xref: /linux/net/sched/cls_u32.c (revision fd639726bf15fca8ee1a00dce8e0096d0ad9bd18)
1 /*
2  * net/sched/cls_u32.c	Ugly (or Universal) 32bit key Packet Classifier.
3  *
4  *		This program is free software; you can redistribute it and/or
5  *		modify it under the terms of the GNU General Public License
6  *		as published by the Free Software Foundation; either version
7  *		2 of the License, or (at your option) any later version.
8  *
9  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  *
11  *	The filters are packed to hash tables of key nodes
12  *	with a set of 32bit key/mask pairs at every node.
13  *	Nodes reference next level hash tables etc.
14  *
15  *	This scheme is the best universal classifier I managed to
16  *	invent; it is not super-fast, but it is not slow (provided you
17  *	program it correctly), and general enough.  And its relative
18  *	speed grows as the number of rules becomes larger.
19  *
20  *	It seems that it represents the best middle point between
21  *	speed and manageability both by human and by machine.
22  *
23  *	It is especially useful for link sharing combined with QoS;
24  *	pure RSVP doesn't need such a general approach and can use
25  *	much simpler (and faster) schemes, sort of cls_rsvp.c.
26  *
27  *	JHS: We should remove the CONFIG_NET_CLS_IND from here
28  *	eventually when the meta match extension is made available
29  *
30  *	nfmark match added by Catalin(ux aka Dino) BOIE <catab at umbrella.ro>
31  */
32 
33 #include <linux/module.h>
34 #include <linux/slab.h>
35 #include <linux/types.h>
36 #include <linux/kernel.h>
37 #include <linux/string.h>
38 #include <linux/errno.h>
39 #include <linux/percpu.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/skbuff.h>
42 #include <linux/bitmap.h>
43 #include <linux/netdevice.h>
44 #include <linux/hash.h>
45 #include <net/netlink.h>
46 #include <net/act_api.h>
47 #include <net/pkt_cls.h>
48 #include <linux/netdevice.h>
49 #include <linux/idr.h>
50 
51 struct tc_u_knode {
52 	struct tc_u_knode __rcu	*next;
53 	u32			handle;
54 	struct tc_u_hnode __rcu	*ht_up;
55 	struct tcf_exts		exts;
56 #ifdef CONFIG_NET_CLS_IND
57 	int			ifindex;
58 #endif
59 	u8			fshift;
60 	struct tcf_result	res;
61 	struct tc_u_hnode __rcu	*ht_down;
62 #ifdef CONFIG_CLS_U32_PERF
63 	struct tc_u32_pcnt __percpu *pf;
64 #endif
65 	u32			flags;
66 #ifdef CONFIG_CLS_U32_MARK
67 	u32			val;
68 	u32			mask;
69 	u32 __percpu		*pcpu_success;
70 #endif
71 	struct tcf_proto	*tp;
72 	union {
73 		struct work_struct	work;
74 		struct rcu_head		rcu;
75 	};
76 	/* The 'sel' field MUST be the last field in structure to allow for
77 	 * tc_u32_keys allocated at end of structure.
78 	 */
79 	struct tc_u32_sel	sel;
80 };
81 
82 struct tc_u_hnode {
83 	struct tc_u_hnode __rcu	*next;
84 	u32			handle;
85 	u32			prio;
86 	struct tc_u_common	*tp_c;
87 	int			refcnt;
88 	unsigned int		divisor;
89 	struct idr		handle_idr;
90 	struct rcu_head		rcu;
91 	/* The 'ht' field MUST be the last field in structure to allow for
92 	 * more entries allocated at end of structure.
93 	 */
94 	struct tc_u_knode __rcu	*ht[1];
95 };
96 
97 struct tc_u_common {
98 	struct tc_u_hnode __rcu	*hlist;
99 	struct tcf_block	*block;
100 	int			refcnt;
101 	struct idr		handle_idr;
102 	struct hlist_node	hnode;
103 	struct rcu_head		rcu;
104 };
105 
106 static inline unsigned int u32_hash_fold(__be32 key,
107 					 const struct tc_u32_sel *sel,
108 					 u8 fshift)
109 {
110 	unsigned int h = ntohl(key & sel->hmask) >> fshift;
111 
112 	return h;
113 }
114 
115 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp,
116 			struct tcf_result *res)
117 {
118 	struct {
119 		struct tc_u_knode *knode;
120 		unsigned int	  off;
121 	} stack[TC_U32_MAXDEPTH];
122 
123 	struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
124 	unsigned int off = skb_network_offset(skb);
125 	struct tc_u_knode *n;
126 	int sdepth = 0;
127 	int off2 = 0;
128 	int sel = 0;
129 #ifdef CONFIG_CLS_U32_PERF
130 	int j;
131 #endif
132 	int i, r;
133 
134 next_ht:
135 	n = rcu_dereference_bh(ht->ht[sel]);
136 
137 next_knode:
138 	if (n) {
139 		struct tc_u32_key *key = n->sel.keys;
140 
141 #ifdef CONFIG_CLS_U32_PERF
142 		__this_cpu_inc(n->pf->rcnt);
143 		j = 0;
144 #endif
145 
146 		if (tc_skip_sw(n->flags)) {
147 			n = rcu_dereference_bh(n->next);
148 			goto next_knode;
149 		}
150 
151 #ifdef CONFIG_CLS_U32_MARK
152 		if ((skb->mark & n->mask) != n->val) {
