xref: /linux/net/sched/cls_u32.c (revision 0e2b2a76278153d1ac312b0691cb65dabb9aef3e)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * net/sched/cls_u32.c	Ugly (or Universal) 32bit key Packet Classifier.
4  *
5  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6  *
7  *	The filters are packed to hash tables of key nodes
8  *	with a set of 32bit key/mask pairs at every node.
9  *	Nodes reference next level hash tables etc.
10  *
11  *	This scheme is the best universal classifier I managed to
12  *	invent; it is not super-fast, but it is not slow (provided you
13  *	program it correctly), and general enough.  And its relative
14  *	speed grows as the number of rules becomes larger.
15  *
16  *	It seems that it represents the best middle point between
17  *	speed and manageability both by human and by machine.
18  *
19  *	It is especially useful for link sharing combined with QoS;
20  *	pure RSVP doesn't need such a general approach and can use
21  *	much simpler (and faster) schemes, sort of cls_rsvp.c.
22  *
23  *	nfmark match added by Catalin(ux aka Dino) BOIE <catab at umbrella.ro>
24  */
25 
26 #include <linux/module.h>
27 #include <linux/slab.h>
28 #include <linux/types.h>
29 #include <linux/kernel.h>
30 #include <linux/string.h>
31 #include <linux/errno.h>
32 #include <linux/percpu.h>
33 #include <linux/rtnetlink.h>
34 #include <linux/skbuff.h>
35 #include <linux/bitmap.h>
36 #include <linux/netdevice.h>
37 #include <linux/hash.h>
38 #include <net/netlink.h>
39 #include <net/act_api.h>
40 #include <net/pkt_cls.h>
41 #include <linux/idr.h>
42 #include <net/tc_wrapper.h>
43 
44 struct tc_u_knode {
45 	struct tc_u_knode __rcu	*next;
46 	u32			handle;
47 	struct tc_u_hnode __rcu	*ht_up;
48 	struct tcf_exts		exts;
49 	int			ifindex;
50 	u8			fshift;
51 	struct tcf_result	res;
52 	struct tc_u_hnode __rcu	*ht_down;
53 #ifdef CONFIG_CLS_U32_PERF
54 	struct tc_u32_pcnt __percpu *pf;
55 #endif
56 	u32			flags;
57 	unsigned int		in_hw_count;
58 #ifdef CONFIG_CLS_U32_MARK
59 	u32			val;
60 	u32			mask;
61 	u32 __percpu		*pcpu_success;
62 #endif
63 	struct rcu_work		rwork;
64 	/* The 'sel' field MUST be the last field in structure to allow for
65 	 * tc_u32_keys allocated at end of structure.
66 	 */
67 	struct tc_u32_sel	sel;
68 };
69 
70 struct tc_u_hnode {
71 	struct tc_u_hnode __rcu	*next;
72 	u32			handle;
73 	u32			prio;
74 	int			refcnt;
75 	unsigned int		divisor;
76 	struct idr		handle_idr;
77 	bool			is_root;
78 	struct rcu_head		rcu;
79 	u32			flags;
80 	/* The 'ht' field MUST be the last field in structure to allow for
81 	 * more entries allocated at end of structure.
82 	 */
83 	struct tc_u_knode __rcu	*ht[];
84 };
85 
86 struct tc_u_common {
87 	struct tc_u_hnode __rcu	*hlist;
88 	void			*ptr;
89 	int			refcnt;
90 	struct idr		handle_idr;
91 	struct hlist_node	hnode;
92 	long			knodes;
93 };
94 
95 static inline unsigned int u32_hash_fold(__be32 key,
96 					 const struct tc_u32_sel *sel,
97 					 u8 fshift)
98 {
99 	unsigned int h = ntohl(key & sel->hmask) >> fshift;
100 
101 	return h;
102 }
103 
104 TC_INDIRECT_SCOPE int u32_classify(struct sk_buff *skb,
105 				   const struct tcf_proto *tp,
106 				   struct tcf_result *res)
107 {
108 	struct {
109 		struct tc_u_knode *knode;
110 		unsigned int	  off;
111 	} stack[TC_U32_MAXDEPTH];
112 
113 	struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
114 	unsigned int off = skb_network_offset(skb);
115 	struct tc_u_knode *n;
116 	int sdepth = 0;
117 	int off2 = 0;
118 	int sel = 0;
119 #ifdef CONFIG_CLS_U32_PERF
120 	int j;
121 #endif
122 	int i, r;
123 
124 next_ht:
125 	n = rcu_dereference_bh(ht->ht[sel]);
126 
127 next_knode:
128 	if (n) {
129 		struct tc_u32_key *key = n->sel.keys;
130 
131 #ifdef CONFIG_CLS_U32_PERF
132 		__this_cpu_inc(n->pf->rcnt);
133 		j = 0;
134 #endif
135 
136 		if (tc_skip_sw(n->flags)) {
137 			n = rcu_dereference_bh(n->next);
138 			goto next_knode;
139 		}
140 
141 #ifdef CONFIG_CLS_U32_MARK
142 		if ((skb->mark & n->mask) != n->val) {
143 			n = rcu_dereference_bh(n->next);
144 			goto next_knode;
145 		} else {
146 			__this_cpu_inc(*n->pcpu_success);
147 		}
148 #endif
149 
150 		for (i = n->sel.nkeys; i > 0; i--, key++) {
151 			int toff = off + key->off + (off2 & key->offmask);
152 			__be32 *data, hdata;
153 
154 			if (skb_headroom(skb) + toff > INT_MAX)
155 				goto out;
156 
157 			data = skb_header_pointer(skb, toff, 4, &hdata);
158 			if (!data)
159 				goto out;
160 			if ((*data ^ key->val) & key->mask) {
161 				n = rcu_dereference_bh(n->next);
162 				goto next_knode;
163 			}
164 #ifdef CONFIG_CLS_U32_PERF
165 			__this_cpu_inc(n->pf->kcnts[j]);
166 			j++;
167 #endif
168 		}
169 
170 		ht = rcu_dereference_bh(n->ht_down);
171 		if (!ht) {
172 check_terminal:
173 			if (n->sel.flags & TC_U32_TERMINAL) {
174 
175 				*res = n->res;
176 				if (!tcf_match_indev(skb, n->ifindex)) {
177 					n = rcu_dereference_bh(n->next);
178 					goto next_knode;
179 				}
180 #ifdef CONFIG_CLS_U32_PERF
181 				__this_cpu_inc(n->pf->rhit);
182 #endif
183 				r = tcf_exts_exec(skb, &n->exts, res);
184 				if (r < 0) {
185 					n = rcu_dereference_bh(n->next);
186 					goto next_knode;
187 				}
188 
189 				return r;
190 			}
191 			n = rcu_dereference_bh(n->next);
192 			goto next_knode;
193 		}
194 
195 		/* PUSH */
