GNU Linux-libre 4.14.328-gnu1
[releases.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/netdevice.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 rcu_head         rcu;
89         /* The 'ht' field MUST be the last field in structure to allow for
90          * more entries allocated at end of structure.
91          */
92         struct tc_u_knode __rcu *ht[1];
93 };
94
95 struct tc_u_common {
96         struct tc_u_hnode __rcu *hlist;
97         struct Qdisc            *q;
98         int                     refcnt;
99         u32                     hgenerator;
100         struct hlist_node       hnode;
101         struct rcu_head         rcu;
102 };
103
104 static inline unsigned int u32_hash_fold(__be32 key,
105                                          const struct tc_u32_sel *sel,
106                                          u8 fshift)
107 {
108         unsigned int h = ntohl(key & sel->hmask) >> fshift;
109
110         return h;
111 }
112
113 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp,
114                         struct tcf_result *res)
115 {
116         struct {
117                 struct tc_u_knode *knode;
118                 unsigned int      off;
119         } stack[TC_U32_MAXDEPTH];
120
121         struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
122         unsigned int off = skb_network_offset(skb);
123         struct tc_u_knode *n;
124         int sdepth = 0;
125         int off2 = 0;
126         int sel = 0;
127 #ifdef CONFIG_CLS_U32_PERF
128         int j;
129 #endif
130         int i, r;
131
132 next_ht:
133         n = rcu_dereference_bh(ht->ht[sel]);
134
135 next_knode:
136         if (n) {
137                 struct tc_u32_key *key = n->sel.keys;
138
139 #ifdef CONFIG_CLS_U32_PERF
140                 __this_cpu_inc(n->pf->rcnt);
141                 j = 0;
142 #endif
143
144                 if (tc_skip_sw(n->flags)) {
145                         n = rcu_dereference_bh(n->next);
146                         goto next_knode;
147                 }
148
149 #ifdef CONFIG_CLS_U32_MARK
150                 if ((skb->mark & n->mask) != n->val) {
151                         n = rcu_dereference_bh(n->next);
152                         goto next_knode;
153                 } else {
154                         __this_cpu_inc(*n->pcpu_success);
155                 }
156 #endif
157
158                 for (i = n->sel.nkeys; i > 0; i--, key++) {
159                         int toff = off + key->off + (off2 & key->offmask);
160                         __be32 *data, hdata;
161
162                         if (skb_headroom(skb) + toff > INT_MAX)
163                                 goto out;
164
165                         data = skb_header_pointer(skb, toff, 4, &hdata);
166                         if (!data)
167                                 goto out;
168                         if ((*data ^ key->val) & key->mask) {
169                                 n = rcu_dereference_bh(n->next);
170                                 goto next_knode;
171                         }
172 #ifdef CONFIG_CLS_U32_PERF
173                         __this_cpu_inc(n->pf->kcnts[j]);
174                         j++;
175 #endif
176                 }
177
178                 ht = rcu_dereference_bh(n->ht_down);
179                 if (!ht) {
180 check_terminal:
181                         if (n->sel.flags & TC_U32_TERMINAL) {
182
183                                 *res = n->res;
184 #ifdef CONFIG_NET_CLS_IND
185                                 if (!tcf_match_indev(skb, n->ifindex)) {
186                                         n = rcu_dereference_bh(n->next);
187                                         goto next_knode;
188                                 }
189 #endif
190 #ifdef CONFIG_CLS_U32_PERF
191                                 __this_cpu_inc(n->pf->rhit);
192 #endif
193                                 r = tcf_exts_exec(skb, &n->exts, res);
194                                 if (r < 0) {
195                                         n = rcu_dereference_bh(n->next);
196                                         goto next_knode;
197                                 }
198
199                                 return r;
200                         }
201                         n = rcu_dereference_bh(n->next);
202                         goto next_knode;
203                 }
204
205                 /* PUSH */
206                 if (sdepth >= TC_U32_MAXDEPTH)
207                         goto deadloop;
208                 stack[sdepth].knode = n;
209                 stack[sdepth].off = off;
210                 sdepth++;
211
212                 ht = rcu_dereference_bh(n->ht_down);
213                 sel = 0;
214                 if (ht->divisor) {
215                         __be32 *data, hdata;
216
217                         data = skb_header_pointer(skb, off + n->sel.hoff, 4,
218                                                   &hdata);
219                         if (!data)
220                                 goto out;
221                         sel = ht->divisor & u32_hash_fold(*data, &n->sel,
222                                                           n->fshift);
223                 }
224                 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
225                         goto next_ht;
226
227                 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
228                         off2 = n->sel.off + 3;
229                         if (n->sel.flags & TC_U32_VAROFFSET) {
230                                 __be16 *data, hdata;
231
232                                 data = skb_header_pointer(skb,
233                                                           off + n->sel.offoff,
234                                                           2, &hdata);
235                                 if (!data)