153 			n = rcu_dereference_bh(n->next);
154 			goto next_knode;
155 		} else {
156 			__this_cpu_inc(*n->pcpu_success);
157 		}
158 #endif
159 
160 		for (i = n->sel.nkeys; i > 0; i--, key++) {
161 			int toff = off + key->off + (off2 & key->offmask);
162 			__be32 *data, hdata;
163 
164 			if (skb_headroom(skb) + toff > INT_MAX)
165 				goto out;
166 
167 			data = skb_header_pointer(skb, toff, 4, &hdata);
168 			if (!data)
169 				goto out;
170 			if ((*data ^ key->val) & key->mask) {
171 				n = rcu_dereference_bh(n->next);
172 				goto next_knode;
173 			}
174 #ifdef CONFIG_CLS_U32_PERF
175 			__this_cpu_inc(n->pf->kcnts[j]);
176 			j++;
177 #endif
178 		}
179 
180 		ht = rcu_dereference_bh(n->ht_down);
181 		if (!ht) {
182 check_terminal:
183 			if (n->sel.flags & TC_U32_TERMINAL) {
184 
185 				*res = n->res;
186 #ifdef CONFIG_NET_CLS_IND
187 				if (!tcf_match_indev(skb, n->ifindex)) {
188 					n = rcu_dereference_bh(n->next);
189 					goto next_knode;
190 				}
191 #endif
192 #ifdef CONFIG_CLS_U32_PERF
193 				__this_cpu_inc(n->pf->rhit);
194 #endif
195 				r = tcf_exts_exec(skb, &n->exts, res);
196 				if (r < 0) {
197 					n = rcu_dereference_bh(n->next);
198 					goto next_knode;
199 				}
200 
201 				return r;
202 			}
203 			n = rcu_dereference_bh(n->next);
204 			goto next_knode;
205 		}
206 
207 		/* PUSH */
208 		if (sdepth >= TC_U32_MAXDEPTH)
209 			goto deadloop;
210 		stack[sdepth].knode = n;
211 		stack[sdepth].off = off;
212 		sdepth++;
213 
214 		ht = rcu_dereference_bh(n->ht_down);
215 		sel = 0;
216 		if (ht->divisor) {
217 			__be32 *data, hdata;
218 
219 			data = skb_header_pointer(skb, off + n->sel.hoff, 4,
220 						  &hdata);
221 			if (!data)
222 				goto out;
223 			sel = ht->divisor & u32_hash_fold(*data, &n->sel,
224 							  n->fshift);
225 		}
226 		if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
227 			goto next_ht;
228 
229 		if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
230 			off2 = n->sel.off + 3;
231 			if (n->sel.flags & TC_U32_VAROFFSET) {
232 				__be16 *data, hdata;
233 
234 				data = skb_header_pointer(skb,
235 							  off + n->sel.offoff,
236 							  2, &hdata);
237 				if (!data)
238 					goto out;
239 				off2 += ntohs(n->sel.offmask & *data) >>
240 					n->sel.offshift;
241 			}
242 			off2 &= ~3;
243 		}
244 		if (n->sel.flags & TC_U32_EAT) {
245 			off += off2;
246 			off2 = 0;
247 		}
248 
249 		if (off < skb->len)
250 			goto next_ht;
251 	}
252 
253 	/* POP */
254 	if (sdepth--) {
255 		n = stack[sdepth].knode;
256 		ht = rcu_dereference_bh(n->ht_up);
257 		off = stack[sdepth].off;
258 		goto check_terminal;
259 	}
260 out:
261 	return -1;
262 
263 deadloop:
264 	net_warn_ratelimited("cls_u32: dead loop\n");
265 	return -1;
266 }
267 
268 static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
269 {
270 	struct tc_u_hnode *ht;
271 
272 	for (ht = rtnl_dereference(tp_c->hlist);
273 	     ht;
274 	     ht = rtnl_dereference(ht->next))
275 		if (ht->handle == handle)
276 			break;
277 
278 	return ht;
279 }
280 
281 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
282 {
283 	unsigned int sel;
284 	struct tc_u_knode *n = NULL;
285 
286 	sel = TC_U32_HASH(handle);
287 	if (sel > ht->divisor)
288 		goto out;
289 
290 	for (n = rtnl_dereference(ht->ht[sel]);
291 	     n;
292 	     n = rtnl_dereference(n->next))
293 		if (n->handle == handle)
294 			break;
295 out:
296 	return n;
297 }
298 
299 
300 static void *u32_get(struct tcf_proto *tp, u32 handle)
301 {
302 	struct tc_u_hnode *ht;
303 	struct tc_u_common *tp_c = tp->data;
304 
305 	if (TC_U32_HTID(handle) == TC_U32_ROOT)
306 		ht = rtnl_dereference(tp->root);
307 	else
308 		ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
309 
310 	if (!ht)
311 		return NULL;
312 
313 	if (TC_U32_KEY(handle) == 0)
314 		return ht;
315 
316 	return u32_lookup_key(ht, handle);
317 }
318 
319 static u32 gen_new_htid(struct tc_u_common *tp_c, struct tc_u_hnode *ptr)
320 {
321 	unsigned long idr_index;
322 	int err;
323 
324 	/* This is only used inside rtnl lock it is safe to increment
325 	 * without read _copy_ update semantics
326 	 */
327 	err = idr_alloc_ext(&tp_c->handle_idr, ptr, &idr_index,
328 			    1, 0x7FF, GFP_KERNEL);
329 	if (err)
330 		return 0;
331 	return (u32)(idr_index | 0x800) << 20;
332 }
333 
334 static struct hlist_head *tc_u_common_hash;
335 
336 #define U32_HASH_SHIFT 10
337 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
338 
