]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - net/sched/cls_tcindex.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:


72608baaef02c25d863b6320d2939db285ff2daf
[mirror_ubuntu-bionic-kernel.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 union {
32 struct work_struct work;
33 struct rcu_head rcu;
34 };
35 };
36
37 struct tcindex_filter {
38 u16 key;
39 struct tcindex_filter_result result;
40 struct tcindex_filter __rcu *next;
41 union {
42 struct work_struct work;
43 struct rcu_head rcu;
44 };
45 };
46
47
48 struct tcindex_data {
49 struct tcindex_filter_result *perfect; /* perfect hash; NULL if none */
50 struct tcindex_filter __rcu **h; /* imperfect hash; */
51 struct tcf_proto *tp;
52 u16 mask; /* AND key with mask */
53 u32 shift; /* shift ANDed key to the right */
54 u32 hash; /* hash table size; 0 if undefined */
55 u32 alloc_hash; /* allocated size */
56 u32 fall_through; /* 0: only classify if explicit match */
57 struct rcu_head rcu;
58 };
59
60 static inline int tcindex_filter_is_set(struct tcindex_filter_result *r)
61 {
62 return tcf_exts_has_actions(&r->exts) || r->res.classid;
63 }
64
65 static struct tcindex_filter_result *tcindex_lookup(struct tcindex_data *p,
66 u16 key)
67 {
68 if (p->perfect) {
69 struct tcindex_filter_result *f = p->perfect + key;
70
71 return tcindex_filter_is_set(f) ? f : NULL;
72 } else if (p->h) {
73 struct tcindex_filter __rcu **fp;
74 struct tcindex_filter *f;
75
76 fp = &p->h[key % p->hash];
77 for (f = rcu_dereference_bh_rtnl(*fp);
78 f;
79 fp = &f->next, f = rcu_dereference_bh_rtnl(*fp))
80 if (f->key == key)
81 return &f->result;
82 }
83
84 return NULL;
85 }
86
87
88 static int tcindex_classify(struct sk_buff *skb, const struct tcf_proto *tp,
89 struct tcf_result *res)
90 {
91 struct tcindex_data *p = rcu_dereference_bh(tp->root);
92 struct tcindex_filter_result *f;
93 int key = (skb->tc_index & p->mask) >> p->shift;
94
95 pr_debug("tcindex_classify(skb %p,tp %p,res %p),p %p\n",
96 skb, tp, res, p);
97
98 f = tcindex_lookup(p, key);
99 if (!f) {
100 struct Qdisc *q = tcf_block_q(tp->chain->block);
101
102 if (!p->fall_through)
103 return -1;
104 res->classid = TC_H_MAKE(TC_H_MAJ(q->handle), key);
105 res->class = 0;
106 pr_debug("alg 0x%x\n", res->classid);
107 return 0;
108 }
109 *res = f->res;
110 pr_debug("map 0x%x\n", res->classid);
111
112 return tcf_exts_exec(skb, &f->exts, res);
113 }
114
115
116 static void *tcindex_get(struct tcf_proto *tp, u32 handle)
117 {
118 struct tcindex_data *p = rtnl_dereference(tp->root);
119 struct tcindex_filter_result *r;
120
121 pr_debug("tcindex_get(tp %p,handle 0x%08x)\n", tp, handle);
122 if (p->perfect && handle >= p->alloc_hash)
123 return NULL;
124 r = tcindex_lookup(p, handle);
125 return r && tcindex_filter_is_set(r) ? r : NULL;
126 }
127
128 static int tcindex_init(struct tcf_proto *tp)
129 {
130 struct tcindex_data *p;
