) {
138                 struct tc_u32_key *key = n->sel.keys;
139
140 #ifdef CONFIG_CLS_U32_PERF
141                 __this_cpu_inc(n->pf->rcnt);
142                 j = 0;
143 #endif
144
145                 if (tc_skip_sw(n->flags)) {
146                         n = rcu_dereference_bh(n->next);
147                         goto next_knode;
148                 }
149
150 #ifdef CONFIG_CLS_U32_MARK
151                 if ((skb->mark & n->mask) != n->val) {
152                         n = rcu_dereference_bh(n->next);
153                         goto next_knode;
154                 } else {
155                         __this_cpu_inc(*n->pcpu_success);
156                 }
157 #endif
158
159                 for (i = n->sel.nkeys; i > 0; i--, key++) {
160                         int toff = off + key->off + (off2 & key->offmask);
161                         __be32 *data, hdata;
162
163                         if (skb_headroom(skb) + toff > INT_MAX)
164                                 goto out;
165
166                         data = skb_header_pointer(skb, toff, 4, &hdata);
167                         if (!data)
168                                 goto out;
169                         if ((*data ^ key->val) & key->mask) {
170                                 n = rcu_dereference_bh(n->next);
171                                 goto next_knode;
172                         }
173 #ifdef CONFIG_CLS_U32_PERF
174                         __this_cpu_inc(n->pf->kcnts[j]);
175                         j++;
176 #endif
177                 }
178
179                 ht = rcu_dereference_bh(n->ht_down);
180                 if (!ht) {
181 check_terminal:
182                         if (n->sel.flags & TC_U32_TERMINAL) {
183
184                                 *res = n->res;
185 #ifdef CONFIG_NET_CLS_IND
186                                 if (!tcf_match_indev(skb, n->ifindex)) {
187                                         n = rcu_dereference_bh(n->next);
188                                         goto next_knode;
189                                 }
190 #endif
191 #ifdef CONFIG_CLS_U32_PERF
192                                 __this_cpu_inc(n->pf->rhit);
193 #endif
194                                 r = tcf_exts_exec(skb, &n->exts, res);
195                                 if (r < 0) {
196                                         n = rcu_dereference_bh(n->next);
197                                         goto next_knode;
198                                 }
199
200                                 return r;
201                         }
202                         n = rcu_dereference_bh(n->next);
203                         goto next_knode;
204                 }
205
206                 /* PUSH */
207                 if (sdepth >= TC_U32_MAXDEPTH)
208                         goto deadloop;
209                 stack[sdepth].knode = n;
210                 stack[sdepth].off = off;
211                 sdepth++;
212
213                 ht = rcu_dereference_bh(n->ht_down);
214                 sel = 0;
215                 if (ht->divisor) {
216                         __be32 *data, hdata;
217
218                         data = skb_header_pointer(skb, off + n->sel.hoff, 4,
219                                                   &hdata);
220                         if (!data)
221                                 goto out;
222                         sel = ht->divisor & u32_hash_fold(*data, &n->sel,
223                                                           n->fshift);
224                 }
225                 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
226                         goto next_ht;
227
228                 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
229                         off2 = n->sel.off + 3;
230                         if (n->sel.flags & TC_U32_VAROFFSET) {
231                                 __be16 *data, hdata;
232
233                                 data = skb_header_pointer(skb,
234                                                           off + n->sel.offoff,
235                                                           2, &hdata);
236                                 if (!data)
237                                         goto out;
238                                 off2 += ntohs(n->sel.offmask & *data) >>
239                                         n->sel.offshift;
240                         }
241                         off2 &= ~3;
242                 }
243                 if (n->sel.flags & TC_U32_EAT) {
244                         off += off2;
245                         off2 = 0;
246                 }
247
248                 if (off < skb->len)
249                         goto next_ht;
250         }
251
252         /* POP */
253         if (sdepth--) {
254                 n = stack[sdepth].knode;
255                 ht = rcu_dereference_bh(n->ht_up);
256                 off = stack[sdepth].off;
257                 goto check_terminal;
258         }
259 out:
260         return -1;
261
262 deadloop:
