xref: /linux/net/sched/cls_u32.c (revision 110e6f26af80dfd90b6e5c645b1aed7228aa580d)
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 <net/netlink.h>
44 #include <net/act_api.h>
45 #include <net/pkt_cls.h>
46 #include <linux/netdevice.h>
47 
48 struct tc_u_knode {
49 	struct tc_u_knode __rcu	*next;
50 	u32			handle;
51 	struct tc_u_hnode __rcu	*ht_up;
52 	struct tcf_exts		exts;
53 #ifdef CONFIG_NET_CLS_IND
54 	int			ifindex;
55 #endif
56 	u8			fshift;
57 	struct tcf_result	res;
58 	struct tc_u_hnode __rcu	*ht_down;
59 #ifdef CONFIG_CLS_U32_PERF
60 	struct tc_u32_pcnt __percpu *pf;
61 #endif
62 	u32			flags;
63 #ifdef CONFIG_CLS_U32_MARK
64 	u32			val;
65 	u32			mask;
66 	u32 __percpu		*pcpu_success;
67 #endif
68 	struct tcf_proto	*tp;
69 	struct rcu_head		rcu;
70 	/* The 'sel' field MUST be the last field in structure to allow for
71 	 * tc_u32_keys allocated at end of structure.
72 	 */
73 	struct tc_u32_sel	sel;
74 };
75 
76 struct tc_u_hnode {
77 	struct tc_u_hnode __rcu	*next;
78 	u32			handle;
79 	u32			prio;
80 	struct tc_u_common	*tp_c;
81 	int			refcnt;
82 	unsigned int		divisor;
83 	struct rcu_head		rcu;
84 	/* The 'ht' field MUST be the last field in structure to allow for
85 	 * more entries allocated at end of structure.
86 	 */
87 	struct tc_u_knode __rcu	*ht[1];
88 };
89 
90 struct tc_u_common {
91 	struct tc_u_hnode __rcu	*hlist;
92 	struct Qdisc		*q;
93 	int			refcnt;
94 	u32			hgenerator;
95 	struct rcu_head		rcu;
96 };
97 
98 static inline unsigned int u32_hash_fold(__be32 key,
99 					 const struct tc_u32_sel *sel,
100 					 u8 fshift)
101 {
102 	unsigned int h = ntohl(key & sel->hmask) >> fshift;
103 
104 	return h;
105 }
106 
107 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp, struct tcf_result *res)
108 {
109 	struct {
110 		struct tc_u_knode *knode;
111 		unsigned int	  off;
112 	} stack[TC_U32_MAXDEPTH];
113 
114 	struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
115 	unsigned int off = skb_network_offset(skb);
116 	struct tc_u_knode *n;
117 	int sdepth = 0;
118 	int off2 = 0;
119 	int sel = 0;
120 #ifdef CONFIG_CLS_U32_PERF
121 	int j;
122 #endif
123 	int i, r;
124 
125 next_ht:
126 	n = rcu_dereference_bh(ht->ht[sel]);
127 
128 next_knode:
129 	if (n) {
130 		struct tc_u32_key *key = n->sel.keys;
131 
132 #ifdef CONFIG_CLS_U32_PERF
133 		__this_cpu_inc(n->pf->rcnt);
134 		j = 0;
135 #endif
136 
137 #ifdef CONFIG_CLS_U32_MARK
138 		if ((skb->mark & n->mask) != n->val) {
139 			n = rcu_dereference_bh(n->next);
140 			goto next_knode;
141 		} else {
142 			__this_cpu_inc(*n->pcpu_success);
143 		}
144 #endif
145 
146 		for (i = n->sel.nkeys; i > 0; i--, key++) {
147 			int toff = off + key->off + (off2 & key->offmask);
148 			__be32 *data, hdata;
149 
150 			if (skb_headroom(skb) + toff > INT_MAX)
151 				goto out;
152 
153 			data = skb_header_pointer(skb, toff, 4, &hdata);
154 			if (!data)
155 				goto out;
156 			if ((*data ^ key->val) & key->mask) {
157 				n = rcu_dereference_bh(n->next);
158 				goto next_knode;
159 			}
160 #ifdef CONFIG_CLS_U32_PERF
161 			__this_cpu_inc(n->pf->kcnts[j]);
162 			j++;
163 #endif
164 		}
165 
166 		ht = rcu_dereference_bh(n->ht_down);
167 		if (!ht) {
168 check_terminal:
169 			if (n->sel.flags & TC_U32_TERMINAL) {
170 
171 				*res = n->res;
172 #ifdef CONFIG_NET_CLS_IND
173 				if (!tcf_match_indev(skb, n->ifindex)) {
174 					n = rcu_dereference_bh(n->next);
175 					goto next_knode;
176 				}
177 #endif
178 #ifdef CONFIG_CLS_U32_PERF
179 				__this_cpu_inc(n->pf->rhit);
180 #endif
181 				r = tcf_exts_exec(skb, &n->exts, res);
182 				if (r < 0) {
183 					n = rcu_dereference_bh(n->next);
184 					goto next_knode;
185 				}
186 
187 				return r;
188 			}
189 			n = rcu_dereference_bh(n->next);
190 			goto next_knode;
191 		}
192 
193 		/* PUSH */
194 		if (sdepth >= TC_U32_MAXDEPTH)
195 			goto deadloop;
196 		stack[sdepth].knode = n;
197 		stack[sdepth].off = off;
198 		sdepth++;
199 
200 		ht = rcu_dereference_bh(n->ht_down);
201 		sel = 0;
202 		if (ht->divisor) {
203 			__be32 *data, hdata;
204 
205 			data = skb_header_pointer(skb, off + n->sel.hoff, 4,
206 						  &hdata);
207 			if (!data)
208 				goto out;
209 			sel = ht->divisor & u32_hash_fold(*data, &n->sel,
210 							  n->fshift);
211 		}
212 		if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