236                                         goto out;
237                                 off2 += ntohs(n->sel.offmask & *data) >>
238                                         n->sel.offshift;
239                         }
240                         off2 &= ~3;
241                 }
242                 if (n->sel.flags & TC_U32_EAT) {
243                         off += off2;
244                         off2 = 0;
245                 }
246
247                 if (off < skb->len)
248                         goto next_ht;
249         }
250
251         /* POP */
252         if (sdepth--) {
253                 n = stack[sdepth].knode;
254                 ht = rcu_dereference_bh(n->ht_up);
255                 off = stack[sdepth].off;
256                 goto check_terminal;
257         }
258 out:
259         return -1;
260
261 deadloop:
262         net_warn_ratelimited("cls_u32: dead loop\n");
263         return -1;
264 }
265
266 static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
267 {
268         struct tc_u_hnode *ht;
269
270         for (ht = rtnl_dereference(tp_c->hlist);
271              ht;
272              ht = rtnl_dereference(ht->next))
273                 if (ht->handle == handle)
274                         break;
275
276         return ht;
277 }
278
279 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
280 {
281         unsigned int sel;
282         struct tc_u_knode *n = NULL;
283
284         sel = TC_U32_HASH(handle);
285         if (sel > ht->divisor)
286                 goto out;
287
288         for (n = rtnl_dereference(ht->ht[sel]);
289              n;
290              n = rtnl_dereference(n->next))
291                 if (n->handle == handle)
292                         break;
293 out:
294         return n;
295 }
296
297
298 static void *u32_get(struct tcf_proto *tp, u32 handle)
299 {
300         struct tc_u_hnode *ht;
301         struct tc_u_common *tp_c = tp->data;
302
303         if (TC_U32_HTID(handle) == TC_U32_ROOT)
304                 ht = rtnl_dereference(tp->root);
305         else
306                 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
307
308         if (!ht)
309                 return NULL;
310
311         if (TC_U32_KEY(handle) == 0)
312                 return ht;
313
314         return u32_lookup_key(ht, handle);
315 }
316
317 static u32 gen_new_htid(struct tc_u_common *tp_c)
318 {
319         int i = 0x800;
320
321         /* hgenerator only used inside rtnl lock it is safe to increment
322          * without read _copy_ update semantics
323          */
324         do {
325                 if (++tp_c->hgenerator == 0x7FF)
326                         tp_c->hgenerator = 1;
327         } while (--i > 0 && u32_lookup_ht(tp_c, (tp_c->hgenerator|0x800)<<20));
328
329         return i > 0 ? (tp_c->hgenerator|0x800)<<20 : 0;
330 }
331
332 static struct hlist_head *tc_u_common_hash;
333
334 #define U32_HASH_SHIFT 10
335 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
336
337 static unsigned int tc_u_hash(const struct tcf_proto *tp)
338 {
339         struct net_device *dev = tp->q->dev_queue->dev;
340         u32 qhandle = tp->q->handle;
341         int ifindex = dev->ifindex;
342
343         return hash_64((u64)ifindex << 32 | qhandle, U32_HASH_SHIFT);
344 }
345
346 static struct tc_u_common *tc_u_common_find(const struct tcf_proto *tp)
347 {
348         struct tc_u_common *tc;
349         unsigned int h;
350
351         h = tc_u_hash(tp);
352         hlist_for_each_entry(tc, &tc_u_common_hash[h], hnode) {
353                 if (tc->q == tp->q)
354                         return tc;
355         }
356         return NULL;
357 }
358
359 static int u32_init(struct tcf_proto *tp)
360 {
361         struct tc_u_hnode *root_ht;
362         struct tc_u_common *tp_c;
363         unsigned int h;
364
365         tp_c = tc_u_common_find(tp);
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) : 0x80000000;
373         root_ht->prio = tp->prio;
374
375         if (tp_c == NULL) {
376                 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
377                 if (tp_c == NULL) {
378                         kfree(root_ht);
379                         return -ENOBUFS;
380                 }
381                 tp_c->q = tp->q;
382                 INIT_HLIST_NODE(&tp_c->hnode);
383
384                 h = tc_u_hash(tp);
385                 hlist_add_head(&tp_c->hnode, &tc_u_common_hash[h]);
386         }
387
388         tp_c->refcnt++;
389         RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
390         rcu_assign_pointer(tp_c->hlist, root_ht);
391         root_ht->tp_c = tp_c;
392
393         rcu_assign_pointer(tp->root, root_ht);
394         tp->data = tp_c;
395         return 0;
396 }
397
398 static void __u32_destroy_key(struct tc_u_knode *n)
399 {
400         struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
401
402         tcf_exts_destroy(&n->exts);
403         if (ht && --ht->refcnt == 0)
404                 kfree(ht);
405         kfree(n);
406 }
407
408 static void u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n,
409                            bool free_pf)
410 {
411         tcf_exts_put_net(&n->exts);
412 #ifdef CONFIG_CLS_U32_PERF
413         if (free_pf)
414                 free_percpu(n->pf);
415 #endif
416 #ifdef CONFIG_CLS_U32_MARK
417         if (free_pf)
418                 free_percpu(n->pcpu_success);
419 #endif
420         __u32_destroy_key(n);
421 }
422
423 /* u32_delete_key_rcu should be called when free'ing a copied
424  * version of a tc_u_knode obtained from u32_init_knode(). When
425  * copies are obtained from u32_init_knode() the statistics are
426  * shared between the old and new copies to allow readers to
427  * continue to update the statistics during the copy. To support
428  * this the u32_delete_key_rcu variant does not free the percpu
429  * statistics.