196 		if (sdepth >= TC_U32_MAXDEPTH)
197 			goto deadloop;
198 		stack[sdepth].knode = n;
199 		stack[sdepth].off = off;
200 		sdepth++;
201 
202 		ht = rcu_dereference_bh(n->ht_down);
203 		sel = 0;
204 		if (ht->divisor) {
205 			__be32 *data, hdata;
206 
207 			data = skb_header_pointer(skb, off + n->sel.hoff, 4,
208 						  &hdata);
209 			if (!data)
210 				goto out;
211 			sel = ht->divisor & u32_hash_fold(*data, &n->sel,
212 							  n->fshift);
213 		}
214 		if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
215 			goto next_ht;
216 
217 		if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
218 			off2 = n->sel.off + 3;
219 			if (n->sel.flags & TC_U32_VAROFFSET) {
220 				__be16 *data, hdata;
221 
222 				data = skb_header_pointer(skb,
223 							  off + n->sel.offoff,
224 							  2, &hdata);
225 				if (!data)
226 					goto out;
227 				off2 += ntohs(n->sel.offmask & *data) >>
228 					n->sel.offshift;
229 			}
230 			off2 &= ~3;
231 		}
232 		if (n->sel.flags & TC_U32_EAT) {
233 			off += off2;
234 			off2 = 0;
235 		}
236 
237 		if (off < skb->len)
238 			goto next_ht;
239 	}
240 
241 	/* POP */
242 	if (sdepth--) {
243 		n = stack[sdepth].knode;
244 		ht = rcu_dereference_bh(n->ht_up);
245 		off = stack[sdepth].off;
246 		goto check_terminal;
247 	}
248 out:
249 	return -1;
250 
251 deadloop:
252 	net_warn_ratelimited("cls_u32: dead loop\n");
253 	return -1;
254 }
255 
256 static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
257 {
258 	struct tc_u_hnode *ht;
259 
260 	for (ht = rtnl_dereference(tp_c->hlist);
261 	     ht;
262 	     ht = rtnl_dereference(ht->next))
263 		if (ht->handle == handle)
264 			break;
265 
266 	return ht;
267 }
268 
269 static struct tc_u_knode *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 void *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 NULL;
300 
301 	if (TC_U32_KEY(handle) == 0)
302 		return ht;
303 
304 	return u32_lookup_key(ht, handle);
305 }
306 
307 /* Protected by rtnl lock */
308 static u32 gen_new_htid(struct tc_u_common *tp_c, struct tc_u_hnode *ptr)
309 {
310 	int id = idr_alloc_cyclic(&tp_c->handle_idr, ptr, 1, 0x7FF, GFP_KERNEL);
311 	if (id < 0)
312 		return 0;
313 	return (id | 0x800U) << 20;
314 }
315 
316 static struct hlist_head *tc_u_common_hash;
317 
318 #define U32_HASH_SHIFT 10
319 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
320 
321 static void *tc_u_common_ptr(const struct tcf_proto *tp)
322 {
323 	struct tcf_block *block = tp->chain->block;
324 
325 	/* The block sharing is currently supported only
326 	 * for classless qdiscs. In that case we use block
327 	 * for tc_u_common identification. In case the
328 	 * block is not shared, block->q is a valid pointer
329 	 * and we can use that. That works for classful qdiscs.
330 	 */
331 	if (tcf_block_shared(block))
332 		return block;
333 	else
334 		return block->q;
335 }
336 
337 static struct hlist_head *tc_u_hash(void *key)
338 {
339 	return tc_u_common_hash + hash_ptr(key, U32_HASH_SHIFT);
340 }
341 
342 static struct tc_u_common *tc_u_common_find(void *key)
343 {
344 	struct tc_u_common *tc;
345 	hlist_for_each_entry(tc, tc_u_hash(key), hnode) {
346 		if (tc->ptr == key)
347 			return tc;
348 	}
349 	return NULL;
350 }
351 
352 static int u32_init(struct tcf_proto *tp)
353 {
354 	struct tc_u_hnode *root_ht;
355 	void *key = tc_u_common_ptr(tp);
356 	struct tc_u_common *tp_c = tc_u_common_find(key);
357 
358 	root_ht = kzalloc(struct_size(root_ht, ht, 1), GFP_KERNEL);
359 	if (root_ht == NULL)
360 		return -ENOBUFS;
361 
362 	root_ht->refcnt++;
363 	root_ht->handle = tp_c ? gen_new_htid(tp_c, root_ht) : 0x80000000;
364 	root_ht->prio = tp->prio;
365 	root_ht->is_root = true;
366 	idr_init(&root_ht->handle_idr);
367 
368 	if (tp_c == NULL) {
369 		tp_c = kzalloc(struct_size(tp_c, hlist->ht, 1), GFP_KERNEL);
370 		if (tp_c == NULL) {
371 			kfree(root_ht);
372 			return -ENOBUFS;
373 		}
374 		tp_c->ptr = key;
375 		INIT_HLIST_NODE(&tp_c->hnode);
376 		idr_init(&tp_c->handle_idr);
377 
378 		hlist_add_head(&tp_c->hnode, tc_u_hash(key));
379 	}
380 
381 	tp_c->refcnt++;
382 	RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
383 	rcu_assign_pointer(tp_c->hlist, root_ht);
384 
385 	root_ht->refcnt++;
386 	rcu_assign_pointer(tp->root, root_ht);
387 	tp->data = tp_c;
388 	return 0;
389 }
390 
391 static void __u32_destroy_key(struct tc_u_knode *n)
392 {
393 	struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
394 
395 	tcf_exts_destroy(&n->exts);
396 	if (ht && --ht->refcnt == 0)
397 		kfree(ht);
398 	kfree(n);
399 }
400 
401 static void u32_destroy_key(struct tc_u_knode *n, bool free_pf)
402 {
403 	tcf_exts_put_net(&n->exts);
404 #ifdef CONFIG_CLS_U32_PERF
405 	if (free_pf)
406 		free_percpu(n->pf);
407 #endif
408 #ifdef CONFIG_CLS_U32_MARK
409 	if (free_pf)
410 		free_percpu(n->pcpu_success);
411 #endif
412 	__u32_destroy_key(n);
413 }
414 
415 /* u32_delete_key_rcu should be called when free'ing a copied
416  * version of a tc_u_knode obtained from u32_init_knode(). When
417  * copies are obtained from u32_init_knode() the statistics are
418  * shared between the old and new copies to allow readers to
419  * continue to update the statistics during the copy. To support
420  * this the u32_delete_key_rcu variant does not free the percpu
421  * statistics.
