1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * net/sched/cls_route.c ROUTE4 classifier.
4 *
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6 */
7
8 #include <linux/module.h>
9 #include <linux/slab.h>
10 #include <linux/types.h>
11 #include <linux/kernel.h>
12 #include <linux/string.h>
13 #include <linux/errno.h>
14 #include <linux/skbuff.h>
15 #include <net/dst.h>
16 #include <net/route.h>
17 #include <net/netlink.h>
18 #include <net/act_api.h>
19 #include <net/pkt_cls.h>
20
21 /*
22 * 1. For now we assume that route tags < 256.
23 * It allows to use direct table lookups, instead of hash tables.
24 * 2. For now we assume that "from TAG" and "fromdev DEV" statements
25 * are mutually exclusive.
26 * 3. "to TAG from ANY" has higher priority, than "to ANY from XXX"
27 */
28 struct route4_fastmap {
29 struct route4_filter *filter;
30 u32 id;
31 int iif;
32 };
33
34 struct route4_head {
35 struct route4_fastmap fastmap[16];
36 struct route4_bucket __rcu *table[256 + 1];
37 struct rcu_head rcu;
38 };
39
40 struct route4_bucket {
41 /* 16 FROM buckets + 16 IIF buckets + 1 wildcard bucket */
42 struct route4_filter __rcu *ht[16 + 16 + 1];
43 struct rcu_head rcu;
44 };
45
46 struct route4_filter {
47 struct route4_filter __rcu *next;
48 u32 id;
49 int iif;
50
51 struct tcf_result res;
52 struct tcf_exts exts;
53 u32 handle;
54 struct route4_bucket *bkt;
55 struct tcf_proto *tp;
56 struct rcu_work rwork;
57 };
58
59 #define ROUTE4_FAILURE ((struct route4_filter *)(-1L))
60
route4_fastmap_hash(u32 id,int iif)61 static inline int route4_fastmap_hash(u32 id, int iif)
62 {
63 return id & 0xF;
64 }
65
66 static DEFINE_SPINLOCK(fastmap_lock);
67 static void
route4_reset_fastmap(struct route4_head * head)68 route4_reset_fastmap(struct route4_head *head)
69 {
70 spin_lock_bh(&fastmap_lock);
71 memset(head->fastmap, 0, sizeof(head->fastmap));
72 spin_unlock_bh(&fastmap_lock);
73 }
74
75 static void
route4_set_fastmap(struct route4_head * head,u32 id,int iif,struct route4_filter * f)76 route4_set_fastmap(struct route4_head *head, u32 id, int iif,
77 struct route4_filter *f)
78 {
79 int h = route4_fastmap_hash(id, iif);
80
81 /* fastmap updates must look atomic to aling id, iff, filter */
82 spin_lock_bh(&fastmap_lock);
83 head->fastmap[h].id = id;
84 head->fastmap[h].iif = iif;
85 head->fastmap[h].filter = f;
86 spin_unlock_bh(&fastmap_lock);
87 }
88
route4_hash_to(u32 id)89 static inline int route4_hash_to(u32 id)
90 {
91 return id & 0xFF;
92 }
93
route4_hash_from(u32 id)94 static inline int route4_hash_from(u32 id)
95 {
96 return (id >> 16) & 0xF;
97 }
98
route4_hash_iif(int iif)99 static inline int route4_hash_iif(int iif)
100 {
101 return 16 + ((iif >> 16) & 0xF);
102 }
103
route4_hash_wild(void)104 static inline int route4_hash_wild(void)
105 {
106 return 32;
107 }
108
109 #define ROUTE4_APPLY_RESULT() \
110 { \
111 *res = f->res; \