263         net_warn_ratelimited("cls_u32: dead loop\n");
264         return -1;
265 }
266
267 static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
268 {
269         struct tc_u_hnode *ht;
270
271         for (ht = rtnl_dereference(tp_c->hlist);
272              ht;
273              ht = rtnl_dereference(ht->next))
274                 if (ht->handle == handle)
275                         break;
276
277         return ht;
278 }
279
280 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
281 {
282         unsigned int sel;
283         struct tc_u_knode *n = NULL;
284
285         sel = TC_U32_HASH(handle);
286         if (sel > ht->divisor)
287                 goto out;
288
289         for (n = rtnl_dereference(ht->ht[sel]);
290              n;
291              n = rtnl_dereference(n->next))
292                 if (n->handle == handle)
293                         break;
294 out:
295         return n;
296 }
297
298
299 static void *u32_get(struct tcf_proto *tp, u32 handle)
300 {
301         struct tc_u_hnode *ht;
302         struct tc_u_common *tp_c = tp->data;
303
304         if (TC_U32_HTID(handle) == TC_U32_ROOT)
305                 ht = rtnl_dereference(tp->root);
306         else
307                 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
308
309         if (!ht)
310                 return NULL;
311
312         if (TC_U32_KEY(handle) == 0)
313                 return ht;
314
315         return u32_lookup_key(ht, handle);
316 }
317
318 static u32 gen_new_htid(struct tc_u_common *tp_c, struct tc_u_hnode *ptr)
319 {
320         unsigned long idr_index;
321         int err;
322
323         /* This is only used inside rtnl lock it is safe to increment
324          * without read _copy_ update semantics
325          */
326         err = idr_alloc_ext(&tp_c->handle_idr, ptr, &idr_index,
327                             1, 0x7FF, GFP_KERNEL);
328         if (err)
329                 return 0;
330         return (u32)(idr_index | 0x800) << 20;
331 }
332
333 static struct hlist_head *tc_u_common_hash;
334
335 #define U32_HASH_SHIFT 10
336 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
337
338 static unsigned int tc_u_hash(const struct tcf_proto *tp)
339 {
340         return hash_ptr(tp->chain->block, U32_HASH_SHIFT);
341 }
342
343 static struct tc_u_common *tc_u_common_find(const struct tcf_proto *tp)
344 {
345         struct tc_u_common *tc;
346         unsigned int h;
347
348         h = tc_u_hash(tp);
349         hlist_for_each_entry(tc, &tc_u_common_hash[h], hnode) {
350                 if (tc->block == tp->chain->block)
351                         return tc;
352         }
353         return NULL;
354 }
355
356 static int u32_init(struct tcf_proto *tp)
357 {
358         struct tc_u_hnode *root_ht;
359         struct tc_u_common *tp_c;
360         unsigned int h;
361
362         tp_c = tc_u_common_find(tp);
363
364         root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
365         if (root_ht == NULL)
366                 return -ENOBUFS;
367
368         root_ht->refcnt++;
369         root_ht->handle = tp_c ? gen_new_htid(tp_c, root_ht) : 0x80000000;
370         root_ht->prio = tp->prio;
371         idr_init(&root_ht->handle_idr);
372
373         if (tp_c == NULL) {
374                 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
375                 if (tp_c == NULL) {
376                         kfree(root_ht);
377                         return -ENOBUFS;
378                 }
379                 tp_c->block = tp->chain->block;
380                 INIT_HLIST_NODE(&tp_c->hnode);
381                 idr_init(&tp_c->handle_idr);
382
383                 h = tc_u_hash(tp);
384                 hlist_add_head(&tp_c->hnode, &tc_u_common_hash[h]);
385         }
386
387         tp_c->refcnt++;
388         RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
389         rcu_assign_pointer(tp_c->hlist, root_ht);
390         root_ht->tp_c = tp_c;
391
392         rcu_assign_pointer(tp->root, root_ht);
393         tp->data = tp_c;
394         return 0;
395 }
396
397 static int u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n,
398                            bool free_pf)
399 {
400         tcf_exts_destroy(&n->exts);
401         tcf_exts_put_net(&n->exts);
402         if (n->ht_down)
403                 n->ht_down->refcnt--;
404 #ifdef CONFIG_CLS_U32_PERF
405         if (free_pf)
406                 free_percpu(n->pf);
407 #endif
408 #ifdef CONFIG_CLS_U32_MARK
409         if (free_pf)
410                 free_percpu(n->pcpu_success);
411 #endif
412         kfree(n);
413         return 0;
414 }
415
416 /* u32_delete_key_rcu should be called when free'ing a copied
417  * version of a tc_u_knode obtained from u32_init_knode(). When
418  * copies are obtained from u32_init_knode() the statistics are
419  * shared between the old and new copies to allow readers to
420  * continue to update the statistics during the copy. To support
421  * this the u32_delete_key_rcu variant does not free the percpu
422  * statistics.