430  */
431 static void u32_delete_key_work(struct work_struct *work)
432 {
433         struct tc_u_knode *key = container_of(work, struct tc_u_knode, work);
434
435         rtnl_lock();
436         u32_destroy_key(key->tp, key, false);
437         rtnl_unlock();
438 }
439
440 static void u32_delete_key_rcu(struct rcu_head *rcu)
441 {
442         struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
443
444         INIT_WORK(&key->work, u32_delete_key_work);
445         tcf_queue_work(&key->work);
446 }
447
448 /* u32_delete_key_freepf_rcu is the rcu callback variant
449  * that free's the entire structure including the statistics
450  * percpu variables. Only use this if the key is not a copy
451  * returned by u32_init_knode(). See u32_delete_key_rcu()
452  * for the variant that should be used with keys return from
453  * u32_init_knode()
454  */
455 static void u32_delete_key_freepf_work(struct work_struct *work)
456 {
457         struct tc_u_knode *key = container_of(work, struct tc_u_knode, work);
458
459         rtnl_lock();
460         u32_destroy_key(key->tp, key, true);
461         rtnl_unlock();
462 }
463
464 static void u32_delete_key_freepf_rcu(struct rcu_head *rcu)
465 {
466         struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
467
468         INIT_WORK(&key->work, u32_delete_key_freepf_work);
469         tcf_queue_work(&key->work);
470 }
471
472 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
473 {
474         struct tc_u_knode __rcu **kp;
475         struct tc_u_knode *pkp;
476         struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
477
478         if (ht) {
479                 kp = &ht->ht[TC_U32_HASH(key->handle)];
480                 for (pkp = rtnl_dereference(*kp); pkp;
481                      kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
482                         if (pkp == key) {
483                                 RCU_INIT_POINTER(*kp, key->next);
484
485                                 tcf_unbind_filter(tp, &key->res);
486                                 tcf_exts_get_net(&key->exts);
487                                 call_rcu(&key->rcu, u32_delete_key_freepf_rcu);
488                                 return 0;
489                         }
490                 }
491         }
492         WARN_ON(1);
493         return 0;
494 }
495
496 static void u32_remove_hw_knode(struct tcf_proto *tp, u32 handle)
497 {
498         struct net_device *dev = tp->q->dev_queue->dev;
499         struct tc_cls_u32_offload cls_u32 = {};
500
501         if (!tc_should_offload(dev, 0))
502                 return;
503
504         tc_cls_common_offload_init(&cls_u32.common, tp);
505         cls_u32.command = TC_CLSU32_DELETE_KNODE;
506         cls_u32.knode.handle = handle;
507
508         dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_CLSU32, &cls_u32);
509 }
510
511 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
512                                 u32 flags)
513 {
514         struct net_device *dev = tp->q->dev_queue->dev;
515         struct tc_cls_u32_offload cls_u32 = {};
516         int err;
517
518         if (!tc_should_offload(dev, flags))
519                 return tc_skip_sw(flags) ? -EINVAL : 0;
520
521         tc_cls_common_offload_init(&cls_u32.common, tp);
522         cls_u32.command = TC_CLSU32_NEW_HNODE;
523         cls_u32.hnode.divisor = h->divisor;
524         cls_u32.hnode.handle = h->handle;
525         cls_u32.hnode.prio = h->prio;
526
527         err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_CLSU32, &cls_u32);
528         if (tc_skip_sw(flags))
529                 return err;
530
531         return 0;
532 }
533
534 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h)
535 {
536         struct net_device *dev = tp->q->dev_queue->dev;
537         struct tc_cls_u32_offload cls_u32 = {};
538
539         if (!tc_should_offload(dev, 0))
540                 return;
541
542         tc_cls_common_offload_init(&cls_u32.common, tp);
543         cls_u32.command = TC_CLSU32_DELETE_HNODE;
544         cls_u32.hnode.divisor = h->divisor;
545         cls_u32.hnode.handle = h->handle;
546         cls_u32.hnode.prio = h->prio;
547
548         dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_CLSU32, &cls_u32);
549 }
550
551 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
552                                 u32 flags)
553 {
554         struct net_device *dev = tp->q->dev_queue->dev;
555         struct tc_cls_u32_offload cls_u32 = {};
556         int err;
557
558         if (!tc_should_offload(dev, flags))
559                 return tc_skip_sw(flags) ? -EINVAL : 0;
560
561         tc_cls_common_offload_init(&cls_u32.common, tp);
562         cls_u32.command = TC_CLSU32_REPLACE_KNODE;
563         cls_u32.knode.handle = n->handle;
564         cls_u32.knode.fshift = n->fshift;
565 #ifdef CONFIG_CLS_U32_MARK
566         cls_u32.knode.val = n->val;
567         cls_u32.knode.mask = n->mask;
568 #else
569         cls_u32.knode.val = 0;
570         cls_u32.knode.mask = 0;
571 #endif
572         cls_u32.knode.sel = &n->sel;
573         cls_u32.knode.exts = &n->exts;
574         if (n->ht_down)
