xref: /linux/net/sched/cls_u32.c (revision 24bce201d79807b668bf9d9e0aca801c5c0d5f78)
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 
43 struct tc_u_knode {
44 	struct tc_u_knode __rcu	*next;
45 	u32			handle;
46 	struct tc_u_hnode __rcu	*ht_up;
47 	struct tcf_exts		exts;
48 	int			ifindex;
49 	u8			fshift;
50 	struct tcf_result	res;
51 	struct tc_u_hnode __rcu	*ht_down;
52 #ifdef CONFIG_CLS_U32_PERF
53 	struct tc_u32_pcnt __percpu *pf;
54 #endif
55 	u32			flags;
56 	unsigned int		in_hw_count;
57 #ifdef CONFIG_CLS_U32_MARK
58 	u32			val;
59 	u32			mask;
60 	u32 __percpu		*pcpu_success;
61 #endif
62 	struct rcu_work		rwork;
63 	/* The 'sel' field MUST be the last field in structure to allow for
64 	 * tc_u32_keys allocated at end of structure.
65 	 */
66 	struct tc_u32_sel	sel;
67 };
68 
69 struct tc_u_hnode {
70 	struct tc_u_hnode __rcu	*next;
71 	u32			handle;
72 	u32			prio;
73 	int			refcnt;
74 	unsigned int		divisor;
75 	struct idr		handle_idr;
76 	bool			is_root;
77 	struct rcu_head		rcu;
78 	u32			flags;
79 	/* The 'ht' field MUST be the last field in structure to allow for
80 	 * more entries allocated at end of structure.
81 	 */
82 	struct tc_u_knode __rcu	*ht[];
83 };
84 
85 struct tc_u_common {
86 	struct tc_u_hnode __rcu	*hlist;
87 	void			*ptr;
88 	int			refcnt;
89 	struct idr		handle_idr;
90 	struct hlist_node	hnode;
91 	long			knodes;
92 };
93 
94 static inline unsigned int u32_hash_fold(__be32 key,
95 					 const struct tc_u32_sel *sel,
96 					 u8 fshift)
97 {
98 	unsigned int h = ntohl(key & sel->hmask) >> fshift;
99 
100 	return h;
101 }
102 
103 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp,
104 			struct tcf_result *res)
105 {
106 	struct {
107 		struct tc_u_knode *knode;
108 		unsigned int	  off;
109 	} stack[TC_U32_MAXDEPTH];
110 
111 	struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
112 	unsigned int off = skb_network_offset(skb);
113 	struct tc_u_knode *n;
114 	int sdepth = 0;
115 	int off2 = 0;
116 	int sel = 0;
117 #ifdef CONFIG_CLS_U32_PERF
118 	int j;
119 #endif
120 	int i, r;
121 
122 next_ht:
123 	n = rcu_dereference_bh(ht->ht[sel]);
124 
125 next_knode:
126 	if (n) {
127 		struct tc_u32_key *key = n->sel.keys;
128 
129 #ifdef CONFIG_CLS_U32_PERF
130 		__this_cpu_inc(n->pf->rcnt);
131 		j = 0;
132 #endif
133 
134 		if (tc_skip_sw(n->flags)) {
135 			n = rcu_dereference_bh(n->next);
136 			goto next_knode;
137 		}
138 
139 #ifdef CONFIG_CLS_U32_MARK
140 		if ((skb->mark & n->mask) != n->val) {
141 			n = rcu_dereference_bh(n->next);
142 			goto next_knode;
143 		} else {
144 			__this_cpu_inc(*n->pcpu_success);
145 		}
146 #endif
147 
148 		for (i = n->sel.nkeys; i > 0; i--, key++) {
149 			int toff = off + key->off + (off2 & key->offmask);
150 			__be32 *data, hdata;
151 
152 			if (skb_headroom(skb) + toff > INT_MAX)
153 				goto out;
154 
155 			data = skb_header_pointer(skb, toff, 4, &hdata);
156 			if (!data)
157 				goto out;
158 			if ((*data ^ key->val) & key->mask) {
159 				n = rcu_dereference_bh(n->next);
160 				goto next_knode;
161 			}
162 #ifdef CONFIG_CLS_U32_PERF
163 			__this_cpu_inc(n->pf->kcnts[j]);
164 			j++;
165 #endif
166 		}
167 
168 		ht = rcu_dereference_bh(n->ht_down);
169 		if (!ht) {
170 check_terminal:
171 			if (n->sel.flags & TC_U32_TERMINAL) {
172 
173 				*res = n->res;
174 				if (!tcf_match_indev(skb, n->ifindex)) {
175 					n = rcu_dereference_bh(n->next);
176 					goto next_knode;
177 				}
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 *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
255 {
256 	struct tc_u_hnode *ht;
257 
258 	for (ht = rtnl_dereference(tp_c->hlist);
259 	     ht;
260 	     ht = rtnl_dereference(ht->next))
261 		if (ht->handle == handle)
262 			break;
263 
264 	return ht;
265 }
266 
267 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
268 {
269 	unsigned int sel;
270 	struct tc_u_knode *n = NULL;
271 
272 	sel = TC_U32_HASH(handle);
273 	if (sel > ht->divisor)
274 		goto out;
275 
276 	for (n = rtnl_dereference(ht->ht[sel]);
277 	     n;
278 	     n = rtnl_dereference(n->next))
279 		if (n->handle == handle)
280 			break;
281 out:
282 	return n;
283 }
284 
285 
286 static void *u32_get(struct tcf_proto *tp, u32 handle)
287 {
288 	struct tc_u_hnode *ht;
289 	struct tc_u_common *tp_c = tp->data;
290 
291 	if (TC_U32_HTID(handle) == TC_U32_ROOT)
292 		ht = rtnl_dereference(tp->root);
293 	else
294 		ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
295 
296 	if (!ht)
297 		return NULL;
298 
299 	if (TC_U32_KEY(handle) == 0)
300 		return ht;
301 
302 	return u32_lookup_key(ht, handle);
303 }
304 
305 /* Protected by rtnl lock */
306 static u32 gen_new_htid(struct tc_u_common *tp_c, struct tc_u_hnode *ptr)
307 {
308 	int id = idr_alloc_cyclic(&tp_c->handle_idr, ptr, 1, 0x7FF, GFP_KERNEL);
309 	if (id < 0)
310 		return 0;
311 	return (id | 0x800U) << 20;
312 }
313 
314 static struct hlist_head *tc_u_common_hash;
315 
316 #define U32_HASH_SHIFT 10
317 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
318 
319 static void *tc_u_common_ptr(const struct tcf_proto *tp)
320 {
321 	struct tcf_block *block = tp->chain->block;
322 
323 	/* The block sharing is currently supported only
324 	 * for classless qdiscs. In that case we use block
325 	 * for tc_u_common identification. In case the
326 	 * block is not shared, block->q is a valid pointer
327 	 * and we can use that. That works for classful qdiscs.
