Linux Audio

Check our new training course

Loading...
v3.1
 
  1/*
  2 * net/sched/cls_u32.c	Ugly (or Universal) 32bit key Packet Classifier.
  3 *
  4 *		This program is free software; you can redistribute it and/or
  5 *		modify it under the terms of the GNU General Public License
  6 *		as published by the Free Software Foundation; either version
  7 *		2 of the License, or (at your option) any later version.
  8 *
  9 * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
 10 *
 11 *	The filters are packed to hash tables of key nodes
 12 *	with a set of 32bit key/mask pairs at every node.
 13 *	Nodes reference next level hash tables etc.
 14 *
 15 *	This scheme is the best universal classifier I managed to
 16 *	invent; it is not super-fast, but it is not slow (provided you
 17 *	program it correctly), and general enough.  And its relative
 18 *	speed grows as the number of rules becomes larger.
 19 *
 20 *	It seems that it represents the best middle point between
 21 *	speed and manageability both by human and by machine.
 22 *
 23 *	It is especially useful for link sharing combined with QoS;
 24 *	pure RSVP doesn't need such a general approach and can use
 25 *	much simpler (and faster) schemes, sort of cls_rsvp.c.
 26 *
 27 *	JHS: We should remove the CONFIG_NET_CLS_IND from here
 28 *	eventually when the meta match extension is made available
 29 *
 30 *	nfmark match added by Catalin(ux aka Dino) BOIE <catab at umbrella.ro>
 31 */
 32
 33#include <linux/module.h>
 34#include <linux/slab.h>
 35#include <linux/types.h>
 36#include <linux/kernel.h>
 37#include <linux/string.h>
 38#include <linux/errno.h>
 
 39#include <linux/rtnetlink.h>
 40#include <linux/skbuff.h>
 
 
 
 41#include <net/netlink.h>
 42#include <net/act_api.h>
 43#include <net/pkt_cls.h>
 
 44
 45struct tc_u_knode {
 46	struct tc_u_knode	*next;
 47	u32			handle;
 48	struct tc_u_hnode	*ht_up;
 49	struct tcf_exts		exts;
 50#ifdef CONFIG_NET_CLS_IND
 51	char                     indev[IFNAMSIZ];
 52#endif
 53	u8			fshift;
 54	struct tcf_result	res;
 55	struct tc_u_hnode	*ht_down;
 56#ifdef CONFIG_CLS_U32_PERF
 57	struct tc_u32_pcnt	*pf;
 58#endif
 
 
 59#ifdef CONFIG_CLS_U32_MARK
 60	struct tc_u32_mark	mark;
 
 
 61#endif
 
 
 
 
 62	struct tc_u32_sel	sel;
 63};
 64
 65struct tc_u_hnode {
 66	struct tc_u_hnode	*next;
 67	u32			handle;
 68	u32			prio;
 69	struct tc_u_common	*tp_c;
 70	int			refcnt;
 71	unsigned int		divisor;
 72	struct tc_u_knode	*ht[1];
 
 
 
 
 
 
 
 73};
 74
 75struct tc_u_common {
 76	struct tc_u_hnode	*hlist;
 77	struct Qdisc		*q;
 78	int			refcnt;
 79	u32			hgenerator;
 80};
 81
 82static const struct tcf_ext_map u32_ext_map = {
 83	.action = TCA_U32_ACT,
 84	.police = TCA_U32_POLICE
 85};
 86
 87static inline unsigned int u32_hash_fold(__be32 key,
 88					 const struct tc_u32_sel *sel,
 89					 u8 fshift)
 90{
 91	unsigned int h = ntohl(key & sel->hmask) >> fshift;
 92
 93	return h;
 94}
 95
 96static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp, struct tcf_result *res)
 
 97{
 98	struct {
 99		struct tc_u_knode *knode;
100		unsigned int	  off;
101	} stack[TC_U32_MAXDEPTH];
102
103	struct tc_u_hnode *ht = (struct tc_u_hnode *)tp->root;
104	unsigned int off = skb_network_offset(skb);
105	struct tc_u_knode *n;
106	int sdepth = 0;
107	int off2 = 0;
108	int sel = 0;
109#ifdef CONFIG_CLS_U32_PERF
110	int j;
111#endif
112	int i, r;
113
114next_ht:
115	n = ht->ht[sel];
116
117next_knode:
118	if (n) {
119		struct tc_u32_key *key = n->sel.keys;
120
121#ifdef CONFIG_CLS_U32_PERF
122		n->pf->rcnt += 1;
123		j = 0;
124#endif
125
 
 
 
 
 
126#ifdef CONFIG_CLS_U32_MARK
127		if ((skb->mark & n->mark.mask) != n->mark.val) {
128			n = n->next;
129			goto next_knode;
130		} else {
131			n->mark.success++;
132		}
133#endif
134
135		for (i = n->sel.nkeys; i > 0; i--, key++) {
136			int toff = off + key->off + (off2 & key->offmask);
137			__be32 *data, hdata;
138
139			if (skb_headroom(skb) + toff > INT_MAX)
140				goto out;
141
142			data = skb_header_pointer(skb, toff, 4, &hdata);
143			if (!data)
144				goto out;
145			if ((*data ^ key->val) & key->mask) {
146				n = n->next;
147				goto next_knode;
148			}
149#ifdef CONFIG_CLS_U32_PERF
150			n->pf->kcnts[j] += 1;
151			j++;
152#endif
153		}
154		if (n->ht_down == NULL) {
 
 
155check_terminal:
156			if (n->sel.flags & TC_U32_TERMINAL) {
157
158				*res = n->res;
159#ifdef CONFIG_NET_CLS_IND
160				if (!tcf_match_indev(skb, n->indev)) {
161					n = n->next;
162					goto next_knode;
163				}
164#endif
165#ifdef CONFIG_CLS_U32_PERF
166				n->pf->rhit += 1;
167#endif
168				r = tcf_exts_exec(skb, &n->exts, res);
169				if (r < 0) {
170					n = n->next;
171					goto next_knode;
172				}
173
174				return r;
175			}
176			n = n->next;
177			goto next_knode;
178		}
179
180		/* PUSH */
181		if (sdepth >= TC_U32_MAXDEPTH)
182			goto deadloop;
183		stack[sdepth].knode = n;
184		stack[sdepth].off = off;
185		sdepth++;
186
187		ht = n->ht_down;
188		sel = 0;
189		if (ht->divisor) {
190			__be32 *data, hdata;
191
192			data = skb_header_pointer(skb, off + n->sel.hoff, 4,
193						  &hdata);
194			if (!data)
195				goto out;
196			sel = ht->divisor & u32_hash_fold(*data, &n->sel,
197							  n->fshift);
198		}
199		if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
200			goto next_ht;
201
202		if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
203			off2 = n->sel.off + 3;
204			if (n->sel.flags & TC_U32_VAROFFSET) {
205				__be16 *data, hdata;
206
207				data = skb_header_pointer(skb,
208							  off + n->sel.offoff,
209							  2, &hdata);
210				if (!data)
211					goto out;
212				off2 += ntohs(n->sel.offmask & *data) >>
213					n->sel.offshift;
214			}
215			off2 &= ~3;
216		}
217		if (n->sel.flags & TC_U32_EAT) {
218			off += off2;
219			off2 = 0;
220		}
221
222		if (off < skb->len)
223			goto next_ht;
224	}
225
226	/* POP */
227	if (sdepth--) {
228		n = stack[sdepth].knode;
229		ht = n->ht_up;
230		off = stack[sdepth].off;
231		goto check_terminal;
232	}
233out:
234	return -1;
235
236deadloop:
237	if (net_ratelimit())
238		pr_warning("cls_u32: dead loop\n");
239	return -1;
240}
241
242static struct tc_u_hnode *
243u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
244{
245	struct tc_u_hnode *ht;
246
247	for (ht = tp_c->hlist; ht; ht = ht->next)
 