339 static unsigned int tc_u_hash(const struct tcf_proto *tp)
340 {
341 	return hash_ptr(tp->chain->block, U32_HASH_SHIFT);
342 }
343 
344 static struct tc_u_common *tc_u_common_find(const struct tcf_proto *tp)
345 {
346 	struct tc_u_common *tc;
347 	unsigned int h;
348 
349 	h = tc_u_hash(tp);
350 	hlist_for_each_entry(tc, &tc_u_common_hash[h], hnode) {
351 		if (tc->block == tp->chain->block)
352 			return tc;
353 	}
354 	return NULL;
355 }
356 
357 static int u32_init(struct tcf_proto *tp)
358 {
359 	struct tc_u_hnode *root_ht;
360 	struct tc_u_common *tp_c;
361 	unsigned int h;
362 
363 	tp_c = tc_u_common_find(tp);
364 
365 	root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
366 	if (root_ht == NULL)
367 		return -ENOBUFS;
368 
369 	root_ht->refcnt++;
370 	root_ht->handle = tp_c ? gen_new_htid(tp_c, root_ht) : 0x80000000;
371 	root_ht->prio = tp->prio;
372 	idr_init(&root_ht->handle_idr);
373 
374 	if (tp_c == NULL) {
375 		tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
376 		if (tp_c == NULL) {
377 			kfree(root_ht);
378 			return -ENOBUFS;
379 		}
380 		tp_c->block = tp->chain->block;
381 		INIT_HLIST_NODE(&tp_c->hnode);
382 		idr_init(&tp_c->handle_idr);
383 
384 		h = tc_u_hash(tp);
385 		hlist_add_head(&tp_c->hnode, &tc_u_common_hash[h]);
386 	}
387 
388 	tp_c->refcnt++;
389 	RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
390 	rcu_assign_pointer(tp_c->hlist, root_ht);
391 	root_ht->tp_c = tp_c;
392 
393 	rcu_assign_pointer(tp->root, root_ht);
394 	tp->data = tp_c;
395 	return 0;
396 }
397 
398 static int u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n,
399 			   bool free_pf)
400 {
401 	tcf_exts_destroy(&n->exts);
402 	tcf_exts_put_net(&n->exts);
403 	if (n->ht_down)
404 		n->ht_down->refcnt--;
405 #ifdef CONFIG_CLS_U32_PERF
406 	if (free_pf)
407 		free_percpu(n->pf);
408 #endif
409 #ifdef CONFIG_CLS_U32_MARK
410 	if (free_pf)
411 		free_percpu(n->pcpu_success);
412 #endif
413 	kfree(n);
414 	return 0;
415 }
416 
417 /* u32_delete_key_rcu should be called when free'ing a copied
418  * version of a tc_u_knode obtained from u32_init_knode(). When
419  * copies are obtained from u32_init_knode() the statistics are
420  * shared between the old and new copies to allow readers to
421  * continue to update the statistics during the copy. To support
422  * this the u32_delete_key_rcu variant does not free the percpu
423  * statistics.
424  */
425 static void u32_delete_key_work(struct work_struct *work)
426 {
427 	struct tc_u_knode *key = container_of(work, struct tc_u_knode, work);
428 
429 	rtnl_lock();
430 	u32_destroy_key(key->tp, key, false);
431 	rtnl_unlock();
432 }
433 
434 static void u32_delete_key_rcu(struct rcu_head *rcu)
435 {
436 	struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
437 
438 	INIT_WORK(&key->work, u32_delete_key_work);
439 	tcf_queue_work(&key->work);
440 }
441 
442 /* u32_delete_key_freepf_rcu is the rcu callback variant
443  * that free's the entire structure including the statistics
444  * percpu variables. Only use this if the key is not a copy
445  * returned by u32_init_knode(). See u32_delete_key_rcu()
446  * for the variant that should be used with keys return from
447  * u32_init_knode()
448  */
449 static void u32_delete_key_freepf_work(struct work_struct *work)
450 {
451 	struct tc_u_knode *key = container_of(work, struct tc_u_knode, work);
452 
453 	rtnl_lock();
454 	u32_destroy_key(key->tp, key, true);
455 	rtnl_unlock();
456 }
457 
458 static void u32_delete_key_freepf_rcu(struct rcu_head *rcu)
459 {
460 	struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
461 
462 	INIT_WORK(&key->work, u32_delete_key_freepf_work);
463 	tcf_queue_work(&key->work);
464 }
465 
466 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
467 {
468 	struct tc_u_knode __rcu **kp;
469 	struct tc_u_knode *pkp;
470 	struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
471 
472 	if (ht) {
473 		kp = &ht->ht[TC_U32_HASH(key->handle)];
474 		for (pkp = rtnl_dereference(*kp); pkp;
475 		     kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
476 			if (pkp == key) {
477 				RCU_INIT_POINTER(*kp, key->next);
478 
479 				tcf_unbind_filter(tp, &key->res);
480 				tcf_exts_get_net(&key->exts);
481 				call_rcu(&key->rcu, u32_delete_key_freepf_rcu);
482 				return 0;
483 			}
484 		}
485 	}
486 	WARN_ON(1);
487 	return 0;
488 }
489 
490 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h)
491 {