112 if (tcf_exts_has_actions(&f->exts)) { \
113 int r = tcf_exts_exec(skb, &f->exts, res); \
114 if (r < 0) { \
115 dont_cache = 1; \
116 continue; \
117 } \
118 return r; \
119 } else if (!dont_cache) \
120 route4_set_fastmap(head, id, iif, f); \
121 return 0; \
122 }
123
route4_classify(struct sk_buff * skb,const struct tcf_proto * tp,struct tcf_result * res)124 static int route4_classify(struct sk_buff *skb, const struct tcf_proto *tp,
125 struct tcf_result *res)
126 {
127 struct route4_head *head = rcu_dereference_bh(tp->root);
128 struct dst_entry *dst;
129 struct route4_bucket *b;
130 struct route4_filter *f;
131 u32 id, h;
132 int iif, dont_cache = 0;
133
134 dst = skb_dst(skb);
135 if (!dst)
136 goto failure;
137
138 id = dst->tclassid;
139
140 iif = inet_iif(skb);
141
142 h = route4_fastmap_hash(id, iif);
143
144 spin_lock(&fastmap_lock);
145 if (id == head->fastmap[h].id &&
146 iif == head->fastmap[h].iif &&
147 (f = head->fastmap[h].filter) != NULL) {
148 if (f == ROUTE4_FAILURE) {
149 spin_unlock(&fastmap_lock);
150 goto failure;
151 }
152
153 *res = f->res;
154 spin_unlock(&fastmap_lock);
155 return 0;
156 }
157 spin_unlock(&fastmap_lock);
158
159 h = route4_hash_to(id);
160
161 restart:
162 b = rcu_dereference_bh(head->table[h]);
163 if (b) {
164 for (f = rcu_dereference_bh(b->ht[route4_hash_from(id)]);
165 f;
166 f = rcu_dereference_bh(f->next))
167 if (f->id == id)
168 ROUTE4_APPLY_RESULT();
169
170 for (f = rcu_dereference_bh(b->ht[route4_hash_iif(iif)]);
171 f;
172 f = rcu_dereference_bh(f->next))
173 if (f->iif == iif)
174 ROUTE4_APPLY_RESULT();
175
176 for (f = rcu_dereference_bh(b->ht[route4_hash_wild()]);
177 f;
178 f = rcu_dereference_bh(f->next))
179 ROUTE4_APPLY_RESULT();
180 }
181 if (h < 256) {
182 h = 256;
183 id &= ~0xFFFF;
184 goto restart;
185 }
186
187 if (!dont_cache)
188 route4_set_fastmap(head, id, iif, ROUTE4_FAILURE);
189 failure:
190 return -1;
191 }
192
to_hash(u32 id)193 static inline u32 to_hash(u32 id)
194 {
195 u32 h = id & 0xFF;
196
197 if (id & 0x8000)
198 h += 256;
199 return h;
200 }
201
from_hash(u32 id)202 static inline u32 from_hash(u32 id)
203 {
204 id &= 0xFFFF;
205 if (id == 0xFFFF)
206 return 32;
207 if (!(id & 0x8000)) {
208 if (id > 255)
209 return 256;
210 return id & 0xF;
211 }
212 return 16 + (id & 0xF);
213 }
214
route4_get(struct tcf_proto * tp,u32 handle)215 static void *route4_get(struct tcf_proto *tp, u32 handle)
216 {
217 struct route4_head *head = rtnl_dereference(tp->root);
218 struct route4_bucket *b;
219 struct route4_filter *f;
220 unsigned int h1, h2;
221
222 h1 = to_hash(handle);
223 if (h1 > 256)
224 return NULL;
225
226 h2 = from_hash(handle >> 16);
227 if (h2 > 32)
228 return NULL;
229
230 b = rtnl_dereference(head->table[h1]);
231 if (b) {
232 for (f = rtnl_dereference(b->ht[h2]);
233 f;
234 f = rtnl_dereference(f->next))
235 if (f->handle == handle)