422  */
423 static void u32_delete_key_work(struct work_struct *work)
424 {
425 	struct tc_u_knode *key = container_of(to_rcu_work(work),
426 					      struct tc_u_knode,
427 					      rwork);
428 	rtnl_lock();
429 	u32_destroy_key(key, false);
430 	rtnl_unlock();
431 }
432 
433 /* u32_delete_key_freepf_rcu is the rcu callback variant
434  * that free's the entire structure including the statistics
435  * percpu variables. Only use this if the key is not a copy
436  * returned by u32_init_knode(). See u32_delete_key_rcu()
437  * for the variant that should be used with keys return from
438  * u32_init_knode()
439  */
440 static void u32_delete_key_freepf_work(struct work_struct *work)
441 {
442 	struct tc_u_knode *key = container_of(to_rcu_work(work),
443 					      struct tc_u_knode,
444 					      rwork);
445 	rtnl_lock();
446 	u32_destroy_key(key, true);
447 	rtnl_unlock();
448 }
449 
450 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
451 {
452 	struct tc_u_common *tp_c = tp->data;
453 	struct tc_u_knode __rcu **kp;
454 	struct tc_u_knode *pkp;
455 	struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
456 
457 	if (ht) {
458 		kp = &ht->ht[TC_U32_HASH(key->handle)];
459 		for (pkp = rtnl_dereference(*kp); pkp;
460 		     kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
461 			if (pkp == key) {
462 				RCU_INIT_POINTER(*kp, key->next);
463 				tp_c->knodes--;
464 
465 				tcf_unbind_filter(tp, &key->res);
466 				idr_remove(&ht->handle_idr, key->handle);
467 				tcf_exts_get_net(&key->exts);
468 				tcf_queue_work(&key->rwork, u32_delete_key_freepf_work);
469 				return 0;
470 			}
471 		}
472 	}
473 	WARN_ON(1);
474 	return 0;
475 }
476 
477 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
478 			       struct netlink_ext_ack *extack)
479 {
480 	struct tcf_block *block = tp->chain->block;
481 	struct tc_cls_u32_offload cls_u32 = {};
482 
483 	tc_cls_common_offload_init(&cls_u32.common, tp, h->flags, extack);
484 	cls_u32.command = TC_CLSU32_DELETE_HNODE;
485 	cls_u32.hnode.divisor = h->divisor;
486 	cls_u32.hnode.handle = h->handle;
487 	cls_u32.hnode.prio = h->prio;
488 
489 	tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, false, true);
490 }
491 
492 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
493 				u32 flags, struct netlink_ext_ack *extack)
494 {
495 	struct tcf_block *block = tp->chain->block;
496 	struct tc_cls_u32_offload cls_u32 = {};
497 	bool skip_sw = tc_skip_sw(flags);
498 	bool offloaded = false;
499 	int err;
500 
501 	tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
502 	cls_u32.command = TC_CLSU32_NEW_HNODE;
503 	cls_u32.hnode.divisor = h->divisor;
504 	cls_u32.hnode.handle = h->handle;
505 	cls_u32.hnode.prio = h->prio;
506 
507 	err = tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, skip_sw, true);
508 	if (err < 0) {
509 		u32_clear_hw_hnode(tp, h, NULL);
510 		return err;
511 	} else if (err > 0) {
512 		offloaded = true;
513 	}
514 
515 	if (skip_sw && !offloaded)
516 		return -EINVAL;
517 
518 	return 0;
519 }
520 
521 static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
522 				struct netlink_ext_ack *extack)
523 {
524 	struct tcf_block *block = tp->chain->block;
525 	struct tc_cls_u32_offload cls_u32 = {};
526 
527 	tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
528 	cls_u32.command = TC_CLSU32_DELETE_KNODE;
529 	cls_u32.knode.handle = n->handle;
530 
531 	tc_setup_cb_destroy(block, tp, TC_SETUP_CLSU32, &cls_u32, false,
532 			    &n->flags, &n->in_hw_count, true);
533 }
534 
535 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
536 				u32 flags, struct netlink_ext_ack *extack)
537 {
538 	struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
539 	struct tcf_block *block = tp->chain->block;
540 	struct tc_cls_u32_offload cls_u32 = {};
541 	bool skip_sw = tc_skip_sw(flags);
542 	int err;
543 
544 	tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
545 	cls_u32.command = TC_CLSU32_REPLACE_KNODE;
546 	cls_u32.knode.handle = n->handle;
547 	cls_u32.knode.fshift = n->fshift;
548 #ifdef CONFIG_CLS_U32_MARK
549 	cls_u32.knode.val = n->val;
550 	cls_u32.knode.mask = n->mask;
551 #else
552 	cls_u32.knode.val = 0;
553 	cls_u32.knode.mask = 0;
554 #endif
555 	cls_u32.knode.sel = &n->sel;
556 	cls_u32.knode.res = &n->res;
557 	cls_u32.knode.exts = &n->exts;
558 	if (n->ht_down)
559 		cls_u32.knode.link_handle = ht->handle;
560 
561 	err = tc_setup_cb_add(block, tp, TC_SETUP_CLSU32, &cls_u32, skip_sw,
562 			      &n->flags, &n->in_hw_count, true);
563 	if (err) {
564 		u32_remove_hw_knode(tp, n, NULL);
565 		return err;
566 	}
567 
568 	if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
569 		return -EINVAL;
570 
571 	return 0;
572 }
573 
574 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
575 			    struct netlink_ext_ack *extack)
576 {
577 	struct tc_u_common *tp_c = tp->data;
578 	struct tc_u_knode *n;
579 	unsigned int h;
580 
581 	for (h = 0; h <= ht->divisor; h++) {
582 		while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
583 			RCU_INIT_POINTER(ht->ht[h],
584 					 rtnl_dereference(n->next));
585 			tp_c->knodes--;
586 			tcf_unbind_filter(tp, &n->res);
587 			u32_remove_hw_knode(tp, n, extack);
588 			idr_remove(&ht->handle_idr, n->handle);
589 			if (tcf_exts_get_net(&n->exts))
590 				tcf_queue_work(&n->rwork, u32_delete_key_freepf_work);
591 			else
592 				u32_destroy_key(n, true);
593 		}
594 	}
595 }
596 