213 			goto next_ht;
214 
215 		if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
216 			off2 = n->sel.off + 3;
217 			if (n->sel.flags & TC_U32_VAROFFSET) {
218 				__be16 *data, hdata;
219 
220 				data = skb_header_pointer(skb,
221 							  off + n->sel.offoff,
222 							  2, &hdata);
223 				if (!data)
224 					goto out;
225 				off2 += ntohs(n->sel.offmask & *data) >>
226 					n->sel.offshift;
227 			}
228 			off2 &= ~3;
229 		}
230 		if (n->sel.flags & TC_U32_EAT) {
231 			off += off2;
232 			off2 = 0;
233 		}
234 
235 		if (off < skb->len)
236 			goto next_ht;
237 	}
238 
239 	/* POP */
240 	if (sdepth--) {
241 		n = stack[sdepth].knode;
242 		ht = rcu_dereference_bh(n->ht_up);
243 		off = stack[sdepth].off;
244 		goto check_terminal;
245 	}
246 out:
247 	return -1;
248 
249 deadloop:
250 	net_warn_ratelimited("cls_u32: dead loop\n");
251 	return -1;
252 }
253 
254 static struct tc_u_hnode *
255 u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
256 {
257 	struct tc_u_hnode *ht;
258 
259 	for (ht = rtnl_dereference(tp_c->hlist);
260 	     ht;
261 	     ht = rtnl_dereference(ht->next))
262 		if (ht->handle == handle)
263 			break;
264 
265 	return ht;
266 }
267 
268 static struct tc_u_knode *
269 u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
270 {
271 	unsigned int sel;
272 	struct tc_u_knode *n = NULL;
273 
274 	sel = TC_U32_HASH(handle);
275 	if (sel > ht->divisor)
276 		goto out;
277 
278 	for (n = rtnl_dereference(ht->ht[sel]);
279 	     n;
280 	     n = rtnl_dereference(n->next))
281 		if (n->handle == handle)
282 			break;
283 out:
284 	return n;
285 }
286 
287 
288 static unsigned long u32_get(struct tcf_proto *tp, u32 handle)
289 {
290 	struct tc_u_hnode *ht;
291 	struct tc_u_common *tp_c = tp->data;
292 
293 	if (TC_U32_HTID(handle) == TC_U32_ROOT)
294 		ht = rtnl_dereference(tp->root);
295 	else
296 		ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
297 
298 	if (!ht)
299 		return 0;
300 
301 	if (TC_U32_KEY(handle) == 0)
302 		return (unsigned long)ht;
303 
304 	return (unsigned long)u32_lookup_key(ht, handle);
305 }
306 
307 static u32 gen_new_htid(struct tc_u_common *tp_c)
308 {
309 	int i = 0x800;
310 
311 	/* hgenerator only used inside rtnl lock it is safe to increment
312 	 * without read _copy_ update semantics
313 	 */
314 	do {
315 		if (++tp_c->hgenerator == 0x7FF)
316 			tp_c->hgenerator = 1;
317 	} while (--i > 0 && u32_lookup_ht(tp_c, (tp_c->hgenerator|0x800)<<20));
318 
319 	return i > 0 ? (tp_c->hgenerator|0x800)<<20 : 0;
320 }
321 
322 static int u32_init(struct tcf_proto *tp)
323 {
324 	struct tc_u_hnode *root_ht;
325 	struct tc_u_common *tp_c;
326 
327 	tp_c = tp->q->u32_node;
328 
329 	root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
330 	if (root_ht == NULL)
331 		return -ENOBUFS;
332 
333 	root_ht->divisor = 0;
334 	root_ht->refcnt++;
335 	root_ht->handle = tp_c ? gen_new_htid(tp_c) : 0x80000000;
336 	root_ht->prio = tp->prio;
337 
338 	if (tp_c == NULL) {
339 		tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
340 		if (tp_c == NULL) {
341 			kfree(root_ht);
342 			return -ENOBUFS;
343 		}
344 		tp_c->q = tp->q;
345 		tp->q->u32_node = tp_c;
346 	}
347 
348 	tp_c->refcnt++;
349 	RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
350 	rcu_assign_pointer(tp_c->hlist, root_ht);
351 	root_ht->tp_c = tp_c;
352 
353 	rcu_assign_pointer(tp->root, root_ht);
354 	tp->data = tp_c;
355 	return 0;
356 }
357 
358 static int u32_destroy_key(struct tcf_proto *tp,
359 			   struct tc_u_knode *n,
360 			   bool free_pf)
361 {
362 	tcf_exts_destroy(&n->exts);
363 	if (n->ht_down)
364 		n->ht_down->refcnt--;
365 #ifdef CONFIG_CLS_U32_PERF
366 	if (free_pf)
367 		free_percpu(n->pf);
368 #endif
369 #ifdef CONFIG_CLS_U32_MARK
370 	if (free_pf)
371 		free_percpu(n->pcpu_success);
372 #endif
373 	kfree(n);
374 	return 0;
375 }
376 
377 /* u32_delete_key_rcu should be called when free'ing a copied
378  * version of a tc_u_knode obtained from u32_init_knode(). When
379  * copies are obtained from u32_init_knode() the statistics are
380  * shared between the old and new copies to allow readers to
381  * continue to update the statistics during the copy. To support
382  * this the u32_delete_key_rcu variant does not free the percpu
383  * statistics.
