xref: /kernel/linux/linux-5.10/net/sched/cls_u32.c (revision 8c2ecf20)
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
43struct 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
69struct 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
85struct 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
94static 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
103static 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
122next_ht:
123	n = rcu_dereference_bh(ht->ht[sel]);
124
125next_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) {
170check_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	}
246out:
247	return -1;
248
249deadloop:
250	net_warn_ratelimited("cls_u32: dead loop\n");
251	return -1;
252}
253
254static 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
267static 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;
281out:
282	return n;
283}
284
285
286static 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 */
306static 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
314static struct hlist_head *tc_u_common_hash;
315
316#define U32_HASH_SHIFT 10
317#define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
318
319static 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
335static struct hlist_head *tc_u_hash(void *key)
336{
337	return tc_u_common_hash + hash_ptr(key, U32_HASH_SHIFT);
338}
339
340static 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
350static 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(sizeof(*tp_c), 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
389static 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
399static 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 */
421static 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 */
438static 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
448static 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
475static 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
490static 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
519static 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
533static 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
572static 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
595static 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
623static 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
657static 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
682out:
683	*last = tp_c->refcnt == 1 && tp_c->knodes == 0;
684	return ret;
685}
686
687static 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
702static 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
713static 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, bool ovr,
717			 struct netlink_ext_ack *extack)
718{
719	int err, ifindex = -1;
720
721	err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr, true, extack);
722	if (err < 0)
723		return err;
724
725	if (tb[TCA_U32_INDEV]) {
726		ifindex = tcf_change_indev(net, tb[TCA_U32_INDEV], extack);
727		if (ifindex < 0)
728			return -EINVAL;
729	}
730
731	if (tb[TCA_U32_LINK]) {
732		u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
733		struct tc_u_hnode *ht_down = NULL, *ht_old;
734
735		if (TC_U32_KEY(handle)) {
736			NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table");
737			return -EINVAL;
738		}
739
740		if (handle) {
741			ht_down = u32_lookup_ht(tp->data, handle);
742
743			if (!ht_down) {
744				NL_SET_ERR_MSG_MOD(extack, "Link hash table not found");
745				return -EINVAL;
746			}
747			if (ht_down->is_root) {
748				NL_SET_ERR_MSG_MOD(extack, "Not linking to root node");
749				return -EINVAL;
750			}
751			ht_down->refcnt++;
752		}
753
754		ht_old = rtnl_dereference(n->ht_down);
755		rcu_assign_pointer(n->ht_down, ht_down);
756
757		if (ht_old)
758			ht_old->refcnt--;
759	}
760	if (tb[TCA_U32_CLASSID]) {
761		n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
762		tcf_bind_filter(tp, &n->res, base);
763	}
764
765	if (ifindex >= 0)
766		n->ifindex = ifindex;
767
768	return 0;
769}
770
771static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
772			      struct tc_u_knode *n)
773{
774	struct tc_u_knode __rcu **ins;
775	struct tc_u_knode *pins;
776	struct tc_u_hnode *ht;
777
778	if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
779		ht = rtnl_dereference(tp->root);
780	else
781		ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
782
783	ins = &ht->ht[TC_U32_HASH(n->handle)];
784
785	/* The node must always exist for it to be replaced if this is not the
786	 * case then something went very wrong elsewhere.