328 	 */
329 	if (tcf_block_shared(block))
330 		return block;
331 	else
332 		return block->q;
333 }
334 
335 static struct hlist_head *tc_u_hash(void *key)
336 {
337 	return tc_u_common_hash + hash_ptr(key, U32_HASH_SHIFT);
338 }
339 
340 static struct tc_u_common *tc_u_common_find(void *key)
341 {
342 	struct tc_u_common *tc;
343 	hlist_for_each_entry(tc, tc_u_hash(key), hnode) {
344 		if (tc->ptr == key)
345 			return tc;
346 	}
347 	return NULL;
348 }
349 
350 static int u32_init(struct tcf_proto *tp)
351 {
352 	struct tc_u_hnode *root_ht;
353 	void *key = tc_u_common_ptr(tp);
354 	struct tc_u_common *tp_c = tc_u_common_find(key);
355 
356 	root_ht = kzalloc(struct_size(root_ht, ht, 1), GFP_KERNEL);
357 	if (root_ht == NULL)
358 		return -ENOBUFS;
359 
360 	root_ht->refcnt++;
361 	root_ht->handle = tp_c ? gen_new_htid(tp_c, root_ht) : 0x80000000;
362 	root_ht->prio = tp->prio;
363 	root_ht->is_root = true;
364 	idr_init(&root_ht->handle_idr);
365 
366 	if (tp_c == NULL) {
367 		tp_c = kzalloc(struct_size(tp_c, hlist->ht, 1), GFP_KERNEL);
368 		if (tp_c == NULL) {
369 			kfree(root_ht);
370 			return -ENOBUFS;
371 		}
372 		tp_c->ptr = key;
373 		INIT_HLIST_NODE(&tp_c->hnode);
374 		idr_init(&tp_c->handle_idr);
375 
376 		hlist_add_head(&tp_c->hnode, tc_u_hash(key));
377 	}
378 
379 	tp_c->refcnt++;
380 	RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
381 	rcu_assign_pointer(tp_c->hlist, root_ht);
382 
383 	root_ht->refcnt++;
384 	rcu_assign_pointer(tp->root, root_ht);
385 	tp->data = tp_c;
386 	return 0;
387 }
388 
389 static void __u32_destroy_key(struct tc_u_knode *n)
390 {
391 	struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
392 
393 	tcf_exts_destroy(&n->exts);
394 	if (ht && --ht->refcnt == 0)
395 		kfree(ht);
396 	kfree(n);
397 }
398 
399 static void u32_destroy_key(struct tc_u_knode *n, bool free_pf)
400 {
401 	tcf_exts_put_net(&n->exts);
402 #ifdef CONFIG_CLS_U32_PERF
403 	if (free_pf)
404 		free_percpu(n->pf);
405 #endif
406 #ifdef CONFIG_CLS_U32_MARK
407 	if (free_pf)
408 		free_percpu(n->pcpu_success);
409 #endif
410 	__u32_destroy_key(n);
411 }
412 
413 /* u32_delete_key_rcu should be called when free'ing a copied
414  * version of a tc_u_knode obtained from u32_init_knode(). When
415  * copies are obtained from u32_init_knode() the statistics are
416  * shared between the old and new copies to allow readers to
417  * continue to update the statistics during the copy. To support
418  * this the u32_delete_key_rcu variant does not free the percpu
419  * statistics.