575                 cls_u32.knode.link_handle = n->ht_down->handle;
576
577         err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_CLSU32, &cls_u32);
578
579         if (!err)
580                 n->flags |= TCA_CLS_FLAGS_IN_HW;
581
582         if (tc_skip_sw(flags))
583                 return err;
584
585         return 0;
586 }
587
588 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
589 {
590         struct tc_u_knode *n;
591         unsigned int h;
592
593         for (h = 0; h <= ht->divisor; h++) {
594                 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
595                         RCU_INIT_POINTER(ht->ht[h],
596                                          rtnl_dereference(n->next));
597                         tcf_unbind_filter(tp, &n->res);
598                         u32_remove_hw_knode(tp, n->handle);
599                         if (tcf_exts_get_net(&n->exts))
600                                 call_rcu(&n->rcu, u32_delete_key_freepf_rcu);
601                         else
602                                 u32_destroy_key(n->tp, n, true);
603                 }
604         }
605 }
606
607 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
608 {
609         struct tc_u_common *tp_c = tp->data;
610         struct tc_u_hnode __rcu **hn;
611         struct tc_u_hnode *phn;
612
613         WARN_ON(ht->refcnt);
614
615         u32_clear_hnode(tp, ht);
616
617         hn = &tp_c->hlist;
618         for (phn = rtnl_dereference(*hn);
619              phn;
620              hn = &phn->next, phn = rtnl_dereference(*hn)) {
621                 if (phn == ht) {
622                         u32_clear_hw_hnode(tp, ht);
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 bool ht_empty(struct tc_u_hnode *ht)
633 {
634         unsigned int h;
635
636         for (h = 0; h <= ht->divisor; h++)
637                 if (rcu_access_pointer(ht->ht[h]))
638                         return false;
639
640         return true;
641 }
642
643 static void u32_destroy(struct tcf_proto *tp)
644 {
645         struct tc_u_common *tp_c = tp->data;
646         struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
647
648         WARN_ON(root_ht == NULL);
649
650         if (root_ht && --root_ht->refcnt == 0)
651                 u32_destroy_hnode(tp, root_ht);
652
653         if (--tp_c->refcnt == 0) {
654                 struct tc_u_hnode *ht;
655
656                 hlist_del(&tp_c->hnode);
657
658                 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
659                         u32_clear_hnode(tp, ht);
660                         RCU_INIT_POINTER(tp_c->hlist, ht->next);
661
662                         /* u32_destroy_key() will later free ht for us, if it's
663                          * still referenced by some knode
664                          */
665                         if (--ht->refcnt == 0)
666                                 kfree_rcu(ht, rcu);
667                 }
668
669                 kfree(tp_c);
670         }
671
672         tp->data = NULL;
673 }
674
675 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last)
676 {
677         struct tc_u_hnode *ht = arg;
678         struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
679         struct tc_u_common *tp_c = tp->data;
680         int ret = 0;
681
682         if (ht == NULL)
683                 goto out;
684
685         if (TC_U32_KEY(ht->handle)) {
686                 u32_remove_hw_knode(tp, ht->handle);
687                 ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
688                 goto out;
689         }
690
691         if (root_ht == ht)
692                 return -EINVAL;
693
694         if (ht->refcnt == 1) {
695                 ht->refcnt--;
696                 u32_destroy_hnode(tp, ht);
697         } else {
698                 return -EBUSY;
699         }
700
701 out:
702         *last = true;
703         if (root_ht) {
704                 if (root_ht->refcnt > 1) {
705                         *last = false;
706                         goto ret;
707                 }
708                 if (root_ht->refcnt == 1) {
709                         if (!ht_empty(root_ht)) {
710                                 *last = false;
711                                 goto ret;
712                         }
713                 }
714         }
715
716         if (tp_c->refcnt > 1) {
717                 *last = false;
718                 goto ret;
719         }
720
721         if (tp_c->refcnt == 1) {
722                 struct tc_u_hnode *ht;
723
724                 for (ht = rtnl_dereference(tp_c->hlist);
725                      ht;
726                      ht = rtnl_dereference(ht->next))
727                         if (!ht_empty(ht)) {
728                                 *last = false;
729                                 break;
730                         }
731         }
732
733 ret:
734         return ret;
735 }
736
737 #define NR_U32_NODE (1<<12)
738 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 handle)
739 {
740         struct tc_u_knode *n;
741         unsigned long i;
742         unsigned long *bitmap = kzalloc(BITS_TO_LONGS(NR_U32_NODE) * sizeof(unsigned long),