131
132 pr_debug("tcindex_init(tp %p)\n", tp);
133 p = kzalloc(sizeof(struct tcindex_data), GFP_KERNEL);
134 if (!p)
135 return -ENOMEM;
136
137 p->mask = 0xffff;
138 p->hash = DEFAULT_HASH_SIZE;
139 p->fall_through = 1;
140
141 rcu_assign_pointer(tp->root, p);
142 return 0;
143 }
144
145 static void __tcindex_destroy_rexts(struct tcindex_filter_result *r)
146 {
147 tcf_exts_destroy(&r->exts);
148 tcf_exts_put_net(&r->exts);
149 }
150
151 static void tcindex_destroy_rexts_work(struct work_struct *work)
152 {
153 struct tcindex_filter_result *r;
154
155 r = container_of(work, struct tcindex_filter_result, work);
156 rtnl_lock();
157 __tcindex_destroy_rexts(r);
158 rtnl_unlock();
159 }
160
161 static void tcindex_destroy_rexts(struct rcu_head *head)
162 {
163 struct tcindex_filter_result *r;
164
165 r = container_of(head, struct tcindex_filter_result, rcu);
166 INIT_WORK(&r->work, tcindex_destroy_rexts_work);
167 tcf_queue_work(&r->work);
168 }
169
170 static void __tcindex_destroy_fexts(struct tcindex_filter *f)
171 {
172 tcf_exts_destroy(&f->result.exts);
173 tcf_exts_put_net(&f->result.exts);
174 kfree(f);
175 }
176
177 static void tcindex_destroy_fexts_work(struct work_struct *work)
178 {
179 struct tcindex_filter *f = container_of(work, struct tcindex_filter,
180 work);
181
182 rtnl_lock();
183 __tcindex_destroy_fexts(f);
184 rtnl_unlock();
185 }
186
187 static void tcindex_destroy_fexts(struct rcu_head *head)
188 {
189 struct tcindex_filter *f = container_of(head, struct tcindex_filter,
190 rcu);
191
192 INIT_WORK(&f->work, tcindex_destroy_fexts_work);
193 tcf_queue_work(&f->work);
194 }
195
196 static int tcindex_delete(struct tcf_proto *tp, void *arg, bool *last)
197 {
198 struct tcindex_data *p = rtnl_dereference(tp->root);
199 struct tcindex_filter_result *r = arg;
200 struct tcindex_filter __rcu **walk;
201 struct tcindex_filter *f = NULL;
202
203 pr_debug("tcindex_delete(tp %p,arg %p),p %p\n", tp, arg, p);
204 if (p->perfect) {
205 if (!r->res.class)
206 return -ENOENT;
207 } else {
208 int i;
209
210 for (i = 0; i < p->hash; i++) {
211 walk = p->h + i;
212 for (f = rtnl_dereference(*walk); f;
213 walk = &f->next, f = rtnl_dereference(*walk)) {
214 if (&f->result == r)
215 goto found;
216 }
217 }
218 return -ENOENT;
219
220 found:
221 rcu_assign_pointer(*walk, rtnl_dereference(f->next));
222 }
223 tcf_unbind_filter(tp, &r->res);
224 /* all classifiers are required to call tcf_exts_destroy() after rcu
225 * grace period, since converted-to-rcu actions are relying on that
226 * in cleanup() callback
227 */
228 if (f) {
229 if (tcf_exts_get_net(&f->result.exts))
230 call_rcu(&f->rcu, tcindex_destroy_fexts);
231 else
232 __tcindex_destroy_fexts(f);
233 } else {
234 if (tcf_exts_get_net(&r->exts))
235 call_rcu(&r->rcu, tcindex_destroy_rexts);
236 else
237 __tcindex_destroy_rexts(r);
238 }
239
240 *last = false;
241 return 0;