420  */
421 static void u32_delete_key_work(struct work_struct *work)
422 {
423 	struct tc_u_knode *key = container_of(to_rcu_work(work),
424 					      struct tc_u_knode,
425 					      rwork);
426 	rtnl_lock();
427 	u32_destroy_key(key, false);
428 	rtnl_unlock();
429 }
430 
431 /* u32_delete_key_freepf_rcu is the rcu callback variant
432  * that free's the entire structure including the statistics
433  * percpu variables. Only use this if the key is not a copy
434  * returned by u32_init_knode(). See u32_delete_key_rcu()
435  * for the variant that should be used with keys return from
436  * u32_init_knode()
437  */
438 static void u32_delete_key_freepf_work(struct work_struct *work)
439 {
440 	struct tc_u_knode *key = container_of(to_rcu_work(work),
441 					      struct tc_u_knode,
442 					      rwork);
443 	rtnl_lock();
444 	u32_destroy_key(key, true);
445 	rtnl_unlock();
446 }
447 
448 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
449 {
450 	struct tc_u_common *tp_c = tp->data;
451 	struct tc_u_knode __rcu **kp;
452 	struct tc_u_knode *pkp;
453 	struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
454 
455 	if (ht) {
456 		kp = &ht->ht[TC_U32_HASH(key->handle)];
457 		for (pkp = rtnl_dereference(*kp); pkp;
458 		     kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
459 			if (pkp == key) {
460 				RCU_INIT_POINTER(*kp, key->next);
461 				tp_c->knodes--;
462 
463 				tcf_unbind_filter(tp, &key->res);
464 				idr_remove(&ht->handle_idr, key->handle);
465 				tcf_exts_get_net(&key->exts);
466 				tcf_queue_work(&key->rwork, u32_delete_key_freepf_work);
467 				return 0;
468 			}
469 		}
470 	}
471 	WARN_ON(1);
472 	return 0;
473 }
474 
475 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
476 			       struct netlink_ext_ack *extack)
477 {
478 	struct tcf_block *block = tp->chain->block;
479 	struct tc_cls_u32_offload cls_u32 = {};
480 
481 	tc_cls_common_offload_init(&cls_u32.common, tp, h->flags, extack);
482 	cls_u32.command = TC_CLSU32_DELETE_HNODE;
483 	cls_u32.hnode.divisor = h->divisor;
484 	cls_u32.hnode.handle = h->handle;
485 	cls_u32.hnode.prio = h->prio;
486 
487 	tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, false, true);
488 }
489 
490 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
491 				u32 flags, struct netlink_ext_ack *extack)
492 {
493 	struct tcf_block *block = tp->chain->block;
494 	struct tc_cls_u32_offload cls_u32 = {};
495 	bool skip_sw = tc_skip_sw(flags);
496 	bool offloaded = false;
497 	int err;
498 
499 	tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
500 	cls_u32.command = TC_CLSU32_NEW_HNODE;
501 	cls_u32.hnode.divisor = h->divisor;
502 	cls_u32.hnode.handle = h->handle;
503 	cls_u32.hnode.prio = h->prio;
504 
505 	err = tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, skip_sw, true);
506 	if (err < 0) {
507 		u32_clear_hw_hnode(tp, h, NULL);
508 		return err;
509 	} else if (err > 0) {
510 		offloaded = true;
511 	}
512 
513 	if (skip_sw && !offloaded)
514 		return -EINVAL;
515 
516 	return 0;
517 }
518 
519 static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
520 				struct netlink_ext_ack *extack)
521 {
522 	struct tcf_block *block = tp->chain->block;
523 	struct tc_cls_u32_offload cls_u32 = {};
524 
525 	tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
526 	cls_u32.command = TC_CLSU32_DELETE_KNODE;
527 	cls_u32.knode.handle = n->handle;
528 
529 	tc_setup_cb_destroy(block, tp, TC_SETUP_CLSU32, &cls_u32, false,
530 			    &n->flags, &n->in_hw_count, true);
531 }
532 
533 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
534 				u32 flags, struct netlink_ext_ack *extack)
535 {
536 	struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
537 	struct tcf_block *block = tp->chain->block;
538 	struct tc_cls_u32_offload cls_u32 = {};
539 	bool skip_sw = tc_skip_sw(flags);
540 	int err;
541 
542 	tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
543 	cls_u32.command = TC_CLSU32_REPLACE_KNODE;
544 	cls_u32.knode.handle = n->handle;
545 	cls_u32.knode.fshift = n->fshift;
546 #ifdef CONFIG_CLS_U32_MARK
547 	cls_u32.knode.val = n->val;
548 	cls_u32.knode.mask = n->mask;
549 #else
550 	cls_u32.knode.val = 0;
551 	cls_u32.knode.mask = 0;
552 #endif
553 	cls_u32.knode.sel = &n->sel;
554 	cls_u32.knode.res = &n->res;
555 	cls_u32.knode.exts = &n->exts;
556 	if (n->ht_down)
557 		cls_u32.knode.link_handle = ht->handle;
558 
559 	err = tc_setup_cb_add(block, tp, TC_SETUP_CLSU32, &cls_u32, skip_sw,
560 			      &n->flags, &n->in_hw_count, true);
561 	if (err) {
562 		u32_remove_hw_knode(tp, n, NULL);
563 		return err;
564 	}
565 
566 	if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
567 		return -EINVAL;
568 
569 	return 0;
570 }
571 
572 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
573 			    struct netlink_ext_ack *extack)
574 {
575 	struct tc_u_common *tp_c = tp->data;
576 	struct tc_u_knode *n;
577 	unsigned int h;
578 
579 	for (h = 0; h <= ht->divisor; h++) {
580 		while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
581 			RCU_INIT_POINTER(ht->ht[h],
582 					 rtnl_dereference(n->next));
583 			tp_c->knodes--;
584 			tcf_unbind_filter(tp, &n->res);
585 			u32_remove_hw_knode(tp, n, extack);
586 			idr_remove(&ht->handle_idr, n->handle);
587 			if (tcf_exts_get_net(&n->exts))
588 				tcf_queue_work(&n->rwork, u32_delete_key_freepf_work);
589 			else
590 				u32_destroy_key(n, true);
