Merge branch 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[sfrench/cifs-2.6.git] / net / sched / cls_tcindex.c
1 /*
2  * net/sched/cls_tcindex.c      Packet classifier for skb->tc_index
3  *
4  * Written 1998,1999 by Werner Almesberger, EPFL ICA
5  */
6
7 #include <linux/module.h>
8 #include <linux/types.h>
9 #include <linux/kernel.h>
10 #include <linux/skbuff.h>
11 #include <linux/errno.h>
12 #include <linux/slab.h>
13 #include <net/act_api.h>
14 #include <net/netlink.h>
15 #include <net/pkt_cls.h>
16 #include <net/sch_generic.h>
17
18 /*
19  * Passing parameters to the root seems to be done more awkwardly than really
20  * necessary. At least, u32 doesn't seem to use such dirty hacks. To be
21  * verified. FIXME.
22  */
23
24 #define PERFECT_HASH_THRESHOLD  64      /* use perfect hash if not bigger */
25 #define DEFAULT_HASH_SIZE       64      /* optimized for diffserv */
26
27
28 struct tcindex_filter_result {
29         struct tcf_exts         exts;
30         struct tcf_result       res;
31         struct rcu_work         rwork;
32 };
33
34 struct tcindex_filter {
35         u16 key;
36         struct tcindex_filter_result result;
37         struct tcindex_filter __rcu *next;
38         struct rcu_work rwork;
39 };
40
41
42 struct tcindex_data {
43         struct tcindex_filter_result *perfect; /* perfect hash; NULL if none */
44         struct tcindex_filter __rcu **h; /* imperfect hash; */
45         struct tcf_proto *tp;
46         u16 mask;               /* AND key with mask */
47         u32 shift;              /* shift ANDed key to the right */
48         u32 hash;               /* hash table size; 0 if undefined */
49         u32 alloc_hash;         /* allocated size */
50         u32 fall_through;       /* 0: only classify if explicit match */
51         struct rcu_head rcu;
52 };
53
54 static inline int tcindex_filter_is_set(struct tcindex_filter_result *r)
55 {
56         return tcf_exts_has_actions(&r->exts) || r->res.classid;
57 }
58
59 static struct tcindex_filter_result *tcindex_lookup(struct tcindex_data *p,
60                                                     u16 key)
61 {
62         if (p->perfect) {
63                 struct tcindex_filter_result *f = p->perfect + key;
64
65                 return tcindex_filter_is_set(f) ? f : NULL;
66         } else if (p->h) {
67                 struct tcindex_filter __rcu **fp;
68                 struct tcindex_filter *f;
69
70                 fp = &p->h[key % p->hash];
71                 for (f = rcu_dereference_bh_rtnl(*fp);
72                      f;
73                      fp = &f->next, f = rcu_dereference_bh_rtnl(*fp))
74                         if (f->key == key)
75                                 return &f->result;
76         }
77
78         return NULL;
79 }
80
81
82 static int tcindex_classify(struct sk_buff *skb, const struct tcf_proto *tp,
83                             struct tcf_result *res)
84 {
85         struct tcindex_data *p = rcu_dereference_bh(tp->root);
86         struct tcindex_filter_result *f;
87         int key = (skb->tc_index & p->mask) >> p->shift;
88
89         pr_debug("tcindex_classify(skb %p,tp %p,res %p),p %p\n",
90                  skb, tp, res, p);
91
92         f = tcindex_lookup(p, key);
93         if (!f) {
94                 struct Qdisc *q = tcf_block_q(tp->chain->block);
95
96                 if (!p->fall_through)
97                         return -1;
98                 res->classid = TC_H_MAKE(TC_H_MAJ(q->handle), key);
99                 res->class = 0;
100                 pr_debug("alg 0x%x\n", res->classid);
101                 return 0;
102         }
103         *res = f->res;
104         pr_debug("map 0x%x\n", res->classid);
105
106         return tcf_exts_exec(skb, &f->exts, res);
107 }
108
109