423  */
424 static void u32_delete_key_work(struct work_struct *work)
425 {
426         struct tc_u_knode *key = container_of(work, struct tc_u_knode, work);
427
428         rtnl_lock();
429         u32_destroy_key(key->tp, key, false);
430         rtnl_unlock();
431 }
432
433 static void u32_delete_key_rcu(struct rcu_head *rcu)
434 {
435         struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
436
437         INIT_WORK(&key->work, u32_delete_key_work);
438         tcf_queue_work(&key->work);
439 }
440
441 /* u32_delete_key_freepf_rcu is the rcu callback variant
442  * that free's the entire structure including the statistics
443  * percpu variables. Only use this if the key is not a copy
444  * returned by u32_init_knode(). See u32_delete_key_rcu()
445  * for the variant that should be used with keys return from
446  * u32_init_knode()
447  */
448 static void u32_delete_key_freepf_work(struct work_struct *work)
449 {
450         struct tc_u_knode *key = container_of(work, struct tc_u_knode, work);
451
452         rtnl_lock();
453         u32_destroy_key(key->tp, key, true);
454         rtnl_unlock();
455 }
456
457 static void u32_delete_key_freepf_rcu(struct rcu_head *rcu)
458 {
459         struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
460
461         INIT_WORK(&key->work, u32_delete_key_freepf_work);
462         tcf_queue_work(&key->work);
463 }
464
465 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
466 {
467         struct tc_u_knode __rcu **kp;
468         struct tc_u_knode *pkp;
469         struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
470
471         if (ht) {
472                 kp = &ht->ht[TC_U32_HASH(key->handle)];
473                 for (pkp = rtnl_dereference(*kp); pkp;
474                      kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
475                         if (pkp == key) {
476                                 RCU_INIT_POINTER(*kp, key->next);
477
478                                 tcf_unbind_filter(tp, &key->res);
479                                 tcf_exts_get_net(&key->exts);
480                                 call_rcu(&key->rcu, u32_delete_key_freepf_rcu);
481                                 return 0;
482                         }
483                 }
484         }
485         WARN_ON(1);
486         return 0;
487 }
488
489 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h)
490 {
491         struct tcf_block *block = tp->chain->block;
492         struct tc_cls_u32_offload cls_u32 = {};
493
494         tc_cls_common_offload_init(&cls_u32.common, tp);
495         cls_u32.command = TC_CLSU32_DELETE_HNODE;
496         cls_u32.hnode.divisor = h->divisor;
497         cls_u32.hnode.handle = h->handle;
498         cls_u32.hnode.prio = h->prio;
499
500         tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false);
501 }
502
503 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
504                                 u32 flags)
505 {
506         struct tcf_block *block = tp->chain->block;
507         struct tc_cls_u32_offload cls_u32 = {};
508         bool skip_sw = tc_skip_sw(flags);
509         bool offloaded = false;
510         int err;
511
512         tc_cls_common_offload_init(&cls_u32.common, tp);
513         cls_u32.command = TC_CLSU32_NEW_HNODE;
514         cls_u32.hnode.divisor = h->divisor;
515         cls_u32.hnode.handle = h->handle;
516         cls_u32.hnode.prio = h->prio;
517
518         err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw);
519         if (err < 0) {
520                 u32_clear_hw_hnode(tp, h);
521                 return err;
522         } else if (err > 0) {
523                 offloaded = true;
524         }
525
526         if (skip_sw && !offloaded)
527                 return -EINVAL;
528
529         return 0;
530 }
531
532 static void u32_remove_hw_knode(struct tcf_proto *tp, u32 handle)
533 {
534         struct tcf_block *block = tp->chain->block;
535         struct tc_cls_u32_offload cls_u32 = {};
536
537         tc_cls_common_offload_init(&cls_u32.common, tp);
538         cls_u32.command = TC_CLSU32_DELETE_KNODE;
539         cls_u32.knode.handle = handle;
540
541         tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false);
542 }
543
544 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
545                                 u32 flags)
546 {
547         struct tcf_block *block = tp->chain->block;
548         struct tc_cls_u32_offload cls_u32 = {};
549         bool skip_sw = tc_skip_sw(flags);
550         int err;
551
552         tc_cls_common_offload_init(&cls_u32.common, tp);
553         cls_u32.command = TC_CLSU32_REPLACE_KNODE;
554         cls_u32.knode.handle = n->handle;
555         cls_u32.knode.fshift = n->fshift;
556 #ifdef CONFIG_CLS_U32_MARK
557         cls_u32.knode.val = n->val;
558         cls_u32.knode.mask = n->mask;
559 #else
560         cls_u32.knode.val = 0;
561         cls_u32.knode.mask = 0;
562 #endif
563         cls_u32.knode.sel = &n->sel;
564         cls_u32.knode.exts = &n->exts;
565         if (n->ht_down)
566                 cls_u32.knode.link_handle = n->ht_down->handle;
567
568         err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw);
569         if (err < 0) {