787	 */
788	for (pins = rtnl_dereference(*ins); ;
789	     ins = &pins->next, pins = rtnl_dereference(*ins))
790		if (pins->handle == n->handle)
791			break;
792
793	idr_replace(&ht->handle_idr, n, n->handle);
794	RCU_INIT_POINTER(n->next, pins->next);
795	rcu_assign_pointer(*ins, n);
796}
797
798static struct tc_u_knode *u32_init_knode(struct net *net, struct tcf_proto *tp,
799					 struct tc_u_knode *n)
800{
801	struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
802	struct tc_u32_sel *s = &n->sel;
803	struct tc_u_knode *new;
804
805	new = kzalloc(struct_size(new, sel.keys, s->nkeys), GFP_KERNEL);
806	if (!new)
807		return NULL;
808
809	RCU_INIT_POINTER(new->next, n->next);
810	new->handle = n->handle;
811	RCU_INIT_POINTER(new->ht_up, n->ht_up);
812
813	new->ifindex = n->ifindex;
814	new->fshift = n->fshift;
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
846static 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, bool ovr, bool rtnl_held,
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, flags = 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		flags = nla_get_u32(tb[TCA_U32_FLAGS]);
877		if (!tc_flags_valid(flags)) {
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 ^ flags) &
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], ovr, extack);
904
905		if (err) {
906			__u32_destroy_key(new);
907			return err;
908		}
909
910		err = u32_replace_hw_knode(tp, new, flags, extack);
911		if (err) {
912			__u32_destroy_key(new);
913			return err;
914		}
915
916		if (!tc_in_hw(new->flags))
917			new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
918
919		u32_replace_knode(tp, tp_c, new);
920		tcf_unbind_filter(tp, &n->res);
921		tcf_exts_get_net(&n->exts);
922		tcf_queue_work(&n->rwork, u32_delete_key_work);
923		return 0;
924	}
925
926	if (tb[TCA_U32_DIVISOR]) {
927		unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
928
929		if (!is_power_of_2(divisor)) {
930			NL_SET_ERR_MSG_MOD(extack, "Divisor is not a power of 2");
931			return -EINVAL;
932		}
933		if (divisor-- > 0x100) {
934			NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets");
935			return -EINVAL;
936		}
937		if (TC_U32_KEY(handle)) {
938			NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table");
939			return -EINVAL;
940		}
941		ht = kzalloc(struct_size(ht, ht, divisor + 1), GFP_KERNEL);
942		if (ht == NULL)
943			return -ENOBUFS;
944		if (handle == 0) {
945			handle = gen_new_htid(tp->data, ht);
946			if (handle == 0) {
947				kfree(ht);
948				return -ENOMEM;
949			}
950		} else {
951			err = idr_alloc_u32(&tp_c->handle_idr, ht, &handle,
952					    handle, GFP_KERNEL);
953			if (err) {
954				kfree(ht);
955				return err;
956			}
957		}
958		ht->refcnt = 1;
959		ht->divisor = divisor;
960		ht->handle = handle;
961		ht->prio = tp->prio;
962		idr_init(&ht->handle_idr);
963		ht->flags = flags;
964
965		err = u32_replace_hw_hnode(tp, ht, flags, extack);
966		if (err) {
967			idr_remove(&tp_c->handle_idr, handle);
968			kfree(ht);
969			return err;
970		}
971
972		RCU_INIT_POINTER(ht->next, tp_c->hlist);
973		rcu_assign_pointer(tp_c->hlist, ht);
974		*arg = ht;
975
976		return 0;
977	}
978
979	if (tb[TCA_U32_HASH]) {
980		htid = nla_get_u32(tb[TCA_U32_HASH]);
981		if (TC_U32_HTID(htid) == TC_U32_ROOT) {
982			ht = rtnl_dereference(tp->root);
983			htid = ht->handle;
984		} else {
985			ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
986			if (!ht) {
987				NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found");
988				return -EINVAL;
989			}
990		}
991	} else {
992		ht = rtnl_dereference(tp->root);
993		htid = ht->handle;
994	}
995
996	if (ht->divisor < TC_U32_HASH(htid)) {
997		NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value");
998		return -EINVAL;
999	}
1000
1001	/* At this point, we need to derive the new handle that will be used to
1002	 * uniquely map the identity of this table match entry. The
1003	 * identity of the entry that we need to construct is 32 bits made of:
1004	 *     htid(12b):bucketid(8b):node/entryid(12b)
1005	 *
1006	 * At this point _we have the table(ht)_ in which we will insert this
1007	 * entry. We carry the table's id in variable "htid".
1008	 * Note that earlier code picked the ht selection either by a) the user
1009	 * providing the htid specified via TCA_U32_HASH attribute or b) when
1010	 * no such attribute is passed then the root ht, is default to at ID
1011	 * 0x[800][00][000]. Rule: the root table has a single bucket with ID 0.