591 		}
592 	}
593 }
594 
595 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
596 			     struct netlink_ext_ack *extack)
597 {
598 	struct tc_u_common *tp_c = tp->data;
599 	struct tc_u_hnode __rcu **hn;
600 	struct tc_u_hnode *phn;
601 
602 	WARN_ON(--ht->refcnt);
603 
604 	u32_clear_hnode(tp, ht, extack);
605 
606 	hn = &tp_c->hlist;
607 	for (phn = rtnl_dereference(*hn);
608 	     phn;
609 	     hn = &phn->next, phn = rtnl_dereference(*hn)) {
610 		if (phn == ht) {
611 			u32_clear_hw_hnode(tp, ht, extack);
612 			idr_destroy(&ht->handle_idr);
613 			idr_remove(&tp_c->handle_idr, ht->handle);
614 			RCU_INIT_POINTER(*hn, ht->next);
615 			kfree_rcu(ht, rcu);
616 			return 0;
617 		}
618 	}
619 
620 	return -ENOENT;
621 }
622 
623 static void u32_destroy(struct tcf_proto *tp, bool rtnl_held,
624 			struct netlink_ext_ack *extack)
625 {
626 	struct tc_u_common *tp_c = tp->data;
627 	struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
628 
629 	WARN_ON(root_ht == NULL);
630 
631 	if (root_ht && --root_ht->refcnt == 1)
632 		u32_destroy_hnode(tp, root_ht, extack);
633 
634 	if (--tp_c->refcnt == 0) {
635 		struct tc_u_hnode *ht;
636 
637 		hlist_del(&tp_c->hnode);
638 
639 		while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
640 			u32_clear_hnode(tp, ht, extack);
641 			RCU_INIT_POINTER(tp_c->hlist, ht->next);
642 
643 			/* u32_destroy_key() will later free ht for us, if it's
644 			 * still referenced by some knode
645 			 */
646 			if (--ht->refcnt == 0)
647 				kfree_rcu(ht, rcu);
648 		}
649 
650 		idr_destroy(&tp_c->handle_idr);
651 		kfree(tp_c);
652 	}
653 
654 	tp->data = NULL;
655 }
656 
657 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last,
658 		      bool rtnl_held, struct netlink_ext_ack *extack)
659 {
660 	struct tc_u_hnode *ht = arg;
661 	struct tc_u_common *tp_c = tp->data;
662 	int ret = 0;
663 
664 	if (TC_U32_KEY(ht->handle)) {
665 		u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack);
666 		ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
667 		goto out;
668 	}
669 
670 	if (ht->is_root) {
671 		NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node");
672 		return -EINVAL;
673 	}
674 
675 	if (ht->refcnt == 1) {
676 		u32_destroy_hnode(tp, ht, extack);
677 	} else {
678 		NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter");
679 		return -EBUSY;
680 	}
681 
682 out:
683 	*last = tp_c->refcnt == 1 && tp_c->knodes == 0;
684 	return ret;
685 }
686 
687 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
688 {
689 	u32 index = htid | 0x800;
690 	u32 max = htid | 0xFFF;
691 
692 	if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max, GFP_KERNEL)) {
693 		index = htid + 1;
694 		if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max,
695 				 GFP_KERNEL))
696 			index = max;
697 	}
698 
699 	return index;
700 }
701 
702 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
703 	[TCA_U32_CLASSID]	= { .type = NLA_U32 },
704 	[TCA_U32_HASH]		= { .type = NLA_U32 },
705 	[TCA_U32_LINK]		= { .type = NLA_U32 },
706 	[TCA_U32_DIVISOR]	= { .type = NLA_U32 },
707 	[TCA_U32_SEL]		= { .len = sizeof(struct tc_u32_sel) },
708 	[TCA_U32_INDEV]		= { .type = NLA_STRING, .len = IFNAMSIZ },
709 	[TCA_U32_MARK]		= { .len = sizeof(struct tc_u32_mark) },
710 	[TCA_U32_FLAGS]		= { .type = NLA_U32 },
711 };
712 
713 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
714 			 unsigned long base,
715 			 struct tc_u_knode *n, struct nlattr **tb,
716 			 struct nlattr *est, u32 flags, u32 fl_flags,
717 			 struct netlink_ext_ack *extack)
718 {
719 	int err;
720 
721 	err = tcf_exts_validate_ex(net, tp, tb, est, &n->exts, flags,
722 				   fl_flags, extack);
723 	if (err < 0)
724 		return err;
725 
726 	if (tb[TCA_U32_LINK]) {
727 		u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
728 		struct tc_u_hnode *ht_down = NULL, *ht_old;
729 
730 		if (TC_U32_KEY(handle)) {
731 			NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table");
732 			return -EINVAL;
733 		}
734 
735 		if (handle) {
736 			ht_down = u32_lookup_ht(tp->data, handle);
737 
738 			if (!ht_down) {
739 				NL_SET_ERR_MSG_MOD(extack, "Link hash table not found");
740 				return -EINVAL;
741 			}
742 			if (ht_down->is_root) {
743 				NL_SET_ERR_MSG_MOD(extack, "Not linking to root node");
744 				return -EINVAL;
745 			}
746 			ht_down->refcnt++;
747 		}
748 
749 		ht_old = rtnl_dereference(n->ht_down);
750 		rcu_assign_pointer(n->ht_down, ht_down);
751 
752 		if (ht_old)
753 			ht_old->refcnt--;
754 	}
755 	if (tb[TCA_U32_CLASSID]) {
756 		n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
757 		tcf_bind_filter(tp, &n->res, base);
758 	}
759 
760 	if (tb[TCA_U32_INDEV]) {
761 		int ret;
762 		ret = tcf_change_indev(net, tb[TCA_U32_INDEV], extack);
763 		if (ret < 0)
764 			return -EINVAL;
765 		n->ifindex = ret;
766 	}
767 	return 0;
768 }
769 
770 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
771 			      struct tc_u_knode *n)
772 {
773 	struct tc_u_knode __rcu **ins;
774 	struct tc_u_knode *pins;
775 	struct tc_u_hnode *ht;
776 
777 	if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
778 		ht = rtnl_dereference(tp->root);
779 	else
780 		ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
781 
782 	ins = &ht->ht[TC_U32_HASH(n->handle)];
783 
784 	/* The node must always exist for it to be replaced if this is not the
785 	 * case then something went very wrong elsewhere.