492 	struct tcf_block *block = tp->chain->block;
493 	struct tc_cls_u32_offload cls_u32 = {};
494 
495 	tc_cls_common_offload_init(&cls_u32.common, tp);
496 	cls_u32.command = TC_CLSU32_DELETE_HNODE;
497 	cls_u32.hnode.divisor = h->divisor;
498 	cls_u32.hnode.handle = h->handle;
499 	cls_u32.hnode.prio = h->prio;
500 
501 	tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false);
502 }
503 
504 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
505 				u32 flags)
506 {
507 	struct tcf_block *block = tp->chain->block;
508 	struct tc_cls_u32_offload cls_u32 = {};
509 	bool skip_sw = tc_skip_sw(flags);
510 	bool offloaded = false;
511 	int err;
512 
513 	tc_cls_common_offload_init(&cls_u32.common, tp);
514 	cls_u32.command = TC_CLSU32_NEW_HNODE;
515 	cls_u32.hnode.divisor = h->divisor;
516 	cls_u32.hnode.handle = h->handle;
517 	cls_u32.hnode.prio = h->prio;
518 
519 	err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw);
520 	if (err < 0) {
521 		u32_clear_hw_hnode(tp, h);
522 		return err;
523 	} else if (err > 0) {
524 		offloaded = true;
525 	}
526 
527 	if (skip_sw && !offloaded)
528 		return -EINVAL;
529 
530 	return 0;
531 }
532 
533 static void u32_remove_hw_knode(struct tcf_proto *tp, u32 handle)
534 {
535 	struct tcf_block *block = tp->chain->block;
536 	struct tc_cls_u32_offload cls_u32 = {};
537 
538 	tc_cls_common_offload_init(&cls_u32.common, tp);
539 	cls_u32.command = TC_CLSU32_DELETE_KNODE;
540 	cls_u32.knode.handle = handle;
541 
542 	tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false);
543 }
544 
545 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
546 				u32 flags)
547 {
548 	struct tcf_block *block = tp->chain->block;
549 	struct tc_cls_u32_offload cls_u32 = {};
550 	bool skip_sw = tc_skip_sw(flags);
551 	int err;
552 
553 	tc_cls_common_offload_init(&cls_u32.common, tp);
554 	cls_u32.command = TC_CLSU32_REPLACE_KNODE;
555 	cls_u32.knode.handle = n->handle;
556 	cls_u32.knode.fshift = n->fshift;
557 #ifdef CONFIG_CLS_U32_MARK
558 	cls_u32.knode.val = n->val;
559 	cls_u32.knode.mask = n->mask;
560 #else
561 	cls_u32.knode.val = 0;
562 	cls_u32.knode.mask = 0;
563 #endif
564 	cls_u32.knode.sel = &n->sel;
565 	cls_u32.knode.exts = &n->exts;
566 	if (n->ht_down)
567 		cls_u32.knode.link_handle = n->ht_down->handle;
568 
569 	err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw);
570 	if (err < 0) {
571 		u32_remove_hw_knode(tp, n->handle);
572 		return err;
573 	} else if (err > 0) {
574 		n->flags |= TCA_CLS_FLAGS_IN_HW;
575 	}
576 
577 	if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
578 		return -EINVAL;
579 
580 	return 0;
581 }
582 
583 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
584 {
585 	struct tc_u_knode *n;
586 	unsigned int h;
587 
588 	for (h = 0; h <= ht->divisor; h++) {
589 		while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
590 			RCU_INIT_POINTER(ht->ht[h],
591 					 rtnl_dereference(n->next));
592 			tcf_unbind_filter(tp, &n->res);
593 			u32_remove_hw_knode(tp, n->handle);
594 			idr_remove_ext(&ht->handle_idr, n->handle);
595 			if (tcf_exts_get_net(&n->exts))
596 				call_rcu(&n->rcu, u32_delete_key_freepf_rcu);
597 			else
598 				u32_destroy_key(n->tp, n, true);
599 		}
600 	}
601 }
602 
603 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
604 {
605 	struct tc_u_common *tp_c = tp->data;
606 	struct tc_u_hnode __rcu **hn;
607 	struct tc_u_hnode *phn;
608 
609 	WARN_ON(ht->refcnt);
610 
611 	u32_clear_hnode(tp, ht);
612 
613 	hn = &tp_c->hlist;
614 	for (phn = rtnl_dereference(*hn);
615 	     phn;
616 	     hn = &phn->next, phn = rtnl_dereference(*hn)) {
617 		if (phn == ht) {
618 			u32_clear_hw_hnode(tp, ht);
619 			idr_destroy(&ht->handle_idr);
620 			idr_remove_ext(&tp_c->handle_idr, ht->handle);
621 			RCU_INIT_POINTER(*hn, ht->next);
622 			kfree_rcu(ht, rcu);
623 			return 0;
624 		}
625 	}
626 
627 	return -ENOENT;
628 }
629 
630 static bool ht_empty(struct tc_u_hnode *ht)
631 {
632 	unsigned int h;
633 
634 	for (h = 0; h <= ht->divisor; h++)
635 		if (rcu_access_pointer(ht->ht[h]))
636 			return false;
637 
638 	return true;
639 }
640 
641 static void u32_destroy(struct tcf_proto *tp)
642 {
643 	struct tc_u_common *tp_c = tp->data;
644 	struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
645 
646 	WARN_ON(root_ht == NULL);
647 
648 	if (root_ht && --root_ht->refcnt == 0)
649 		u32_destroy_hnode(tp, root_ht);
650 
651 	if (--tp_c->refcnt == 0) {
652 		struct tc_u_hnode *ht;