570                 u32_remove_hw_knode(tp, n->handle);
571                 return err;
572         } else if (err > 0) {
573                 n->flags |= TCA_CLS_FLAGS_IN_HW;
574         }
575
576         if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
577                 return -EINVAL;
578
579         return 0;
580 }
581
582 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
583 {
584         struct tc_u_knode *n;
585         unsigned int h;
586
587         for (h = 0; h <= ht->divisor; h++) {
588                 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
589                         RCU_INIT_POINTER(ht->ht[h],
590                                          rtnl_dereference(n->next));
591                         tcf_unbind_filter(tp, &n->res);
592                         u32_remove_hw_knode(tp, n->handle);
593                         idr_remove_ext(&ht->handle_idr, n->handle);
594                         if (tcf_exts_get_net(&n->exts))
595                                 call_rcu(&n->rcu, u32_delete_key_freepf_rcu);
596                         else
597                                 u32_destroy_key(n->tp, n, true);
598                 }
599         }
600 }
601
602 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
603 {
604         struct tc_u_common *tp_c = tp->data;
605         struct tc_u_hnode __rcu **hn;
606         struct tc_u_hnode *phn;
607
608         WARN_ON(ht->refcnt);
609
610         u32_clear_hnode(tp, ht);
611
612         hn = &tp_c->hlist;
613         for (phn = rtnl_dereference(*hn);
614              phn;
615              hn = &phn->next, phn = rtnl_dereference(*hn)) {
616                 if (phn == ht) {
617                         u32_clear_hw_hnode(tp, ht);
618                         idr_destroy(&ht->handle_idr);
619                         idr_remove_ext(&tp_c->handle_idr, ht->handle);
620                         RCU_INIT_POINTER(*hn, ht->next);
621                         kfree_rcu(ht, rcu);
622                         return 0;
623                 }
624         }
625
626         return -ENOENT;
627 }
628
629 static bool ht_empty(struct tc_u_hnode *ht)
630 {
631         unsigned int h;
632
633         for (h = 0; h <= ht->divisor; h++)
634                 if (rcu_access_pointer(ht->ht[h]))
635                         return false;
636
637         return true;
638 }
639
640 static void u32_destroy(struct tcf_proto *tp)
641 {
642         struct tc_u_common *tp_c = tp->data;
643         struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
644
645         WARN_ON(root_ht == NULL);
646
647         if (root_ht && --root_ht->refcnt == 0)
648                 u32_destroy_hnode(tp, root_ht);
649
650         if (--tp_c->refcnt == 0) {
651                 struct tc_u_hnode *ht;
652
653                 hlist_del(&tp_c->hnode);
654
655                 for (ht = rtnl_dereference(tp_c->hlist);
656                      ht;
657                      ht = rtnl_dereference(ht->next)) {
658                         ht->refcnt--;
659                         u32_clear_hnode(tp, ht);
660                 }
661
662                 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
663                         RCU_INIT_POINTER(tp_c->hlist, ht->next);
664                         kfree_rcu(ht, rcu);
665                 }
666
667                 idr_destroy(&tp_c->handle_idr);
668                 kfree(tp_c);
669         }
670
671         tp->data = NULL;
672 }
673
674 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last)
675 {
676         struct tc_u_hnode *ht = arg;
677         struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
678         struct tc_u_common *tp_c = tp->data;
679         int ret = 0;
680
681         if (ht == NULL)
682                 goto out;
683
684         if (TC_U32_KEY(ht->handle)) {
685                 u32_remove_hw_knode(tp, ht->handle);
686                 ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
687                 goto out;
688         }
689
690         if (root_ht == ht)
691                 return -EINVAL;
692
693         if (ht->refcnt == 1) {
694                 ht->refcnt--;
695                 u32_destroy_hnode(tp, ht);
696         } else {
697                 return -EBUSY;
698         }
699
700 out:
701         *last = true;
702         if (root_ht) {
703                 if (root_ht->refcnt > 1) {
704                         *last = false;
705                         goto ret;
706                 }
707                 if (root_ht->refcnt == 1) {
708                         if (!ht_empty(root_ht)) {
709                                 *last = false;
710                                 goto ret;
711                         }
712                 }
713         }
714
715         if (tp_c->refcnt > 1) {
716                 *last = false;
717                 goto ret;
718         }
719
720         if (tp_c->refcnt == 1) {
721                 struct tc_u_hnode *ht;
722
723                 for (ht = rtnl_dereference(tp_c->hlist);
724                      ht;
725                      ht = rtnl_dereference(ht->next))
726                         if (!ht_empty(ht)) {
727                                 *last = false;
728                                 break;