384  */
385 static void u32_delete_key_rcu(struct rcu_head *rcu)
386 {
387 	struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
388 
389 	u32_destroy_key(key->tp, key, false);
390 }
391 
392 /* u32_delete_key_freepf_rcu is the rcu callback variant
393  * that free's the entire structure including the statistics
394  * percpu variables. Only use this if the key is not a copy
395  * returned by u32_init_knode(). See u32_delete_key_rcu()
396  * for the variant that should be used with keys return from
397  * u32_init_knode()
398  */
399 static void u32_delete_key_freepf_rcu(struct rcu_head *rcu)
400 {
401 	struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
402 
403 	u32_destroy_key(key->tp, key, true);
404 }
405 
406 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
407 {
408 	struct tc_u_knode __rcu **kp;
409 	struct tc_u_knode *pkp;
410 	struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
411 
412 	if (ht) {
413 		kp = &ht->ht[TC_U32_HASH(key->handle)];
414 		for (pkp = rtnl_dereference(*kp); pkp;
415 		     kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
416 			if (pkp == key) {
417 				RCU_INIT_POINTER(*kp, key->next);
418 
419 				tcf_unbind_filter(tp, &key->res);
420 				call_rcu(&key->rcu, u32_delete_key_freepf_rcu);
421 				return 0;
422 			}
423 		}
424 	}
425 	WARN_ON(1);
426 	return 0;
427 }
428 
429 static void u32_remove_hw_knode(struct tcf_proto *tp, u32 handle)
430 {
431 	struct net_device *dev = tp->q->dev_queue->dev;
432 	struct tc_cls_u32_offload u32_offload = {0};
433 	struct tc_to_netdev offload;
434 
435 	offload.type = TC_SETUP_CLSU32;
436 	offload.cls_u32 = &u32_offload;
437 
438 	if (tc_should_offload(dev, 0)) {
439 		offload.cls_u32->command = TC_CLSU32_DELETE_KNODE;
440 		offload.cls_u32->knode.handle = handle;
441 		dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle,
442 					      tp->protocol, &offload);
443 	}
444 }
445 
446 static void u32_replace_hw_hnode(struct tcf_proto *tp,
447 				 struct tc_u_hnode *h,
448 				 u32 flags)
449 {
450 	struct net_device *dev = tp->q->dev_queue->dev;
451 	struct tc_cls_u32_offload u32_offload = {0};
452 	struct tc_to_netdev offload;
453 
454 	offload.type = TC_SETUP_CLSU32;
455 	offload.cls_u32 = &u32_offload;
456 
457 	if (tc_should_offload(dev, flags)) {
458 		offload.cls_u32->command = TC_CLSU32_NEW_HNODE;
459 		offload.cls_u32->hnode.divisor = h->divisor;
460 		offload.cls_u32->hnode.handle = h->handle;
461 		offload.cls_u32->hnode.prio = h->prio;
462 
463 		dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle,
464 					      tp->protocol, &offload);
465 	}
466 }
467 
468 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h)
469 {
470 	struct net_device *dev = tp->q->dev_queue->dev;
471 	struct tc_cls_u32_offload u32_offload = {0};
472 	struct tc_to_netdev offload;
473 
474 	offload.type = TC_SETUP_CLSU32;
475 	offload.cls_u32 = &u32_offload;
476 
477 	if (tc_should_offload(dev, 0)) {
478 		offload.cls_u32->command = TC_CLSU32_DELETE_HNODE;
479 		offload.cls_u32->hnode.divisor = h->divisor;
480 		offload.cls_u32->hnode.handle = h->handle;
481 		offload.cls_u32->hnode.prio = h->prio;
482 
483 		dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle,
484 					      tp->protocol, &offload);
485 	}
486 }
487 
488 static void u32_replace_hw_knode(struct tcf_proto *tp,
489 				 struct tc_u_knode *n,
490 				 u32 flags)
491 {
492 	struct net_device *dev = tp->q->dev_queue->dev;
493 	struct tc_cls_u32_offload u32_offload = {0};
494 	struct tc_to_netdev offload;
495 
496 	offload.type = TC_SETUP_CLSU32;