786 	 */
787 	for (pins = rtnl_dereference(*ins); ;
788 	     ins = &pins->next, pins = rtnl_dereference(*ins))
789 		if (pins->handle == n->handle)
790 			break;
791 
792 	idr_replace(&ht->handle_idr, n, n->handle);
793 	RCU_INIT_POINTER(n->next, pins->next);
794 	rcu_assign_pointer(*ins, n);
795 }
796 
797 static struct tc_u_knode *u32_init_knode(struct net *net, struct tcf_proto *tp,
798 					 struct tc_u_knode *n)
799 {
800 	struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
801 	struct tc_u32_sel *s = &n->sel;
802 	struct tc_u_knode *new;
803 
804 	new = kzalloc(struct_size(new, sel.keys, s->nkeys), GFP_KERNEL);
805 	if (!new)
806 		return NULL;
807 
808 	RCU_INIT_POINTER(new->next, n->next);
809 	new->handle = n->handle;
810 	RCU_INIT_POINTER(new->ht_up, n->ht_up);
811 
812 	new->ifindex = n->ifindex;
813 	new->fshift = n->fshift;
814 	new->res = n->res;
815 	new->flags = n->flags;
816 	RCU_INIT_POINTER(new->ht_down, ht);
817 
818 #ifdef CONFIG_CLS_U32_PERF
819 	/* Statistics may be incremented by readers during update
820 	 * so we must keep them in tact. When the node is later destroyed
821 	 * a special destroy call must be made to not free the pf memory.
822 	 */
823 	new->pf = n->pf;
824 #endif
825 
826 #ifdef CONFIG_CLS_U32_MARK
827 	new->val = n->val;
828 	new->mask = n->mask;
829 	/* Similarly success statistics must be moved as pointers */
830 	new->pcpu_success = n->pcpu_success;
831 #endif
832 	memcpy(&new->sel, s, struct_size(s, keys, s->nkeys));
833 
834 	if (tcf_exts_init(&new->exts, net, TCA_U32_ACT, TCA_U32_POLICE)) {
835 		kfree(new);
836 		return NULL;
837 	}
838 
839 	/* bump reference count as long as we hold pointer to structure */
840 	if (ht)
841 		ht->refcnt++;
842 
843 	return new;
844 }
845 
846 static int u32_change(struct net *net, struct sk_buff *in_skb,
847 		      struct tcf_proto *tp, unsigned long base, u32 handle,
848 		      struct nlattr **tca, void **arg, u32 flags,
849 		      struct netlink_ext_ack *extack)
850 {
851 	struct tc_u_common *tp_c = tp->data;
852 	struct tc_u_hnode *ht;
853 	struct tc_u_knode *n;
854 	struct tc_u32_sel *s;
855 	struct nlattr *opt = tca[TCA_OPTIONS];
856 	struct nlattr *tb[TCA_U32_MAX + 1];
857 	u32 htid, userflags = 0;
858 	size_t sel_size;
859 	int err;
860 
861 	if (!opt) {
862 		if (handle) {
863 			NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options");
864 			return -EINVAL;
865 		} else {
866 			return 0;
867 		}
868 	}
869 
870 	err = nla_parse_nested_deprecated(tb, TCA_U32_MAX, opt, u32_policy,
871 					  extack);
872 	if (err < 0)
873 		return err;
874 
875 	if (tb[TCA_U32_FLAGS]) {
876 		userflags = nla_get_u32(tb[TCA_U32_FLAGS]);
877 		if (!tc_flags_valid(userflags)) {
878 			NL_SET_ERR_MSG_MOD(extack, "Invalid filter flags");
879 			return -EINVAL;
880 		}
881 	}
882 
883 	n = *arg;
884 	if (n) {
885 		struct tc_u_knode *new;
886 
887 		if (TC_U32_KEY(n->handle) == 0) {
888 			NL_SET_ERR_MSG_MOD(extack, "Key node id cannot be zero");
889 			return -EINVAL;
890 		}
891 
892 		if ((n->flags ^ userflags) &
893 		    ~(TCA_CLS_FLAGS_IN_HW | TCA_CLS_FLAGS_NOT_IN_HW)) {
894 			NL_SET_ERR_MSG_MOD(extack, "Key node flags do not match passed flags");
895 			return -EINVAL;
896 		}
897 
898 		new = u32_init_knode(net, tp, n);
899 		if (!new)
900 			return -ENOMEM;
901 
902 		err = u32_set_parms(net, tp, base, new, tb,
903 				    tca[TCA_RATE], flags, new->flags,
904 				    extack);
905 
906 		if (err) {
907 			__u32_destroy_key(new);
908 			return err;
909 		}
910 
911 		err = u32_replace_hw_knode(tp, new, flags, extack);
912 		if (err) {
913 			__u32_destroy_key(new);
914 			return err;
915 		}
916 
917 		if (!tc_in_hw(new->flags))
918 			new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
919 
920 		u32_replace_knode(tp, tp_c, new);
921 		tcf_unbind_filter(tp, &n->res);
922 		tcf_exts_get_net(&n->exts);
923 		tcf_queue_work(&n->rwork, u32_delete_key_work);
924 		return 0;
925 	}
926 
927 	if (tb[TCA_U32_DIVISOR]) {
928 		unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
929 
930 		if (!is_power_of_2(divisor)) {
931 			NL_SET_ERR_MSG_MOD(extack, "Divisor is not a power of 2");