729                         }
730         }
731
732 ret:
733         return ret;
734 }
735
736 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
737 {
738         unsigned long idr_index;
739         u32 start = htid | 0x800;
740         u32 max = htid | 0xFFF;
741         u32 min = htid;
742
743         if (idr_alloc_ext(&ht->handle_idr, NULL, &idr_index,
744                           start, max + 1, GFP_KERNEL)) {
745                 if (idr_alloc_ext(&ht->handle_idr, NULL, &idr_index,
746                                   min + 1, max + 1, GFP_KERNEL))
747                         return max;
748         }
749
750         return (u32)idr_index;
751 }
752
753 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
754         [TCA_U32_CLASSID]       = { .type = NLA_U32 },
755         [TCA_U32_HASH]          = { .type = NLA_U32 },
756         [TCA_U32_LINK]          = { .type = NLA_U32 },
757         [TCA_U32_DIVISOR]       = { .type = NLA_U32 },
758         [TCA_U32_SEL]           = { .len = sizeof(struct tc_u32_sel) },
759         [TCA_U32_INDEV]         = { .type = NLA_STRING, .len = IFNAMSIZ },
760         [TCA_U32_MARK]          = { .len = sizeof(struct tc_u32_mark) },
761         [TCA_U32_FLAGS]         = { .type = NLA_U32 },
762 };
763
764 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
765                          unsigned long base, struct tc_u_hnode *ht,
766                          struct tc_u_knode *n, struct nlattr **tb,
767                          struct nlattr *est, bool ovr)
768 {
769         int err;
770
771         err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr);
772         if (err < 0)
773                 return err;
774
775         if (tb[TCA_U32_LINK]) {
776                 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
777                 struct tc_u_hnode *ht_down = NULL, *ht_old;
778
779                 if (TC_U32_KEY(handle))
780                         return -EINVAL;
781
782                 if (handle) {
783                         ht_down = u32_lookup_ht(ht->tp_c, handle);
784
785                         if (ht_down == NULL)
786                                 return -EINVAL;
787                         ht_down->refcnt++;
788                 }
789
790                 ht_old = rtnl_dereference(n->ht_down);
791                 rcu_assign_pointer(n->ht_down, ht_down);
792
793                 if (ht_old)
794                         ht_old->refcnt--;
795         }
796         if (tb[TCA_U32_CLASSID]) {
797                 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
798                 tcf_bind_filter(tp, &n->res, base);
799         }
800
801 #ifdef CONFIG_NET_CLS_IND
802         if (tb[TCA_U32_INDEV]) {
803                 int ret;
804                 ret = tcf_change_indev(net, tb[TCA_U32_INDEV]);
805                 if (ret < 0)
806                         return -EINVAL;
807                 n->ifindex = ret;
808         }
809 #endif
810         return 0;
811 }
812
813 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
814                               struct tc_u_knode *n)
815 {
816         struct tc_u_knode __rcu **ins;
817         struct tc_u_knode *pins;
818         struct tc_u_hnode *ht;
819
820         if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
821                 ht = rtnl_dereference(tp->root);
822         else
823                 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
824
825         ins = &ht->ht[TC_U32_HASH(n->handle)];
826
827         /* The node must always exist for it to be replaced if this is not the
828          * case then something went very wrong elsewhere.
829          */
830         for (pins = rtnl_dereference(*ins); ;
831              ins = &pins->next, pins = rtnl_dereference(*ins))
832                 if (pins->handle == n->handle)
833                         break;
834
835         idr_replace_ext(&ht->handle_idr, n, n->handle);
836         RCU_INIT_POINTER(n->next, pins->next);
837         rcu_assign_pointer(*ins, n);
838 }
839
840 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp,
841                                          struct tc_u_knode *n)
842 {
843         struct tc_u_knode *new;
844         struct tc_u32_sel *s = &n->sel;
845
846         new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key),
847                       GFP_KERNEL);
848
849         if (!new)
850                 return NULL;
851
852         RCU_INIT_POINTER(new->next, n->next);
853         new->handle = n->handle;
854         RCU_INIT_POINTER(new->ht_up, n->ht_up);
855
856 #ifdef CONFIG_NET_CLS_IND
857         new->ifindex = n->ifindex;
858 #endif
859         new->fshift = n->fshift;
860         new->res = n->res;
861         new->flags = n->flags;
862         RCU_INIT_POINTER(new->ht_down, n->ht_down);
863
864         /* bump reference count as long as we hold pointer to structure */
865         if (new->ht_down)
866                 new->ht_down->refcnt++;
867
868 #ifdef CONFIG_CLS_U32_PERF
869         /* Statistics may be incremented by readers during update
870          * so we must keep them in tact. When the node is later destroyed
871          * a special destroy call must be made to not free the pf memory.