497 	offload.cls_u32 = &u32_offload;
498 
499 	if (tc_should_offload(dev, flags)) {
500 		offload.cls_u32->command = TC_CLSU32_REPLACE_KNODE;
501 		offload.cls_u32->knode.handle = n->handle;
502 		offload.cls_u32->knode.fshift = n->fshift;
503 #ifdef CONFIG_CLS_U32_MARK
504 		offload.cls_u32->knode.val = n->val;
505 		offload.cls_u32->knode.mask = n->mask;
506 #else
507 		offload.cls_u32->knode.val = 0;
508 		offload.cls_u32->knode.mask = 0;
509 #endif
510 		offload.cls_u32->knode.sel = &n->sel;
511 		offload.cls_u32->knode.exts = &n->exts;
512 		if (n->ht_down)
513 			offload.cls_u32->knode.link_handle = n->ht_down->handle;
514 
515 		dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle,
516 					      tp->protocol, &offload);
517 	}
518 }
519 
520 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
521 {
522 	struct tc_u_knode *n;
523 	unsigned int h;
524 
525 	for (h = 0; h <= ht->divisor; h++) {
526 		while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
527 			RCU_INIT_POINTER(ht->ht[h],
528 					 rtnl_dereference(n->next));
529 			tcf_unbind_filter(tp, &n->res);
530 			u32_remove_hw_knode(tp, n->handle);
531 			call_rcu(&n->rcu, u32_delete_key_freepf_rcu);
532 		}
533 	}
534 }
535 
536 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
537 {
538 	struct tc_u_common *tp_c = tp->data;
539 	struct tc_u_hnode __rcu **hn;
540 	struct tc_u_hnode *phn;
541 
542 	WARN_ON(ht->refcnt);
543 
544 	u32_clear_hnode(tp, ht);
545 
546 	hn = &tp_c->hlist;
547 	for (phn = rtnl_dereference(*hn);
548 	     phn;
549 	     hn = &phn->next, phn = rtnl_dereference(*hn)) {
550 		if (phn == ht) {
551 			u32_clear_hw_hnode(tp, ht);
552 			RCU_INIT_POINTER(*hn, ht->next);
553 			kfree_rcu(ht, rcu);
554 			return 0;
555 		}
556 	}
557 
558 	return -ENOENT;
559 }
560 
561 static bool ht_empty(struct tc_u_hnode *ht)
562 {
563 	unsigned int h;
564 
565 	for (h = 0; h <= ht->divisor; h++)
566 		if (rcu_access_pointer(ht->ht[h]))
567 			return false;
568 
569 	return true;
570 }
571 
572 static bool u32_destroy(struct tcf_proto *tp, bool force)
573 {
574 	struct tc_u_common *tp_c = tp->data;
575 	struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
576 
577 	WARN_ON(root_ht == NULL);
578 
579 	if (!force) {
580 		if (root_ht) {
581 			if (root_ht->refcnt > 1)
582 				return false;
583 			if (root_ht->refcnt == 1) {
584 				if (!ht_empty(root_ht))
585 					return false;
586 			}
587 		}
588 
589 		if (tp_c->refcnt > 1)
590 			return false;
591 
592 		if (tp_c->refcnt == 1) {
593 			struct tc_u_hnode *ht;
594 
595 			for (ht = rtnl_dereference(tp_c->hlist);
596 			     ht;
597 			     ht = rtnl_dereference(ht->next))
598 				if (!ht_empty(ht))
599 					return false;
600 		}
601 	}
602 
603 	if (root_ht && --root_ht->refcnt == 0)
604 		u32_destroy_hnode(tp, root_ht);
605 
606 	if (--tp_c->refcnt == 0) {
607 		struct tc_u_hnode *ht;
608 
609 		tp->q->u32_node = NULL;
610 
611 		for (ht = rtnl_dereference(tp_c->hlist);
612 		     ht;
613 		     ht = rtnl_dereference(ht->next)) {
614 			ht->refcnt--;
615 			u32_clear_hnode(tp, ht);
616 		}
617 
618 		while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
619 			RCU_INIT_POINTER(tp_c->hlist, ht->next);
620 			kfree_rcu(ht, rcu);
621 		}
622 
623 		kfree(tp_c);
624 	}
625 
626 	tp->data = NULL;
627 	return true;
628 }
629 
630 static int u32_delete(struct tcf_proto *tp, unsigned long arg)
631 {
632 	struct tc_u_hnode *ht = (struct tc_u_hnode *)arg;
633 	struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
634 
635 	if (ht == NULL)
636 		return 0;