110 static void *tcindex_get(struct tcf_proto *tp, u32 handle)
111 {
112         struct tcindex_data *p = rtnl_dereference(tp->root);
113         struct tcindex_filter_result *r;
114
115         pr_debug("tcindex_get(tp %p,handle 0x%08x)\n", tp, handle);
116         if (p->perfect && handle >= p->alloc_hash)
117                 return NULL;
118         r = tcindex_lookup(p, handle);
119         return r && tcindex_filter_is_set(r) ? r : NULL;
120 }
121
122 static int tcindex_init(struct tcf_proto *tp)
123 {
124         struct tcindex_data *p;
125
126         pr_debug("tcindex_init(tp %p)\n", tp);
127         p = kzalloc(sizeof(struct tcindex_data), GFP_KERNEL);
128         if (!p)
129                 return -ENOMEM;
130
131         p->mask = 0xffff;
132         p->hash = DEFAULT_HASH_SIZE;
133         p->fall_through = 1;
134
135         rcu_assign_pointer(tp->root, p);
136         return 0;
137 }
138
139 static void __tcindex_destroy_rexts(struct tcindex_filter_result *r)
140 {
141         tcf_exts_destroy(&r->exts);
142         tcf_exts_put_net(&r->exts);
143 }
144
145 static void tcindex_destroy_rexts_work(struct work_struct *work)
146 {
147         struct tcindex_filter_result *r;
148
149         r = container_of(to_rcu_work(work),
150                          struct tcindex_filter_result,
151                          rwork);
152         rtnl_lock();
153         __tcindex_destroy_rexts(r);
154         rtnl_unlock();
155 }
156
157 static void __tcindex_destroy_fexts(struct tcindex_filter *f)
158 {
159         tcf_exts_destroy(&f->result.exts);
160         tcf_exts_put_net(&f->result.exts);
161         kfree(f);
162 }
163
164 static void tcindex_destroy_fexts_work(struct work_struct *work)
165 {
166         struct tcindex_filter *f = container_of(to_rcu_work(work),
167                                                 struct tcindex_filter,
168                                                 rwork);
169
170         rtnl_lock();
171         __tcindex_destroy_fexts(f);
172         rtnl_unlock();
173 }
174
175 static int tcindex_delete(struct tcf_proto *tp, void *arg, bool *last,
176                           struct netlink_ext_ack *extack)
177 {
178         struct tcindex_data *p = rtnl_dereference(tp->root);
179         struct tcindex_filter_result *r = arg;
180         struct tcindex_filter __rcu **walk;
181         struct tcindex_filter *f = NULL;
182
183         pr_debug("tcindex_delete(tp %p,arg %p),p %p\n", tp, arg, p);
184         if (p->perfect) {
185                 if (!r->res.class)
186                         return -ENOENT;
187         } else {
188                 int i;
189
190                 for (i = 0; i < p->hash; i++) {
191                         walk = p->h + i;
192                         for (f = rtnl_dereference(*walk); f;
193                              walk = &f->next, f = rtnl_dereference(*walk)) {
194                                 if (&f->result == r)
195                                         goto found;
196                         }
197                 }
198                 return -ENOENT;
199
200 found:
201                 rcu_assign_pointer(*walk, rtnl_dereference(f->next));
202         }
203         tcf_unbind_filter(tp, &r->res);
204         /* all classifiers are required to call tcf_exts_destroy() after rcu
205          * grace period, since converted-to-rcu actions are relying on that
206          * in cleanup() callback
207          */
208         if (f) {
209                 if (tcf_exts_get_net(&f->result.exts))
210                         tcf_queue_work(&f->rwork, tcindex_destroy_fexts_work);
211                 else
212                         __tcindex_destroy_fexts(f);
213         } else {
214                 if (tcf_exts_get_net(&r->exts))