872          */
873         new->pf = n->pf;
874 #endif
875
876 #ifdef CONFIG_CLS_U32_MARK
877         new->val = n->val;
878         new->mask = n->mask;
879         /* Similarly success statistics must be moved as pointers */
880         new->pcpu_success = n->pcpu_success;
881 #endif
882         new->tp = tp;
883         memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
884
885         if (tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE)) {
886                 kfree(new);
887                 return NULL;
888         }
889
890         return new;
891 }
892
893 static int u32_change(struct net *net, struct sk_buff *in_skb,
894                       struct tcf_proto *tp, unsigned long base, u32 handle,
895                       struct nlattr **tca, void **arg, bool ovr)
896 {
897         struct tc_u_common *tp_c = tp->data;
898         struct tc_u_hnode *ht;
899         struct tc_u_knode *n;
900         struct tc_u32_sel *s;
901         struct nlattr *opt = tca[TCA_OPTIONS];
902         struct nlattr *tb[TCA_U32_MAX + 1];
903         u32 htid, flags = 0;
904         int err;
905 #ifdef CONFIG_CLS_U32_PERF
906         size_t size;
907 #endif
908
909         if (opt == NULL)
910                 return handle ? -EINVAL : 0;
911
912         err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy, NULL);
913         if (err < 0)
914                 return err;
915
916         if (tb[TCA_U32_FLAGS]) {
917                 flags = nla_get_u32(tb[TCA_U32_FLAGS]);
918                 if (!tc_flags_valid(flags))
919                         return -EINVAL;
920         }
921
922         n = *arg;
923         if (n) {
924                 struct tc_u_knode *new;
925
926                 if (TC_U32_KEY(n->handle) == 0)
927                         return -EINVAL;
928
929                 if (n->flags != flags)
930                         return -EINVAL;
931
932                 new = u32_init_knode(tp, n);
933                 if (!new)
934                         return -ENOMEM;
935
936                 err = u32_set_parms(net, tp, base,
937                                     rtnl_dereference(n->ht_up), new, tb,
938                                     tca[TCA_RATE], ovr);
939
940                 if (err) {
941                         u32_destroy_key(tp, new, false);
942                         return err;
943                 }
944
945                 err = u32_replace_hw_knode(tp, new, flags);
946                 if (err) {
947                         u32_destroy_key(tp, new, false);
948                         return err;
949                 }
950
951                 if (!tc_in_hw(new->flags))
952                         new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
953
954                 u32_replace_knode(tp, tp_c, new);
955                 tcf_unbind_filter(tp, &n->res);
956                 tcf_exts_get_net(&n->exts);
957                 call_rcu(&n->rcu, u32_delete_key_rcu);
958                 return 0;
959         }
960
961         if (tb[TCA_U32_DIVISOR]) {
962                 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
963
964                 if (--divisor > 0x100)
965                         return -EINVAL;
966                 if (TC_U32_KEY(handle))
967                         return -EINVAL;
968                 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
969                 if (ht == NULL)
970                         return -ENOBUFS;
971                 if (handle == 0) {
972                         handle = gen_new_htid(tp->data, ht);
973                         if (handle == 0) {
974                                 kfree(ht);
975                                 return -ENOMEM;
976                         }
977                 } else {
978                         err = idr_alloc_ext(&tp_c->handle_idr, ht, NULL,
979                                             handle, handle + 1, GFP_KERNEL);
980                         if (err) {
981                                 kfree(ht);
982                                 return err;
983                         }
984                 }
985                 ht->tp_c = tp_c;
986                 ht->refcnt = 1;
987                 ht->divisor = divisor;
988                 ht->handle = handle;
989                 ht->prio = tp->prio;
990                 idr_init(&ht->handle_idr);
991
992                 err = u32_replace_hw_hnode(tp, ht, flags);
993                 if (err) {
994                         idr_remove_ext(&tp_c->handle_idr, handle);
995                         kfree(ht);
996                         return err;
997                 }
998
999                 RCU_INIT_POINTER(ht->next, tp_c->hlist);
1000                 rcu_assign_pointer(tp_c->hlist, ht);
1001                 *arg = ht;
1002
1003                 return 0;
1004         }
1005
1006         if (tb[TCA_U32_HASH]) {
1007                 htid = nla_get_u32(tb[TCA_U32_HASH]);
1008                 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
1009                         ht = rtnl_dereference(tp->root);
1010                         htid = ht->handle;
1011                 } else {
1012                         ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