242 }
243
244 static int tcindex_destroy_element(struct tcf_proto *tp,
245 void *arg, struct tcf_walker *walker)
246 {
247 bool last;
248
249 return tcindex_delete(tp, arg, &last);
250 }
251
252 static void __tcindex_destroy(struct rcu_head *head)
253 {
254 struct tcindex_data *p = container_of(head, struct tcindex_data, rcu);
255
256 kfree(p->perfect);
257 kfree(p->h);
258 kfree(p);
259 }
260
261 static inline int
262 valid_perfect_hash(struct tcindex_data *p)
263 {
264 return p->hash > (p->mask >> p->shift);
265 }
266
267 static const struct nla_policy tcindex_policy[TCA_TCINDEX_MAX + 1] = {
268 [TCA_TCINDEX_HASH] = { .type = NLA_U32 },
269 [TCA_TCINDEX_MASK] = { .type = NLA_U16 },
270 [TCA_TCINDEX_SHIFT] = { .type = NLA_U32 },
271 [TCA_TCINDEX_FALL_THROUGH] = { .type = NLA_U32 },
272 [TCA_TCINDEX_CLASSID] = { .type = NLA_U32 },
273 };
274
275 static int tcindex_filter_result_init(struct tcindex_filter_result *r)
276 {
277 memset(r, 0, sizeof(*r));
278 return tcf_exts_init(&r->exts, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
279 }
280
281 static void __tcindex_partial_destroy(struct rcu_head *head)
282 {
283 struct tcindex_data *p = container_of(head, struct tcindex_data, rcu);
284
285 kfree(p->perfect);
286 kfree(p);
287 }
288
289 static void tcindex_free_perfect_hash(struct tcindex_data *cp)
290 {
291 int i;
292
293 for (i = 0; i < cp->hash; i++)
294 tcf_exts_destroy(&cp->perfect[i].exts);
295 kfree(cp->perfect);
296 }
297
298 static int tcindex_alloc_perfect_hash(struct tcindex_data *cp)
299 {
300 int i, err = 0;
301
302 cp->perfect = kcalloc(cp->hash, sizeof(struct tcindex_filter_result),
303 GFP_KERNEL);
304 if (!cp->perfect)
305 return -ENOMEM;
306
307 for (i = 0; i < cp->hash; i++) {
308 err = tcf_exts_init(&cp->perfect[i].exts,
309 TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
310 if (err < 0)
311 goto errout;
312 }
313
314 return 0;
315
316 errout:
317 tcindex_free_perfect_hash(cp);
318 return err;
319 }
320
321 static int
322 tcindex_set_parms(struct net *net, struct tcf_proto *tp, unsigned long base,
323 u32 handle, struct tcindex_data *p,
324 struct tcindex_filter_result *r, struct nlattr **tb,
325 struct nlattr *est, bool ovr)
326 {
327 struct tcindex_filter_result new_filter_result, *old_r = r;
328 struct tcindex_filter_result cr;
329 struct tcindex_data *cp = NULL, *oldp;
330 struct tcindex_filter *f = NULL; /* make gcc behave */
331 int err, balloc = 0;
332 struct tcf_exts e;
333
334 err = tcf_exts_init(&e, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
335 if (err < 0)
336 return err;
337 err = tcf_exts_validate(net, tp, tb, est, &e, ovr);
338 if (err < 0)
339 goto errout;
340
341 err = -ENOMEM;
342 /* tcindex_data attributes must look atomic to classifier/lookup so
343 * allocate new tcindex data and RCU assign it onto root. Keeping
344 * perfect hash and hash pointers from old data.