653 
654 		hlist_del(&tp_c->hnode);
655 
656 		for (ht = rtnl_dereference(tp_c->hlist);
657 		     ht;
658 		     ht = rtnl_dereference(ht->next)) {
659 			ht->refcnt--;
660 			u32_clear_hnode(tp, ht);
661 		}
662 
663 		while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
664 			RCU_INIT_POINTER(tp_c->hlist, ht->next);
665 			kfree_rcu(ht, rcu);
666 		}
667 
668 		idr_destroy(&tp_c->handle_idr);
669 		kfree(tp_c);
670 	}
671 
672 	tp->data = NULL;
673 }
674 
675 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last)
676 {
677 	struct tc_u_hnode *ht = arg;
678 	struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
679 	struct tc_u_common *tp_c = tp->data;
680 	int ret = 0;
681 
682 	if (ht == NULL)
683 		goto out;
684 
685 	if (TC_U32_KEY(ht->handle)) {
686 		u32_remove_hw_knode(tp, ht->handle);
687 		ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
688 		goto out;
689 	}
690 
691 	if (root_ht == ht)
692 		return -EINVAL;
693 
694 	if (ht->refcnt == 1) {
695 		ht->refcnt--;
696 		u32_destroy_hnode(tp, ht);
697 	} else {
698 		return -EBUSY;
699 	}
700 
701 out:
702 	*last = true;
703 	if (root_ht) {
704 		if (root_ht->refcnt > 1) {
705 			*last = false;
706 			goto ret;
707 		}
708 		if (root_ht->refcnt == 1) {
709 			if (!ht_empty(root_ht)) {
710 				*last = false;
711 				goto ret;
712 			}
713 		}
714 	}
715 
716 	if (tp_c->refcnt > 1) {
717 		*last = false;
718 		goto ret;
719 	}
720 
721 	if (tp_c->refcnt == 1) {
722 		struct tc_u_hnode *ht;
723 
724 		for (ht = rtnl_dereference(tp_c->hlist);
725 		     ht;
726 		     ht = rtnl_dereference(ht->next))
727 			if (!ht_empty(ht)) {
728 				*last = false;
729 				break;
730 			}
731 	}
732 
733 ret:
734 	return ret;
735 }
736 
737 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
738 {
739 	unsigned long idr_index;
740 	u32 start = htid | 0x800;
741 	u32 max = htid | 0xFFF;
742 	u32 min = htid;
743 
744 	if (idr_alloc_ext(&ht->handle_idr, NULL, &idr_index,
745 			  start, max + 1, GFP_KERNEL)) {
746 		if (idr_alloc_ext(&ht->handle_idr, NULL, &idr_index,
747 				  min + 1, max + 1, GFP_KERNEL))
748 			return max;
749 	}
750 
751 	return (u32)idr_index;
752 }
753 
754 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
755 	[TCA_U32_CLASSID]	= { .type = NLA_U32 },
756 	[TCA_U32_HASH]		= { .type = NLA_U32 },
757 	[TCA_U32_LINK]		= { .type = NLA_U32 },
758 	[TCA_U32_DIVISOR]	= { .type = NLA_U32 },
759 	[TCA_U32_SEL]		= { .len = sizeof(struct tc_u32_sel) },
760 	[TCA_U32_INDEV]		= { .type = NLA_STRING, .len = IFNAMSIZ },
761 	[TCA_U32_MARK]		= { .len = sizeof(struct tc_u32_mark) },
762 	[TCA_U32_FLAGS]		= { .type = NLA_U32 },
763 };
764 
765 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
766 			 unsigned long base, struct tc_u_hnode *ht,
767 			 struct tc_u_knode *n, struct nlattr **tb,
768 			 struct nlattr *est, bool ovr)
769 {
770 	int err;
771 
772 	err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr);
773 	if (err < 0)
774 		return err;
775 
776 	if (tb[TCA_U32_LINK]) {
777 		u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
778 		struct tc_u_hnode *ht_down = NULL, *ht_old;
779 
780 		if (TC_U32_KEY(handle))
781 			return -EINVAL;
782 
783 		if (handle) {
784 			ht_down = u32_lookup_ht(ht->tp_c, handle);
785 
786 			if (ht_down == NULL)
787 				return -EINVAL;
788 			ht_down->refcnt++;
789 		}
790 
791 		ht_old = rtnl_dereference(n->ht_down);
792 		rcu_assign_pointer(n->ht_down, ht_down);
793 
794 		if (ht_old)
795 			ht_old->refcnt--;
796 	}
797 	if (tb[TCA_U32_CLASSID]) {
798 		n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
799 		tcf_bind_filter(tp, &n->res, base);
800 	}
801 
802 #ifdef CONFIG_NET_CLS_IND
803 	if (tb[TCA_U32_INDEV]) {
804 		int ret;
805 		ret = tcf_change_indev(net, tb[TCA_U32_INDEV]);
806 		if (ret < 0)
807 			return -EINVAL;
808 		n->ifindex = ret;
809 	}
810 #endif
811 	return 0;
812 }
813 
814 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
815 			      struct tc_u_knode *n)
816 {
817 	struct tc_u_knode __rcu **ins;
818 	struct tc_u_knode *pins;
819 	struct tc_u_hnode *ht;
820 
821 	if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
822 		ht = rtnl_dereference(tp->root);
823 	else
824 		ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
825 
826 	ins = &ht->ht[TC_U32_HASH(n->handle)];
827 
828 	/* The node must always exist for it to be replaced if this is not the
829 	 * case then something went very wrong elsewhere.