 
248		if (ht->handle == handle)
249			break;
250
251	return ht;
252}
253
254static struct tc_u_knode *
255u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
256{
257	unsigned int sel;
258	struct tc_u_knode *n = NULL;
259
260	sel = TC_U32_HASH(handle);
261	if (sel > ht->divisor)
262		goto out;
263
264	for (n = ht->ht[sel]; n; n = n->next)
 
 
265		if (n->handle == handle)
266			break;
267out:
268	return n;
269}
270
271
272static unsigned long u32_get(struct tcf_proto *tp, u32 handle)
273{
274	struct tc_u_hnode *ht;
275	struct tc_u_common *tp_c = tp->data;
276
277	if (TC_U32_HTID(handle) == TC_U32_ROOT)
278		ht = tp->root;
279	else
280		ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
281
282	if (!ht)
283		return 0;
284
285	if (TC_U32_KEY(handle) == 0)
286		return (unsigned long)ht;
287
288	return (unsigned long)u32_lookup_key(ht, handle);
289}
290
291static void u32_put(struct tcf_proto *tp, unsigned long f)
 
292{
 
 
 
 
293}
294
295static u32 gen_new_htid(struct tc_u_common *tp_c)
 
 
 
 
 
296{
297	int i = 0x800;
298
299	do {
300		if (++tp_c->hgenerator == 0x7FF)
301			tp_c->hgenerator = 1;
302	} while (--i > 0 && u32_lookup_ht(tp_c, (tp_c->hgenerator|0x800)<<20));
 
 
 
 
 
 
 
303
304	return i > 0 ? (tp_c->hgenerator|0x800)<<20 : 0;
 
 
 
 
 
 
 
 
 
 
 
 
305}
306
307static int u32_init(struct tcf_proto *tp)
308{
309	struct tc_u_hnode *root_ht;
310	struct tc_u_common *tp_c;
311
312	tp_c = tp->q->u32_node;
313
314	root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
315	if (root_ht == NULL)
316		return -ENOBUFS;
317
318	root_ht->divisor = 0;
319	root_ht->refcnt++;
320	root_ht->handle = tp_c ? gen_new_htid(tp_c) : 0x80000000;
321	root_ht->prio = tp->prio;
 
 
322
323	if (tp_c == NULL) {
324		tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
325		if (tp_c == NULL) {
326			kfree(root_ht);
327			return -ENOBUFS;
328		}
329		tp_c->q = tp->q;
330		tp->q->u32_node = tp_c;
 
 
 
331	}
332
333	tp_c->refcnt++;
334	root_ht->next = tp_c->hlist;
335	tp_c->hlist = root_ht;
336	root_ht->tp_c = tp_c;
337
338	tp->root = root_ht;
 
339	tp->data = tp_c;
340	return 0;
341}
342
343static int u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n)
344{
345	tcf_unbind_filter(tp, &n->res);
346	tcf_exts_destroy(tp, &n->exts);
347	if (n->ht_down)
348		n->ht_down->refcnt--;
 
 
349#ifdef CONFIG_CLS_U32_PERF
350	kfree(n->pf);
 
 
 
 
 
351#endif
352	kfree(n);
353	return 0;
354}
355
356static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode* key)
 
 
 
 
 
 
 
 
357{
358	struct tc_u_knode **kp;
359	struct tc_u_hnode *ht = key->ht_up;
 
 
 
 
 
360
361	if (ht) {
362		for (kp = &ht->ht[TC_U32_HASH(key->handle)]; *kp; kp = &(*kp)->next) {
363			if (*kp == key) {
364				tcf_tree_lock(tp);
365				*kp = key->next;
366				tcf_tree_unlock(tp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
367
368				u32_destroy_key(tp, key);
 
 
 
 
 
 
 
 
 
 
 
369				return 0;
370			}
371		}
372	}
373	WARN_ON(1);
374	return 0;
375}
376
377static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
378{
 
379	struct tc_u_knode *n;
380	unsigned int h;
381
382	for (h = 0; h <= ht->divisor; h++) {
383		while ((n = ht->ht[h]) != NULL) {
384			ht->ht[h] = n->next;
385
386			u32_destroy_key(tp, n);
 
 
 
 
 
 
 
387		}
388	}
389}
390
391static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
 
392{
393	struct tc_u_common *tp_c = tp->data;
394	struct tc_u_hnode **hn;
 
395
396	WARN_ON(ht->refcnt);
397
398	u32_clear_hnode(tp, ht);
399
400	for (hn = &tp_c->hlist; *hn; hn = &(*hn)->next) {
401		if (*hn == ht) {
402			*hn = ht->next;
403			kfree(ht);
 
 
 
 
 
 
404			return 0;
405		}
406	}
407
408	WARN_ON(1);
409	return -ENOENT;
410}
411
412static void u32_destroy(struct tcf_proto *tp)
 
413{
414	struct tc_u_common *tp_c = tp->data;
415	struct tc_u_hnode *root_ht = tp->root;
416
417	WARN_ON(root_ht == NULL);
418
419	if (root_ht && --root_ht->refcnt == 0)
420		u32_destroy_hnode(tp, root_ht);
421
422	if (--tp_c->refcnt == 0) {
423		struct tc_u_hnode *ht;
424
425		tp->q->u32_node = NULL;
426
427		for (ht = tp_c->hlist; ht; ht = ht->next) {
428			ht->refcnt--;
429			u32_clear_hnode(tp, ht);
430		}
431
432		while ((ht = tp_c->hlist) != NULL) {
433			tp_c->hlist = ht->next;
434
435			WARN_ON(ht->refcnt != 0);
436
437			kfree(ht);
438		}
439
 
440		kfree(tp_c);
441	}
442
443	tp->data = NULL;
444}
445
446static int u32_delete(struct tcf_proto *tp, unsigned long arg)
 
447{
448	struct tc_u_hnode *ht = (struct tc_u_hnode *)arg;
449
450	if (ht == NULL)
451		return 0;
452
453	if (TC_U32_KEY(ht->handle))
454		return u32_delete_key(tp, (struct tc_u_knode *)ht);
 
 
 
455
456	if (tp->root == ht)
 
457		return -EINVAL;
 
458
459	if (ht->refcnt == 1) {
460		ht->refcnt--;
461		u32_destroy_hnode(tp, ht);
462	} else {
 
463		return -EBUSY;
464	}
465
466	return 0;
 
 
467}
468
469static u32 gen_new_kid(struct tc_u_hnode *ht, u32 handle)
470{
471	struct tc_u_knode *n;
472	unsigned int i = 0x7FF;
473
474	for (n = ht->ht[TC_U32_HASH(handle)]; n; n = n->next)
475		if (i < TC_U32_NODE(n->handle))
476			i = TC_U32_NODE(n->handle);
477	i++;
 
 
478
479	return handle | (i > 0xFFF ? 0xFFF : i);
480}
481
482static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
483	[TCA_U32_CLASSID]	= { .type = NLA_U32 },
484	[TCA_U32_HASH]		= { .type = NLA_U32 },
485	[TCA_U32_LINK]		= { .type = NLA_U32 },
486	[TCA_U32_DIVISOR]	= { .type = NLA_U32 },
487	[TCA_U32_SEL]		= { .len = sizeof(struct tc_u32_sel) },
488	[TCA_U32_INDEV]		= { .type = NLA_STRING, .len = IFNAMSIZ },
489	[TCA_U32_MARK]		= { .len = sizeof(struct tc_u32_mark) },
 
490};
491
492static int u32_set_parms(struct tcf_proto *tp, unsigned long base,
493			 struct tc_u_hnode *ht,
494			 struct tc_u_knode *n, struct nlattr **tb,
495			 struct nlattr *est)
 