236 return f;
237 }
238 return NULL;
239 }
240
route4_init(struct tcf_proto * tp)241 static int route4_init(struct tcf_proto *tp)
242 {
243 struct route4_head *head;
244
245 head = kzalloc(sizeof(struct route4_head), GFP_KERNEL);
246 if (head == NULL)
247 return -ENOBUFS;
248
249 rcu_assign_pointer(tp->root, head);
250 return 0;
251 }
252
__route4_delete_filter(struct route4_filter * f)253 static void __route4_delete_filter(struct route4_filter *f)
254 {
255 tcf_exts_destroy(&f->exts);
256 tcf_exts_put_net(&f->exts);
257 kfree(f);
258 }
259
route4_delete_filter_work(struct work_struct * work)260 static void route4_delete_filter_work(struct work_struct *work)
261 {
262 struct route4_filter *f = container_of(to_rcu_work(work),
263 struct route4_filter,
264 rwork);
265 rtnl_lock();
266 __route4_delete_filter(f);
267 rtnl_unlock();
268 }
269
route4_queue_work(struct route4_filter * f)270 static void route4_queue_work(struct route4_filter *f)
271 {
272 tcf_queue_work(&f->rwork, route4_delete_filter_work);
273 }
274
route4_destroy(struct tcf_proto * tp,bool rtnl_held,struct netlink_ext_ack * extack)275 static void route4_destroy(struct tcf_proto *tp, bool rtnl_held,
276 struct netlink_ext_ack *extack)
277 {
278 struct route4_head *head = rtnl_dereference(tp->root);
279 int h1, h2;
280
281 if (head == NULL)
282 return;
283
284 for (h1 = 0; h1 <= 256; h1++) {
285 struct route4_bucket *b;
286
287 b = rtnl_dereference(head->table[h1]);
288 if (b) {
289 for (h2 = 0; h2 <= 32; h2++) {
290 struct route4_filter *f;
291
292 while ((f = rtnl_dereference(b->ht[h2])) != NULL) {
293 struct route4_filter *next;
294
295 next = rtnl_dereference(f->next);
296 RCU_INIT_POINTER(b->ht[h2], next);
297 tcf_unbind_filter(tp, &f->res);
298 if (tcf_exts_get_net(&f->exts))
299 route4_queue_work(f);
300 else
301 __route4_delete_filter(f);
302 }
303 }
304 RCU_INIT_POINTER(head->table[h1], NULL);
305 kfree_rcu(b, rcu);
306 }
307 }
308 kfree_rcu(head, rcu);
309 }
310
route4_delete(struct tcf_proto * tp,void * arg,bool * last,bool rtnl_held,struct netlink_ext_ack * extack)311 static int route4_delete(struct tcf_proto *tp, void *arg, bool *last,
312 bool rtnl_held, struct netlink_ext_ack *extack)
313 {
314 struct route4_head *head = rtnl_dereference(tp->root);
315 struct route4_filter *f = arg;
316 struct route4_filter __rcu **fp;
317 struct route4_filter *nf;
318 struct route4_bucket *b;
319 unsigned int h = 0;
320 int i, h1;
321
322 if (!head || !f)
323 return -EINVAL;
324
325 h = f->handle;
326 b = f->bkt;
327
328 fp = &b->ht[from_hash(h >> 16)];
329 for (nf = rtnl_dereference(*fp); nf;
330 fp = &nf->next, nf = rtnl_dereference(*fp)) {
331 if (nf == f) {
332 /* unlink it */
333 RCU_INIT_POINTER(*fp, rtnl_dereference(f->next));
334
335 /* Remove any fastmap lookups that might ref filter
336 * notice we unlink'd the filter so we can't get it
337 * back in the fastmap.
