Linux Audio

Check our new training course

Loading...
v3.1
 
  1/*
  2 * net/sched/cls_tcindex.c	Packet classifier for skb->tc_index
  3 *
  4 * Written 1998,1999 by Werner Almesberger, EPFL ICA
  5 */
  6
  7#include <linux/module.h>
  8#include <linux/types.h>
  9#include <linux/kernel.h>
 10#include <linux/skbuff.h>
 11#include <linux/errno.h>
 12#include <linux/slab.h>
 
 13#include <net/act_api.h>
 14#include <net/netlink.h>
 15#include <net/pkt_cls.h>
 
 16
 17/*
 18 * Passing parameters to the root seems to be done more awkwardly than really
 19 * necessary. At least, u32 doesn't seem to use such dirty hacks. To be
 20 * verified. FIXME.
 21 */
 22
 23#define PERFECT_HASH_THRESHOLD	64	/* use perfect hash if not bigger */
 24#define DEFAULT_HASH_SIZE	64	/* optimized for diffserv */
 25
 26
 27#define	PRIV(tp)	((struct tcindex_data *) (tp)->root)
 28
 29
 30struct tcindex_filter_result {
 31	struct tcf_exts		exts;
 32	struct tcf_result	res;
 
 
 33};
 34
 35struct tcindex_filter {
 36	u16 key;
 37	struct tcindex_filter_result result;
 38	struct tcindex_filter *next;
 
 39};
 40
 41
 42struct tcindex_data {
 43	struct tcindex_filter_result *perfect; /* perfect hash; NULL if none */
 44	struct tcindex_filter **h; /* imperfect hash; only used if !perfect;
 45				      NULL if unused */
 46	u16 mask;		/* AND key with mask */
 47	int shift;		/* shift ANDed key to the right */
 48	int hash;		/* hash table size; 0 if undefined */
 49	int alloc_hash;		/* allocated size */
 50	int fall_through;	/* 0: only classify if explicit match */
 
 
 51};
 52
 53static const struct tcf_ext_map tcindex_ext_map = {
 54	.police = TCA_TCINDEX_POLICE,
 55	.action = TCA_TCINDEX_ACT
 56};
 57
 58static inline int
 59tcindex_filter_is_set(struct tcindex_filter_result *r)
 60{
 61	return tcf_exts_is_predicative(&r->exts) || r->res.classid;
 62}
 63
 64static struct tcindex_filter_result *
 65tcindex_lookup(struct tcindex_data *p, u16 key)
 66{
 67	struct tcindex_filter *f;
 
 
 
 
 
 
 
 
 
 
 
 68
 69	if (p->perfect)
 70		return tcindex_filter_is_set(p->perfect + key) ?
 71			p->perfect + key : NULL;
 72	else if (p->h) {
 73		for (f = p->h[key % p->hash]; f; f = f->next)
 
 
 
 
 74			if (f->key == key)
 75				return &f->result;
 76	}
 77
 78	return NULL;
 79}
 80
 81
 82static int tcindex_classify(struct sk_buff *skb, const struct tcf_proto *tp,
 83			    struct tcf_result *res)
 84{
 85	struct tcindex_data *p = PRIV(tp);
 86	struct tcindex_filter_result *f;
 87	int key = (skb->tc_index & p->mask) >> p->shift;
 88
 89	pr_debug("tcindex_classify(skb %p,tp %p,res %p),p %p\n",
 90		 skb, tp, res, p);
 91
 92	f = tcindex_lookup(p, key);
 93	if (!f) {
 
 
 94		if (!p->fall_through)
 95			return -1;
 96		res->classid = TC_H_MAKE(TC_H_MAJ(tp->q->handle), key);
 97		res->class = 0;
 98		pr_debug("alg 0x%x\n", res->classid);
 99		return 0;
100	}
101	*res = f->res;
102	pr_debug("map 0x%x\n", res->classid);
103
104	return tcf_exts_exec(skb, &f->exts, res);
105}
106
107
108static unsigned long tcindex_get(struct tcf_proto *tp, u32 handle)
109{
110	struct tcindex_data *p = PRIV(tp);
111	struct tcindex_filter_result *r;
112
113	pr_debug("tcindex_get(tp %p,handle 0x%08x)\n", tp, handle);
114	if (p->perfect && handle >= p->alloc_hash)
115		return 0;
116	r = tcindex_lookup(p, handle);
117	return r && tcindex_filter_is_set(r) ? (unsigned long) r : 0UL;
118}
119
120
121static void tcindex_put(struct tcf_proto *tp, unsigned long f)
122{
123	pr_debug("tcindex_put(tp %p,f 0x%lx)\n", tp, f);
124}
125
126
127static int tcindex_init(struct tcf_proto *tp)
128{
129	struct tcindex_data *p;
130
131	pr_debug("tcindex_init(tp %p)\n", tp);
132	p = kzalloc(sizeof(struct tcindex_data), GFP_KERNEL);
133	if (!p)
134		return -ENOMEM;
135
136	p->mask = 0xffff;
137	p->hash = DEFAULT_HASH_SIZE;
138	p->fall_through = 1;
 