496{
497	int err;
498	struct tcf_exts e;
499
500	err = tcf_exts_validate(tp, tb, est, &e, &u32_ext_map);
501	if (err < 0)
502		return err;
503
504	err = -EINVAL;
505	if (tb[TCA_U32_LINK]) {
506		u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
507		struct tc_u_hnode *ht_down = NULL, *ht_old;
508
509		if (TC_U32_KEY(handle))
510			goto errout;
 
 
511
512		if (handle) {
513			ht_down = u32_lookup_ht(ht->tp_c, handle);
514
515			if (ht_down == NULL)
516				goto errout;
 
 
 
 
 
 
517			ht_down->refcnt++;
518		}
519
520		tcf_tree_lock(tp);
521		ht_old = n->ht_down;
522		n->ht_down = ht_down;
523		tcf_tree_unlock(tp);
524
525		if (ht_old)
526			ht_old->refcnt--;
527	}
528	if (tb[TCA_U32_CLASSID]) {
529		n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
530		tcf_bind_filter(tp, &n->res, base);
531	}
532
533#ifdef CONFIG_NET_CLS_IND
534	if (tb[TCA_U32_INDEV]) {
535		err = tcf_change_indev(tp, n->indev, tb[TCA_U32_INDEV]);
536		if (err < 0)
537			goto errout;
 
 
538	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
539#endif
540	tcf_exts_change(tp, &n->exts, &e);
541
542	return 0;
543errout:
544	tcf_exts_destroy(tp, &e);
545	return err;
 
 
 
 
 
 
 
 
 
 
546}
547
548static int u32_change(struct tcf_proto *tp, unsigned long base, u32 handle,
549		      struct nlattr **tca,
550		      unsigned long *arg)
 
551{
552	struct tc_u_common *tp_c = tp->data;
553	struct tc_u_hnode *ht;
554	struct tc_u_knode *n;
555	struct tc_u32_sel *s;
556	struct nlattr *opt = tca[TCA_OPTIONS];
557	struct nlattr *tb[TCA_U32_MAX + 1];
558	u32 htid;
 
559	int err;
560
561	if (opt == NULL)
562		return handle ? -EINVAL : 0;
 
 
 
 
 
 
563
564	err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy);
 
565	if (err < 0)
566		return err;
567
568	n = (struct tc_u_knode *)*arg;
 
 
 
 
 
 
 
 
569	if (n) {
570		if (TC_U32_KEY(n->handle) == 0)
 
 
 
571			return -EINVAL;
 
 
 
 
 
 
 
572
573		return u32_set_parms(tp, base, n->ht_up, n, tb, tca[TCA_RATE]);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
574	}
575
576	if (tb[TCA_U32_DIVISOR]) {
577		unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
578
579		if (--divisor > 0x100)
 
580			return -EINVAL;
581		if (TC_U32_KEY(handle))
 
 
 
 
 
 
582			return -EINVAL;
583		if (handle == 0) {
584			handle = gen_new_htid(tp->data);
585			if (handle == 0)
586				return -ENOMEM;
587		}
588		ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
589		if (ht == NULL)
590			return -ENOBUFS;
591		ht->tp_c = tp_c;
 
 
 
 
 
 
 
 
 
 
 
 
 
592		ht->refcnt = 1;
593		ht->divisor = divisor;
594		ht->handle = handle;
595		ht->prio = tp->prio;
596		ht->next = tp_c->hlist;
597		tp_c->hlist = ht;
598		*arg = (unsigned long)ht;
 
 
 
 
 
 
 
 
 
 
 
599		return 0;
600	}
601
602	if (tb[TCA_U32_HASH]) {
603		htid = nla_get_u32(tb[TCA_U32_HASH]);
604		if (TC_U32_HTID(htid) == TC_U32_ROOT) {
605			ht = tp->root;
606			htid = ht->handle;
607		} else {
608			ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
609			if (ht == NULL)
 
610				return -EINVAL;
 
611		}
612	} else {
613		ht = tp->root;
614		htid = ht->handle;
615	}
616
617	if (ht->divisor < TC_U32_HASH(htid))
 
618		return -EINVAL;
 
619
620	if (handle) {
621		if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid))
 
622			return -EINVAL;
 
623		handle = htid | TC_U32_NODE(handle);
 
 
 
 
624	} else
625		handle = gen_new_kid(ht, htid);
626
627	if (tb[TCA_U32_SEL] == NULL)
628		return -EINVAL;
 
 
 
629
630	s = nla_data(tb[TCA_U32_SEL]);
 
 
 
 
 
631
632	n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL);
633	if (n == NULL)
634		return -ENOBUFS;
 
 
635
636#ifdef CONFIG_CLS_U32_PERF
637	n->pf = kzalloc(sizeof(struct tc_u32_pcnt) + s->nkeys*sizeof(u64), GFP_KERNEL);
638	if (n->pf == NULL) {
639		kfree(n);
640		return -ENOBUFS;
 
641	}
642#endif
643
644	memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
645	n->ht_up = ht;
646	n->handle = handle;
647	n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
 
 
 
 
 
648
649#ifdef CONFIG_CLS_U32_MARK
 
 
 
 
 
 
650	if (tb[TCA_U32_MARK]) {
651		struct tc_u32_mark *mark;
652
653		mark = nla_data(tb[TCA_U32_MARK]);
654		memcpy(&n->mark, mark, sizeof(struct tc_u32_mark));
655		n->mark.success = 0;
656	}
657#endif
658
659	err = u32_set_parms(tp, base, ht, n, tb, tca[TCA_RATE]);
 
660	if (err == 0) {
661		struct tc_u_knode **ins;
662		for (ins = &ht->ht[TC_U32_HASH(handle)]; *ins; ins = &(*ins)->next)
663			if (TC_U32_NODE(handle) < TC_U32_NODE((*ins)->handle))
664				break;
665
666		n->next = *ins;
667		tcf_tree_lock(tp);
668		*ins = n;
669		tcf_tree_unlock(tp);
 
 
 
 
 
 
 
 
670
671		*arg = (unsigned long)n;
 
 
 
672		return 0;
673	}
 
 
 
 
 
 
 
 
674#ifdef CONFIG_CLS_U32_PERF
675	kfree(n->pf);
 
676#endif
677	kfree(n);
 
 
678	return err;
679}
680
681static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg)
 
682{
683	struct tc_u_common *tp_c = tp->data;
684	struct tc_u_hnode *ht;
685	struct tc_u_knode *n;
686	unsigned int h;
687
688	if (arg->stop)
689		return;
690
691	for (ht = tp_c->hlist; ht; ht = ht->next) {
 
 
692		if (ht->prio != tp->prio)
693			continue;
694		if (arg->count >= arg->skip) {
695			if (arg->fn(tp, (unsigned long)ht, arg) < 0) {
696				arg->stop = 1;
697				return;
698			}
699		}
700		arg->count++;
701		for (h = 0; h <= ht->divisor; h++) {
702			for (n = ht->ht[h]; n; n = n->next) {
 
 
703				if (arg->count < arg->skip) {
704					arg->count++;
705					continue;
706				}
707				if (arg->fn(tp, (unsigned long)n, arg) < 0) {
708					arg->stop = 1;
709					return;
710				}
711				arg->count++;
712			}
713		}
714	}
715}
716
717static int u32_dump(struct tcf_proto *tp, unsigned long fh,
718		     struct sk_buff *skb, struct tcmsg *t)
 
719{
720	struct tc_u_knode *n = (struct tc_u_knode *)fh;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
721	struct nlattr *nest;
722
723	if (n == NULL)
724		return skb->len;
725
726	t->tcm_handle = n->handle;
727
728	nest = nla_nest_start(skb, TCA_OPTIONS);
729	if (nest == NULL)
730		goto nla_put_failure;
731
732	if (TC_U32_KEY(n->handle) == 0) {
733		struct tc_u_hnode *ht = (struct tc_u_hnode *)fh;
734		u32 divisor = ht->divisor + 1;
735
736		NLA_PUT_U32(skb, TCA_U32_DIVISOR, divisor);
 