338 */
339 route4_reset_fastmap(head);
340
341 /* Delete it */
342 tcf_unbind_filter(tp, &f->res);
343 tcf_exts_get_net(&f->exts);
344 tcf_queue_work(&f->rwork, route4_delete_filter_work);
345
346 /* Strip RTNL protected tree */
347 for (i = 0; i <= 32; i++) {
348 struct route4_filter *rt;
349
350 rt = rtnl_dereference(b->ht[i]);
351 if (rt)
352 goto out;
353 }
354
355 /* OK, session has no flows */
356 RCU_INIT_POINTER(head->table[to_hash(h)], NULL);
357 kfree_rcu(b, rcu);
358 break;
359 }
360 }
361
362 out:
363 *last = true;
364 for (h1 = 0; h1 <= 256; h1++) {
365 if (rcu_access_pointer(head->table[h1])) {
366 *last = false;
367 break;
368 }
369 }
370
371 return 0;
372 }
373
374 static const struct nla_policy route4_policy[TCA_ROUTE4_MAX + 1] = {
375 [TCA_ROUTE4_CLASSID] = { .type = NLA_U32 },
376 [TCA_ROUTE4_TO] = { .type = NLA_U32 },
377 [TCA_ROUTE4_FROM] = { .type = NLA_U32 },
378 [TCA_ROUTE4_IIF] = { .type = NLA_U32 },
379 };
380
route4_set_parms(struct net * net,struct tcf_proto * tp,unsigned long base,struct route4_filter * f,u32 handle,struct route4_head * head,struct nlattr ** tb,struct nlattr * est,int new,bool ovr,struct netlink_ext_ack * extack)381 static int route4_set_parms(struct net *net, struct tcf_proto *tp,
382 unsigned long base, struct route4_filter *f,
383 u32 handle, struct route4_head *head,
384 struct nlattr **tb, struct nlattr *est, int new,
385 bool ovr, struct netlink_ext_ack *extack)
386 {
387 u32 id = 0, to = 0, nhandle = 0x8000;
388 struct route4_filter *fp;
389 unsigned int h1;
390 struct route4_bucket *b;
391 int err;
392
393 err = tcf_exts_validate(net, tp, tb, est, &f->exts, ovr, true, extack);
394 if (err < 0)
395 return err;
396
397 if (tb[TCA_ROUTE4_TO]) {
398 if (new && handle & 0x8000)
399 return -EINVAL;
400 to = nla_get_u32(tb[TCA_ROUTE4_TO]);
401 if (to > 0xFF)
402 return -EINVAL;
403 nhandle = to;
404 }
405
406 if (tb[TCA_ROUTE4_FROM]) {
407 if (tb[TCA_ROUTE4_IIF])
408 return -EINVAL;
409 id = nla_get_u32(tb[TCA_ROUTE4_FROM]);
410 if (id > 0xFF)
411 return -EINVAL;
412 nhandle |= id << 16;
413 } else if (tb[TCA_ROUTE4_IIF]) {
414 id = nla_get_u32(tb[TCA_ROUTE4_IIF]);
415 if (id > 0x7FFF)
416 return -EINVAL;
417 nhandle |= (id | 0x8000) << 16;
418 } else
419 nhandle |= 0xFFFF << 16;
420
421 if (handle && new) {
422 nhandle |= handle & 0x7F00;
423 if (nhandle != handle)
424 return -EINVAL;
425 }
426
427 h1 = to_hash(nhandle);
428 b = rtnl_dereference(head->table[h1]);
429 if (!b) {
430 b = kzalloc(sizeof(struct route4_bucket), GFP_KERNEL);
431 if (b == NULL)
432 return -ENOBUFS;
433
434 rcu_assign_pointer(head->table[h1], b);
435 } else {
436 unsigned int h2 = from_hash(nhandle >> 16);
437
438 for (fp = rtnl_dereference(b->ht[h2]);
439 fp;
440 fp = rtnl_dereference(fp->next))
441 if (fp->handle == f->handle)
442 return -EEXIST;
443 }
444
445 if (tb[TCA_ROUTE4_TO])
446 f->id = to;
447
448 if (tb[TCA_ROUTE4_FROM])
449 f->id = to | id<<16;
450 else if (tb[TCA_ROUTE4_IIF])
451 f->iif = id;
452
453 f->handle = nhandle;