215                         tcf_queue_work(&r->rwork, tcindex_destroy_rexts_work);
216                 else
217                         __tcindex_destroy_rexts(r);
218         }
219
220         *last = false;
221         return 0;
222 }
223
224 static int tcindex_destroy_element(struct tcf_proto *tp,
225                                    void *arg, struct tcf_walker *walker)
226 {
227         bool last;
228
229         return tcindex_delete(tp, arg, &last, NULL);
230 }
231
232 static void __tcindex_destroy(struct rcu_head *head)
233 {
234         struct tcindex_data *p = container_of(head, struct tcindex_data, rcu);
235
236         kfree(p->perfect);
237         kfree(p->h);
238         kfree(p);
239 }
240
241 static inline int
242 valid_perfect_hash(struct tcindex_data *p)
243 {
244         return  p->hash > (p->mask >> p->shift);
245 }
246
247 static const struct nla_policy tcindex_policy[TCA_TCINDEX_MAX + 1] = {
248         [TCA_TCINDEX_HASH]              = { .type = NLA_U32 },
249         [TCA_TCINDEX_MASK]              = { .type = NLA_U16 },
250         [TCA_TCINDEX_SHIFT]             = { .type = NLA_U32 },
251         [TCA_TCINDEX_FALL_THROUGH]      = { .type = NLA_U32 },
252         [TCA_TCINDEX_CLASSID]           = { .type = NLA_U32 },
253 };
254
255 static int tcindex_filter_result_init(struct tcindex_filter_result *r)
256 {
257         memset(r, 0, sizeof(*r));
258         return tcf_exts_init(&r->exts, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
259 }
260
261 static void __tcindex_partial_destroy(struct rcu_head *head)
262 {
263         struct tcindex_data *p = container_of(head, struct tcindex_data, rcu);
264
265         kfree(p->perfect);
266         kfree(p);
267 }
268
269 static void tcindex_free_perfect_hash(struct tcindex_data *cp)
270 {
271         int i;
272
273         for (i = 0; i < cp->hash; i++)
274                 tcf_exts_destroy(&cp->perfect[i].exts);
275         kfree(cp->perfect);
276 }
277
278 static int tcindex_alloc_perfect_hash(struct tcindex_data *cp)
279 {
280         int i, err = 0;
281
282         cp->perfect = kcalloc(cp->hash, sizeof(struct tcindex_filter_result),
283                               GFP_KERNEL);
284         if (!cp->perfect)
285                 return -ENOMEM;
286
287         for (i = 0; i < cp->hash; i++) {
288                 err = tcf_exts_init(&cp->perfect[i].exts,
289                                     TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
290                 if (err < 0)
291                         goto errout;
292         }
293
294         return 0;
295
296 errout:
297         tcindex_free_perfect_hash(cp);
298         return err;
299 }
300
301 static int
302 tcindex_set_parms(struct net *net, struct tcf_proto *tp, unsigned long base,
303                   u32 handle, struct tcindex_data *p,
304                   struct tcindex_filter_result *r, struct nlattr **tb,
305                   struct nlattr *est, bool ovr, struct netlink_ext_ack *extack)
306 {
307         struct tcindex_filter_result new_filter_result, *old_r = r;
308         struct tcindex_filter_result cr;
309         struct tcindex_data *cp = NULL, *oldp;
310         struct tcindex_filter *f = NULL; /* make gcc behave */
311         int err, balloc = 0;
312         struct tcf_exts e;
313
314         err = tcf_exts_init(&e, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
315         if (err < 0)
316                 return err;
317         err = tcf_exts_validate(net, tp, tb, est, &e, ovr, extack);
318         if (err < 0)
319                 goto errout;
320
321         err = -ENOMEM;
322         /* tcindex_data attributes must look atomic to classifier/lookup so
323          * allocate new tcindex data and RCU assign it onto root. Keeping
324          * perfect hash and hash pointers from old data.