932 			return -EINVAL;
933 		}
934 		if (divisor-- > 0x100) {
935 			NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets");
936 			return -EINVAL;
937 		}
938 		if (TC_U32_KEY(handle)) {
939 			NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table");
940 			return -EINVAL;
941 		}
942 		ht = kzalloc(struct_size(ht, ht, divisor + 1), GFP_KERNEL);
943 		if (ht == NULL)
944 			return -ENOBUFS;
945 		if (handle == 0) {
946 			handle = gen_new_htid(tp->data, ht);
947 			if (handle == 0) {
948 				kfree(ht);
949 				return -ENOMEM;
950 			}
951 		} else {
952 			err = idr_alloc_u32(&tp_c->handle_idr, ht, &handle,
953 					    handle, GFP_KERNEL);
954 			if (err) {
955 				kfree(ht);
956 				return err;
957 			}
958 		}
959 		ht->refcnt = 1;
960 		ht->divisor = divisor;
961 		ht->handle = handle;
962 		ht->prio = tp->prio;
963 		idr_init(&ht->handle_idr);
964 		ht->flags = userflags;
965 
966 		err = u32_replace_hw_hnode(tp, ht, userflags, extack);
967 		if (err) {
968 			idr_remove(&tp_c->handle_idr, handle);
969 			kfree(ht);
970 			return err;
971 		}
972 
973 		RCU_INIT_POINTER(ht->next, tp_c->hlist);
974 		rcu_assign_pointer(tp_c->hlist, ht);
975 		*arg = ht;
976 
977 		return 0;
978 	}
979 
980 	if (tb[TCA_U32_HASH]) {
981 		htid = nla_get_u32(tb[TCA_U32_HASH]);
982 		if (TC_U32_HTID(htid) == TC_U32_ROOT) {
983 			ht = rtnl_dereference(tp->root);
984 			htid = ht->handle;
985 		} else {
986 			ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
987 			if (!ht) {
988 				NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found");
989 				return -EINVAL;
990 			}
991 		}
992 	} else {
993 		ht = rtnl_dereference(tp->root);
994 		htid = ht->handle;
995 	}
996 
997 	if (ht->divisor < TC_U32_HASH(htid)) {
998 		NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value");
999 		return -EINVAL;
1000 	}
1001 
1002 	if (handle) {
1003 		if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) {
1004 			NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch");
1005 			return -EINVAL;
1006 		}
1007 		handle = htid | TC_U32_NODE(handle);
1008 		err = idr_alloc_u32(&ht->handle_idr, NULL, &handle, handle,
1009 				    GFP_KERNEL);
1010 		if (err)
1011 			return err;
1012 	} else
1013 		handle = gen_new_kid(ht, htid);
1014 
1015 	if (tb[TCA_U32_SEL] == NULL) {
1016 		NL_SET_ERR_MSG_MOD(extack, "Selector not specified");
1017 		err = -EINVAL;
1018 		goto erridr;
1019 	}
1020 
1021 	s = nla_data(tb[TCA_U32_SEL]);
1022 	sel_size = struct_size(s, keys, s->nkeys);
1023 	if (nla_len(tb[TCA_U32_SEL]) < sel_size) {
1024 		err = -EINVAL;
1025 		goto erridr;
1026 	}
1027 
1028 	n = kzalloc(struct_size(n, sel.keys, s->nkeys), GFP_KERNEL);
1029 	if (n == NULL) {
1030 		err = -ENOBUFS;
1031 		goto erridr;
1032 	}
1033 
1034 #ifdef CONFIG_CLS_U32_PERF
1035 	n->pf = __alloc_percpu(struct_size(n->pf, kcnts, s->nkeys),
1036 			       __alignof__(struct tc_u32_pcnt));
1037 	if (!n->pf) {
1038 		err = -ENOBUFS;
1039 		goto errfree;
1040 	}
1041 #endif
1042 
1043 	memcpy(&n->sel, s, sel_size);
1044 	RCU_INIT_POINTER(n->ht_up, ht);
1045 	n->handle = handle;
1046 	n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1047 	n->flags = userflags;
1048 
1049 	err = tcf_exts_init(&n->exts, net, TCA_U32_ACT, TCA_U32_POLICE);
1050 	if (err < 0)
1051 		goto errout;
1052 
1053 #ifdef CONFIG_CLS_U32_MARK
1054 	n->pcpu_success = alloc_percpu(u32);
1055 	if (!n->pcpu_success) {
1056 		err = -ENOMEM;
1057 		goto errout;
1058 	}
1059 
1060 	if (tb[TCA_U32_MARK]) {
1061 		struct tc_u32_mark *mark;
1062 
1063 		mark = nla_data(tb[TCA_U32_MARK]);
1064 		n->val = mark->val;
1065 		n->mask = mark->mask;
1066 	}
1067 #endif
1068 
1069 	err = u32_set_parms(net, tp, base, n, tb, tca[TCA_RATE],
1070 			    flags, n->flags, extack);
1071 	if (err == 0) {
1072 		struct tc_u_knode __rcu **ins;
1073 		struct tc_u_knode *pins;
1074 
1075 		err = u32_replace_hw_knode(tp, n, flags, extack);
1076 		if (err)
1077 			goto errhw;
1078 
1079 		if (!tc_in_hw(n->flags))
1080 			n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1081 
1082 		ins = &ht->ht[TC_U32_HASH(handle)];