737	} else {
738		NLA_PUT(skb, TCA_U32_SEL,
739			sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
740			&n->sel);
741		if (n->ht_up) {
 
 
 
 
 
 
 
742			u32 htid = n->handle & 0xFFFFF000;
743			NLA_PUT_U32(skb, TCA_U32_HASH, htid);
 
744		}
745		if (n->res.classid)
746			NLA_PUT_U32(skb, TCA_U32_CLASSID, n->res.classid);
747		if (n->ht_down)
748			NLA_PUT_U32(skb, TCA_U32_LINK, n->ht_down->handle);
 
 
 
 
 
 
 
749
750#ifdef CONFIG_CLS_U32_MARK
751		if (n->mark.val || n->mark.mask)
752			NLA_PUT(skb, TCA_U32_MARK, sizeof(n->mark), &n->mark);
 
 
 
 
 
 
 
 
 
 
 
 
 
753#endif
754
755		if (tcf_exts_dump(skb, &n->exts, &u32_ext_map) < 0)
756			goto nla_put_failure;
757
758#ifdef CONFIG_NET_CLS_IND
759		if (strlen(n->indev))
760			NLA_PUT_STRING(skb, TCA_U32_INDEV, n->indev);
761#endif
 
 
762#ifdef CONFIG_CLS_U32_PERF
763		NLA_PUT(skb, TCA_U32_PCNT,
764		sizeof(struct tc_u32_pcnt) + n->sel.nkeys*sizeof(u64),
765			n->pf);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
766#endif
767	}
768
769	nla_nest_end(skb, nest);
770
771	if (TC_U32_KEY(n->handle))
772		if (tcf_exts_dump_stats(skb, &n->exts, &u32_ext_map) < 0)
773			goto nla_put_failure;
774	return skb->len;
775
776nla_put_failure:
777	nla_nest_cancel(skb, nest);
778	return -1;
779}
780
781static struct tcf_proto_ops cls_u32_ops __read_mostly = {
782	.kind		=	"u32",
783	.classify	=	u32_classify,
784	.init		=	u32_init,
785	.destroy	=	u32_destroy,
786	.get		=	u32_get,
787	.put		=	u32_put,
788	.change		=	u32_change,
789	.delete		=	u32_delete,
790	.walk		=	u32_walk,
 
791	.dump		=	u32_dump,
 
792	.owner		=	THIS_MODULE,
793};
794
795static int __init init_u32(void)
796{
 
 
797	pr_info("u32 classifier\n");
798#ifdef CONFIG_CLS_U32_PERF
799	pr_info("    Performance counters on\n");
800#endif
801#ifdef CONFIG_NET_CLS_IND
802	pr_info("    input device check on\n");
803#endif
804#ifdef CONFIG_NET_CLS_ACT
805	pr_info("    Actions configured\n");
806#endif
807	return register_tcf_proto_ops(&cls_u32_ops);
 
 
 
 
 
 
 
 
 
 
 
 
808}
809
810static void __exit exit_u32(void)
811{
812	unregister_tcf_proto_ops(&cls_u32_ops);
 
813}
814
815module_init(init_u32)
816module_exit(exit_u32)
817MODULE_LICENSE("GPL");
v5.9
   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[1];
  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(sizeof(*root_ht), 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 int u32_destroy_key(struct tc_u_knode *n, bool free_pf)
 390{
 391	struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
 392
 393	tcf_exts_destroy(&n->exts);
 394	tcf_exts_put_net(&n->exts);
 395	if (ht && --ht->refcnt == 0)
 396		kfree(ht);
 397#ifdef CONFIG_CLS_U32_PERF
 398	if (free_pf)
 399		free_percpu(n->pf);
 400#endif
 401#ifdef CONFIG_CLS_U32_MARK
 402	if (free_pf)
 403		free_percpu(n->pcpu_success);
 404#endif
 405	kfree(n);
 406	return 0;
 407}
 408
 409/* u32_delete_key_rcu should be called when free'ing a copied
 410 * version of a tc_u_knode obtained from u32_init_knode(). When
 411 * copies are obtained from u32_init_knode() the statistics are
 412 * shared between the old and new copies to allow readers to
 413 * continue to update the statistics during the copy. To support
 414 * this the u32_delete_key_rcu variant does not free the percpu
 415 * statistics.
 416 */
 417static void u32_delete_key_work(struct work_struct *work)
 418{
 419	struct tc_u_knode *key = container_of(to_rcu_work(work),
 420					      struct tc_u_knode,
 421					      rwork);
 422	rtnl_lock();
 423	u32_destroy_key(key, false);
 424	rtnl_unlock();
 425}
 426
 427/* u32_delete_key_freepf_rcu is the rcu callback variant
 428 * that free's the entire structure including the statistics
 429 * percpu variables. Only use this if the key is not a copy
 430 * returned by u32_init_knode(). See u32_delete_key_rcu()
 431 * for the variant that should be used with keys return from
 432 * u32_init_knode()
 433 */
 434static void u32_delete_key_freepf_work(struct work_struct *work)
 435{
 436	struct tc_u_knode *key = container_of(to_rcu_work(work),
 437					      struct tc_u_knode,
 438					      rwork);
 439	rtnl_lock();
 440	u32_destroy_key(key, true);
 441	rtnl_unlock();
 442}
 443
 444static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
 445{
 446	struct tc_u_common *tp_c = tp->data;
 447	struct tc_u_knode __rcu **kp;
 448	struct tc_u_knode *pkp;
 449	struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
 450
 451	if (ht) {
 452		kp = &ht->ht[TC_U32_HASH(key->handle)];
 453		for (pkp = rtnl_dereference(*kp); pkp;
 454		     kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
 455			if (pkp == key) {
 456				RCU_INIT_POINTER(*kp, key->next);
 457				tp_c->knodes--;
 458
 459				tcf_unbind_filter(tp, &key->res);
 460				idr_remove(&ht->handle_idr, key->handle);
 461				tcf_exts_get_net(&key->exts);
 462				tcf_queue_work(&key->rwork, u32_delete_key_freepf_work);
 463				return 0;
 464			}
 465		}
 466	}
 467	WARN_ON(1);
 468	return 0;
 469}
 470
 471static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
 472			       struct netlink_ext_ack *extack)
 473{
 474	struct tcf_block *block = tp->chain->block;
 475	struct tc_cls_u32_offload cls_u32 = {};
 476
 477	tc_cls_common_offload_init(&cls_u32.common, tp, h->flags, extack);
 478	cls_u32.command = TC_CLSU32_DELETE_HNODE;
 479	cls_u32.hnode.divisor = h->divisor;
 480	cls_u32.hnode.handle = h->handle;
 481	cls_u32.hnode.prio = h->prio;
 482
 483	tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, false, true);
 484}
 485
 486static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
 487				u32 flags, struct netlink_ext_ack *extack)
 488{
 489	struct tcf_block *block = tp->chain->block;
 490	struct tc_cls_u32_offload cls_u32 = {};
 491	bool skip_sw = tc_skip_sw(flags);
 492	bool offloaded = false;
 493	int err;
 494
 495	tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
 496	cls_u32.command = TC_CLSU32_NEW_HNODE;
 497	cls_u32.hnode.divisor = h->divisor;
 498	cls_u32.hnode.handle = h->handle;
 499	cls_u32.hnode.prio = h->prio;
 500
 501	err = tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, skip_sw, true);
 502	if (err < 0) {
 503		u32_clear_hw_hnode(tp, h, NULL);
 504		return err;
 505	} else if (err > 0) {
 506		offloaded = true;
 507	}
 508
 509	if (skip_sw && !offloaded)
 510		return -EINVAL;
 511
 512	return 0;
 513}
 514
 515static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
 516				struct netlink_ext_ack *extack)
 517{
 518	struct tcf_block *block = tp->chain->block;
 519	struct tc_cls_u32_offload cls_u32 = {};
 520
 521	tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
 522	cls_u32.command = TC_CLSU32_DELETE_KNODE;
 523	cls_u32.knode.handle = n->handle;
 524
 525	tc_setup_cb_destroy(block, tp, TC_SETUP_CLSU32, &cls_u32, false,
 526			    &n->flags, &n->in_hw_count, true);
 527}
 528
 529static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
 530				u32 flags, struct netlink_ext_ack *extack)
 531{
 532	struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
 533	struct tcf_block *block = tp->chain->block;
 534	struct tc_cls_u32_offload cls_u32 = {};
 535	bool skip_sw = tc_skip_sw(flags);
 536	int err;
 537
 538	tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
 539	cls_u32.command = TC_CLSU32_REPLACE_KNODE;
 540	cls_u32.knode.handle = n->handle;
 541	cls_u32.knode.fshift = n->fshift;
 542#ifdef CONFIG_CLS_U32_MARK
 543	cls_u32.knode.val = n->val;
 544	cls_u32.knode.mask = n->mask;
 545#else
 546	cls_u32.knode.val = 0;
 547	cls_u32.knode.mask = 0;
 548#endif
 549	cls_u32.knode.sel = &n->sel;
 550	cls_u32.knode.res = &n->res;
 551	cls_u32.knode.exts = &n->exts;
 552	if (n->ht_down)
 553		cls_u32.knode.link_handle = ht->handle;
 554
 555	err = tc_setup_cb_add(block, tp, TC_SETUP_CLSU32, &cls_u32, skip_sw,
 556			      &n->flags, &n->in_hw_count, true);
 557	if (err) {
 558		u32_remove_hw_knode(tp, n, NULL);
 559		return err;
 560	}
 561
 562	if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
 563		return -EINVAL;
 564
 565	return 0;
 566}
 567
 568static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
 569			    struct netlink_ext_ack *extack)
 570{
 571	struct tc_u_common *tp_c = tp->data;
 572	struct tc_u_knode *n;
 573	unsigned int h;
 574
 575	for (h = 0; h <= ht->divisor; h++) {
 576		while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
 577			RCU_INIT_POINTER(ht->ht[h],
 578					 rtnl_dereference(n->next));
 579			tp_c->knodes--;
 580			tcf_unbind_filter(tp, &n->res);
 581			u32_remove_hw_knode(tp, n, extack);
 582			idr_remove(&ht->handle_idr, n->handle);
 583			if (tcf_exts_get_net(&n->exts))
 584				tcf_queue_work(&n->rwork, u32_delete_key_freepf_work);
 585			else
 586				u32_destroy_key(n, true);
 587		}
 588	}
 589}
 590
 591static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
 592			     struct netlink_ext_ack *extack)
 593{
 594	struct tc_u_common *tp_c = tp->data;
 595	struct tc_u_hnode __rcu **hn;
 596	struct tc_u_hnode *phn;
 597
 598	WARN_ON(--ht->refcnt);
 599
 600	u32_clear_hnode(tp, ht, extack);
 601
 602	hn = &tp_c->hlist;
 603	for (phn = rtnl_dereference(*hn);
 604	     phn;
 605	     hn = &phn->next, phn = rtnl_dereference(*hn)) {
 606		if (phn == ht) {
 607			u32_clear_hw_hnode(tp, ht, extack);
 608			idr_destroy(&ht->handle_idr);
 609			idr_remove(&tp_c->handle_idr, ht->handle);
 610			RCU_INIT_POINTER(*hn, ht->next);
 611			kfree_rcu(ht, rcu);
 612			return 0;
 613		}
 614	}
 615
 