325          */
326         cp = kzalloc(sizeof(*cp), GFP_KERNEL);
327         if (!cp)
328                 goto errout;
329
330         cp->mask = p->mask;
331         cp->shift = p->shift;
332         cp->hash = p->hash;
333         cp->alloc_hash = p->alloc_hash;
334         cp->fall_through = p->fall_through;
335         cp->tp = tp;
336
337         if (p->perfect) {
338                 int i;
339
340                 if (tcindex_alloc_perfect_hash(cp) < 0)
341                         goto errout;
342                 for (i = 0; i < cp->hash; i++)
343                         cp->perfect[i].res = p->perfect[i].res;
344                 balloc = 1;
345         }
346         cp->h = p->h;
347
348         err = tcindex_filter_result_init(&new_filter_result);
349         if (err < 0)
350                 goto errout1;
351         err = tcindex_filter_result_init(&cr);
352         if (err < 0)
353                 goto errout1;
354         if (old_r)
355                 cr.res = r->res;
356
357         if (tb[TCA_TCINDEX_HASH])
358                 cp->hash = nla_get_u32(tb[TCA_TCINDEX_HASH]);
359
360         if (tb[TCA_TCINDEX_MASK])
361                 cp->mask = nla_get_u16(tb[TCA_TCINDEX_MASK]);
362
363         if (tb[TCA_TCINDEX_SHIFT])
364                 cp->shift = nla_get_u32(tb[TCA_TCINDEX_SHIFT]);
365
366         err = -EBUSY;
367
368         /* Hash already allocated, make sure that we still meet the
369          * requirements for the allocated hash.
370          */
371         if (cp->perfect) {
372                 if (!valid_perfect_hash(cp) ||
373                     cp->hash > cp->alloc_hash)
374                         goto errout_alloc;
375         } else if (cp->h && cp->hash != cp->alloc_hash) {
376                 goto errout_alloc;
377         }
378
379         err = -EINVAL;
380         if (tb[TCA_TCINDEX_FALL_THROUGH])
381                 cp->fall_through = nla_get_u32(tb[TCA_TCINDEX_FALL_THROUGH]);
382
383         if (!cp->hash) {
384                 /* Hash not specified, use perfect hash if the upper limit
385                  * of the hashing index is below the threshold.
386                  */
387                 if ((cp->mask >> cp->shift) < PERFECT_HASH_THRESHOLD)
388                         cp->hash = (cp->mask >> cp->shift) + 1;
389                 else
390                         cp->hash = DEFAULT_HASH_SIZE;
391         }
392
393         if (!cp->perfect && !cp->h)
394                 cp->alloc_hash = cp->hash;
395
396         /* Note: this could be as restrictive as if (handle & ~(mask >> shift))
397          * but then, we'd fail handles that may become valid after some future
398          * mask change. While this is extremely unlikely to ever matter,
399          * the check below is safer (and also more backwards-compatible).
400          */
401         if (cp->perfect || valid_perfect_hash(cp))
402                 if (handle >= cp->alloc_hash)
403                         goto errout_alloc;
404
405
406         err = -ENOMEM;
407         if (!cp->perfect && !cp->h) {
408                 if (valid_perfect_hash(cp)) {
409                         if (tcindex_alloc_perfect_hash(cp) < 0)
410                                 goto errout_alloc;
411                         balloc = 1;
412                 } else {
413                         struct tcindex_filter __rcu **hash;
414
415                         hash = kcalloc(cp->hash,
416                                        sizeof(struct tcindex_filter *),
417                                        GFP_KERNEL);
418
419                         if (!hash)
420                                 goto errout_alloc;
421
422                         cp->h = hash;
423                         balloc = 2;
424                 }
425         }
426
427         if (cp->perfect)
428                 r = cp->perfect + handle;
429         else
430                 r = tcindex_lookup(cp, handle) ? : &new_filter_result;
431
432         if (r == &new_filter_result) {
433                 f = kzalloc(sizeof(*f), GFP_KERNEL);
434                 if (!f)
435                         goto errout_alloc;
436                 f->key = handle;
437                 f->next = NULL;
438                 err = tcindex_filter_result_init(&f->result);
439                 if (err < 0) {
440                         kfree(f);
441                         goto errout_alloc;