1013                         if (ht == NULL)
1014                                 return -EINVAL;
1015                 }
1016         } else {
1017                 ht = rtnl_dereference(tp->root);
1018                 htid = ht->handle;
1019         }
1020
1021         if (ht->divisor < TC_U32_HASH(htid))
1022                 return -EINVAL;
1023
1024         if (handle) {
1025                 if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid))
1026                         return -EINVAL;
1027                 handle = htid | TC_U32_NODE(handle);
1028                 err = idr_alloc_ext(&ht->handle_idr, NULL, NULL,
1029                                     handle, handle + 1,
1030                                     GFP_KERNEL);
1031                 if (err)
1032                         return err;
1033         } else
1034                 handle = gen_new_kid(ht, htid);
1035
1036         if (tb[TCA_U32_SEL] == NULL) {
1037                 err = -EINVAL;
1038                 goto erridr;
1039         }
1040
1041         s = nla_data(tb[TCA_U32_SEL]);
1042
1043         n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL);
1044         if (n == NULL) {
1045                 err = -ENOBUFS;
1046                 goto erridr;
1047         }
1048
1049 #ifdef CONFIG_CLS_U32_PERF
1050         size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64);
1051         n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt));
1052         if (!n->pf) {
1053                 err = -ENOBUFS;
1054                 goto errfree;
1055         }
1056 #endif
1057
1058         memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
1059         RCU_INIT_POINTER(n->ht_up, ht);
1060         n->handle = handle;
1061         n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1062         n->flags = flags;
1063         n->tp = tp;
1064
1065         err = tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE);
1066         if (err < 0)
1067                 goto errout;
1068
1069 #ifdef CONFIG_CLS_U32_MARK
1070         n->pcpu_success = alloc_percpu(u32);
1071         if (!n->pcpu_success) {
1072                 err = -ENOMEM;
1073                 goto errout;
1074         }
1075
1076         if (tb[TCA_U32_MARK]) {
1077                 struct tc_u32_mark *mark;
1078
1079                 mark = nla_data(tb[TCA_U32_MARK]);
1080                 n->val = mark->val;
1081                 n->mask = mark->mask;
1082         }
1083 #endif
1084
1085         err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr);
1086         if (err == 0) {
1087                 struct tc_u_knode __rcu **ins;
1088                 struct tc_u_knode *pins;
1089
1090                 err = u32_replace_hw_knode(tp, n, flags);
1091                 if (err)
1092                         goto errhw;
1093
1094                 if (!tc_in_hw(n->flags))
1095                         n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1096
1097                 ins = &ht->ht[TC_U32_HASH(handle)];
1098                 for (pins = rtnl_dereference(*ins); pins;
1099                      ins = &pins->next, pins = rtnl_dereference(*ins))
1100                         if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1101                                 break;
1102
1103                 RCU_INIT_POINTER(n->next, pins);
1104                 rcu_assign_pointer(*ins, n);
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_ext(&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 void u32_bind_class(void *fh, u32 classid, unsigned long cl)
1167 {
1168         struct tc_u_knode *n = fh;
1169
1170         if (n && n->res.classid == classid)
1171                 n->res.class = cl;
1172 }
1173
1174 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1175                     struct sk_buff *skb, struct tcmsg *t)
1176 {
1177         struct tc_u_knode *n = fh;
1178         struct tc_u_hnode *ht_up, *ht_down;
1179         struct nlattr *nest;
1180
1181         if (n == NULL)
1182                 return skb->len;
1183
1184         t->tcm_handle = n->handle;
1185
1186         nest = nla_nest_start(skb, TCA_OPTIONS);
1187         if (nest == NULL)
1188                 goto nla_put_failure;
1189
1190         if (TC_U32_KEY(n->handle) == 0) {
1191                 struct tc_u_hnode *ht = fh;
1192                 u32 divisor = ht->divisor + 1;
1193
1194                 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1195                         goto nla_put_failure;
1196         } else {
1197 #ifdef CONFIG_CLS_U32_PERF
1198                 struct tc_u32_pcnt *gpf;
1199                 int cpu;
1200 #endif
1201
1202                 if (nla_put(skb, TCA_U32_SEL,
1203                             sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
1204                             &n->sel))
1205                         goto nla_put_failure;
1206
1207                 ht_up = rtnl_dereference(n->ht_up);
1208                 if (ht_up) {
1209                         u32 htid = n->handle & 0xFFFFF000;
1210                         if (nla_put_u32(skb, TCA_U32_HASH, htid))
1211                                 goto nla_put_failure;