 616	return -ENOENT;
 617}
 618
 619static void u32_destroy(struct tcf_proto *tp, bool rtnl_held,
 620			struct netlink_ext_ack *extack)
 621{
 622	struct tc_u_common *tp_c = tp->data;
 623	struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
 624
 625	WARN_ON(root_ht == NULL);
 626
 627	if (root_ht && --root_ht->refcnt == 1)
 628		u32_destroy_hnode(tp, root_ht, extack);
 629
 630	if (--tp_c->refcnt == 0) {
 631		struct tc_u_hnode *ht;
 632
 633		hlist_del(&tp_c->hnode);
 634
 635		while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
 636			u32_clear_hnode(tp, ht, extack);
 637			RCU_INIT_POINTER(tp_c->hlist, ht->next);
 638
 639			/* u32_destroy_key() will later free ht for us, if it's
 640			 * still referenced by some knode
 641			 */
 642			if (--ht->refcnt == 0)
 643				kfree_rcu(ht, rcu);
 
 
 644		}
 645
 646		idr_destroy(&tp_c->handle_idr);
 647		kfree(tp_c);
 648	}
 649
 650	tp->data = NULL;
 651}
 652
 653static int u32_delete(struct tcf_proto *tp, void *arg, bool *last,
 654		      bool rtnl_held, struct netlink_ext_ack *extack)
 655{
 656	struct tc_u_hnode *ht = arg;
 657	struct tc_u_common *tp_c = tp->data;
 658	int ret = 0;
 
 659
 660	if (TC_U32_KEY(ht->handle)) {
 661		u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack);
 662		ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
 663		goto out;
 664	}
 665
 666	if (ht->is_root) {
 667		NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node");
 668		return -EINVAL;
 669	}
 670
 671	if (ht->refcnt == 1) {
 672		u32_destroy_hnode(tp, ht, extack);
 
 673	} else {
 674		NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter");
 675		return -EBUSY;
 676	}
 677
 678out:
 679	*last = tp_c->refcnt == 1 && tp_c->knodes == 0;
 680	return ret;
 681}
 682
 683static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
 684{
 685	u32 index = htid | 0x800;
 686	u32 max = htid | 0xFFF;
 687
 688	if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max, GFP_KERNEL)) {
 689		index = htid + 1;
 690		if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max,
 691				 GFP_KERNEL))
 692			index = max;
 693	}
 694
 695	return index;
 696}
 697
 698static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
 699	[TCA_U32_CLASSID]	= { .type = NLA_U32 },
 700	[TCA_U32_HASH]		= { .type = NLA_U32 },
 701	[TCA_U32_LINK]		= { .type = NLA_U32 },
 702	[TCA_U32_DIVISOR]	= { .type = NLA_U32 },
 703	[TCA_U32_SEL]		= { .len = sizeof(struct tc_u32_sel) },
 704	[TCA_U32_INDEV]		= { .type = NLA_STRING, .len = IFNAMSIZ },
 705	[TCA_U32_MARK]		= { .len = sizeof(struct tc_u32_mark) },
 706	[TCA_U32_FLAGS]		= { .type = NLA_U32 },
 707};
 708
 709static int u32_set_parms(struct net *net, struct tcf_proto *tp,
 710			 unsigned long base,
 711			 struct tc_u_knode *n, struct nlattr **tb,
 712			 struct nlattr *est, bool ovr,
 713			 struct netlink_ext_ack *extack)
 714{
 715	int err;
 
 716
 717	err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr, true, extack);
 718	if (err < 0)
 719		return err;
 720
 
 721	if (tb[TCA_U32_LINK]) {
 722		u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
 723		struct tc_u_hnode *ht_down = NULL, *ht_old;
 724
 725		if (TC_U32_KEY(handle)) {
 726			NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table");
 727			return -EINVAL;
 728		}
 729
 730		if (handle) {
 731			ht_down = u32_lookup_ht(tp->data, handle);
 732
 733			if (!ht_down) {
 734				NL_SET_ERR_MSG_MOD(extack, "Link hash table not found");
 735				return -EINVAL;
 736			}
 737			if (ht_down->is_root) {
 738				NL_SET_ERR_MSG_MOD(extack, "Not linking to root node");
 739				return -EINVAL;
 740			}
 741			ht_down->refcnt++;
 742		}
 743
 744		ht_old = rtnl_dereference(n->ht_down);
 745		rcu_assign_pointer(n->ht_down, ht_down);
 