597 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
598 			     struct netlink_ext_ack *extack)
599 {
600 	struct tc_u_common *tp_c = tp->data;
601 	struct tc_u_hnode __rcu **hn;
602 	struct tc_u_hnode *phn;
603 
604 	WARN_ON(--ht->refcnt);
605 
606 	u32_clear_hnode(tp, ht, extack);
607 
608 	hn = &tp_c->hlist;
609 	for (phn = rtnl_dereference(*hn);
610 	     phn;
611 	     hn = &phn->next, phn = rtnl_dereference(*hn)) {
612 		if (phn == ht) {
613 			u32_clear_hw_hnode(tp, ht, extack);
614 			idr_destroy(&ht->handle_idr);
615 			idr_remove(&tp_c->handle_idr, ht->handle);
616 			RCU_INIT_POINTER(*hn, ht->next);
617 			kfree_rcu(ht, rcu);
618 			return 0;
619 		}
620 	}
621 
622 	return -ENOENT;
623 }
624 
625 static void u32_destroy(struct tcf_proto *tp, bool rtnl_held,
626 			struct netlink_ext_ack *extack)
627 {
628 	struct tc_u_common *tp_c = tp->data;
629 	struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
630 
631 	WARN_ON(root_ht == NULL);
632 
633 	if (root_ht && --root_ht->refcnt == 1)
634 		u32_destroy_hnode(tp, root_ht, extack);
635 
636 	if (--tp_c->refcnt == 0) {
637 		struct tc_u_hnode *ht;
638 
639 		hlist_del(&tp_c->hnode);
640 
641 		while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
642 			u32_clear_hnode(tp, ht, extack);
643 			RCU_INIT_POINTER(tp_c->hlist, ht->next);
644 
645 			/* u32_destroy_key() will later free ht for us, if it's
646 			 * still referenced by some knode
647 			 */
648 			if (--ht->refcnt == 0)
649 				kfree_rcu(ht, rcu);
650 		}
651 
652 		idr_destroy(&tp_c->handle_idr);
653 		kfree(tp_c);
654 	}
655 
656 	tp->data = NULL;
657 }
658 
659 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last,
660 		      bool rtnl_held, struct netlink_ext_ack *extack)
661 {
662 	struct tc_u_hnode *ht = arg;
663 	struct tc_u_common *tp_c = tp->data;
664 	int ret = 0;
665 
666 	if (TC_U32_KEY(ht->handle)) {
667 		u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack);
668 		ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
669 		goto out;
670 	}
671 
672 	if (ht->is_root) {
673 		NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node");
674 		return -EINVAL;
675 	}
676 
677 	if (ht->refcnt == 1) {
678 		u32_destroy_hnode(tp, ht, extack);
679 	} else {
680 		NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter");
681 		return -EBUSY;
682 	}
683 
684 out:
685 	*last = tp_c->refcnt == 1 && tp_c->knodes == 0;
686 	return ret;
687 }
688 
689 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
690 {
691 	u32 index = htid | 0x800;
692 	u32 max = htid | 0xFFF;
693 
694 	if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max, GFP_KERNEL)) {
695 		index = htid + 1;
696 		if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max,
697 				 GFP_KERNEL))
698 			index = max;
699 	}
700 
701 	return index;
702 }
703 
704 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
705 	[TCA_U32_CLASSID]	= { .type = NLA_U32 },
706 	[TCA_U32_HASH]		= { .type = NLA_U32 },
707 	[TCA_U32_LINK]		= { .type = NLA_U32 },
708 	[TCA_U32_DIVISOR]	= { .type = NLA_U32 },
709 	[TCA_U32_SEL]		= { .len = sizeof(struct tc_u32_sel) },
710 	[TCA_U32_INDEV]		= { .type = NLA_STRING, .len = IFNAMSIZ },
711 	[TCA_U32_MARK]		= { .len = sizeof(struct tc_u32_mark) },
712 	[TCA_U32_FLAGS]		= { .type = NLA_U32 },
713 };
714 
715 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
716 			 unsigned long base,
717 			 struct tc_u_knode *n, struct nlattr **tb,
718 			 struct nlattr *est, u32 flags, u32 fl_flags,
719 			 struct netlink_ext_ack *extack)
720 {
721 	int err, ifindex = -1;
722 
723 	err = tcf_exts_validate_ex(net, tp, tb, est, &n->exts, flags,
724 				   fl_flags, extack);
725 	if (err < 0)
726 		return err;
727 
728 	if (tb[TCA_U32_INDEV]) {
729 		ifindex = tcf_change_indev(net, tb[TCA_U32_INDEV], extack);
730 		if (ifindex < 0)
731 			return -EINVAL;
732 	}
733 
734 	if (tb[TCA_U32_LINK]) {
735 		u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
736 		struct tc_u_hnode *ht_down = NULL, *ht_old;
737 
738 		if (TC_U32_KEY(handle)) {
739 			NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table");
740 			return -EINVAL;
741 		}
742 
743 		if (handle) {
744 			ht_down = u32_lookup_ht(tp->data, handle);
745 
746 			if (!ht_down) {
747 				NL_SET_ERR_MSG_MOD(extack, "Link hash table not found");
748 				return -EINVAL;
749 			}
750 			if (ht_down->is_root) {
751 				NL_SET_ERR_MSG_MOD(extack, "Not linking to root node");
752 				return -EINVAL;
753 			}
754 			ht_down->refcnt++;
755 		}
756 
757 		ht_old = rtnl_dereference(n->ht_down);
758 		rcu_assign_pointer(n->ht_down, ht_down);
759 
760 		if (ht_old)
761 			ht_old->refcnt--;
762 	}
763 	if (tb[TCA_U32_CLASSID]) {
764 		n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
765 		tcf_bind_filter(tp, &n->res, base);
766 	}
767 
768 	if (ifindex >= 0)
769 		n->ifindex = ifindex;
770 
771 	return 0;
772 }
773 
774 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
775 			      struct tc_u_knode *n)
776 {
777 	struct tc_u_knode __rcu **ins;
778 	struct tc_u_knode *pins;
779 	struct tc_u_hnode *ht;
780 
781 	if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
782 		ht = rtnl_dereference(tp->root);
783 	else
784 		ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
785 
786 	ins = &ht->ht[TC_U32_HASH(n->handle)];
787 
788 	/* The node must always exist for it to be replaced if this is not the
789 	 * case then something went very wrong elsewhere.