830 	 */
831 	for (pins = rtnl_dereference(*ins); ;
832 	     ins = &pins->next, pins = rtnl_dereference(*ins))
833 		if (pins->handle == n->handle)
834 			break;
835 
836 	idr_replace_ext(&ht->handle_idr, n, n->handle);
837 	RCU_INIT_POINTER(n->next, pins->next);
838 	rcu_assign_pointer(*ins, n);
839 }
840 
841 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp,
842 					 struct tc_u_knode *n)
843 {
844 	struct tc_u_knode *new;
845 	struct tc_u32_sel *s = &n->sel;
846 
847 	new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key),
848 		      GFP_KERNEL);
849 
850 	if (!new)
851 		return NULL;
852 
853 	RCU_INIT_POINTER(new->next, n->next);
854 	new->handle = n->handle;
855 	RCU_INIT_POINTER(new->ht_up, n->ht_up);
856 
857 #ifdef CONFIG_NET_CLS_IND
858 	new->ifindex = n->ifindex;
859 #endif
860 	new->fshift = n->fshift;
861 	new->res = n->res;
862 	new->flags = n->flags;
863 	RCU_INIT_POINTER(new->ht_down, n->ht_down);
864 
865 	/* bump reference count as long as we hold pointer to structure */
866 	if (new->ht_down)
867 		new->ht_down->refcnt++;
868 
869 #ifdef CONFIG_CLS_U32_PERF
870 	/* Statistics may be incremented by readers during update
871 	 * so we must keep them in tact. When the node is later destroyed
872 	 * a special destroy call must be made to not free the pf memory.
873 	 */
874 	new->pf = n->pf;
875 #endif
876 
877 #ifdef CONFIG_CLS_U32_MARK
878 	new->val = n->val;
879 	new->mask = n->mask;
880 	/* Similarly success statistics must be moved as pointers */
881 	new->pcpu_success = n->pcpu_success;
882 #endif
883 	new->tp = tp;
884 	memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
885 
886 	if (tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE)) {
887 		kfree(new);
888 		return NULL;
889 	}
890 
891 	return new;
892 }
893 
894 static int u32_change(struct net *net, struct sk_buff *in_skb,
895 		      struct tcf_proto *tp, unsigned long base, u32 handle,
896 		      struct nlattr **tca, void **arg, bool ovr)
897 {
898 	struct tc_u_common *tp_c = tp->data;
899 	struct tc_u_hnode *ht;
900 	struct tc_u_knode *n;
901 	struct tc_u32_sel *s;
902 	struct nlattr *opt = tca[TCA_OPTIONS];
903 	struct nlattr *tb[TCA_U32_MAX + 1];
904 	u32 htid, flags = 0;
905 	int err;
906 #ifdef CONFIG_CLS_U32_PERF
907 	size_t size;
908 #endif
909 
910 	if (opt == NULL)
911 		return handle ? -EINVAL : 0;
912 
913 	err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy, NULL);
914 	if (err < 0)
915 		return err;
916 
917 	if (tb[TCA_U32_FLAGS]) {
918 		flags = nla_get_u32(tb[TCA_U32_FLAGS]);
919 		if (!tc_flags_valid(flags))
920 			return -EINVAL;
921 	}
922 
923 	n = *arg;
924 	if (n) {
925 		struct tc_u_knode *new;
926 
927 		if (TC_U32_KEY(n->handle) == 0)
928 			return -EINVAL;
929 
930 		if (n->flags != flags)
931 			return -EINVAL;
932 
933 		new = u32_init_knode(tp, n);
934 		if (!new)
935 			return -ENOMEM;
936 
937 		err = u32_set_parms(net, tp, base,
938 				    rtnl_dereference(n->ht_up), new, tb,
939 				    tca[TCA_RATE], ovr);
940 
941 		if (err) {
942 			u32_destroy_key(tp, new, false);
943 			return err;
944 		}
945 
946 		err = u32_replace_hw_knode(tp, new, flags);
947 		if (err) {
948 			u32_destroy_key(tp, new, false);
949 			return err;
950 		}
951 
952 		if (!tc_in_hw(new->flags))
953 			new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
954 
955 		u32_replace_knode(tp, tp_c, new);
956 		tcf_unbind_filter(tp, &n->res);
957 		tcf_exts_get_net(&n->exts);
958 		call_rcu(&n->rcu, u32_delete_key_rcu);
959 		return 0;
960 	}
961 
962 	if (tb[TCA_U32_DIVISOR]) {
963 		unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
964 
965 		if (--divisor > 0x100)
966 			return -EINVAL;
967 		if (TC_U32_KEY(handle))
968 			return -EINVAL;
969 		ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
970 		if (ht == NULL)
971 			return -ENOBUFS;
972 		if (handle == 0) {
973 			handle = gen_new_htid(tp->data, ht);
974 			if (handle == 0) {
975 				kfree(ht);
976 				return -ENOMEM;
977 			}
978 		} else {
979 			err = idr_alloc_ext(&tp_c->handle_idr, ht, NULL,
980 					    handle, handle + 1, GFP_KERNEL);
981 			if (err) {
982 				kfree(ht);
983 				return err;
984 			}
985 		}
986 		ht->tp_c = tp_c;
987 		ht->refcnt = 1;
988 		ht->divisor = divisor;
989 		ht->handle = handle;
990 		ht->prio = tp->prio;
991 		idr_init(&ht->handle_idr);
992 
993 		err = u32_replace_hw_hnode(tp, ht, flags);
994 		if (err) {
995 			idr_remove_ext(&tp_c->handle_idr, handle);
996 			kfree(ht);
997 			return err;
998 		}
999 
1000 		RCU_INIT_POINTER(ht->next, tp_c->hlist);