 
 746
 747		if (ht_old)
 748			ht_old->refcnt--;
 749	}
 750	if (tb[TCA_U32_CLASSID]) {
 751		n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
 752		tcf_bind_filter(tp, &n->res, base);
 753	}
 754
 
 755	if (tb[TCA_U32_INDEV]) {
 756		int ret;
 757		ret = tcf_change_indev(net, tb[TCA_U32_INDEV], extack);
 758		if (ret < 0)
 759			return -EINVAL;
 760		n->ifindex = ret;
 761	}
 762	return 0;
 763}
 764
 765static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
 766			      struct tc_u_knode *n)
 767{
 768	struct tc_u_knode __rcu **ins;
 769	struct tc_u_knode *pins;
 770	struct tc_u_hnode *ht;
 771
 772	if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
 773		ht = rtnl_dereference(tp->root);
 774	else
 775		ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
 776
 777	ins = &ht->ht[TC_U32_HASH(n->handle)];
 778
 779	/* The node must always exist for it to be replaced if this is not the
 780	 * case then something went very wrong elsewhere.
 781	 */
 782	for (pins = rtnl_dereference(*ins); ;
 783	     ins = &pins->next, pins = rtnl_dereference(*ins))
 784		if (pins->handle == n->handle)
 785			break;
 786
 787	idr_replace(&ht->handle_idr, n, n->handle);
 788	RCU_INIT_POINTER(n->next, pins->next);
 789	rcu_assign_pointer(*ins, n);
 790}
 791
 792static struct tc_u_knode *u32_init_knode(struct net *net, struct tcf_proto *tp,
 793					 struct tc_u_knode *n)
 794{
 795	struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
 796	struct tc_u32_sel *s = &n->sel;
 797	struct tc_u_knode *new;
 798
 799	new = kzalloc(struct_size(new, sel.keys, s->nkeys), GFP_KERNEL);
 800	if (!new)
 801		return NULL;
 802
 803	RCU_INIT_POINTER(new->next, n->next);
 804	new->handle = n->handle;
 805	RCU_INIT_POINTER(new->ht_up, n->ht_up);
 806
 807	new->ifindex = n->ifindex;
 808	new->fshift = n->fshift;
 809	new->res = n->res;
 810	new->flags = n->flags;
 811	RCU_INIT_POINTER(new->ht_down, ht);
 812
 813	/* bump reference count as long as we hold pointer to structure */
 814	if (ht)
 815		ht->refcnt++;
 816
 817#ifdef CONFIG_CLS_U32_PERF
 818	/* Statistics may be incremented by readers during update
 819	 * so we must keep them in tact. When the node is later destroyed
 820	 * a special destroy call must be made to not free the pf memory.
 821	 */
 822	new->pf = n->pf;
 823#endif
 
 824
 825#ifdef CONFIG_CLS_U32_MARK
 826	new->val = n->val;
 827	new->mask = n->mask;
 828	/* Similarly success statistics must be moved as pointers */
 829	new->pcpu_success = n->pcpu_success;
 830#endif
 831	memcpy(&new->sel, s, struct_size(s, keys, s->nkeys));
 832
 833	if (tcf_exts_init(&new->exts, net, TCA_U32_ACT, TCA_U32_POLICE)) {
 834		kfree(new);
 835		return NULL;
 836	}
 837
 838	return new;
 839}
 840
 841static int u32_change(struct net *net, struct sk_buff *in_skb,
 842		      struct tcf_proto *tp, unsigned long base, u32 handle,
 843		      struct nlattr **tca, void **arg, bool ovr, bool rtnl_held,
 844		      struct netlink_ext_ack *extack)
 845{
 846	struct tc_u_common *tp_c = tp->data;
 847	struct tc_u_hnode *ht;
 848	struct tc_u_knode *n;
 849	struct tc_u32_sel *s;
 850	struct nlattr *opt = tca[TCA_OPTIONS];
 851	struct nlattr *tb[TCA_U32_MAX + 1];
 852	u32 htid, flags = 0;
 853	size_t sel_size;
 854	int err;
 855
 856	if (!opt) {
 857		if (handle) {
 858			NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options");
 859			return -EINVAL;
 860		} else {
 861			return 0;
 862		}
 863	}
 864
 865	err = nla_parse_nested_deprecated(tb, TCA_U32_MAX, opt, u32_policy,
 866					  extack);
 867	if (err < 0)
 868		return err;
 869
 870	if (tb[TCA_U32_FLAGS]) {
 871		flags = nla_get_u32(tb[TCA_U32_FLAGS]);
 872		if (!tc_flags_valid(flags)) {
 873			NL_SET_ERR_MSG_MOD(extack, "Invalid filter flags");
 874			return -EINVAL;
 875		}
 876	}
 877
 878	n = *arg;
 879	if (n) {
 880		struct tc_u_knode *new;
 881
 882		if (TC_U32_KEY(n->handle) == 0) {
 883			NL_SET_ERR_MSG_MOD(extack, "Key node id cannot be zero");
 884			return -EINVAL;
 885		}
 886
 887		if ((n->flags ^ flags) &
 888		    ~(TCA_CLS_FLAGS_IN_HW | TCA_CLS_FLAGS_NOT_IN_HW)) {
 889			NL_SET_ERR_MSG_MOD(extack, "Key node flags do not match passed flags");
 890			return -EINVAL;
 891		}
 892
 893		new = u32_init_knode(net, tp, n);
 894		if (!new)
 895			return -ENOMEM;
 896
 897		err = u32_set_parms(net, tp, base, new, tb,
 898				    tca[TCA_RATE], ovr, extack);
 899
 900		if (err) {
 901			u32_destroy_key(new, false);
 902			return err;
 903		}
 904
 905		err = u32_replace_hw_knode(tp, new, flags, extack);
 906		if (err) {
 907			u32_destroy_key(new, false);
 908			return err;
 909		}
 910
 911		if (!tc_in_hw(new->flags))
 912			new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
 913
 914		u32_replace_knode(tp, tp_c, new);
 915		tcf_unbind_filter(tp, &n->res);
 916		tcf_exts_get_net(&n->exts);
 917		tcf_queue_work(&n->rwork, u32_delete_key_work);
 918		return 0;
 919	}
 920
 921	if (tb[TCA_U32_DIVISOR]) {
 922		unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
 923
 924		if (!is_power_of_2(divisor)) {
 925			NL_SET_ERR_MSG_MOD(extack, "Divisor is not a power of 2");
 926			return -EINVAL;
 927		}
 928		if (divisor-- > 0x100) {
 929			NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets");
 930			return -EINVAL;
 931		}
 932		if (TC_U32_KEY(handle)) {
 933			NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table");
 934			return -EINVAL;
 
 
 