790 	 */
791 	for (pins = rtnl_dereference(*ins); ;
792 	     ins = &pins->next, pins = rtnl_dereference(*ins))
793 		if (pins->handle == n->handle)
794 			break;
795 
796 	idr_replace(&ht->handle_idr, n, n->handle);
797 	RCU_INIT_POINTER(n->next, pins->next);
798 	rcu_assign_pointer(*ins, n);
799 }
800 
801 static struct tc_u_knode *u32_init_knode(struct net *net, struct tcf_proto *tp,
802 					 struct tc_u_knode *n)
803 {
804 	struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
805 	struct tc_u32_sel *s = &n->sel;
806 	struct tc_u_knode *new;
807 
808 	new = kzalloc(struct_size(new, sel.keys, s->nkeys), GFP_KERNEL);
809 	if (!new)
810 		return NULL;
811 
812 	RCU_INIT_POINTER(new->next, n->next);
813 	new->handle = n->handle;
814 	RCU_INIT_POINTER(new->ht_up, n->ht_up);
815 
816 	new->ifindex = n->ifindex;
817 	new->fshift = n->fshift;
818 	new->res = n->res;
819 	new->flags = n->flags;
820 	RCU_INIT_POINTER(new->ht_down, ht);
821 
822 #ifdef CONFIG_CLS_U32_PERF
823 	/* Statistics may be incremented by readers during update
824 	 * so we must keep them in tact. When the node is later destroyed
825 	 * a special destroy call must be made to not free the pf memory.
826 	 */
827 	new->pf = n->pf;
828 #endif
829 
830 #ifdef CONFIG_CLS_U32_MARK
831 	new->val = n->val;
832 	new->mask = n->mask;
833 	/* Similarly success statistics must be moved as pointers */
834 	new->pcpu_success = n->pcpu_success;
835 #endif
836 	memcpy(&new->sel, s, struct_size(s, keys, s->nkeys));
837 
838 	if (tcf_exts_init(&new->exts, net, TCA_U32_ACT, TCA_U32_POLICE)) {
839 		kfree(new);
840 		return NULL;
841 	}
842 
843 	/* bump reference count as long as we hold pointer to structure */
844 	if (ht)
845 		ht->refcnt++;
846 
847 	return new;
848 }
849 
850 static int u32_change(struct net *net, struct sk_buff *in_skb,
851 		      struct tcf_proto *tp, unsigned long base, u32 handle,
852 		      struct nlattr **tca, void **arg, u32 flags,
853 		      struct netlink_ext_ack *extack)
854 {
855 	struct tc_u_common *tp_c = tp->data;
856 	struct tc_u_hnode *ht;
857 	struct tc_u_knode *n;
858 	struct tc_u32_sel *s;
859 	struct nlattr *opt = tca[TCA_OPTIONS];
860 	struct nlattr *tb[TCA_U32_MAX + 1];
861 	u32 htid, userflags = 0;
862 	size_t sel_size;
863 	int err;
864 
865 	if (!opt) {
866 		if (handle) {
867 			NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options");
868 			return -EINVAL;
869 		} else {
870 			return 0;
871 		}
872 	}
873 
874 	err = nla_parse_nested_deprecated(tb, TCA_U32_MAX, opt, u32_policy,
875 					  extack);
876 	if (err < 0)
877 		return err;
878 
879 	if (tb[TCA_U32_FLAGS]) {
880 		userflags = nla_get_u32(tb[TCA_U32_FLAGS]);
881 		if (!tc_flags_valid(userflags)) {
882 			NL_SET_ERR_MSG_MOD(extack, "Invalid filter flags");
883 			return -EINVAL;
884 		}
885 	}
886 
887 	n = *arg;
888 	if (n) {
889 		struct tc_u_knode *new;
890 
891 		if (TC_U32_KEY(n->handle) == 0) {
892 			NL_SET_ERR_MSG_MOD(extack, "Key node id cannot be zero");
893 			return -EINVAL;
894 		}
895 
896 		if ((n->flags ^ userflags) &
897 		    ~(TCA_CLS_FLAGS_IN_HW | TCA_CLS_FLAGS_NOT_IN_HW)) {
898 			NL_SET_ERR_MSG_MOD(extack, "Key node flags do not match passed flags");
899 			return -EINVAL;
900 		}
901 
902 		new = u32_init_knode(net, tp, n);
903 		if (!new)
904 			return -ENOMEM;
905 
906 		err = u32_set_parms(net, tp, base, new, tb,
907 				    tca[TCA_RATE], flags, new->flags,
908 				    extack);
909 
910 		if (err) {
911 			__u32_destroy_key(new);
912 			return err;
913 		}
914 
915 		err = u32_replace_hw_knode(tp, new, flags, extack);
916 		if (err) {
917 			__u32_destroy_key(new);
918 			return err;
919 		}
920 
921 		if (!tc_in_hw(new->flags))
922 			new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
923 
924 		u32_replace_knode(tp, tp_c, new);
925 		tcf_unbind_filter(tp, &n->res);
926 		tcf_exts_get_net(&n->exts);
927 		tcf_queue_work(&n->rwork, u32_delete_key_work);
928 		return 0;
929 	}
930 
931 	if (tb[TCA_U32_DIVISOR]) {
932 		unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
933 
934 		if (!is_power_of_2(divisor)) {