345 */
346 cp = kzalloc(sizeof(*cp), GFP_KERNEL);
347 if (!cp)
348 goto errout;
349
350 cp->mask = p->mask;
351 cp->shift = p->shift;
352 cp->hash = p->hash;
353 cp->alloc_hash = p->alloc_hash;
354 cp->fall_through = p->fall_through;
355 cp->tp = tp;
356
357 if (p->perfect) {
358 int i;
359
360 if (tcindex_alloc_perfect_hash(cp) < 0)
361 goto errout;
362 for (i = 0; i < cp->hash; i++)
363 cp->perfect[i].res = p->perfect[i].res;
364 balloc = 1;
365 }
366 cp->h = p->h;
367
368 err = tcindex_filter_result_init(&new_filter_result);
369 if (err < 0)
370 goto errout1;
371 err = tcindex_filter_result_init(&cr);
372 if (err < 0)
373 goto errout1;
374 if (old_r)
375 cr.res = r->res;
376
377 if (tb[TCA_TCINDEX_HASH])
378 cp->hash = nla_get_u32(tb[TCA_TCINDEX_HASH]);
379
380 if (tb[TCA_TCINDEX_MASK])
381 cp->mask = nla_get_u16(tb[TCA_TCINDEX_MASK]);
382
383 if (tb[TCA_TCINDEX_SHIFT])
384 cp->shift = nla_get_u32(tb[TCA_TCINDEX_SHIFT]);
385
386 err = -EBUSY;
387
388 /* Hash already allocated, make sure that we still meet the
389 * requirements for the allocated hash.
390 */
391 if (cp->perfect) {
392 if (!valid_perfect_hash(cp) ||
393 cp->hash > cp->alloc_hash)
394 goto errout_alloc;
395 } else if (cp->h && cp->hash != cp->alloc_hash) {
396 goto errout_alloc;
397 }
398
399 err = -EINVAL;
400 if (tb[TCA_TCINDEX_FALL_THROUGH])
401 cp->fall_through = nla_get_u32(tb[TCA_TCINDEX_FALL_THROUGH]);
402
403 if (!cp->hash) {
404 /* Hash not specified, use perfect hash if the upper limit
405 * of the hashing index is below the threshold.
406 */
407 if ((cp->mask >> cp->shift) < PERFECT_HASH_THRESHOLD)
408 cp->hash = (cp->mask >> cp->shift) + 1;
409 else
410 cp->hash = DEFAULT_HASH_SIZE;
411 }
412
413 if (!cp->perfect && !cp->h)
414 cp->alloc_hash = cp->hash;
415
416 /* Note: this could be as restrictive as if (handle & ~(mask >> shift))
417 * but then, we'd fail handles that may become valid after some future
418 * mask change. While this is extremely unlikely to ever matter,
419 * the check below is safer (and also more backwards-compatible).
420 */
421 if (cp->perfect || valid_perfect_hash(cp))
422 if (handle >= cp->alloc_hash)
423 goto errout_alloc;
424
425
426 err = -ENOMEM;
427 if (!cp->perfect && !cp->h) {
428 if (valid_perfect_hash(cp)) {
429 if (tcindex_alloc_perfect_hash(cp) < 0)
430 goto errout_alloc;
431 balloc = 1;
432 } else {
433 struct tcindex_filter __rcu **hash;
434
435 hash = kcalloc(cp->hash,
436 sizeof(struct tcindex_filter *),
437 GFP_KERNEL);
438
439 if (!hash)
440 goto errout_alloc;
441
442 cp->h = hash;
443 balloc = 2;
444 }
445 }
446
447 if (cp->perfect)
448 r = cp->perfect + handle;
449 else
450 r = tcindex_lookup(cp, handle) ? : &new_filter_result;
451
452 if (r == &new_filter_result) {
453 f = kzalloc(sizeof(*f), GFP_KERNEL);
454 if (!f)
455 goto errout_alloc;
456 f->key = handle;
457 f->next = NULL;
458 err = tcindex_filter_result_init(&f->result);
459 if (err < 0) {
460 kfree(f);