 
 935		}
 936		ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
 937		if (ht == NULL)
 938			return -ENOBUFS;
 939		if (handle == 0) {
 940			handle = gen_new_htid(tp->data, ht);
 941			if (handle == 0) {
 942				kfree(ht);
 943				return -ENOMEM;
 944			}
 945		} else {
 946			err = idr_alloc_u32(&tp_c->handle_idr, ht, &handle,
 947					    handle, GFP_KERNEL);
 948			if (err) {
 949				kfree(ht);
 950				return err;
 951			}
 952		}
 953		ht->refcnt = 1;
 954		ht->divisor = divisor;
 955		ht->handle = handle;
 956		ht->prio = tp->prio;
 957		idr_init(&ht->handle_idr);
 958		ht->flags = flags;
 959
 960		err = u32_replace_hw_hnode(tp, ht, flags, extack);
 961		if (err) {
 962			idr_remove(&tp_c->handle_idr, handle);
 963			kfree(ht);
 964			return err;
 965		}
 966
 967		RCU_INIT_POINTER(ht->next, tp_c->hlist);
 968		rcu_assign_pointer(tp_c->hlist, ht);
 969		*arg = ht;
 970
 971		return 0;
 972	}
 973
 974	if (tb[TCA_U32_HASH]) {
 975		htid = nla_get_u32(tb[TCA_U32_HASH]);
 976		if (TC_U32_HTID(htid) == TC_U32_ROOT) {
 977			ht = rtnl_dereference(tp->root);
 978			htid = ht->handle;
 979		} else {
 980			ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
 981			if (!ht) {
 982				NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found");
 983				return -EINVAL;
 984			}
 985		}
 986	} else {
 987		ht = rtnl_dereference(tp->root);
 988		htid = ht->handle;
 989	}
 990
 991	if (ht->divisor < TC_U32_HASH(htid)) {
 992		NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value");
 993		return -EINVAL;
 994	}
 995
 996	if (handle) {
 997		if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) {
 998			NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch");
 999			return -EINVAL;
1000		}
1001		handle = htid | TC_U32_NODE(handle);
1002		err = idr_alloc_u32(&ht->handle_idr, NULL, &handle, handle,
1003				    GFP_KERNEL);
1004		if (err)
1005			return err;
1006	} else
1007		handle = gen_new_kid(ht, htid);
1008
1009	if (tb[TCA_U32_SEL] == NULL) {
1010		NL_SET_ERR_MSG_MOD(extack, "Selector not specified");
1011		err = -EINVAL;
1012		goto erridr;
1013	}
1014
1015	s = nla_data(tb[TCA_U32_SEL]);
1016	sel_size = struct_size(s, keys, s->nkeys);
1017	if (nla_len(tb[TCA_U32_SEL]) < sel_size) {
1018		err = -EINVAL;
1019		goto erridr;
1020	}
1021
1022	n = kzalloc(struct_size(n, sel.keys, s->nkeys), GFP_KERNEL);
1023	if (n == NULL) {
1024		err = -ENOBUFS;
1025		goto erridr;
1026	}
1027
1028#ifdef CONFIG_CLS_U32_PERF
1029	n->pf = __alloc_percpu(struct_size(n->pf, kcnts, s->nkeys),
1030			       __alignof__(struct tc_u32_pcnt));
1031	if (!n->pf) {
1032		err = -ENOBUFS;
1033		goto errfree;
1034	}
1035#endif
1036
1037	memcpy(&n->sel, s, sel_size);
1038	RCU_INIT_POINTER(n->ht_up, ht);
1039	n->handle = handle;
1040	n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1041	n->flags = flags;
1042
1043	err = tcf_exts_init(&n->exts, net, TCA_U32_ACT, TCA_U32_POLICE);
1044	if (err < 0)
1045		goto errout;
1046
1047#ifdef CONFIG_CLS_U32_MARK
1048	n->pcpu_success = alloc_percpu(u32);
1049	if (!n->pcpu_success) {
1050		err = -ENOMEM;
1051		goto errout;
1052	}
1053
1054	if (tb[TCA_U32_MARK]) {
1055		struct tc_u32_mark *mark;
1056
1057		mark = nla_data(tb[TCA_U32_MARK]);
1058		n->val = mark->val;
1059		n->mask = mark->mask;
1060	}
1061#endif
1062
1063	err = u32_set_parms(net, tp, base, n, tb, tca[TCA_RATE], ovr,
1064			    extack);
1065	if (err == 0) {
1066		struct tc_u_knode __rcu **ins;
1067		struct tc_u_knode *pins;
 