935 			NL_SET_ERR_MSG_MOD(extack, "Divisor is not a power of 2");
936 			return -EINVAL;
937 		}
938 		if (divisor-- > 0x100) {
939 			NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets");
940 			return -EINVAL;
941 		}
942 		if (TC_U32_KEY(handle)) {
943 			NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table");
944 			return -EINVAL;
945 		}
946 		ht = kzalloc(struct_size(ht, ht, divisor + 1), GFP_KERNEL);
947 		if (ht == NULL)
948 			return -ENOBUFS;
949 		if (handle == 0) {
950 			handle = gen_new_htid(tp->data, ht);
951 			if (handle == 0) {
952 				kfree(ht);
953 				return -ENOMEM;
954 			}
955 		} else {
956 			err = idr_alloc_u32(&tp_c->handle_idr, ht, &handle,
957 					    handle, GFP_KERNEL);
958 			if (err) {
959 				kfree(ht);
960 				return err;
961 			}
962 		}
963 		ht->refcnt = 1;
964 		ht->divisor = divisor;
965 		ht->handle = handle;
966 		ht->prio = tp->prio;
967 		idr_init(&ht->handle_idr);
968 		ht->flags = userflags;
969 
970 		err = u32_replace_hw_hnode(tp, ht, userflags, extack);
971 		if (err) {
972 			idr_remove(&tp_c->handle_idr, handle);
973 			kfree(ht);
974 			return err;
975 		}
976 
977 		RCU_INIT_POINTER(ht->next, tp_c->hlist);
978 		rcu_assign_pointer(tp_c->hlist, ht);
979 		*arg = ht;
980 
981 		return 0;
982 	}
983 
984 	if (tb[TCA_U32_HASH]) {
985 		htid = nla_get_u32(tb[TCA_U32_HASH]);
986 		if (TC_U32_HTID(htid) == TC_U32_ROOT) {
987 			ht = rtnl_dereference(tp->root);
988 			htid = ht->handle;
989 		} else {
990 			ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
991 			if (!ht) {
992 				NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found");
993 				return -EINVAL;
994 			}
995 		}
996 	} else {
997 		ht = rtnl_dereference(tp->root);
998 		htid = ht->handle;
999 	}
1000 
1001 	if (ht->divisor < TC_U32_HASH(htid)) {
1002 		NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value");
1003 		return -EINVAL;
1004 	}
1005 
1006 	if (handle) {
1007 		if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) {
1008 			NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch");
1009 			return -EINVAL;
1010 		}
1011 		handle = htid | TC_U32_NODE(handle);
1012 		err = idr_alloc_u32(&ht->handle_idr, NULL, &handle, handle,
1013 				    GFP_KERNEL);
1014 		if (err)
1015 			return err;
1016 	} else
1017 		handle = gen_new_kid(ht, htid);
1018 
1019 	if (tb[TCA_U32_SEL] == NULL) {
1020 		NL_SET_ERR_MSG_MOD(extack, "Selector not specified");
1021 		err = -EINVAL;
1022 		goto erridr;
1023 	}
1024 
1025 	s = nla_data(tb[TCA_U32_SEL]);
1026 	sel_size = struct_size(s, keys, s->nkeys);
1027 	if (nla_len(tb[TCA_U32_SEL]) < sel_size) {
1028 		err = -EINVAL;
1029 		goto erridr;
1030 	}
1031 
1032 	n = kzalloc(struct_size(n, sel.keys, s->nkeys), GFP_KERNEL);
1033 	if (n == NULL) {
1034 		err = -ENOBUFS;
1035 		goto erridr;
1036 	}
1037 
1038 #ifdef CONFIG_CLS_U32_PERF
1039 	n->pf = __alloc_percpu(struct_size(n->pf, kcnts, s->nkeys),
1040 			       __alignof__(struct tc_u32_pcnt));
1041 	if (!n->pf) {
1042 		err = -ENOBUFS;
1043 		goto errfree;
1044 	}
1045 #endif
1046 
1047 	unsafe_memcpy(&n->sel, s, sel_size,
1048 		      /* A composite flex-array structure destination,
1049 		       * which was correctly sized with struct_size(),
1050 		       * bounds-checked against nla_len(), and allocated
1051 		       * above. */);
1052 	RCU_INIT_POINTER(n->ht_up, ht);
1053 	n->handle = handle;
1054 	n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1055 	n->flags = userflags;
1056 
1057 	err = tcf_exts_init(&n->exts, net, TCA_U32_ACT, TCA_U32_POLICE);
1058 	if (err < 0)
1059 		goto errout;
1060 
1061 #ifdef CONFIG_CLS_U32_MARK
1062 	n->pcpu_success = alloc_percpu(u32);
1063 	if (!n->pcpu_success) {
1064 		err = -ENOMEM;
1065 		goto errout;
1066 	}
1067 
1068 	if (tb[TCA_U32_MARK]) {
1069 		struct tc_u32_mark *mark;
1070 
1071 		mark = nla_data(tb[TCA_U32_MARK]);
1072 		n->val = mark->val;
1073 		n->mask = mark->mask;
1074 	}
1075 #endif
1076 
1077 	err = u32_set_parms(net, tp, base, n, tb, tca[TCA_RATE],
1078 			    flags, n->flags, extack);
1079 	if (err == 0) {
1080 		struct tc_u_knode __rcu **ins;
1081 		struct tc_u_knode *pins;
1082 