1001 		rcu_assign_pointer(tp_c->hlist, ht);
1002 		*arg = ht;
1003 
1004 		return 0;
1005 	}
1006 
1007 	if (tb[TCA_U32_HASH]) {
1008 		htid = nla_get_u32(tb[TCA_U32_HASH]);
1009 		if (TC_U32_HTID(htid) == TC_U32_ROOT) {
1010 			ht = rtnl_dereference(tp->root);
1011 			htid = ht->handle;
1012 		} else {
1013 			ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
1014 			if (ht == NULL)
1015 				return -EINVAL;
1016 		}
1017 	} else {
1018 		ht = rtnl_dereference(tp->root);
1019 		htid = ht->handle;
1020 	}
1021 
1022 	if (ht->divisor < TC_U32_HASH(htid))
1023 		return -EINVAL;
1024 
1025 	if (handle) {
1026 		if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid))
1027 			return -EINVAL;
1028 		handle = htid | TC_U32_NODE(handle);
1029 		err = idr_alloc_ext(&ht->handle_idr, NULL, NULL,
1030 				    handle, handle + 1,
1031 				    GFP_KERNEL);
1032 		if (err)
1033 			return err;
1034 	} else
1035 		handle = gen_new_kid(ht, htid);
1036 
1037 	if (tb[TCA_U32_SEL] == NULL) {
1038 		err = -EINVAL;
1039 		goto erridr;
1040 	}
1041 
1042 	s = nla_data(tb[TCA_U32_SEL]);
1043 
1044 	n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL);
1045 	if (n == NULL) {
1046 		err = -ENOBUFS;
1047 		goto erridr;
1048 	}
1049 
1050 #ifdef CONFIG_CLS_U32_PERF
1051 	size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64);
1052 	n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt));
1053 	if (!n->pf) {
1054 		err = -ENOBUFS;
1055 		goto errfree;
1056 	}
1057 #endif
1058 
1059 	memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
1060 	RCU_INIT_POINTER(n->ht_up, ht);
1061 	n->handle = handle;
1062 	n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1063 	n->flags = flags;
1064 	n->tp = tp;
1065 
1066 	err = tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE);
1067 	if (err < 0)
1068 		goto errout;
1069 
1070 #ifdef CONFIG_CLS_U32_MARK
1071 	n->pcpu_success = alloc_percpu(u32);
1072 	if (!n->pcpu_success) {
1073 		err = -ENOMEM;
1074 		goto errout;
1075 	}
1076 
1077 	if (tb[TCA_U32_MARK]) {
1078 		struct tc_u32_mark *mark;
1079 
1080 		mark = nla_data(tb[TCA_U32_MARK]);
1081 		n->val = mark->val;
1082 		n->mask = mark->mask;
1083 	}
1084 #endif
1085 
1086 	err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr);
1087 	if (err == 0) {
1088 		struct tc_u_knode __rcu **ins;
1089 		struct tc_u_knode *pins;
1090 
1091 		err = u32_replace_hw_knode(tp, n, flags);
1092 		if (err)
1093 			goto errhw;
1094 
1095 		if (!tc_in_hw(n->flags))
1096 			n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1097 
1098 		ins = &ht->ht[TC_U32_HASH(handle)];
1099 		for (pins = rtnl_dereference(*ins); pins;
1100 		     ins = &pins->next, pins = rtnl_dereference(*ins))
1101 			if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1102 				break;
1103 
1104 		RCU_INIT_POINTER(n->next, pins);
1105 		rcu_assign_pointer(*ins, n);
1106 		*arg = n;
1107 		return 0;
1108 	}
1109 
1110 errhw:
1111 #ifdef CONFIG_CLS_U32_MARK
1112 	free_percpu(n->pcpu_success);
1113 #endif
1114 
1115 errout:
1116 	tcf_exts_destroy(&n->exts);
1117 #ifdef CONFIG_CLS_U32_PERF
1118 errfree:
1119 	free_percpu(n->pf);
1120 #endif
1121 	kfree(n);
1122 erridr:
1123 	idr_remove_ext(&ht->handle_idr, handle);
1124 	return err;
1125 }
1126 
1127 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg)
1128 {
1129 	struct tc_u_common *tp_c = tp->data;
1130 	struct tc_u_hnode *ht;
1131 	struct tc_u_knode *n;
1132 	unsigned int h;
1133 
1134 	if (arg->stop)
1135 		return;
1136 
1137 	for (ht = rtnl_dereference(tp_c->hlist);
1138 	     ht;
1139 	     ht = rtnl_dereference(ht->next)) {
1140 		if (ht->prio != tp->prio)
1141 			continue;
1142 		if (arg->count >= arg->skip) {
1143 			if (arg->fn(tp, ht, arg) < 0) {
1144 				arg->stop = 1;
1145 				return;
1146 			}
1147 		}
1148 		arg->count++;
1149 		for (h = 0; h <= ht->divisor; h++) {
1150 			for (n = rtnl_dereference(ht->ht[h]);
1151 			     n;
1152 			     n = rtnl_dereference(n->next)) {
1153 				if (arg->count < arg->skip) {
1154 					arg->count++;
1155 					continue;
1156 				}
1157 				if (arg->fn(tp, n, arg) < 0) {
1158 					arg->stop = 1;
1159 					return;
1160 				}
1161 				arg->count++;
1162 			}
1163 		}
1164 	}
1165 }
1166 
1167 static void u32_bind_class(void *fh, u32 classid, unsigned long cl)
1168 {
1169 	struct tc_u_knode *n = fh;
1170 
1171 	if (n && n->res.classid == classid)
1172 		n->res.class = cl;
1173 }
1174 
1175 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1176 		    struct sk_buff *skb, struct tcmsg *t)
1177 {
1178 	struct tc_u_knode *n = fh;