 
1068
1069		err = u32_replace_hw_knode(tp, n, flags, extack);
1070		if (err)
1071			goto errhw;
1072
1073		if (!tc_in_hw(n->flags))
1074			n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1075
1076		ins = &ht->ht[TC_U32_HASH(handle)];
1077		for (pins = rtnl_dereference(*ins); pins;
1078		     ins = &pins->next, pins = rtnl_dereference(*ins))
1079			if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1080				break;
1081
1082		RCU_INIT_POINTER(n->next, pins);
1083		rcu_assign_pointer(*ins, n);
1084		tp_c->knodes++;
1085		*arg = n;
1086		return 0;
1087	}
1088
1089errhw:
1090#ifdef CONFIG_CLS_U32_MARK
1091	free_percpu(n->pcpu_success);
1092#endif
1093
1094errout:
1095	tcf_exts_destroy(&n->exts);
1096#ifdef CONFIG_CLS_U32_PERF
1097errfree:
1098	free_percpu(n->pf);
1099#endif
1100	kfree(n);
1101erridr:
1102	idr_remove(&ht->handle_idr, handle);
1103	return err;
1104}
1105
1106static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg,
1107		     bool rtnl_held)
1108{
1109	struct tc_u_common *tp_c = tp->data;
1110	struct tc_u_hnode *ht;
1111	struct tc_u_knode *n;
1112	unsigned int h;
1113
1114	if (arg->stop)
1115		return;
1116
1117	for (ht = rtnl_dereference(tp_c->hlist);
1118	     ht;
1119	     ht = rtnl_dereference(ht->next)) {
1120		if (ht->prio != tp->prio)
1121			continue;
1122		if (arg->count >= arg->skip) {
1123			if (arg->fn(tp, ht, arg) < 0) {
1124				arg->stop = 1;
1125				return;
1126			}
1127		}
1128		arg->count++;
1129		for (h = 0; h <= ht->divisor; h++) {
1130			for (n = rtnl_dereference(ht->ht[h]);
1131			     n;
1132			     n = rtnl_dereference(n->next)) {
1133				if (arg->count < arg->skip) {
1134					arg->count++;
1135					continue;
1136				}
1137				if (arg->fn(tp, n, arg) < 0) {
1138					arg->stop = 1;
1139					return;
1140				}
1141				arg->count++;
1142			}
1143		}
1144	}
1145}
1146
1147static int u32_reoffload_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
1148			       bool add, flow_setup_cb_t *cb, void *cb_priv,
1149			       struct netlink_ext_ack *extack)
1150{
1151	struct tc_cls_u32_offload cls_u32 = {};
1152	int err;
1153
1154	tc_cls_common_offload_init(&cls_u32.common, tp, ht->flags, extack);
1155	cls_u32.command = add ? TC_CLSU32_NEW_HNODE : TC_CLSU32_DELETE_HNODE;
1156	cls_u32.hnode.divisor = ht->divisor;
1157	cls_u32.hnode.handle = ht->handle;
1158	cls_u32.hnode.prio = ht->prio;
1159
1160	err = cb(TC_SETUP_CLSU32, &cls_u32, cb_priv);
1161	if (err && add && tc_skip_sw(ht->flags))
1162		return err;
1163
1164	return 0;
1165}
1166
1167static int u32_reoffload_knode(struct tcf_proto *tp, struct tc_u_knode *n,
1168			       bool add, flow_setup_cb_t *cb, void *cb_priv,
1169			       struct netlink_ext_ack *extack)
1170{
1171	struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
1172	struct tcf_block *block = tp->chain->block;
1173	struct tc_cls_u32_offload cls_u32 = {};
1174	int err;
1175
1176	tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
1177	cls_u32.command = add ?
1178		TC_CLSU32_REPLACE_KNODE : TC_CLSU32_DELETE_KNODE;
1179	cls_u32.knode.handle = n->handle;
1180
1181	if (add) {
1182		cls_u32.knode.fshift = n->fshift;
1183#ifdef CONFIG_CLS_U32_MARK
1184		cls_u32.knode.val = n->val;
1185		cls_u32.knode.mask = n->mask;
1186#else
1187		cls_u32.knode.val = 0;
1188		cls_u32.knode.mask = 0;
1189#endif
1190		cls_u32.knode.sel = &n->sel;
1191		cls_u32.knode.res = &n->res;
1192		cls_u32.knode.exts = &n->exts;
1193		if (n->ht_down)
1194			cls_u32.knode.link_handle = ht->handle;
1195	}
1196
1197	err = tc_setup_cb_reoffload(block, tp, add, cb, TC_SETUP_CLSU32,
1198				    &cls_u32, cb_priv, &n->flags,
1199				    &n->in_hw_count);
1200	if (err)
1201		return err;
1202
1203	return 0;
1204}
1205
1206static int u32_reoffload(struct tcf_proto *tp, bool add, flow_setup_cb_t *cb,
1207			 void *cb_priv, struct netlink_ext_ack *extack)
1208{
1209	struct tc_u_common *tp_c = tp->data;
1210	struct tc_u_hnode *ht;
1211	struct tc_u_knode *n;
1212	unsigned int h;
1213	int err;
1214
1215	for (ht = rtnl_dereference(tp_c->hlist);
1216	     ht;
1217	     ht = rtnl_dereference(ht->next)) {
1218		if (ht->prio != tp->prio)
1219			continue;
1220
1221		/* When adding filters to a new dev, try to offload the
1222		 * hashtable first. When removing, do the filters before the
1223		 * hashtable.
1224		 */
1225		if (add && !tc_skip_hw(ht->flags)) {
1226			err = u32_reoffload_hnode(tp, ht, add, cb, cb_priv,
1227						  extack);
1228			if (err)
1229				return err;
1230		}
1231
1232		for (h = 0; h <= ht->divisor; h++) {
1233			for (n = rtnl_dereference(ht->ht[h]);
1234			     n;
1235			     n = rtnl_dereference(n->next)) {
1236				if (tc_skip_hw(n->flags))
1237					continue;
1238
1239				err = u32_reoffload_knode(tp, n, add, cb,
1240							  cb_priv, extack);
1241				if (err)
1242					return err;
1243			}
1244		}
1245
1246		if (!add && !tc_skip_hw(ht->flags))
1247			u32_reoffload_hnode(tp, ht, add, cb, cb_priv, extack);
1248	}
1249
1250	return 0;
1251}
1252
1253static void u32_bind_class(void *fh, u32 classid, unsigned long cl, void *q,
1254			   unsigned long base)
1255{
1256	struct tc_u_knode *n = fh;
1257
1258	if (n && n->res.classid == classid) {
1259		if (cl)
1260			__tcf_bind_filter(q, &n->res, base);
1261		else
1262			__tcf_unbind_filter(q, &n->res);
1263	}
1264}
1265
1266static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1267		    struct sk_buff *skb, struct tcmsg *t, bool rtnl_held)
1268{
1269	struct tc_u_knode *n = fh;
1270	struct tc_u_hnode *ht_up, *ht_down;
1271	struct nlattr *nest;
1272
1273	if (n == NULL)
1274		return skb->len;
1275
1276	t->tcm_handle = n->handle;
1277
1278	nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1279	if (nest == NULL)
1280		goto nla_put_failure;
1281
1282	if (TC_U32_KEY(n->handle) == 0) {
1283		struct tc_u_hnode *ht = fh;
1284		u32 divisor = ht->divisor + 1;
1285
1286		if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1287			goto nla_put_failure;
1288	} else {
1289#ifdef CONFIG_CLS_U32_PERF
1290		struct tc_u32_pcnt *gpf;
1291		int cpu;
1292#endif
1293
1294		if (nla_put(skb, TCA_U32_SEL, struct_size(&n->sel, keys, n->sel.nkeys),
1295			    &n->sel))
1296			goto nla_put_failure;
1297
1298		ht_up = rtnl_dereference(n->ht_up);
1299		if (ht_up) {
1300			u32 htid = n->handle & 0xFFFFF000;
1301			if (nla_put_u32(skb, TCA_U32_HASH, htid))
1302				goto nla_put_failure;
1303		}
1304		if (n->res.classid &&
1305		    nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1306			goto nla_put_failure;
1307
1308		ht_down = rtnl_dereference(n->ht_down);
1309		if (ht_down &&
1310		    nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1311			goto nla_put_failure;
1312
1313		if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1314			goto nla_put_failure;
1315
1316#ifdef CONFIG_CLS_U32_MARK
1317		if ((n->val || n->mask)) {
1318			struct tc_u32_mark mark = {.val = n->val,
1319						   .mask = n->mask,
1320						   .success = 0};
1321			int cpum;
1322
1323			for_each_possible_cpu(cpum) {
1324				__u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1325
1326				mark.success += cnt;
1327			}
1328
1329			if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1330				goto nla_put_failure;
1331		}
1332#endif
1333
1334		if (tcf_exts_dump(skb, &n->exts) < 0)
1335			goto nla_put_failure;
1336
1337		if (n->ifindex) {
1338			struct net_device *dev;
1339			dev = __dev_get_by_index(net, n->ifindex);
1340			if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1341				goto nla_put_failure;
1342		}
1343#ifdef CONFIG_CLS_U32_PERF
1344		gpf = kzalloc(struct_size(gpf, kcnts, n->sel.nkeys), GFP_KERNEL);
1345		if (!gpf)
1346			goto nla_put_failure;
1347
1348		for_each_possible_cpu(cpu) {
1349			int i;
1350			struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1351
1352			gpf->rcnt += pf->rcnt;
1353			gpf->rhit += pf->rhit;
1354			for (i = 0; i < n->sel.nkeys; i++)
1355				gpf->kcnts[i] += pf->kcnts[i];
1356		}
1357
1358		if (nla_put_64bit(skb, TCA_U32_PCNT, struct_size(gpf, kcnts, n->sel.nkeys),
1359				  gpf, TCA_U32_PAD)) {
1360			kfree(gpf);
1361			goto nla_put_failure;
1362		}
1363		kfree(gpf);
1364#endif
1365	}
1366
1367	nla_nest_end(skb, nest);
1368
1369	if (TC_U32_KEY(n->handle))
1370		if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1371			goto nla_put_failure;
1372	return skb->len;
1373
1374nla_put_failure:
1375	nla_nest_cancel(skb, nest);
1376	return -1;
1377}
1378
1379static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1380	.kind		=	"u32",
1381	.classify	=	u32_classify,
1382	.init		=	u32_init,
1383	.destroy	=	u32_destroy,
1384	.get		=	u32_get,
 
1385	.change		=	u32_change,
1386	.delete		=	u32_delete,
1387	.walk		=	u32_walk,
1388	.reoffload	=	u32_reoffload,
1389	.dump		=	u32_dump,
1390	.bind_class	=	u32_bind_class,
1391	.owner		=	THIS_MODULE,
1392};
1393
1394static int __init init_u32(void)
1395{
1396	int i, ret;
1397
1398	pr_info("u32 classifier\n");
1399#ifdef CONFIG_CLS_U32_PERF
1400	pr_info("    Performance counters on\n");
1401#endif
 
1402	pr_info("    input device check on\n");
 
1403#ifdef CONFIG_NET_CLS_ACT
1404	pr_info("    Actions configured\n");
1405#endif
1406	tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1407					  sizeof(struct hlist_head),
1408					  GFP_KERNEL);
1409	if (!tc_u_common_hash)
1410		return -ENOMEM;
1411
1412	for (i = 0; i < U32_HASH_SIZE; i++)
1413		INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1414
1415	ret = register_tcf_proto_ops(&cls_u32_ops);
1416	if (ret)
1417		kvfree(tc_u_common_hash);
1418	return ret;
1419}
1420
1421static void __exit exit_u32(void)
1422{
1423	unregister_tcf_proto_ops(&cls_u32_ops);
1424	kvfree(tc_u_common_hash);
1425}
1426
1427module_init(init_u32)
1428module_exit(exit_u32)
1429MODULE_LICENSE("GPL");