442                 }
443         }
444
445         if (tb[TCA_TCINDEX_CLASSID]) {
446                 cr.res.classid = nla_get_u32(tb[TCA_TCINDEX_CLASSID]);
447                 tcf_bind_filter(tp, &cr.res, base);
448         }
449
450         if (old_r)
451                 tcf_exts_change(&r->exts, &e);
452         else
453                 tcf_exts_change(&cr.exts, &e);
454
455         if (old_r && old_r != r) {
456                 err = tcindex_filter_result_init(old_r);
457                 if (err < 0) {
458                         kfree(f);
459                         goto errout_alloc;
460                 }
461         }
462
463         oldp = p;
464         r->res = cr.res;
465         rcu_assign_pointer(tp->root, cp);
466
467         if (r == &new_filter_result) {
468                 struct tcindex_filter *nfp;
469                 struct tcindex_filter __rcu **fp;
470
471                 tcf_exts_change(&f->result.exts, &r->exts);
472
473                 fp = cp->h + (handle % cp->hash);
474                 for (nfp = rtnl_dereference(*fp);
475                      nfp;
476                      fp = &nfp->next, nfp = rtnl_dereference(*fp))
477                                 ; /* nothing */
478
479                 rcu_assign_pointer(*fp, f);
480         }
481
482         if (oldp)
483                 call_rcu(&oldp->rcu, __tcindex_partial_destroy);
484         return 0;
485
486 errout_alloc:
487         if (balloc == 1)
488                 tcindex_free_perfect_hash(cp);
489         else if (balloc == 2)
490                 kfree(cp->h);
491 errout1:
492         tcf_exts_destroy(&cr.exts);
493         tcf_exts_destroy(&new_filter_result.exts);
494 errout:
495         kfree(cp);
496         tcf_exts_destroy(&e);
497         return err;
498 }
499
500 static int
501 tcindex_change(struct net *net, struct sk_buff *in_skb,
502                struct tcf_proto *tp, unsigned long base, u32 handle,
503                struct nlattr **tca, void **arg, bool ovr,
504                struct netlink_ext_ack *extack)
505 {
506         struct nlattr *opt = tca[TCA_OPTIONS];
507         struct nlattr *tb[TCA_TCINDEX_MAX + 1];
508         struct tcindex_data *p = rtnl_dereference(tp->root);
509         struct tcindex_filter_result *r = *arg;
510         int err;
511
512         pr_debug("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
513             "p %p,r %p,*arg %p\n",
514             tp, handle, tca, arg, opt, p, r, arg ? *arg : NULL);
515
516         if (!opt)
517                 return 0;
518
519         err = nla_parse_nested(tb, TCA_TCINDEX_MAX, opt, tcindex_policy, NULL);
520         if (err < 0)
521                 return err;
522
523         return tcindex_set_parms(net, tp, base, handle, p, r, tb,
524                                  tca[TCA_RATE], ovr, extack);
525 }
526
527 static void tcindex_walk(struct tcf_proto *tp, struct tcf_walker *walker)
528 {
529         struct tcindex_data *p = rtnl_dereference(tp->root);
530         struct tcindex_filter *f, *next;
531         int i;
532
533         pr_debug("tcindex_walk(tp %p,walker %p),p %p\n", tp, walker, p);
534         if (p->perfect) {
535                 for (i = 0; i < p->hash; i++) {
536                         if (!p->perfect[i].res.class)
537                                 continue;
538                         if (walker->count >= walker->skip) {
539                                 if (walker->fn(tp, p->perfect + i, walker) < 0) {
540                                         walker->stop = 1;
541                                         return;
542                                 }
543                         }
544                         walker->count++;
545                 }
546         }
547         if (!p->h)
548                 return;
549         for (i = 0; i < p->hash; i++) {
550                 for (f = rtnl_dereference(p->h[i]); f; f = next) {
551                         next = rtnl_dereference(f->next);
552                         if (walker->count >= walker->skip) {
553                                 if (walker->fn(tp, &f->result, walker) < 0) {
554                                         walker->stop = 1;
555                                         return;
556                                 }
557                         }
558                         walker->count++;
559                 }
560         }
561 }