637 
638 	if (TC_U32_KEY(ht->handle)) {
639 		u32_remove_hw_knode(tp, ht->handle);
640 		return u32_delete_key(tp, (struct tc_u_knode *)ht);
641 	}
642 
643 	if (root_ht == ht)
644 		return -EINVAL;
645 
646 	if (ht->refcnt == 1) {
647 		ht->refcnt--;
648 		u32_destroy_hnode(tp, ht);
649 	} else {
650 		return -EBUSY;
651 	}
652 
653 	return 0;
654 }
655 
656 #define NR_U32_NODE (1<<12)
657 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 handle)
658 {
659 	struct tc_u_knode *n;
660 	unsigned long i;
661 	unsigned long *bitmap = kzalloc(BITS_TO_LONGS(NR_U32_NODE) * sizeof(unsigned long),
662 					GFP_KERNEL);
663 	if (!bitmap)
664 		return handle | 0xFFF;
665 
666 	for (n = rtnl_dereference(ht->ht[TC_U32_HASH(handle)]);
667 	     n;
668 	     n = rtnl_dereference(n->next))
669 		set_bit(TC_U32_NODE(n->handle), bitmap);
670 
671 	i = find_next_zero_bit(bitmap, NR_U32_NODE, 0x800);
672 	if (i >= NR_U32_NODE)
673 		i = find_next_zero_bit(bitmap, NR_U32_NODE, 1);
674 
675 	kfree(bitmap);
676 	return handle | (i >= NR_U32_NODE ? 0xFFF : i);
677 }
678 
679 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
680 	[TCA_U32_CLASSID]	= { .type = NLA_U32 },
681 	[TCA_U32_HASH]		= { .type = NLA_U32 },
682 	[TCA_U32_LINK]		= { .type = NLA_U32 },
683 	[TCA_U32_DIVISOR]	= { .type = NLA_U32 },
684 	[TCA_U32_SEL]		= { .len = sizeof(struct tc_u32_sel) },
685 	[TCA_U32_INDEV]		= { .type = NLA_STRING, .len = IFNAMSIZ },
686 	[TCA_U32_MARK]		= { .len = sizeof(struct tc_u32_mark) },
687 	[TCA_U32_FLAGS]		= { .type = NLA_U32 },
688 };
689 
690 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
691 			 unsigned long base, struct tc_u_hnode *ht,
692 			 struct tc_u_knode *n, struct nlattr **tb,
693 			 struct nlattr *est, bool ovr)
694 {
695 	int err;
696 	struct tcf_exts e;
697 
698 	tcf_exts_init(&e, TCA_U32_ACT, TCA_U32_POLICE);
699 	err = tcf_exts_validate(net, tp, tb, est, &e, ovr);
700 	if (err < 0)
701 		return err;
702 
703 	err = -EINVAL;
704 	if (tb[TCA_U32_LINK]) {
705 		u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
706 		struct tc_u_hnode *ht_down = NULL, *ht_old;
707 
708 		if (TC_U32_KEY(handle))
709 			goto errout;
710 
711 		if (handle) {
712 			ht_down = u32_lookup_ht(ht->tp_c, handle);
713 
714 			if (ht_down == NULL)
715 				goto errout;
716 			ht_down->refcnt++;
717 		}
718 
719 		ht_old = rtnl_dereference(n->ht_down);
720 		rcu_assign_pointer(n->ht_down, ht_down);
721 
722 		if (ht_old)
723 			ht_old->refcnt--;
724 	}
725 	if (tb[TCA_U32_CLASSID]) {
726 		n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
727 		tcf_bind_filter(tp, &n->res, base);
728 	}
729 
730 #ifdef CONFIG_NET_CLS_IND
731 	if (tb[TCA_U32_INDEV]) {
732 		int ret;
733 		ret = tcf_change_indev(net, tb[TCA_U32_INDEV]);
734 		if (ret < 0)
735 			goto errout;
736 		n->ifindex = ret;
737 	}
738 #endif
739 	tcf_exts_change(tp, &n->exts, &e);
740 
741 	return 0;
742 errout:
743 	tcf_exts_destroy(&e);
744 	return err;
745 }
746 
747 static void u32_replace_knode(struct tcf_proto *tp,
748 			      struct tc_u_common *tp_c,
749 			      struct tc_u_knode *n)
750 {
751 	struct tc_u_knode __rcu **ins;
752 	struct tc_u_knode *pins;
753 	struct tc_u_hnode *ht;
754 
755 	if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
756 		ht = rtnl_dereference(tp->root);
757 	else
758 		ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
759 
760 	ins = &ht->ht[TC_U32_HASH(n->handle)];
761 
762 	/* The node must always exist for it to be replaced if this is not the
763 	 * case then something went very wrong elsewhere.