461 goto errout_alloc;
462 }
463 }
464
465 if (tb[TCA_TCINDEX_CLASSID]) {
466 cr.res.classid = nla_get_u32(tb[TCA_TCINDEX_CLASSID]);
467 tcf_bind_filter(tp, &cr.res, base);
468 }
469
470 if (old_r && old_r != r) {
471 err = tcindex_filter_result_init(old_r);
472 if (err < 0) {
473 kfree(f);
474 goto errout_alloc;
475 }
476 }
477
478 oldp = p;
479 r->res = cr.res;
480 tcf_exts_change(&r->exts, &e);
481
482 rcu_assign_pointer(tp->root, cp);
483
484 if (r == &new_filter_result) {
485 struct tcindex_filter *nfp;
486 struct tcindex_filter __rcu **fp;
487
488 tcf_exts_change(&f->result.exts, &r->exts);
489
490 fp = cp->h + (handle % cp->hash);
491 for (nfp = rtnl_dereference(*fp);
492 nfp;
493 fp = &nfp->next, nfp = rtnl_dereference(*fp))
494 ; /* nothing */
495
496 rcu_assign_pointer(*fp, f);
497 }
498
499 if (oldp)
500 call_rcu(&oldp->rcu, __tcindex_partial_destroy);
501 return 0;
502
503 errout_alloc:
504 if (balloc == 1)
505 tcindex_free_perfect_hash(cp);
506 else if (balloc == 2)
507 kfree(cp->h);
508 errout1:
509 tcf_exts_destroy(&cr.exts);
510 tcf_exts_destroy(&new_filter_result.exts);
511 errout:
512 kfree(cp);
513 tcf_exts_destroy(&e);
514 return err;
515 }
516
517 static int
518 tcindex_change(struct net *net, struct sk_buff *in_skb,
519 struct tcf_proto *tp, unsigned long base, u32 handle,
520 struct nlattr **tca, void **arg, bool ovr)
521 {
522 struct nlattr *opt = tca[TCA_OPTIONS];
523 struct nlattr *tb[TCA_TCINDEX_MAX + 1];
524 struct tcindex_data *p = rtnl_dereference(tp->root);
525 struct tcindex_filter_result *r = *arg;
526 int err;
527
528 pr_debug("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
529 "p %p,r %p,*arg %p\n",
530 tp, handle, tca, arg, opt, p, r, arg ? *arg : NULL);
531
532 if (!opt)
533 return 0;
534
535 err = nla_parse_nested(tb, TCA_TCINDEX_MAX, opt, tcindex_policy, NULL);
536 if (err < 0)
537 return err;
538
539 return tcindex_set_parms(net, tp, base, handle, p, r, tb,
540 tca[TCA_RATE], ovr);
541 }
542
543 static void tcindex_walk(struct tcf_proto *tp, struct tcf_walker *walker)
544 {
545 struct tcindex_data *p = rtnl_dereference(tp->root);
546 struct tcindex_filter *f, *next;
547 int i;
548
549 pr_debug("tcindex_walk(tp %p,walker %p),p %p\n", tp, walker, p);
550 if (p->perfect) {
551 for (i = 0; i < p->hash; i++) {
552 if (!p->perfect[i].res.class)
553 continue;
554 if (walker->count >= walker->skip) {
555 if (walker->fn(tp, p->perfect + i, walker) < 0) {
556 walker->stop = 1;
557 return;
558 }
559 }
560 walker->count++;
561 }
562 }
563 if (!p->h)
564 return;
565 for (i = 0; i < p->hash; i++) {
566 for (f = rtnl_dereference(p->h[i]); f; f = next) {
567 next = rtnl_dereference(f->next);
568 if (walker->count >= walker->skip) {
569 if (walker->fn(tp, &f->result, walker) < 0) {
570 walker->stop = 1;
571 return;
572 }
573 }
574 walker->count++;
575 }
576 }
577 }
578
579 static void tcindex_destroy(struct tcf_proto *tp)
580 {
581 struct tcindex_data *p = rtnl_dereference(tp->root);