562
563 static void tcindex_destroy(struct tcf_proto *tp,
564                             struct netlink_ext_ack *extack)
565 {
566         struct tcindex_data *p = rtnl_dereference(tp->root);
567         struct tcf_walker walker;
568
569         pr_debug("tcindex_destroy(tp %p),p %p\n", tp, p);
570         walker.count = 0;
571         walker.skip = 0;
572         walker.fn = tcindex_destroy_element;
573         tcindex_walk(tp, &walker);
574
575         call_rcu(&p->rcu, __tcindex_destroy);
576 }
577
578
579 static int tcindex_dump(struct net *net, struct tcf_proto *tp, void *fh,
580                         struct sk_buff *skb, struct tcmsg *t)
581 {
582         struct tcindex_data *p = rtnl_dereference(tp->root);
583         struct tcindex_filter_result *r = fh;
584         struct nlattr *nest;
585
586         pr_debug("tcindex_dump(tp %p,fh %p,skb %p,t %p),p %p,r %p\n",
587                  tp, fh, skb, t, p, r);
588         pr_debug("p->perfect %p p->h %p\n", p->perfect, p->h);
589
590         nest = nla_nest_start(skb, TCA_OPTIONS);
591         if (nest == NULL)
592                 goto nla_put_failure;
593
594         if (!fh) {
595                 t->tcm_handle = ~0; /* whatever ... */
596                 if (nla_put_u32(skb, TCA_TCINDEX_HASH, p->hash) ||
597                     nla_put_u16(skb, TCA_TCINDEX_MASK, p->mask) ||
598                     nla_put_u32(skb, TCA_TCINDEX_SHIFT, p->shift) ||
599                     nla_put_u32(skb, TCA_TCINDEX_FALL_THROUGH, p->fall_through))
600                         goto nla_put_failure;
601                 nla_nest_end(skb, nest);
602         } else {
603                 if (p->perfect) {
604                         t->tcm_handle = r - p->perfect;
605                 } else {
606                         struct tcindex_filter *f;
607                         struct tcindex_filter __rcu **fp;
608                         int i;
609
610                         t->tcm_handle = 0;
611                         for (i = 0; !t->tcm_handle && i < p->hash; i++) {
612                                 fp = &p->h[i];
613                                 for (f = rtnl_dereference(*fp);
614                                      !t->tcm_handle && f;
615                                      fp = &f->next, f = rtnl_dereference(*fp)) {
616                                         if (&f->result == r)
617                                                 t->tcm_handle = f->key;
618                                 }
619                         }
620                 }
621                 pr_debug("handle = %d\n", t->tcm_handle);
622                 if (r->res.class &&
623                     nla_put_u32(skb, TCA_TCINDEX_CLASSID, r->res.classid))
624                         goto nla_put_failure;
625
626                 if (tcf_exts_dump(skb, &r->exts) < 0)
627                         goto nla_put_failure;
628                 nla_nest_end(skb, nest);
629
630                 if (tcf_exts_dump_stats(skb, &r->exts) < 0)
631                         goto nla_put_failure;
632         }
633
634         return skb->len;
635
636 nla_put_failure:
637         nla_nest_cancel(skb, nest);
638         return -1;
639 }
640
641 static void tcindex_bind_class(void *fh, u32 classid, unsigned long cl)
642 {
643         struct tcindex_filter_result *r = fh;
644
645         if (r && r->res.classid == classid)
646                 r->res.class = cl;
647 }
648
649 static struct tcf_proto_ops cls_tcindex_ops __read_mostly = {
650         .kind           =       "tcindex",
651         .classify       =       tcindex_classify,
652         .init           =       tcindex_init,
653         .destroy        =       tcindex_destroy,
654         .get            =       tcindex_get,
655         .change         =       tcindex_change,
656         .delete         =       tcindex_delete,
657         .walk           =       tcindex_walk,
658         .dump           =       tcindex_dump,
659         .bind_class     =       tcindex_bind_class,
660         .owner          =       THIS_MODULE,
661 };
662
663 static int __init init_tcindex(void)
664 {
665         return register_tcf_proto_ops(&cls_tcindex_ops);
666 }
667
668 static void __exit exit_tcindex(void)
669 {
670         unregister_tcf_proto_ops(&cls_tcindex_ops);
671 }
672
673 module_init(init_tcindex)
674 module_exit(exit_tcindex)
675 MODULE_LICENSE("GPL");