764 	 */
765 	for (pins = rtnl_dereference(*ins); ;
766 	     ins = &pins->next, pins = rtnl_dereference(*ins))
767 		if (pins->handle == n->handle)
768 			break;
769 
770 	RCU_INIT_POINTER(n->next, pins->next);
771 	rcu_assign_pointer(*ins, n);
772 }
773 
774 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp,
775 					 struct tc_u_knode *n)
776 {
777 	struct tc_u_knode *new;
778 	struct tc_u32_sel *s = &n->sel;
779 
780 	new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key),
781 		      GFP_KERNEL);
782 
783 	if (!new)
784 		return NULL;
785 
786 	RCU_INIT_POINTER(new->next, n->next);
787 	new->handle = n->handle;
788 	RCU_INIT_POINTER(new->ht_up, n->ht_up);
789 
790 #ifdef CONFIG_NET_CLS_IND
791 	new->ifindex = n->ifindex;
792 #endif
793 	new->fshift = n->fshift;
794 	new->res = n->res;
795 	new->flags = n->flags;
796 	RCU_INIT_POINTER(new->ht_down, n->ht_down);
797 
798 	/* bump reference count as long as we hold pointer to structure */
799 	if (new->ht_down)
800 		new->ht_down->refcnt++;
801 
802 #ifdef CONFIG_CLS_U32_PERF
803 	/* Statistics may be incremented by readers during update
804 	 * so we must keep them in tact. When the node is later destroyed
805 	 * a special destroy call must be made to not free the pf memory.
806 	 */
807 	new->pf = n->pf;
808 #endif
809 
810 #ifdef CONFIG_CLS_U32_MARK
811 	new->val = n->val;
812 	new->mask = n->mask;
813 	/* Similarly success statistics must be moved as pointers */
814 	new->pcpu_success = n->pcpu_success;
815 #endif
816 	new->tp = tp;
817 	memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
818 
819 	tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE);
820 
821 	return new;
822 }
823 
824 static int u32_change(struct net *net, struct sk_buff *in_skb,
825 		      struct tcf_proto *tp, unsigned long base, u32 handle,
826 		      struct nlattr **tca,
827 		      unsigned long *arg, bool ovr)
828 {
829 	struct tc_u_common *tp_c = tp->data;
830 	struct tc_u_hnode *ht;
831 	struct tc_u_knode *n;
832 	struct tc_u32_sel *s;
833 	struct nlattr *opt = tca[TCA_OPTIONS];
834 	struct nlattr *tb[TCA_U32_MAX + 1];
835 	u32 htid, flags = 0;
836 	int err;
837 #ifdef CONFIG_CLS_U32_PERF
838 	size_t size;
839 #endif
840 
841 	if (opt == NULL)
842 		return handle ? -EINVAL : 0;
843 
844 	err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy);
845 	if (err < 0)
846 		return err;
847 
848 	if (tb[TCA_U32_FLAGS])
849 		flags = nla_get_u32(tb[TCA_U32_FLAGS]);
850 
851 	n = (struct tc_u_knode *)*arg;
852 	if (n) {
853 		struct tc_u_knode *new;
854 
855 		if (TC_U32_KEY(n->handle) == 0)
856 			return -EINVAL;
857 
858 		if (n->flags != flags)
859 			return -EINVAL;
860 
861 		new = u32_init_knode(tp, n);
862 		if (!new)
863 			return -ENOMEM;
864 
865 		err = u32_set_parms(net, tp, base,
866 				    rtnl_dereference(n->ht_up), new, tb,
867 				    tca[TCA_RATE], ovr);
868 
869 		if (err) {
870 			u32_destroy_key(tp, new, false);
871 			return err;
872 		}
873 
874 		u32_replace_knode(tp, tp_c, new);
875 		tcf_unbind_filter(tp, &n->res);
876 		call_rcu(&n->rcu, u32_delete_key_rcu);
877 		u32_replace_hw_knode(tp, new, flags);
878 		return 0;
879 	}
880 
881 	if (tb[TCA_U32_DIVISOR]) {
882 		unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
883 
884 		if (--divisor > 0x100)
885 			return -EINVAL;
886 		if (TC_U32_KEY(handle))
887 			return -EINVAL;
888 		if (handle == 0) {
889 			handle = gen_new_htid(tp->data);
890 			if (handle == 0)
891 				return -ENOMEM;
892 		}
893 		ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
894 		if (ht == NULL)
895 			return -ENOBUFS;
896 		ht->tp_c = tp_c;
897 		ht->refcnt = 1;
898 		ht->divisor = divisor;
899 		ht->handle = handle;
900 		ht->prio = tp->prio;
901 		RCU_INIT_POINTER(ht->next, tp_c->hlist);
902 		rcu_assign_pointer(tp_c->hlist, ht);
903 		*arg = (unsigned long)ht;
904 
905 		u32_replace_hw_hnode(tp, ht, flags);
906 		return 0;
907 	}
908 