1083 		for (pins = rtnl_dereference(*ins); pins;
1084 		     ins = &pins->next, pins = rtnl_dereference(*ins))
1085 			if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1086 				break;
1087 
1088 		RCU_INIT_POINTER(n->next, pins);
1089 		rcu_assign_pointer(*ins, n);
1090 		tp_c->knodes++;
1091 		*arg = n;
1092 		return 0;
1093 	}
1094 
1095 errhw:
1096 #ifdef CONFIG_CLS_U32_MARK
1097 	free_percpu(n->pcpu_success);
1098 #endif
1099 
1100 errout:
1101 	tcf_exts_destroy(&n->exts);
1102 #ifdef CONFIG_CLS_U32_PERF
1103 errfree:
1104 	free_percpu(n->pf);
1105 #endif
1106 	kfree(n);
1107 erridr:
1108 	idr_remove(&ht->handle_idr, handle);
1109 	return err;
1110 }
1111 
1112 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg,
1113 		     bool rtnl_held)
1114 {
1115 	struct tc_u_common *tp_c = tp->data;
1116 	struct tc_u_hnode *ht;
1117 	struct tc_u_knode *n;
1118 	unsigned int h;
1119 
1120 	if (arg->stop)
1121 		return;
1122 
1123 	for (ht = rtnl_dereference(tp_c->hlist);
1124 	     ht;
1125 	     ht = rtnl_dereference(ht->next)) {
1126 		if (ht->prio != tp->prio)
1127 			continue;
1128 		if (arg->count >= arg->skip) {
1129 			if (arg->fn(tp, ht, arg) < 0) {
1130 				arg->stop = 1;
1131 				return;
1132 			}
1133 		}
1134 		arg->count++;
1135 		for (h = 0; h <= ht->divisor; h++) {
1136 			for (n = rtnl_dereference(ht->ht[h]);
1137 			     n;
1138 			     n = rtnl_dereference(n->next)) {
1139 				if (arg->count < arg->skip) {
1140 					arg->count++;
1141 					continue;
1142 				}
1143 				if (arg->fn(tp, n, arg) < 0) {
1144 					arg->stop = 1;
1145 					return;
1146 				}
1147 				arg->count++;
1148 			}
1149 		}
1150 	}
1151 }
1152 
1153 static int u32_reoffload_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
1154 			       bool add, flow_setup_cb_t *cb, void *cb_priv,
1155 			       struct netlink_ext_ack *extack)
1156 {
1157 	struct tc_cls_u32_offload cls_u32 = {};
1158 	int err;
1159 
1160 	tc_cls_common_offload_init(&cls_u32.common, tp, ht->flags, extack);
1161 	cls_u32.command = add ? TC_CLSU32_NEW_HNODE : TC_CLSU32_DELETE_HNODE;
1162 	cls_u32.hnode.divisor = ht->divisor;
1163 	cls_u32.hnode.handle = ht->handle;
1164 	cls_u32.hnode.prio = ht->prio;
1165 
1166 	err = cb(TC_SETUP_CLSU32, &cls_u32, cb_priv);
1167 	if (err && add && tc_skip_sw(ht->flags))
1168 		return err;
1169 
1170 	return 0;
1171 }
1172 
1173 static int u32_reoffload_knode(struct tcf_proto *tp, struct tc_u_knode *n,
1174 			       bool add, flow_setup_cb_t *cb, void *cb_priv,
1175 			       struct netlink_ext_ack *extack)
1176 {
1177 	struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
1178 	struct tcf_block *block = tp->chain->block;
1179 	struct tc_cls_u32_offload cls_u32 = {};
1180 
1181 	tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
1182 	cls_u32.command = add ?
1183 		TC_CLSU32_REPLACE_KNODE : TC_CLSU32_DELETE_KNODE;
1184 	cls_u32.knode.handle = n->handle;
1185 
1186 	if (add) {
1187 		cls_u32.knode.fshift = n->fshift;
1188 #ifdef CONFIG_CLS_U32_MARK
1189 		cls_u32.knode.val = n->val;
1190 		cls_u32.knode.mask = n->mask;
1191 #else
1192 		cls_u32.knode.val = 0;
1193 		cls_u32.knode.mask = 0;
1194 #endif
1195 		cls_u32.knode.sel = &n->sel;
1196 		cls_u32.knode.res = &n->res;
1197 		cls_u32.knode.exts = &n->exts;
1198 		if (n->ht_down)
1199 			cls_u32.knode.link_handle = ht->handle;
1200 	}
1201 
1202 	return tc_setup_cb_reoffload(block, tp, add, cb, TC_SETUP_CLSU32,
1203 				     &cls_u32, cb_priv, &n->flags,
1204 				     &n->in_hw_count);
1205 }
1206 
1207 static int u32_reoffload(struct tcf_proto *tp, bool add, flow_setup_cb_t *cb,
1208 			 void *cb_priv, struct netlink_ext_ack *extack)
1209 {
1210 	struct tc_u_common *tp_c = tp->data;
1211 	struct tc_u_hnode *ht;
1212 	struct tc_u_knode *n;
1213 	unsigned int h;
1214 	int err;
1215 
1216 	for (ht = rtnl_dereference(tp_c->hlist);
1217 	     ht;
1218 	     ht = rtnl_dereference(ht->next)) {
1219 		if (ht->prio != tp->prio)
1220 			continue;
1221 
1222 		/* When adding filters to a new dev, try to offload the
1223 		 * hashtable first. When removing, do the filters before the
1224 		 * hashtable.