743                                         GFP_KERNEL);
744         if (!bitmap)
745                 return handle | 0xFFF;
746
747         for (n = rtnl_dereference(ht->ht[TC_U32_HASH(handle)]);
748              n;
749              n = rtnl_dereference(n->next))
750                 set_bit(TC_U32_NODE(n->handle), bitmap);
751
752         i = find_next_zero_bit(bitmap, NR_U32_NODE, 0x800);
753         if (i >= NR_U32_NODE)
754                 i = find_next_zero_bit(bitmap, NR_U32_NODE, 1);
755
756         kfree(bitmap);
757         return handle | (i >= NR_U32_NODE ? 0xFFF : i);
758 }
759
760 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
761         [TCA_U32_CLASSID]       = { .type = NLA_U32 },
762         [TCA_U32_HASH]          = { .type = NLA_U32 },
763         [TCA_U32_LINK]          = { .type = NLA_U32 },
764         [TCA_U32_DIVISOR]       = { .type = NLA_U32 },
765         [TCA_U32_SEL]           = { .len = sizeof(struct tc_u32_sel) },
766         [TCA_U32_INDEV]         = { .type = NLA_STRING, .len = IFNAMSIZ },
767         [TCA_U32_MARK]          = { .len = sizeof(struct tc_u32_mark) },
768         [TCA_U32_FLAGS]         = { .type = NLA_U32 },
769 };
770
771 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
772                          unsigned long base, struct tc_u_hnode *ht,
773                          struct tc_u_knode *n, struct nlattr **tb,
774                          struct nlattr *est, bool ovr)
775 {
776         int err;
777 #ifdef CONFIG_NET_CLS_IND
778         int ifindex = -1;
779 #endif
780
781         err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr);
782         if (err < 0)
783                 return err;
784
785 #ifdef CONFIG_NET_CLS_IND
786         if (tb[TCA_U32_INDEV]) {
787                 ifindex = tcf_change_indev(net, tb[TCA_U32_INDEV]);
788                 if (ifindex < 0)
789                         return -EINVAL;
790         }
791 #endif
792
793         if (tb[TCA_U32_LINK]) {
794                 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
795                 struct tc_u_hnode *ht_down = NULL, *ht_old;
796
797                 if (TC_U32_KEY(handle))
798                         return -EINVAL;
799
800                 if (handle) {
801                         ht_down = u32_lookup_ht(ht->tp_c, handle);
802
803                         if (ht_down == NULL)
804                                 return -EINVAL;
805                         ht_down->refcnt++;
806                 }
807
808                 ht_old = rtnl_dereference(n->ht_down);
809                 rcu_assign_pointer(n->ht_down, ht_down);
810
811                 if (ht_old)
812                         ht_old->refcnt--;
813         }
814         if (tb[TCA_U32_CLASSID]) {
815                 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
816                 tcf_bind_filter(tp, &n->res, base);
817         }
818
819 #ifdef CONFIG_NET_CLS_IND
820         if (ifindex >= 0)
821                 n->ifindex = ifindex;
822 #endif
823
824         return 0;
825 }
826
827 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
828                               struct tc_u_knode *n)
829 {
830         struct tc_u_knode __rcu **ins;
831         struct tc_u_knode *pins;
832         struct tc_u_hnode *ht;
833
834         if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
835                 ht = rtnl_dereference(tp->root);
836         else
837                 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
838
839         ins = &ht->ht[TC_U32_HASH(n->handle)];
840
841         /* The node must always exist for it to be replaced if this is not the
842          * case then something went very wrong elsewhere.
843          */
844         for (pins = rtnl_dereference(*ins); ;
845              ins = &pins->next, pins = rtnl_dereference(*ins))
846                 if (pins->handle == n->handle)
847                         break;
848
849         RCU_INIT_POINTER(n->next, pins->next);
850         rcu_assign_pointer(*ins, n);
851 }
852
853 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp,
854                                          struct tc_u_knode *n)
855 {
856         struct tc_u_knode *new;
857         struct tc_u32_sel *s = &n->sel;
858
859         new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key),
860                       GFP_KERNEL);
861
862         if (!new)
863                 return NULL;
864
865         RCU_INIT_POINTER(new->next, n->next);
866         new->handle = n->handle;
867         RCU_INIT_POINTER(new->ht_up, n->ht_up);
868
869 #ifdef CONFIG_NET_CLS_IND
870         new->ifindex = n->ifindex;
871 #endif
872         new->fshift = n->fshift;
873         new->flags = n->flags;
874         RCU_INIT_POINTER(new->ht_down, n->ht_down);
875
876         /* bump reference count as long as we hold pointer to structure */
877         if (new->ht_down)
878                 new->ht_down->refcnt++;
879
880 #ifdef CONFIG_CLS_U32_PERF
881         /* Statistics may be incremented by readers during update
882          * so we must keep them in tact. When the node is later destroyed
883          * a special destroy call must be made to not free the pf memory.