1012	 * If OTOH the user passed us the htid, they may also pass a bucketid of
1013	 * choice. 0 is fine. For example a user htid is 0x[600][01][000] it is
1014	 * indicating hash bucketid of 1. Rule: the entry/node ID _cannot_ be
1015	 * passed via the htid, so even if it was non-zero it will be ignored.
1016	 *
1017	 * We may also have a handle, if the user passed one. The handle also
1018	 * carries the same addressing of htid(12b):bucketid(8b):node/entryid(12b).
1019	 * Rule: the bucketid on the handle is ignored even if one was passed;
1020	 * rather the value on "htid" is always assumed to be the bucketid.
1021	 */
1022	if (handle) {
1023		/* Rule: The htid from handle and tableid from htid must match */
1024		if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) {
1025			NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch");
1026			return -EINVAL;
1027		}
1028		/* Ok, so far we have a valid htid(12b):bucketid(8b) but we
1029		 * need to finalize the table entry identification with the last
1030		 * part - the node/entryid(12b)). Rule: Nodeid _cannot be 0_ for
1031		 * entries. Rule: nodeid of 0 is reserved only for tables(see
1032		 * earlier code which processes TC_U32_DIVISOR attribute).
1033		 * Rule: The nodeid can only be derived from the handle (and not
1034		 * htid).
1035		 * Rule: if the handle specified zero for the node id example
1036		 * 0x60000000, then pick a new nodeid from the pool of IDs
1037		 * this hash table has been allocating from.
1038		 * If OTOH it is specified (i.e for example the user passed a
1039		 * handle such as 0x60000123), then we use it generate our final
1040		 * handle which is used to uniquely identify the match entry.
1041		 */
1042		if (!TC_U32_NODE(handle)) {
1043			handle = gen_new_kid(ht, htid);
1044		} else {
1045			handle = htid | TC_U32_NODE(handle);
1046			err = idr_alloc_u32(&ht->handle_idr, NULL, &handle,
1047					    handle, GFP_KERNEL);
1048			if (err)
1049				return err;
1050		}
1051	} else {
1052		/* The user did not give us a handle; lets just generate one
1053		 * from the table's pool of nodeids.
1054		 */
1055		handle = gen_new_kid(ht, htid);
1056	}
1057
1058	if (tb[TCA_U32_SEL] == NULL) {
1059		NL_SET_ERR_MSG_MOD(extack, "Selector not specified");
1060		err = -EINVAL;
1061		goto erridr;
1062	}
1063
1064	s = nla_data(tb[TCA_U32_SEL]);
1065	sel_size = struct_size(s, keys, s->nkeys);
1066	if (nla_len(tb[TCA_U32_SEL]) < sel_size) {
1067		err = -EINVAL;
1068		goto erridr;
1069	}
1070
1071	n = kzalloc(struct_size(n, sel.keys, s->nkeys), GFP_KERNEL);
1072	if (n == NULL) {
1073		err = -ENOBUFS;
1074		goto erridr;
1075	}
1076
1077#ifdef CONFIG_CLS_U32_PERF
1078	n->pf = __alloc_percpu(struct_size(n->pf, kcnts, s->nkeys),
1079			       __alignof__(struct tc_u32_pcnt));
1080	if (!n->pf) {
1081		err = -ENOBUFS;
1082		goto errfree;
1083	}
1084#endif
1085
1086	memcpy(&n->sel, s, sel_size);
1087	RCU_INIT_POINTER(n->ht_up, ht);
1088	n->handle = handle;
1089	n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1090	n->flags = flags;
1091
1092	err = tcf_exts_init(&n->exts, net, TCA_U32_ACT, TCA_U32_POLICE);
1093	if (err < 0)
1094		goto errout;
1095
1096#ifdef CONFIG_CLS_U32_MARK
1097	n->pcpu_success = alloc_percpu(u32);
1098	if (!n->pcpu_success) {
1099		err = -ENOMEM;
1100		goto errout;
1101	}
1102