1225 		 */
1226 		if (add && !tc_skip_hw(ht->flags)) {
1227 			err = u32_reoffload_hnode(tp, ht, add, cb, cb_priv,
1228 						  extack);
1229 			if (err)
1230 				return err;
1231 		}
1232 
1233 		for (h = 0; h <= ht->divisor; h++) {
1234 			for (n = rtnl_dereference(ht->ht[h]);
1235 			     n;
1236 			     n = rtnl_dereference(n->next)) {
1237 				if (tc_skip_hw(n->flags))
1238 					continue;
1239 
1240 				err = u32_reoffload_knode(tp, n, add, cb,
1241 							  cb_priv, extack);
1242 				if (err)
1243 					return err;
1244 			}
1245 		}
1246 
1247 		if (!add && !tc_skip_hw(ht->flags))
1248 			u32_reoffload_hnode(tp, ht, add, cb, cb_priv, extack);
1249 	}
1250 
1251 	return 0;
1252 }
1253 
1254 static void u32_bind_class(void *fh, u32 classid, unsigned long cl, void *q,
1255 			   unsigned long base)
1256 {
1257 	struct tc_u_knode *n = fh;
1258 
1259 	if (n && n->res.classid == classid) {
1260 		if (cl)
1261 			__tcf_bind_filter(q, &n->res, base);
1262 		else
1263 			__tcf_unbind_filter(q, &n->res);
1264 	}
1265 }
1266 
1267 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1268 		    struct sk_buff *skb, struct tcmsg *t, bool rtnl_held)
1269 {
1270 	struct tc_u_knode *n = fh;
1271 	struct tc_u_hnode *ht_up, *ht_down;
1272 	struct nlattr *nest;
1273 
1274 	if (n == NULL)
1275 		return skb->len;
1276 
1277 	t->tcm_handle = n->handle;
1278 
1279 	nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1280 	if (nest == NULL)
1281 		goto nla_put_failure;
1282 
1283 	if (TC_U32_KEY(n->handle) == 0) {
1284 		struct tc_u_hnode *ht = fh;
1285 		u32 divisor = ht->divisor + 1;
1286 
1287 		if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1288 			goto nla_put_failure;
1289 	} else {
1290 #ifdef CONFIG_CLS_U32_PERF
1291 		struct tc_u32_pcnt *gpf;
1292 		int cpu;
1293 #endif
1294 
1295 		if (nla_put(skb, TCA_U32_SEL, struct_size(&n->sel, keys, n->sel.nkeys),
1296 			    &n->sel))
1297 			goto nla_put_failure;
1298 
1299 		ht_up = rtnl_dereference(n->ht_up);
1300 		if (ht_up) {
1301 			u32 htid = n->handle & 0xFFFFF000;
1302 			if (nla_put_u32(skb, TCA_U32_HASH, htid))
1303 				goto nla_put_failure;
1304 		}
1305 		if (n->res.classid &&
1306 		    nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1307 			goto nla_put_failure;
1308 
1309 		ht_down = rtnl_dereference(n->ht_down);
1310 		if (ht_down &&
1311 		    nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1312 			goto nla_put_failure;
1313 
1314 		if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1315 			goto nla_put_failure;
1316 
1317 #ifdef CONFIG_CLS_U32_MARK
1318 		if ((n->val || n->mask)) {
1319 			struct tc_u32_mark mark = {.val = n->val,
1320 						   .mask = n->mask,
1321 						   .success = 0};
1322 			int cpum;
1323 
1324 			for_each_possible_cpu(cpum) {
1325 				__u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1326 
1327 				mark.success += cnt;
1328 			}
1329 
1330 			if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1331 				goto nla_put_failure;
1332 		}
1333 #endif
1334 
1335 		if (tcf_exts_dump(skb, &n->exts) < 0)
1336 			goto nla_put_failure;
1337 
1338 		if (n->ifindex) {
1339 			struct net_device *dev;
1340 			dev = __dev_get_by_index(net, n->ifindex);
1341 			if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1342 				goto nla_put_failure;
1343 		}
1344 #ifdef CONFIG_CLS_U32_PERF
1345 		gpf = kzalloc(struct_size(gpf, kcnts, n->sel.nkeys), GFP_KERNEL);
1346 		if (!gpf)
1347 			goto nla_put_failure;
1348 
1349 		for_each_possible_cpu(cpu) {
1350 			int i;
1351 			struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1352 
1353 			gpf->rcnt += pf->rcnt;
1354 			gpf->rhit += pf->rhit;
1355 			for (i = 0; i < n->sel.nkeys; i++)
1356 				gpf->kcnts[i] += pf->kcnts[i];
1357 		}
1358 
1359 		if (nla_put_64bit(skb, TCA_U32_PCNT, struct_size(gpf, kcnts, n->sel.nkeys),
1360 				  gpf, TCA_U32_PAD)) {
1361 			kfree(gpf);
1362 			goto nla_put_failure;
1363 		}
1364 		kfree(gpf);
1365 #endif
1366 	}
1367 
1368 	nla_nest_end(skb, nest);
1369 
1370 	if (TC_U32_KEY(n->handle))
1371 		if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1372 			goto nla_put_failure;
1373 	return skb->len;
1374 
1375 nla_put_failure:
1376 	nla_nest_cancel(skb, nest);
1377 	return -1;
1378 }
1379 
1380 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1381 	.kind		=	"u32",
1382 	.classify	=	u32_classify,
1383 	.init		=	u32_init,
1384 	.destroy	=	u32_destroy,
1385 	.get		=	u32_get,
1386 	.change		=	u32_change,
1387 	.delete		=	u32_delete,
1388 	.walk		=	u32_walk,
1389 	.reoffload	=	u32_reoffload,
1390 	.dump		=	u32_dump,
1391 	.bind_class	=	u32_bind_class,
1392 	.owner		=	THIS_MODULE,
1393 };
1394 
1395 static int __init init_u32(void)
1396 {
1397 	int i, ret;
1398 
1399 	pr_info("u32 classifier\n");
1400 #ifdef CONFIG_CLS_U32_PERF
1401 	pr_info("    Performance counters on\n");
1402 #endif
1403 	pr_info("    input device check on\n");
1404 #ifdef CONFIG_NET_CLS_ACT
1405 	pr_info("    Actions configured\n");
1406 #endif
1407 	tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1408 					  sizeof(struct hlist_head),
1409 					  GFP_KERNEL);
1410 	if (!tc_u_common_hash)
1411 		return -ENOMEM;
1412 
1413 	for (i = 0; i < U32_HASH_SIZE; i++)
1414 		INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1415 
1416 	ret = register_tcf_proto_ops(&cls_u32_ops);
1417 	if (ret)
1418 		kvfree(tc_u_common_hash);
1419 	return ret;
1420 }
1421 
1422 static void __exit exit_u32(void)
1423 {
1424 	unregister_tcf_proto_ops(&cls_u32_ops);
1425 	kvfree(tc_u_common_hash);
1426 }
1427 
1428 module_init(init_u32)
1429 module_exit(exit_u32)
1430 MODULE_LICENSE("GPL");
1431