909 	if (tb[TCA_U32_HASH]) {
910 		htid = nla_get_u32(tb[TCA_U32_HASH]);
911 		if (TC_U32_HTID(htid) == TC_U32_ROOT) {
912 			ht = rtnl_dereference(tp->root);
913 			htid = ht->handle;
914 		} else {
915 			ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
916 			if (ht == NULL)
917 				return -EINVAL;
918 		}
919 	} else {
920 		ht = rtnl_dereference(tp->root);
921 		htid = ht->handle;
922 	}
923 
924 	if (ht->divisor < TC_U32_HASH(htid))
925 		return -EINVAL;
926 
927 	if (handle) {
928 		if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid))
929 			return -EINVAL;
930 		handle = htid | TC_U32_NODE(handle);
931 	} else
932 		handle = gen_new_kid(ht, htid);
933 
934 	if (tb[TCA_U32_SEL] == NULL)
935 		return -EINVAL;
936 
937 	s = nla_data(tb[TCA_U32_SEL]);
938 
939 	n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL);
940 	if (n == NULL)
941 		return -ENOBUFS;
942 
943 #ifdef CONFIG_CLS_U32_PERF
944 	size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64);
945 	n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt));
946 	if (!n->pf) {
947 		kfree(n);
948 		return -ENOBUFS;
949 	}
950 #endif
951 
952 	memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
953 	RCU_INIT_POINTER(n->ht_up, ht);
954 	n->handle = handle;
955 	n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
956 	n->flags = flags;
957 	tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE);
958 	n->tp = tp;
959 
960 #ifdef CONFIG_CLS_U32_MARK
961 	n->pcpu_success = alloc_percpu(u32);
962 	if (!n->pcpu_success) {
963 		err = -ENOMEM;
964 		goto errout;
965 	}
966 
967 	if (tb[TCA_U32_MARK]) {
968 		struct tc_u32_mark *mark;
969 
970 		mark = nla_data(tb[TCA_U32_MARK]);
971 		n->val = mark->val;
972 		n->mask = mark->mask;
973 	}
974 #endif
975 
976 	err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr);
977 	if (err == 0) {
978 		struct tc_u_knode __rcu **ins;
979 		struct tc_u_knode *pins;
980 
981 		ins = &ht->ht[TC_U32_HASH(handle)];
982 		for (pins = rtnl_dereference(*ins); pins;
983 		     ins = &pins->next, pins = rtnl_dereference(*ins))
984 			if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
985 				break;
986 
987 		RCU_INIT_POINTER(n->next, pins);
988 		rcu_assign_pointer(*ins, n);
989 		u32_replace_hw_knode(tp, n, flags);
990 		*arg = (unsigned long)n;
991 		return 0;
992 	}
993 
994 #ifdef CONFIG_CLS_U32_MARK
995 	free_percpu(n->pcpu_success);
996 errout:
997 #endif
998 
999 #ifdef CONFIG_CLS_U32_PERF
1000 	free_percpu(n->pf);
1001 #endif
1002 	kfree(n);
1003 	return err;
1004 }
1005 
1006 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg)
1007 {
1008 	struct tc_u_common *tp_c = tp->data;
1009 	struct tc_u_hnode *ht;
1010 	struct tc_u_knode *n;
1011 	unsigned int h;
1012 
1013 	if (arg->stop)
1014 		return;
1015 
1016 	for (ht = rtnl_dereference(tp_c->hlist);
1017 	     ht;
1018 	     ht = rtnl_dereference(ht->next)) {
1019 		if (ht->prio != tp->prio)
1020 			continue;
1021 		if (arg->count >= arg->skip) {
1022 			if (arg->fn(tp, (unsigned long)ht, arg) < 0) {
1023 				arg->stop = 1;
1024 				return;
1025 			}
1026 		}
1027 		arg->count++;
1028 		for (h = 0; h <= ht->divisor; h++) {
1029 			for (n = rtnl_dereference(ht->ht[h]);
1030 			     n;
1031 			     n = rtnl_dereference(n->next)) {
1032 				if (arg->count < arg->skip) {
1033 					arg->count++;
1034 					continue;
1035 				}
1036 				if (arg->fn(tp, (unsigned long)n, arg) < 0) {
1037 					arg->stop = 1;
1038 					return;
1039 				}
1040 				arg->count++;
1041 			}
1042 		}
1043 	}
1044 }
1045 
1046 static int u32_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
1047 		     struct sk_buff *skb, struct tcmsg *t)
1048 {
1049 	struct tc_u_knode *n = (struct tc_u_knode *)fh;
1050 	struct tc_u_hnode *ht_up, *ht_down;
1051 	struct nlattr *nest;
1052 
1053 	if (n == NULL)
1054 		return skb->len;
1055 
1056 	t->tcm_handle = n->handle;
1057 
1058 	nest = nla_nest_start(skb, TCA_OPTIONS);