582 struct tcf_walker walker;
583
584 pr_debug("tcindex_destroy(tp %p),p %p\n", tp, p);
585 walker.count = 0;
586 walker.skip = 0;
587 walker.fn = tcindex_destroy_element;
588 tcindex_walk(tp, &walker);
589
590 call_rcu(&p->rcu, __tcindex_destroy);
591 }
592
593
594 static int tcindex_dump(struct net *net, struct tcf_proto *tp, void *fh,
595 struct sk_buff *skb, struct tcmsg *t)
596 {
597 struct tcindex_data *p = rtnl_dereference(tp->root);
598 struct tcindex_filter_result *r = fh;
599 struct nlattr *nest;
600
601 pr_debug("tcindex_dump(tp %p,fh %p,skb %p,t %p),p %p,r %p\n",
602 tp, fh, skb, t, p, r);
603 pr_debug("p->perfect %p p->h %p\n", p->perfect, p->h);
604
605 nest = nla_nest_start(skb, TCA_OPTIONS);
606 if (nest == NULL)
607 goto nla_put_failure;
608
609 if (!fh) {
610 t->tcm_handle = ~0; /* whatever ... */
611 if (nla_put_u32(skb, TCA_TCINDEX_HASH, p->hash) ||
612 nla_put_u16(skb, TCA_TCINDEX_MASK, p->mask) ||
613 nla_put_u32(skb, TCA_TCINDEX_SHIFT, p->shift) ||
614 nla_put_u32(skb, TCA_TCINDEX_FALL_THROUGH, p->fall_through))
615 goto nla_put_failure;
616 nla_nest_end(skb, nest);
617 } else {
618 if (p->perfect) {
619 t->tcm_handle = r - p->perfect;
620 } else {
621 struct tcindex_filter *f;
622 struct tcindex_filter __rcu **fp;
623 int i;
624
625 t->tcm_handle = 0;
626 for (i = 0; !t->tcm_handle && i < p->hash; i++) {
627 fp = &p->h[i];
628 for (f = rtnl_dereference(*fp);
629 !t->tcm_handle && f;
630 fp = &f->next, f = rtnl_dereference(*fp)) {
631 if (&f->result == r)
632 t->tcm_handle = f->key;
633 }
634 }
635 }
636 pr_debug("handle = %d\n", t->tcm_handle);
637 if (r->res.class &&
638 nla_put_u32(skb, TCA_TCINDEX_CLASSID, r->res.classid))
639 goto nla_put_failure;
640
641 if (tcf_exts_dump(skb, &r->exts) < 0)
642 goto nla_put_failure;
643 nla_nest_end(skb, nest);
644
645 if (tcf_exts_dump_stats(skb, &r->exts) < 0)
646 goto nla_put_failure;
647 }
648
649 return skb->len;
650
651 nla_put_failure:
652 nla_nest_cancel(skb, nest);
653 return -1;
654 }
655
656 static void tcindex_bind_class(void *fh, u32 classid, unsigned long cl)
657 {
658 struct tcindex_filter_result *r = fh;
659
660 if (r && r->res.classid == classid)
661 r->res.class = cl;
662 }
663
664 static struct tcf_proto_ops cls_tcindex_ops __read_mostly = {
665 .kind = "tcindex",
666 .classify = tcindex_classify,
667 .init = tcindex_init,
668 .destroy = tcindex_destroy,
669 .get = tcindex_get,
670 .change = tcindex_change,
671 .delete = tcindex_delete,
672 .walk = tcindex_walk,
673 .dump = tcindex_dump,
674 .bind_class = tcindex_bind_class,
675 .owner = THIS_MODULE,
676 };
677
678 static int __init init_tcindex(void)
679 {
680 return register_tcf_proto_ops(&cls_tcindex_ops);
681 }
682
683 static void __exit exit_tcindex(void)
684 {
685 unregister_tcf_proto_ops(&cls_tcindex_ops);
686 }
687
688 module_init(init_tcindex)
689 module_exit(exit_tcindex)
690 MODULE_LICENSE("GPL");
ubuntu-bionic-kernel mirror