454 f->bkt = b;
455 f->tp = tp;
456
457 if (tb[TCA_ROUTE4_CLASSID]) {
458 f->res.classid = nla_get_u32(tb[TCA_ROUTE4_CLASSID]);
459 tcf_bind_filter(tp, &f->res, base);
460 }
461
462 return 0;
463 }
464
route4_change(struct net * net,struct sk_buff * in_skb,struct tcf_proto * tp,unsigned long base,u32 handle,struct nlattr ** tca,void ** arg,bool ovr,bool rtnl_held,struct netlink_ext_ack * extack)465 static int route4_change(struct net *net, struct sk_buff *in_skb,
466 struct tcf_proto *tp, unsigned long base, u32 handle,
467 struct nlattr **tca, void **arg, bool ovr,
468 bool rtnl_held, struct netlink_ext_ack *extack)
469 {
470 struct route4_head *head = rtnl_dereference(tp->root);
471 struct route4_filter __rcu **fp;
472 struct route4_filter *fold, *f1, *pfp, *f = NULL;
473 struct route4_bucket *b;
474 struct nlattr *opt = tca[TCA_OPTIONS];
475 struct nlattr *tb[TCA_ROUTE4_MAX + 1];
476 unsigned int h, th;
477 int err;
478 bool new = true;
479
480 if (opt == NULL)
481 return handle ? -EINVAL : 0;
482
483 err = nla_parse_nested_deprecated(tb, TCA_ROUTE4_MAX, opt,
484 route4_policy, NULL);
485 if (err < 0)
486 return err;
487
488 fold = *arg;
489 if (fold && handle && fold->handle != handle)
490 return -EINVAL;
491
492 err = -ENOBUFS;
493 f = kzalloc(sizeof(struct route4_filter), GFP_KERNEL);
494 if (!f)
495 goto errout;
496
497 err = tcf_exts_init(&f->exts, net, TCA_ROUTE4_ACT, TCA_ROUTE4_POLICE);
498 if (err < 0)
499 goto errout;
500
501 if (fold) {
502 f->id = fold->id;
503 f->iif = fold->iif;
504 f->res = fold->res;
505 f->handle = fold->handle;
506
507 f->tp = fold->tp;
508 f->bkt = fold->bkt;
509 new = false;
510 }
511
512 err = route4_set_parms(net, tp, base, f, handle, head, tb,
513 tca[TCA_RATE], new, ovr, extack);
514 if (err < 0)
515 goto errout;
516
517 h = from_hash(f->handle >> 16);
518 fp = &f->bkt->ht[h];
519 for (pfp = rtnl_dereference(*fp);
520 (f1 = rtnl_dereference(*fp)) != NULL;
521 fp = &f1->next)
522 if (f->handle < f1->handle)
523 break;
524
525 tcf_block_netif_keep_dst(tp->chain->block);
526 rcu_assign_pointer(f->next, f1);
527 rcu_assign_pointer(*fp, f);
528
529 if (fold && fold->handle && f->handle != fold->handle) {
530 th = to_hash(fold->handle);
531 h = from_hash(fold->handle >> 16);
532 b = rtnl_dereference(head->table[th]);
533 if (b) {
534 fp = &b->ht[h];
535 for (pfp = rtnl_dereference(*fp); pfp;
536 fp = &pfp->next, pfp = rtnl_dereference(*fp)) {
537 if (pfp == fold) {
538 rcu_assign_pointer(*fp, fold->next);
539 break;
540 }
541 }
542 }
543 }
544
545 route4_reset_fastmap(head);
546 *arg = f;
547 if (fold) {
548 tcf_unbind_filter(tp, &fold->res);
549 tcf_exts_get_net(&fold->exts);
550 tcf_queue_work(&fold->rwork, route4_delete_filter_work);
551 }
552 return 0;
553
554 errout:
555 if (f)
556 tcf_exts_destroy(&f->exts);
557 kfree(f);
558 return err;
559 }
560
route4_walk(struct tcf_proto * tp,struct tcf_walker * arg,bool rtnl_held)561 static void route4_walk(struct tcf_proto *tp, struct tcf_walker *arg,
562 bool rtnl_held)
563 {