1103	if (tb[TCA_U32_MARK]) {
1104		struct tc_u32_mark *mark;
1105
1106		mark = nla_data(tb[TCA_U32_MARK]);
1107		n->val = mark->val;
1108		n->mask = mark->mask;
1109	}
1110#endif
1111
1112	err = u32_set_parms(net, tp, base, n, tb, tca[TCA_RATE], ovr,
1113			    extack);
1114	if (err == 0) {
1115		struct tc_u_knode __rcu **ins;
1116		struct tc_u_knode *pins;
1117
1118		err = u32_replace_hw_knode(tp, n, flags, extack);
1119		if (err)
1120			goto errhw;
1121
1122		if (!tc_in_hw(n->flags))
1123			n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1124
1125		ins = &ht->ht[TC_U32_HASH(handle)];
1126		for (pins = rtnl_dereference(*ins); pins;
1127		     ins = &pins->next, pins = rtnl_dereference(*ins))
1128			if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1129				break;
1130
1131		RCU_INIT_POINTER(n->next, pins);
1132		rcu_assign_pointer(*ins, n);
1133		tp_c->knodes++;
1134		*arg = n;
1135		return 0;
1136	}
1137
1138errhw:
1139#ifdef CONFIG_CLS_U32_MARK
1140	free_percpu(n->pcpu_success);
1141#endif
1142
1143errout:
1144	tcf_exts_destroy(&n->exts);
1145#ifdef CONFIG_CLS_U32_PERF
1146errfree:
1147	free_percpu(n->pf);
1148#endif
1149	kfree(n);
1150erridr:
1151	idr_remove(&ht->handle_idr, handle);
1152	return err;
1153}
1154
1155static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg,
1156		     bool rtnl_held)
1157{
1158	struct tc_u_common *tp_c = tp->data;
1159	struct tc_u_hnode *ht;
1160	struct tc_u_knode *n;
1161	unsigned int h;
1162
1163	if (arg->stop)
1164		return;
1165
1166	for (ht = rtnl_dereference(tp_c->hlist);
1167	     ht;
1168	     ht = rtnl_dereference(ht->next)) {
1169		if (ht->prio != tp->prio)
1170			continue;
1171		if (arg->count >= arg->skip) {
1172			if (arg->fn(tp, ht, arg) < 0) {
1173				arg->stop = 1;
1174				return;
1175			}
1176		}
1177		arg->count++;
1178		for (h = 0; h <= ht->divisor; h++) {
1179			for (n = rtnl_dereference(ht->ht[h]);
1180			     n;
1181			     n = rtnl_dereference(n->next)) {
1182				if (arg->count < arg->skip) {
1183					arg->count++;
1184					continue;
1185				}
1186				if (arg->fn(tp, n, arg) < 0) {
1187					arg->stop = 1;
1188					return;
1189				}
1190				arg->count++;
1191			}
1192		}
1193	}
1194}
1195
1196static int u32_reoffload_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
1197			       bool add, flow_setup_cb_t *cb, void *cb_priv,
1198			       struct netlink_ext_ack *extack)
1199{
1200	struct tc_cls_u32_offload cls_u32 = {};
1201	int err;
1202
1203	tc_cls_common_offload_init(&cls_u32.common, tp, ht->flags, extack);
1204	cls_u32.command = add ? TC_CLSU32_NEW_HNODE : TC_CLSU32_DELETE_HNODE;
1205	cls_u32.hnode.divisor = ht->divisor;
1206	cls_u32.hnode.handle = ht->handle;
1207	cls_u32.hnode.prio = ht->prio;
1208
1209	err = cb(TC_SETUP_CLSU32, &cls_u32, cb_priv);
1210	if (err && add && tc_skip_sw(ht->flags))
1211		return err;
1212
1213	return 0;
1214}
1215
1216static int u32_reoffload_knode(struct tcf_proto *tp, struct tc_u_knode *n,
1217			       bool add, flow_setup_cb_t *cb, void *cb_priv,
1218			       struct netlink_ext_ack *extack)
1219{
1220	struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
1221	struct tcf_block *block = tp->chain->block;
1222	struct tc_cls_u32_offload cls_u32 = {};
1223	int err;
1224
1225	tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
1226	cls_u32.command = add ?