884          */
885         new->pf = n->pf;
886 #endif
887
888 #ifdef CONFIG_CLS_U32_MARK
889         new->val = n->val;
890         new->mask = n->mask;
891         /* Similarly success statistics must be moved as pointers */
892         new->pcpu_success = n->pcpu_success;
893 #endif
894         new->tp = tp;
895         memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
896
897         if (tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE)) {
898                 kfree(new);
899                 return NULL;
900         }
901
902         return new;
903 }
904
905 static int u32_change(struct net *net, struct sk_buff *in_skb,
906                       struct tcf_proto *tp, unsigned long base, u32 handle,
907                       struct nlattr **tca, void **arg, bool ovr)
908 {
909         struct tc_u_common *tp_c = tp->data;
910         struct tc_u_hnode *ht;
911         struct tc_u_knode *n;
912         struct tc_u32_sel *s;
913         struct nlattr *opt = tca[TCA_OPTIONS];
914         struct nlattr *tb[TCA_U32_MAX + 1];
915         u32 htid, flags = 0;
916         size_t sel_size;
917         int err;
918 #ifdef CONFIG_CLS_U32_PERF
919         size_t size;
920 #endif
921
922         if (opt == NULL)
923                 return handle ? -EINVAL : 0;
924
925         err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy, NULL);
926         if (err < 0)
927                 return err;
928
929         if (tb[TCA_U32_FLAGS]) {
930                 flags = nla_get_u32(tb[TCA_U32_FLAGS]);
931                 if (!tc_flags_valid(flags))
932                         return -EINVAL;
933         }
934
935         n = *arg;
936         if (n) {
937                 struct tc_u_knode *new;
938
939                 if (TC_U32_KEY(n->handle) == 0)
940                         return -EINVAL;
941
942                 if ((n->flags ^ flags) &
943                     ~(TCA_CLS_FLAGS_IN_HW | TCA_CLS_FLAGS_NOT_IN_HW))
944                         return -EINVAL;
945
946                 new = u32_init_knode(tp, n);
947                 if (!new)
948                         return -ENOMEM;
949
950                 err = u32_set_parms(net, tp, base,
951                                     rtnl_dereference(n->ht_up), new, tb,
952                                     tca[TCA_RATE], ovr);
953
954                 if (err) {
955                         __u32_destroy_key(new);
956                         return err;
957                 }
958
959                 err = u32_replace_hw_knode(tp, new, flags);
960                 if (err) {
961                         __u32_destroy_key(new);
962                         return err;
963                 }
964
965                 if (!tc_in_hw(new->flags))
966                         new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
967
968                 u32_replace_knode(tp, tp_c, new);
969                 tcf_unbind_filter(tp, &n->res);
970                 tcf_exts_get_net(&n->exts);
971                 call_rcu(&n->rcu, u32_delete_key_rcu);
972                 return 0;
973         }
974
975         if (tb[TCA_U32_DIVISOR]) {
976                 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
977
978                 if (--divisor > 0x100)
979                         return -EINVAL;
980                 if (TC_U32_KEY(handle))
981                         return -EINVAL;
982                 if (handle == 0) {
983                         handle = gen_new_htid(tp->data);
984                         if (handle == 0)
985                                 return -ENOMEM;
986                 }
987                 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
988                 if (ht == NULL)
989                         return -ENOBUFS;
990                 ht->tp_c = tp_c;
991                 ht->refcnt = 1;
992                 ht->divisor = divisor;
993                 ht->handle = handle;
994                 ht->prio = tp->prio;
995
996                 err = u32_replace_hw_hnode(tp, ht, flags);
997                 if (err) {
998                         kfree(ht);
999                         return err;
1000                 }
1001
1002                 RCU_INIT_POINTER(ht->next, tp_c->hlist);
1003                 rcu_assign_pointer(tp_c->hlist, ht);
1004                 *arg = ht;
1005
1006                 return 0;
1007         }
1008
1009         if (tb[TCA_U32_HASH]) {
1010                 htid = nla_get_u32(tb[TCA_U32_HASH]);
1011                 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
1012                         ht = rtnl_dereference(tp->root);
1013                         htid = ht->handle;
1014                 } else {
1015                         ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
1016                         if (ht == NULL)
1017                                 return -EINVAL;
1018                 }
1019         } else {
1020                 ht = rtnl_dereference(tp->root);
1021                 htid = ht->handle;
1022         }
1023
1024         if (ht->divisor < TC_U32_HASH(htid))
1025                 return -EINVAL;
1026
1027         if (handle) {
1028                 if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid))
1029                         return -EINVAL;
1030                 handle = htid | TC_U32_NODE(handle);
1031         } else
1032                 handle = gen_new_kid(ht, htid);
1033
1034         if (tb[TCA_U32_SEL] == NULL)
1035                 return -EINVAL;
1036
1037         s = nla_data(tb[TCA_U32_SEL]);
1038         sel_size = sizeof(*s) + sizeof(*s->keys) * s->nkeys;