1179 	struct tc_u_hnode *ht_up, *ht_down;
1180 	struct nlattr *nest;
1181 
1182 	if (n == NULL)
1183 		return skb->len;
1184 
1185 	t->tcm_handle = n->handle;
1186 
1187 	nest = nla_nest_start(skb, TCA_OPTIONS);
1188 	if (nest == NULL)
1189 		goto nla_put_failure;
1190 
1191 	if (TC_U32_KEY(n->handle) == 0) {
1192 		struct tc_u_hnode *ht = fh;
1193 		u32 divisor = ht->divisor + 1;
1194 
1195 		if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1196 			goto nla_put_failure;
1197 	} else {
1198 #ifdef CONFIG_CLS_U32_PERF
1199 		struct tc_u32_pcnt *gpf;
1200 		int cpu;
1201 #endif
1202 
1203 		if (nla_put(skb, TCA_U32_SEL,
1204 			    sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
1205 			    &n->sel))
1206 			goto nla_put_failure;
1207 
1208 		ht_up = rtnl_dereference(n->ht_up);
1209 		if (ht_up) {
1210 			u32 htid = n->handle & 0xFFFFF000;
1211 			if (nla_put_u32(skb, TCA_U32_HASH, htid))
1212 				goto nla_put_failure;
1213 		}
1214 		if (n->res.classid &&
1215 		    nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1216 			goto nla_put_failure;
1217 
1218 		ht_down = rtnl_dereference(n->ht_down);
1219 		if (ht_down &&
1220 		    nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1221 			goto nla_put_failure;
1222 
1223 		if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1224 			goto nla_put_failure;
1225 
1226 #ifdef CONFIG_CLS_U32_MARK
1227 		if ((n->val || n->mask)) {
1228 			struct tc_u32_mark mark = {.val = n->val,
1229 						   .mask = n->mask,
1230 						   .success = 0};
1231 			int cpum;
1232 
1233 			for_each_possible_cpu(cpum) {
1234 				__u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1235 
1236 				mark.success += cnt;
1237 			}
1238 
1239 			if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1240 				goto nla_put_failure;
1241 		}
1242 #endif
1243 
1244 		if (tcf_exts_dump(skb, &n->exts) < 0)
1245 			goto nla_put_failure;
1246 
1247 #ifdef CONFIG_NET_CLS_IND
1248 		if (n->ifindex) {
1249 			struct net_device *dev;
1250 			dev = __dev_get_by_index(net, n->ifindex);
1251 			if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1252 				goto nla_put_failure;
1253 		}
1254 #endif
1255 #ifdef CONFIG_CLS_U32_PERF
1256 		gpf = kzalloc(sizeof(struct tc_u32_pcnt) +
1257 			      n->sel.nkeys * sizeof(u64),
1258 			      GFP_KERNEL);
1259 		if (!gpf)
1260 			goto nla_put_failure;
1261 
1262 		for_each_possible_cpu(cpu) {
1263 			int i;
1264 			struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1265 
1266 			gpf->rcnt += pf->rcnt;
1267 			gpf->rhit += pf->rhit;
1268 			for (i = 0; i < n->sel.nkeys; i++)
1269 				gpf->kcnts[i] += pf->kcnts[i];
1270 		}
1271 
1272 		if (nla_put_64bit(skb, TCA_U32_PCNT,
1273 				  sizeof(struct tc_u32_pcnt) +
1274 				  n->sel.nkeys * sizeof(u64),
1275 				  gpf, TCA_U32_PAD)) {
1276 			kfree(gpf);
1277 			goto nla_put_failure;
1278 		}
1279 		kfree(gpf);
1280 #endif
1281 	}
1282 
1283 	nla_nest_end(skb, nest);
1284 
1285 	if (TC_U32_KEY(n->handle))
1286 		if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1287 			goto nla_put_failure;
1288 	return skb->len;
1289 
1290 nla_put_failure:
1291 	nla_nest_cancel(skb, nest);
1292 	return -1;
1293 }
1294 
1295 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1296 	.kind		=	"u32",
1297 	.classify	=	u32_classify,
1298 	.init		=	u32_init,
1299 	.destroy	=	u32_destroy,
1300 	.get		=	u32_get,
1301 	.change		=	u32_change,
1302 	.delete		=	u32_delete,
1303 	.walk		=	u32_walk,
1304 	.dump		=	u32_dump,
1305 	.bind_class	=	u32_bind_class,
1306 	.owner		=	THIS_MODULE,
1307 };
1308 
1309 static int __init init_u32(void)
1310 {
1311 	int i, ret;
1312 
1313 	pr_info("u32 classifier\n");
1314 #ifdef CONFIG_CLS_U32_PERF
1315 	pr_info("    Performance counters on\n");
1316 #endif
1317 #ifdef CONFIG_NET_CLS_IND
1318 	pr_info("    input device check on\n");
1319 #endif
1320 #ifdef CONFIG_NET_CLS_ACT
1321 	pr_info("    Actions configured\n");
1322 #endif
1323 	tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1324 					  sizeof(struct hlist_head),
1325 					  GFP_KERNEL);
1326 	if (!tc_u_common_hash)
1327 		return -ENOMEM;
1328 
1329 	for (i = 0; i < U32_HASH_SIZE; i++)
1330 		INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1331 
1332 	ret = register_tcf_proto_ops(&cls_u32_ops);
1333 	if (ret)
1334 		kvfree(tc_u_common_hash);
1335 	return ret;
1336 }
1337 
1338 static void __exit exit_u32(void)
1339 {
1340 	unregister_tcf_proto_ops(&cls_u32_ops);
1341 	kvfree(tc_u_common_hash);
1342 }
1343 
1344 module_init(init_u32)
1345 module_exit(exit_u32)
1346 MODULE_LICENSE("GPL");
1347