1227		TC_CLSU32_REPLACE_KNODE : TC_CLSU32_DELETE_KNODE;
1228	cls_u32.knode.handle = n->handle;
1229
1230	if (add) {
1231		cls_u32.knode.fshift = n->fshift;
1232#ifdef CONFIG_CLS_U32_MARK
1233		cls_u32.knode.val = n->val;
1234		cls_u32.knode.mask = n->mask;
1235#else
1236		cls_u32.knode.val = 0;
1237		cls_u32.knode.mask = 0;
1238#endif
1239		cls_u32.knode.sel = &n->sel;
1240		cls_u32.knode.res = &n->res;
1241		cls_u32.knode.exts = &n->exts;
1242		if (n->ht_down)
1243			cls_u32.knode.link_handle = ht->handle;
1244	}
1245
1246	err = tc_setup_cb_reoffload(block, tp, add, cb, TC_SETUP_CLSU32,
1247				    &cls_u32, cb_priv, &n->flags,
1248				    &n->in_hw_count);
1249	if (err)
1250		return err;
1251
1252	return 0;
1253}
1254
1255static int u32_reoffload(struct tcf_proto *tp, bool add, flow_setup_cb_t *cb,
1256			 void *cb_priv, struct netlink_ext_ack *extack)
1257{
1258	struct tc_u_common *tp_c = tp->data;
1259	struct tc_u_hnode *ht;
1260	struct tc_u_knode *n;
1261	unsigned int h;
1262	int err;
1263
1264	for (ht = rtnl_dereference(tp_c->hlist);
1265	     ht;
1266	     ht = rtnl_dereference(ht->next)) {
1267		if (ht->prio != tp->prio)
1268			continue;
1269
1270		/* When adding filters to a new dev, try to offload the
1271		 * hashtable first. When removing, do the filters before the
1272		 * hashtable.
1273		 */
1274		if (add && !tc_skip_hw(ht->flags)) {
1275			err = u32_reoffload_hnode(tp, ht, add, cb, cb_priv,
1276						  extack);
1277			if (err)
1278				return err;
1279		}
1280
1281		for (h = 0; h <= ht->divisor; h++) {
1282			for (n = rtnl_dereference(ht->ht[h]);
1283			     n;
1284			     n = rtnl_dereference(n->next)) {
1285				if (tc_skip_hw(n->flags))
1286					continue;
1287
1288				err = u32_reoffload_knode(tp, n, add, cb,
1289							  cb_priv, extack);
1290				if (err)
1291					return err;
1292			}
1293		}
1294
1295		if (!add && !tc_skip_hw(ht->flags))
1296			u32_reoffload_hnode(tp, ht, add, cb, cb_priv, extack);
1297	}
1298
1299	return 0;
1300}
1301
1302static void u32_bind_class(void *fh, u32 classid, unsigned long cl, void *q,
1303			   unsigned long base)
1304{
1305	struct tc_u_knode *n = fh;
1306
1307	if (n && n->res.classid == classid) {
1308		if (cl)
1309			__tcf_bind_filter(q, &n->res, base);
1310		else
1311			__tcf_unbind_filter(q, &n->res);
1312	}
1313}
1314
1315static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1316		    struct sk_buff *skb, struct tcmsg *t, bool rtnl_held)
1317{
1318	struct tc_u_knode *n = fh;
1319	struct tc_u_hnode *ht_up, *ht_down;
1320	struct nlattr *nest;
1321
1322	if (n == NULL)
1323		return skb->len;
1324
1325	t->tcm_handle = n->handle;
1326
1327	nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1328	if (nest == NULL)
1329		goto nla_put_failure;
1330
1331	if (TC_U32_KEY(n->handle) == 0) {
1332		struct tc_u_hnode *ht = fh;
1333		u32 divisor = ht->divisor + 1;
1334
1335		if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1336			goto nla_put_failure;
1337	} else {
1338#ifdef CONFIG_CLS_U32_PERF
1339		struct tc_u32_pcnt *gpf;
1340		int cpu;
1341#endif
1342
1343		if (nla_put(skb, TCA_U32_SEL, struct_size(&n->sel, keys, n->sel.nkeys),
1344			    &n->sel))
1345			goto nla_put_failure;
1346
1347		ht_up = rtnl_dereference(n->ht_up);
1348		if (ht_up) {
1349			u32 htid = n->handle & 0xFFFFF000;
1350			if (nla_put_u32(skb, TCA_U32_HASH, htid))
1351				goto nla_put_failure;