1039         if (nla_len(tb[TCA_U32_SEL]) < sel_size)
1040                 return -EINVAL;
1041
1042         n = kzalloc(offsetof(typeof(*n), sel) + sel_size, GFP_KERNEL);
1043         if (n == NULL)
1044                 return -ENOBUFS;
1045
1046 #ifdef CONFIG_CLS_U32_PERF
1047         size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64);
1048         n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt));
1049         if (!n->pf) {
1050                 kfree(n);
1051                 return -ENOBUFS;
1052         }
1053 #endif
1054
1055         memcpy(&n->sel, s, sel_size);
1056         RCU_INIT_POINTER(n->ht_up, ht);
1057         n->handle = handle;
1058         n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1059         n->flags = flags;
1060         n->tp = tp;
1061
1062         err = tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE);
1063         if (err < 0)
1064                 goto errout;
1065
1066 #ifdef CONFIG_CLS_U32_MARK
1067         n->pcpu_success = alloc_percpu(u32);
1068         if (!n->pcpu_success) {
1069                 err = -ENOMEM;
1070                 goto errout;
1071         }
1072
1073         if (tb[TCA_U32_MARK]) {
1074                 struct tc_u32_mark *mark;
1075
1076                 mark = nla_data(tb[TCA_U32_MARK]);
1077                 n->val = mark->val;
1078                 n->mask = mark->mask;
1079         }
1080 #endif
1081
1082         err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr);
1083         if (err == 0) {
1084                 struct tc_u_knode __rcu **ins;
1085                 struct tc_u_knode *pins;
1086
1087                 err = u32_replace_hw_knode(tp, n, flags);
1088                 if (err)
1089                         goto errhw;
1090
1091                 if (!tc_in_hw(n->flags))
1092                         n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1093
1094                 ins = &ht->ht[TC_U32_HASH(handle)];
1095                 for (pins = rtnl_dereference(*ins); pins;
1096                      ins = &pins->next, pins = rtnl_dereference(*ins))
1097                         if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1098                                 break;
1099
1100                 RCU_INIT_POINTER(n->next, pins);
1101                 rcu_assign_pointer(*ins, n);
1102                 *arg = n;
1103                 return 0;
1104         }
1105
1106 errhw:
1107 #ifdef CONFIG_CLS_U32_MARK
1108         free_percpu(n->pcpu_success);
1109 #endif
1110
1111 errout:
1112         tcf_exts_destroy(&n->exts);
1113 #ifdef CONFIG_CLS_U32_PERF
1114         free_percpu(n->pf);
1115 #endif
1116         kfree(n);
1117         return err;
1118 }
1119
1120 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg)
1121 {
1122         struct tc_u_common *tp_c = tp->data;
1123         struct tc_u_hnode *ht;
1124         struct tc_u_knode *n;
1125         unsigned int h;
1126
1127         if (arg->stop)
1128                 return;
1129
1130         for (ht = rtnl_dereference(tp_c->hlist);
1131              ht;
1132              ht = rtnl_dereference(ht->next)) {
1133                 if (ht->prio != tp->prio)
1134                         continue;
1135                 if (arg->count >= arg->skip) {
1136                         if (arg->fn(tp, ht, arg) < 0) {
1137                                 arg->stop = 1;
1138                                 return;
1139                         }
1140                 }
1141                 arg->count++;
1142                 for (h = 0; h <= ht->divisor; h++) {
1143                         for (n = rtnl_dereference(ht->ht[h]);
1144                              n;
1145                              n = rtnl_dereference(n->next)) {
1146                                 if (arg->count < arg->skip) {
1147                                         arg->count++;
1148                                         continue;
1149                                 }
1150                                 if (arg->fn(tp, n, arg) < 0) {
1151                                         arg->stop = 1;
1152                                         return;
1153                                 }
1154                                 arg->count++;
1155                         }
1156                 }
1157         }
1158 }
1159
1160 static void u32_bind_class(void *fh, u32 classid, unsigned long cl)
1161 {
1162         struct tc_u_knode *n = fh;
1163
1164         if (n && n->res.classid == classid)
1165                 n->res.class = cl;
1166 }
1167
1168 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1169                     struct sk_buff *skb, struct tcmsg *t)
1170 {
1171         struct tc_u_knode *n = fh;
1172         struct tc_u_hnode *ht_up, *ht_down;
1173         struct nlattr *nest;
1174
1175         if (n == NULL)
1176                 return skb->len;
1177
1178         t->tcm_handle = n->handle;
1179
1180         nest = nla_nest_start(skb, TCA_OPTIONS);
1181         if (nest == NULL)
1182                 goto nla_put_failure;
1183
1184         if (TC_U32_KEY(n->handle) == 0) {
1185                 struct tc_u_hnode *ht = fh;
1186                 u32 divisor = ht->divisor + 1;
1187
1188                 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1189                         goto nla_put_failure;
1190         } else {
1191 #ifdef CONFIG_CLS_U32_PERF
1192                 struct tc_u32_pcnt *gpf;
1193                 int cpu;
1194 #endif
1195
1196                 if (nla_put(skb, TCA_U32_SEL,
1197                             sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