1083 		err = u32_replace_hw_knode(tp, n, flags, extack);
1084 		if (err)
1085 			goto errhw;
1086 
1087 		if (!tc_in_hw(n->flags))
1088 			n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1089 
1090 		ins = &ht->ht[TC_U32_HASH(handle)];
1091 		for (pins = rtnl_dereference(*ins); pins;
1092 		     ins = &pins->next, pins = rtnl_dereference(*ins))
1093 			if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1094 				break;
1095 
1096 		RCU_INIT_POINTER(n->next, pins);
1097 		rcu_assign_pointer(*ins, n);
1098 		tp_c->knodes++;
1099 		*arg = n;
1100 		return 0;
1101 	}
1102 
1103 errhw:
1104 #ifdef CONFIG_CLS_U32_MARK
1105 	free_percpu(n->pcpu_success);
1106 #endif
1107 
1108 errout:
1109 	tcf_exts_destroy(&n->exts);
1110 #ifdef CONFIG_CLS_U32_PERF
1111 errfree:
1112 	free_percpu(n->pf);
1113 #endif
1114 	kfree(n);
1115 erridr:
1116 	idr_remove(&ht->handle_idr, handle);
1117 	return err;
1118 }
1119 
1120 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg,
1121 		     bool rtnl_held)
1122 {
1123 	struct tc_u_common *tp_c = tp->data;
1124 	struct tc_u_hnode *ht;
1125 	struct tc_u_knode *n;
1126 	unsigned int h;
1127 
1128 	if (arg->stop)
1129 		return;
1130 
1131 	for (ht = rtnl_dereference(tp_c->hlist);
1132 	     ht;
1133 	     ht = rtnl_dereference(ht->next)) {
1134 		if (ht->prio != tp->prio)
1135 			continue;
1136 
1137 		if (!tc_cls_stats_dump(tp, arg, ht))
1138 			return;
1139 
1140 		for (h = 0; h <= ht->divisor; h++) {
1141 			for (n = rtnl_dereference(ht->ht[h]);
1142 			     n;
1143 			     n = rtnl_dereference(n->next)) {
1144 				if (!tc_cls_stats_dump(tp, arg, n))
1145 					return;
1146 			}
1147 		}
1148 	}
1149 }
1150 
1151 static int u32_reoffload_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
1152 			       bool add, flow_setup_cb_t *cb, void *cb_priv,
1153 			       struct netlink_ext_ack *extack)
1154 {
1155 	struct tc_cls_u32_offload cls_u32 = {};
1156 	int err;
1157 
1158 	tc_cls_common_offload_init(&cls_u32.common, tp, ht->flags, extack);
1159 	cls_u32.command = add ? TC_CLSU32_NEW_HNODE : TC_CLSU32_DELETE_HNODE;
1160 	cls_u32.hnode.divisor = ht->divisor;
1161 	cls_u32.hnode.handle = ht->handle;
1162 	cls_u32.hnode.prio = ht->prio;
1163 
1164 	err = cb(TC_SETUP_CLSU32, &cls_u32, cb_priv);
1165 	if (err && add && tc_skip_sw(ht->flags))
1166 		return err;
1167 
1168 	return 0;
1169 }
1170 
1171 static int u32_reoffload_knode(struct tcf_proto *tp, struct tc_u_knode *n,
1172 			       bool add, flow_setup_cb_t *cb, void *cb_priv,
1173 			       struct netlink_ext_ack *extack)
1174 {
1175 	struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
1176 	struct tcf_block *block = tp->chain->block;
1177 	struct tc_cls_u32_offload cls_u32 = {};
1178 
1179 	tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
1180 	cls_u32.command = add ?
1181 		TC_CLSU32_REPLACE_KNODE : TC_CLSU32_DELETE_KNODE;
1182 	cls_u32.knode.handle = n->handle;
1183 
1184 	if (add) {
1185 		cls_u32.knode.fshift = n->fshift;
1186 #ifdef CONFIG_CLS_U32_MARK
1187 		cls_u32.knode.val = n->val;
1188 		cls_u32.knode.mask = n->mask;
1189 #else
1190 		cls_u32.knode.val = 0;
1191 		cls_u32.knode.mask = 0;
1192 #endif
1193 		cls_u32.knode.sel = &n->sel;
1194 		cls_u32.knode.res = &n->res;
1195 		cls_u32.knode.exts = &n->exts;
1196 		if (n->ht_down)
1197 			cls_u32.knode.link_handle = ht->handle;
1198 	}
1199 
1200 	return tc_setup_cb_reoffload(block, tp, add, cb, TC_SETUP_CLSU32,
1201 				     &cls_u32, cb_priv, &n->flags,
1202 				     &n->in_hw_count);
1203 }
1204 
1205 static int u32_reoffload(struct tcf_proto *tp, bool add, flow_setup_cb_t *cb,
1206 			 void *cb_priv, struct netlink_ext_ack *extack)
1207 {
1208 	struct tc_u_common *tp_c = tp->data;
1209 	struct tc_u_hnode *ht;
1210 	struct tc_u_knode *n;
1211 	unsigned int h;
1212 	int err;
1213 
1214 	for (ht = rtnl_dereference(tp_c->hlist);
1215 	     ht;
1216 	     ht = rtnl_dereference(ht->next)) {
1217 		if (ht->prio != tp->prio)
1218 			continue;
1219 
1220 		/* When adding filters to a new dev, try to offload the
1221 		 * hashtable first. When removing, do the filters before the
1222 		 * hashtable.