1212                 }
1213                 if (n->res.classid &&
1214                     nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1215                         goto nla_put_failure;
1216
1217                 ht_down = rtnl_dereference(n->ht_down);
1218                 if (ht_down &&
1219                     nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1220                         goto nla_put_failure;
1221
1222                 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1223                         goto nla_put_failure;
1224
1225 #ifdef CONFIG_CLS_U32_MARK
1226                 if ((n->val || n->mask)) {
1227                         struct tc_u32_mark mark = {.val = n->val,
1228                                                    .mask = n->mask,
1229                                                    .success = 0};
1230                         int cpum;
1231
1232                         for_each_possible_cpu(cpum) {
1233                                 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1234
1235                                 mark.success += cnt;
1236                         }
1237
1238                         if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1239                                 goto nla_put_failure;
1240                 }
1241 #endif
1242
1243                 if (tcf_exts_dump(skb, &n->exts) < 0)
1244                         goto nla_put_failure;
1245
1246 #ifdef CONFIG_NET_CLS_IND
1247                 if (n->ifindex) {
1248                         struct net_device *dev;
1249                         dev = __dev_get_by_index(net, n->ifindex);
1250                         if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1251                                 goto nla_put_failure;
1252                 }
1253 #endif
1254 #ifdef CONFIG_CLS_U32_PERF
1255                 gpf = kzalloc(sizeof(struct tc_u32_pcnt) +
1256                               n->sel.nkeys * sizeof(u64),
1257                               GFP_KERNEL);
1258                 if (!gpf)
1259                         goto nla_put_failure;
1260
1261                 for_each_possible_cpu(cpu) {
1262                         int i;
1263                         struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1264
1265                         gpf->rcnt += pf->rcnt;
1266                         gpf->rhit += pf->rhit;
1267                         for (i = 0; i < n->sel.nkeys; i++)
1268                                 gpf->kcnts[i] += pf->kcnts[i];
1269                 }
1270
1271                 if (nla_put_64bit(skb, TCA_U32_PCNT,
1272                                   sizeof(struct tc_u32_pcnt) +
1273                                   n->sel.nkeys * sizeof(u64),
1274                                   gpf, TCA_U32_PAD)) {
1275                         kfree(gpf);
1276                         goto nla_put_failure;
1277                 }
1278                 kfree(gpf);
1279 #endif
1280         }
1281
1282         nla_nest_end(skb, nest);
1283
1284         if (TC_U32_KEY(n->handle))
1285                 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1286                         goto nla_put_failure;
1287         return skb->len;
1288
1289 nla_put_failure:
1290         nla_nest_cancel(skb, nest);
1291         return -1;
1292 }
1293
1294 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1295         .kind           =       "u32",
1296         .classify       =       u32_classify,
1297         .init           =       u32_init,
1298         .destroy        =       u32_destroy,
1299         .get            =       u32_get,
1300         .change         =       u32_change,
1301         .delete         =       u32_delete,
1302         .walk           =       u32_walk,
1303         .dump           =       u32_dump,
1304         .bind_class     =       u32_bind_class,
1305         .owner          =       THIS_MODULE,
1306 };
1307
1308 static int __init init_u32(void)
1309 {
1310         int i, ret;
1311
1312         pr_info("u32 classifier\n");
1313 #ifdef CONFIG_CLS_U32_PERF
1314         pr_info("    Performance counters on\n");
1315 #endif
1316 #ifdef CONFIG_NET_CLS_IND
1317         pr_info("    input device check on\n");
1318 #endif
1319 #ifdef CONFIG_NET_CLS_ACT
1320         pr_info("    Actions configured\n");
1321 #endif
1322         tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1323                                           sizeof(struct hlist_head),
1324                                           GFP_KERNEL);
1325         if (!tc_u_common_hash)
1326                 return -ENOMEM;
1327
1328         for (i = 0; i < U32_HASH_SIZE; i++)
1329                 INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1330
1331         ret = register_tcf_proto_ops(&cls_u32_ops);
1332         if (ret)
1333                 kvfree(tc_u_common_hash);
1334         return ret;
1335 }
1336
1337 static void __exit exit_u32(void)
1338 {
1339         unregister_tcf_proto_ops(&cls_u32_ops);
1340         kvfree(tc_u_common_hash);
1341 }
1342
1343 module_init(init_u32)
1344 module_exit(exit_u32)
1345 MODULE_LICENSE("GPL");