1059 	if (nest == NULL)
1060 		goto nla_put_failure;
1061 
1062 	if (TC_U32_KEY(n->handle) == 0) {
1063 		struct tc_u_hnode *ht = (struct tc_u_hnode *)fh;
1064 		u32 divisor = ht->divisor + 1;
1065 
1066 		if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1067 			goto nla_put_failure;
1068 	} else {
1069 #ifdef CONFIG_CLS_U32_PERF
1070 		struct tc_u32_pcnt *gpf;
1071 		int cpu;
1072 #endif
1073 
1074 		if (nla_put(skb, TCA_U32_SEL,
1075 			    sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
1076 			    &n->sel))
1077 			goto nla_put_failure;
1078 
1079 		ht_up = rtnl_dereference(n->ht_up);
1080 		if (ht_up) {
1081 			u32 htid = n->handle & 0xFFFFF000;
1082 			if (nla_put_u32(skb, TCA_U32_HASH, htid))
1083 				goto nla_put_failure;
1084 		}
1085 		if (n->res.classid &&
1086 		    nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1087 			goto nla_put_failure;
1088 
1089 		ht_down = rtnl_dereference(n->ht_down);
1090 		if (ht_down &&
1091 		    nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1092 			goto nla_put_failure;
1093 
1094 		if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1095 			goto nla_put_failure;
1096 
1097 #ifdef CONFIG_CLS_U32_MARK
1098 		if ((n->val || n->mask)) {
1099 			struct tc_u32_mark mark = {.val = n->val,
1100 						   .mask = n->mask,
1101 						   .success = 0};
1102 			int cpum;
1103 
1104 			for_each_possible_cpu(cpum) {
1105 				__u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1106 
1107 				mark.success += cnt;
1108 			}
1109 
1110 			if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1111 				goto nla_put_failure;
1112 		}
1113 #endif
1114 
1115 		if (tcf_exts_dump(skb, &n->exts) < 0)
1116 			goto nla_put_failure;
1117 
1118 #ifdef CONFIG_NET_CLS_IND
1119 		if (n->ifindex) {
1120 			struct net_device *dev;
1121 			dev = __dev_get_by_index(net, n->ifindex);
1122 			if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1123 				goto nla_put_failure;
1124 		}
1125 #endif
1126 #ifdef CONFIG_CLS_U32_PERF
1127 		gpf = kzalloc(sizeof(struct tc_u32_pcnt) +
1128 			      n->sel.nkeys * sizeof(u64),
1129 			      GFP_KERNEL);
1130 		if (!gpf)
1131 			goto nla_put_failure;
1132 
1133 		for_each_possible_cpu(cpu) {
1134 			int i;
1135 			struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1136 
1137 			gpf->rcnt += pf->rcnt;
1138 			gpf->rhit += pf->rhit;
1139 			for (i = 0; i < n->sel.nkeys; i++)
1140 				gpf->kcnts[i] += pf->kcnts[i];
1141 		}
1142 
1143 		if (nla_put(skb, TCA_U32_PCNT,
1144 			    sizeof(struct tc_u32_pcnt) + n->sel.nkeys*sizeof(u64),
1145 			    gpf)) {
1146 			kfree(gpf);
1147 			goto nla_put_failure;
1148 		}
1149 		kfree(gpf);
1150 #endif
1151 	}
1152 
1153 	nla_nest_end(skb, nest);
1154 
1155 	if (TC_U32_KEY(n->handle))
1156 		if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1157 			goto nla_put_failure;
1158 	return skb->len;
1159 
1160 nla_put_failure:
1161 	nla_nest_cancel(skb, nest);
1162 	return -1;
1163 }
1164 
1165 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1166 	.kind		=	"u32",
1167 	.classify	=	u32_classify,
1168 	.init		=	u32_init,
1169 	.destroy	=	u32_destroy,
1170 	.get		=	u32_get,
1171 	.change		=	u32_change,
1172 	.delete		=	u32_delete,
1173 	.walk		=	u32_walk,
1174 	.dump		=	u32_dump,
1175 	.owner		=	THIS_MODULE,
1176 };
1177 
1178 static int __init init_u32(void)
1179 {
1180 	pr_info("u32 classifier\n");
1181 #ifdef CONFIG_CLS_U32_PERF
1182 	pr_info("    Performance counters on\n");
1183 #endif
1184 #ifdef CONFIG_NET_CLS_IND
1185 	pr_info("    input device check on\n");
1186 #endif
1187 #ifdef CONFIG_NET_CLS_ACT
1188 	pr_info("    Actions configured\n");
1189 #endif
1190 	return register_tcf_proto_ops(&cls_u32_ops);
1191 }
1192 
1193 static void __exit exit_u32(void)
1194 {
1195 	unregister_tcf_proto_ops(&cls_u32_ops);
1196 }
1197 
1198 module_init(init_u32)
1199 module_exit(exit_u32)
1200 MODULE_LICENSE("GPL");
1201