564 struct route4_head *head = rtnl_dereference(tp->root);
565 unsigned int h, h1;
566
567 if (head == NULL || arg->stop)
568 return;
569
570 for (h = 0; h <= 256; h++) {
571 struct route4_bucket *b = rtnl_dereference(head->table[h]);
572
573 if (b) {
574 for (h1 = 0; h1 <= 32; h1++) {
575 struct route4_filter *f;
576
577 for (f = rtnl_dereference(b->ht[h1]);
578 f;
579 f = rtnl_dereference(f->next)) {
580 if (arg->count < arg->skip) {
581 arg->count++;
582 continue;
583 }
584 if (arg->fn(tp, f, arg) < 0) {
585 arg->stop = 1;
586 return;
587 }
588 arg->count++;
589 }
590 }
591 }
592 }
593 }
594
route4_dump(struct net * net,struct tcf_proto * tp,void * fh,struct sk_buff * skb,struct tcmsg * t,bool rtnl_held)595 static int route4_dump(struct net *net, struct tcf_proto *tp, void *fh,
596 struct sk_buff *skb, struct tcmsg *t, bool rtnl_held)
597 {
598 struct route4_filter *f = fh;
599 struct nlattr *nest;
600 u32 id;
601
602 if (f == NULL)
603 return skb->len;
604
605 t->tcm_handle = f->handle;
606
607 nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
608 if (nest == NULL)
609 goto nla_put_failure;
610
611 if (!(f->handle & 0x8000)) {
612 id = f->id & 0xFF;
613 if (nla_put_u32(skb, TCA_ROUTE4_TO, id))
614 goto nla_put_failure;
615 }
616 if (f->handle & 0x80000000) {
617 if ((f->handle >> 16) != 0xFFFF &&
618 nla_put_u32(skb, TCA_ROUTE4_IIF, f->iif))
619 goto nla_put_failure;
620 } else {
621 id = f->id >> 16;
622 if (nla_put_u32(skb, TCA_ROUTE4_FROM, id))
623 goto nla_put_failure;
624 }
625 if (f->res.classid &&
626 nla_put_u32(skb, TCA_ROUTE4_CLASSID, f->res.classid))
627 goto nla_put_failure;
628
629 if (tcf_exts_dump(skb, &f->exts) < 0)
630 goto nla_put_failure;
631
632 nla_nest_end(skb, nest);
633
634 if (tcf_exts_dump_stats(skb, &f->exts) < 0)
635 goto nla_put_failure;
636
637 return skb->len;
638
639 nla_put_failure:
640 nla_nest_cancel(skb, nest);
641 return -1;
642 }
643
route4_bind_class(void * fh,u32 classid,unsigned long cl,void * q,unsigned long base)644 static void route4_bind_class(void *fh, u32 classid, unsigned long cl, void *q,
645 unsigned long base)
646 {
647 struct route4_filter *f = fh;
648
649 if (f && f->res.classid == classid) {
650 if (cl)
651 __tcf_bind_filter(q, &f->res, base);
652 else
653 __tcf_unbind_filter(q, &f->res);
654 }
655 }
656
657 static struct tcf_proto_ops cls_route4_ops __read_mostly = {
658 .kind = "route",
659 .classify = route4_classify,
660 .init = route4_init,
661 .destroy = route4_destroy,
662 .get = route4_get,
663 .change = route4_change,
664 .delete = route4_delete,
665 .walk = route4_walk,
666 .dump = route4_dump,
667 .bind_class = route4_bind_class,
668 .owner = THIS_MODULE,
669 };
670
init_route4(void)671 static int __init init_route4(void)
672 {
673 return register_tcf_proto_ops(&cls_route4_ops);
674 }
675
exit_route4(void)676 static void __exit exit_route4(void)
677 {
678 unregister_tcf_proto_ops(&cls_route4_ops);
679 }
680
681 module_init(init_route4)
682 module_exit(exit_route4)
683 MODULE_LICENSE("GPL");
684