1352		}
1353		if (n->res.classid &&
1354		    nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1355			goto nla_put_failure;
1356
1357		ht_down = rtnl_dereference(n->ht_down);
1358		if (ht_down &&
1359		    nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1360			goto nla_put_failure;
1361
1362		if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1363			goto nla_put_failure;
1364
1365#ifdef CONFIG_CLS_U32_MARK
1366		if ((n->val || n->mask)) {
1367			struct tc_u32_mark mark = {.val = n->val,
1368						   .mask = n->mask,
1369						   .success = 0};
1370			int cpum;
1371
1372			for_each_possible_cpu(cpum) {
1373				__u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1374
1375				mark.success += cnt;
1376			}
1377
1378			if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1379				goto nla_put_failure;
1380		}
1381#endif
1382
1383		if (tcf_exts_dump(skb, &n->exts) < 0)
1384			goto nla_put_failure;
1385
1386		if (n->ifindex) {
1387			struct net_device *dev;
1388			dev = __dev_get_by_index(net, n->ifindex);
1389			if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1390				goto nla_put_failure;
1391		}
1392#ifdef CONFIG_CLS_U32_PERF
1393		gpf = kzalloc(struct_size(gpf, kcnts, n->sel.nkeys), GFP_KERNEL);
1394		if (!gpf)
1395			goto nla_put_failure;
1396
1397		for_each_possible_cpu(cpu) {
1398			int i;
1399			struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1400
1401			gpf->rcnt += pf->rcnt;
1402			gpf->rhit += pf->rhit;
1403			for (i = 0; i < n->sel.nkeys; i++)
1404				gpf->kcnts[i] += pf->kcnts[i];
1405		}
1406
1407		if (nla_put_64bit(skb, TCA_U32_PCNT, struct_size(gpf, kcnts, n->sel.nkeys),
1408				  gpf, TCA_U32_PAD)) {
1409			kfree(gpf);
1410			goto nla_put_failure;
1411		}
1412		kfree(gpf);
1413#endif
1414	}
1415
1416	nla_nest_end(skb, nest);
1417
1418	if (TC_U32_KEY(n->handle))
1419		if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1420			goto nla_put_failure;
1421	return skb->len;
1422
1423nla_put_failure:
1424	nla_nest_cancel(skb, nest);
1425	return -1;
1426}
1427
1428static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1429	.kind		=	"u32",
1430	.classify	=	u32_classify,
1431	.init		=	u32_init,
1432	.destroy	=	u32_destroy,
1433	.get		=	u32_get,
1434	.change		=	u32_change,
1435	.delete		=	u32_delete,
1436	.walk		=	u32_walk,
1437	.reoffload	=	u32_reoffload,
1438	.dump		=	u32_dump,
1439	.bind_class	=	u32_bind_class,
1440	.owner		=	THIS_MODULE,
1441};
1442
1443static int __init init_u32(void)
1444{
1445	int i, ret;
1446
1447	pr_info("u32 classifier\n");
1448#ifdef CONFIG_CLS_U32_PERF
1449	pr_info("    Performance counters on\n");
1450#endif
1451	pr_info("    input device check on\n");
1452#ifdef CONFIG_NET_CLS_ACT
1453	pr_info("    Actions configured\n");
1454#endif
1455	tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1456					  sizeof(struct hlist_head),
1457					  GFP_KERNEL);
1458	if (!tc_u_common_hash)
1459		return -ENOMEM;
1460
1461	for (i = 0; i < U32_HASH_SIZE; i++)
1462		INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1463
1464	ret = register_tcf_proto_ops(&cls_u32_ops);
1465	if (ret)
1466		kvfree(tc_u_common_hash);
1467	return ret;
1468}
1469
1470static void __exit exit_u32(void)
1471{
1472	unregister_tcf_proto_ops(&cls_u32_ops);
1473	kvfree(tc_u_common_hash);
1474}
1475
1476module_init(init_u32)
1477module_exit(exit_u32)
1478MODULE_LICENSE("GPL");
1479