1198                             &n->sel))
1199                         goto nla_put_failure;
1200
1201                 ht_up = rtnl_dereference(n->ht_up);
1202                 if (ht_up) {
1203                         u32 htid = n->handle & 0xFFFFF000;
1204                         if (nla_put_u32(skb, TCA_U32_HASH, htid))
1205                                 goto nla_put_failure;
1206                 }
1207                 if (n->res.classid &&
1208                     nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1209                         goto nla_put_failure;
1210
1211                 ht_down = rtnl_dereference(n->ht_down);
1212                 if (ht_down &&
1213                     nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1214                         goto nla_put_failure;
1215
1216                 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1217                         goto nla_put_failure;
1218
1219 #ifdef CONFIG_CLS_U32_MARK
1220                 if ((n->val || n->mask)) {
1221                         struct tc_u32_mark mark = {.val = n->val,
1222                                                    .mask = n->mask,
1223                                                    .success = 0};
1224                         int cpum;
1225
1226                         for_each_possible_cpu(cpum) {
1227                                 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1228
1229                                 mark.success += cnt;
1230                         }
1231
1232                         if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1233                                 goto nla_put_failure;
1234                 }
1235 #endif
1236
1237                 if (tcf_exts_dump(skb, &n->exts) < 0)
1238                         goto nla_put_failure;
1239
1240 #ifdef CONFIG_NET_CLS_IND
1241                 if (n->ifindex) {
1242                         struct net_device *dev;
1243                         dev = __dev_get_by_index(net, n->ifindex);
1244                         if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1245                                 goto nla_put_failure;
1246                 }
1247 #endif
1248 #ifdef CONFIG_CLS_U32_PERF
1249                 gpf = kzalloc(sizeof(struct tc_u32_pcnt) +
1250                               n->sel.nkeys * sizeof(u64),
1251                               GFP_KERNEL);
1252                 if (!gpf)
1253                         goto nla_put_failure;
1254
1255                 for_each_possible_cpu(cpu) {
1256                         int i;
1257                         struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1258
1259                         gpf->rcnt += pf->rcnt;
1260                         gpf->rhit += pf->rhit;
1261                         for (i = 0; i < n->sel.nkeys; i++)
1262                                 gpf->kcnts[i] += pf->kcnts[i];
1263                 }
1264
1265                 if (nla_put_64bit(skb, TCA_U32_PCNT,
1266                                   sizeof(struct tc_u32_pcnt) +
1267                                   n->sel.nkeys * sizeof(u64),
1268                                   gpf, TCA_U32_PAD)) {
1269                         kfree(gpf);
1270                         goto nla_put_failure;
1271                 }
1272                 kfree(gpf);
1273 #endif
1274         }
1275
1276         nla_nest_end(skb, nest);
1277
1278         if (TC_U32_KEY(n->handle))
1279                 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1280                         goto nla_put_failure;
1281         return skb->len;
1282
1283 nla_put_failure:
1284         nla_nest_cancel(skb, nest);
1285         return -1;
1286 }
1287
1288 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1289         .kind           =       "u32",
1290         .classify       =       u32_classify,
1291         .init           =       u32_init,
1292         .destroy        =       u32_destroy,
1293         .get            =       u32_get,
1294         .change         =       u32_change,
1295         .delete         =       u32_delete,
1296         .walk           =       u32_walk,
1297         .dump           =       u32_dump,
1298         .bind_class     =       u32_bind_class,
1299         .owner          =       THIS_MODULE,
1300 };
1301
1302 static int __init init_u32(void)
1303 {
1304         int i, ret;
1305
1306         pr_info("u32 classifier\n");
1307 #ifdef CONFIG_CLS_U32_PERF
1308         pr_info("    Performance counters on\n");
1309 #endif
1310 #ifdef CONFIG_NET_CLS_IND
1311         pr_info("    input device check on\n");
1312 #endif
1313 #ifdef CONFIG_NET_CLS_ACT
1314         pr_info("    Actions configured\n");
1315 #endif
1316         tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1317                                           sizeof(struct hlist_head),
1318                                           GFP_KERNEL);
1319         if (!tc_u_common_hash)
1320                 return -ENOMEM;
1321
1322         for (i = 0; i < U32_HASH_SIZE; i++)
1323                 INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1324
1325         ret = register_tcf_proto_ops(&cls_u32_ops);
1326         if (ret)
1327                 kvfree(tc_u_common_hash);
1328         return ret;
1329 }
1330
1331 static void __exit exit_u32(void)
1332 {
1333         unregister_tcf_proto_ops(&cls_u32_ops);
1334         kvfree(tc_u_common_hash);
1335 }
1336
1337 module_init(init_u32)
1338 module_exit(exit_u32)
1339 MODULE_LICENSE("GPL");