1223 		 */
1224 		if (add && !tc_skip_hw(ht->flags)) {
1225 			err = u32_reoffload_hnode(tp, ht, add, cb, cb_priv,
1226 						  extack);
1227 			if (err)
1228 				return err;
1229 		}
1230 
1231 		for (h = 0; h <= ht->divisor; h++) {
1232 			for (n = rtnl_dereference(ht->ht[h]);
1233 			     n;
1234 			     n = rtnl_dereference(n->next)) {
1235 				if (tc_skip_hw(n->flags))
1236 					continue;
1237 
1238 				err = u32_reoffload_knode(tp, n, add, cb,
1239 							  cb_priv, extack);
1240 				if (err)
1241 					return err;
1242 			}
1243 		}
1244 
1245 		if (!add && !tc_skip_hw(ht->flags))
1246 			u32_reoffload_hnode(tp, ht, add, cb, cb_priv, extack);
1247 	}
1248 
1249 	return 0;
1250 }
1251 
1252 static void u32_bind_class(void *fh, u32 classid, unsigned long cl, void *q,
1253 			   unsigned long base)
1254 {
1255 	struct tc_u_knode *n = fh;
1256 
1257 	tc_cls_bind_class(classid, cl, q, &n->res, base);
1258 }
1259 
1260 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1261 		    struct sk_buff *skb, struct tcmsg *t, bool rtnl_held)
1262 {
1263 	struct tc_u_knode *n = fh;
1264 	struct tc_u_hnode *ht_up, *ht_down;
1265 	struct nlattr *nest;
1266 
1267 	if (n == NULL)
1268 		return skb->len;
1269 
1270 	t->tcm_handle = n->handle;
1271 
1272 	nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1273 	if (nest == NULL)
1274 		goto nla_put_failure;
1275 
1276 	if (TC_U32_KEY(n->handle) == 0) {
1277 		struct tc_u_hnode *ht = fh;
1278 		u32 divisor = ht->divisor + 1;
1279 
1280 		if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1281 			goto nla_put_failure;
1282 	} else {
1283 #ifdef CONFIG_CLS_U32_PERF
1284 		struct tc_u32_pcnt *gpf;
1285 		int cpu;
1286 #endif
1287 
1288 		if (nla_put(skb, TCA_U32_SEL, struct_size(&n->sel, keys, n->sel.nkeys),
1289 			    &n->sel))
1290 			goto nla_put_failure;
1291 
1292 		ht_up = rtnl_dereference(n->ht_up);
1293 		if (ht_up) {
1294 			u32 htid = n->handle & 0xFFFFF000;
1295 			if (nla_put_u32(skb, TCA_U32_HASH, htid))
1296 				goto nla_put_failure;
1297 		}
1298 		if (n->res.classid &&
1299 		    nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1300 			goto nla_put_failure;
1301 
1302 		ht_down = rtnl_dereference(n->ht_down);
1303 		if (ht_down &&
1304 		    nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1305 			goto nla_put_failure;
1306 
1307 		if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1308 			goto nla_put_failure;
1309 
1310 #ifdef CONFIG_CLS_U32_MARK
1311 		if ((n->val || n->mask)) {
1312 			struct tc_u32_mark mark = {.val = n->val,
1313 						   .mask = n->mask,
1314 						   .success = 0};
1315 			int cpum;
1316 
1317 			for_each_possible_cpu(cpum) {
1318 				__u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1319 
1320 				mark.success += cnt;
1321 			}
1322 
1323 			if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1324 				goto nla_put_failure;
1325 		}
1326 #endif
1327 
1328 		if (tcf_exts_dump(skb, &n->exts) < 0)
1329 			goto nla_put_failure;
1330 
1331 		if (n->ifindex) {
1332 			struct net_device *dev;
1333 			dev = __dev_get_by_index(net, n->ifindex);
1334 			if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1335 				goto nla_put_failure;
1336 		}
1337 #ifdef CONFIG_CLS_U32_PERF
1338 		gpf = kzalloc(struct_size(gpf, kcnts, n->sel.nkeys), GFP_KERNEL);
1339 		if (!gpf)
1340 			goto nla_put_failure;
1341 
1342 		for_each_possible_cpu(cpu) {
1343 			int i;
1344 			struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1345 
1346 			gpf->rcnt += pf->rcnt;
1347 			gpf->rhit += pf->rhit;
1348 			for (i = 0; i < n->sel.nkeys; i++)
1349 				gpf->kcnts[i] += pf->kcnts[i];
1350 		}
1351 
1352 		if (nla_put_64bit(skb, TCA_U32_PCNT, struct_size(gpf, kcnts, n->sel.nkeys),
1353 				  gpf, TCA_U32_PAD)) {
1354 			kfree(gpf);
1355 			goto nla_put_failure;
1356 		}
1357 		kfree(gpf);
1358 #endif
1359 	}
1360 
1361 	nla_nest_end(skb, nest);
1362 
1363 	if (TC_U32_KEY(n->handle))
1364 		if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1365 			goto nla_put_failure;
1366 	return skb->len;
1367 
1368 nla_put_failure:
1369 	nla_nest_cancel(skb, nest);
1370 	return -1;
1371 }
1372 
1373 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1374 	.kind		=	"u32",
1375 	.classify	=	u32_classify,
1376 	.init		=	u32_init,
1377 	.destroy	=	u32_destroy,
1378 	.get		=	u32_get,
1379 	.change		=	u32_change,
1380 	.delete		=	u32_delete,
1381 	.walk		=	u32_walk,
1382 	.reoffload	=	u32_reoffload,
1383 	.dump		=	u32_dump,
1384 	.bind_class	=	u32_bind_class,
1385 	.owner		=	THIS_MODULE,
1386 };
1387 
1388 static int __init init_u32(void)
1389 {
1390 	int i, ret;
1391 
1392 	pr_info("u32 classifier\n");
1393 #ifdef CONFIG_CLS_U32_PERF
1394 	pr_info("    Performance counters on\n");
1395 #endif
1396 	pr_info("    input device check on\n");
1397 #ifdef CONFIG_NET_CLS_ACT
1398 	pr_info("    Actions configured\n");
1399 #endif
1400 	tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1401 					  sizeof(struct hlist_head),
1402 					  GFP_KERNEL);
1403 	if (!tc_u_common_hash)
1404 		return -ENOMEM;
1405 
1406 	for (i = 0; i < U32_HASH_SIZE; i++)
1407 		INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1408 
1409 	ret = register_tcf_proto_ops(&cls_u32_ops);
1410 	if (ret)
1411 		kvfree(tc_u_common_hash);
1412 	return ret;
1413 }
1414 
1415 static void __exit exit_u32(void)
1416 {
1417 	unregister_tcf_proto_ops(&cls_u32_ops);
1418 	kvfree(tc_u_common_hash);
1419 }
1420 
1421 module_init(init_u32)
1422 module_exit(exit_u32)
1423 MODULE_LICENSE("GPL");
1424