139
140	tp->root = p;
141	return 0;
142}
143
 
 
 
 
 
 
 
 
 
 
144
145static int
146__tcindex_delete(struct tcf_proto *tp, unsigned long arg, int lock)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
147{
148	struct tcindex_data *p = PRIV(tp);
149	struct tcindex_filter_result *r = (struct tcindex_filter_result *) arg;
 
 
 
 
 
 
 
 
 
 
 
 
 
150	struct tcindex_filter *f = NULL;
151
152	pr_debug("tcindex_delete(tp %p,arg 0x%lx),p %p,f %p\n", tp, arg, p, f);
153	if (p->perfect) {
154		if (!r->res.class)
155			return -ENOENT;
156	} else {
157		int i;
158		struct tcindex_filter **walk = NULL;
159
160		for (i = 0; i < p->hash; i++)
161			for (walk = p->h+i; *walk; walk = &(*walk)->next)
162				if (&(*walk)->result == r)
 
 
163					goto found;
 
 
164		return -ENOENT;
165
166found:
167		f = *walk;
168		if (lock)
169			tcf_tree_lock(tp);
170		*walk = f->next;
171		if (lock)
172			tcf_tree_unlock(tp);
173	}
174	tcf_unbind_filter(tp, &r->res);
175	tcf_exts_destroy(tp, &r->exts);
176	kfree(f);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
177	return 0;
178}
179
180static int tcindex_delete(struct tcf_proto *tp, unsigned long arg)
181{
182	return __tcindex_delete(tp, arg, 1);
 
 
 
 
183}
184
185static inline int
186valid_perfect_hash(struct tcindex_data *p)
187{
188	return  p->hash > (p->mask >> p->shift);
189}
190
191static const struct nla_policy tcindex_policy[TCA_TCINDEX_MAX + 1] = {
192	[TCA_TCINDEX_HASH]		= { .type = NLA_U32 },
193	[TCA_TCINDEX_MASK]		= { .type = NLA_U16 },
194	[TCA_TCINDEX_SHIFT]		= { .type = NLA_U32 },
195	[TCA_TCINDEX_FALL_THROUGH]	= { .type = NLA_U32 },
196	[TCA_TCINDEX_CLASSID]		= { .type = NLA_U32 },
197};
198
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
199static int
200tcindex_set_parms(struct tcf_proto *tp, unsigned long base, u32 handle,
201		  struct tcindex_data *p, struct tcindex_filter_result *r,
202		  struct nlattr **tb, struct nlattr *est)
 
203{
204	int err, balloc = 0;
205	struct tcindex_filter_result new_filter_result, *old_r = r;
206	struct tcindex_filter_result cr;
207	struct tcindex_data cp;
208	struct tcindex_filter *f = NULL; /* make gcc behave */
 
 
209	struct tcf_exts e;
210
211	err = tcf_exts_validate(tp, tb, est, &e, &tcindex_ext_map);
212	if (err < 0)
213		return err;
 
 
 
214
215	memcpy(&cp, p, sizeof(cp));
216	memset(&new_filter_result, 0, sizeof(new_filter_result));
 
 
 
 
 
 
217
218	if (old_r)
219		memcpy(&cr, r, sizeof(cr));
220	else
221		memset(&cr, 0, sizeof(cr));
 
 
 
222
223	if (tb[TCA_TCINDEX_HASH])
224		cp.hash = nla_get_u32(tb[TCA_TCINDEX_HASH]);
225
226	if (tb[TCA_TCINDEX_MASK])
227		cp.mask = nla_get_u16(tb[TCA_TCINDEX_MASK]);
228
229	if (tb[TCA_TCINDEX_SHIFT])
230		cp.shift = nla_get_u32(tb[TCA_TCINDEX_SHIFT]);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
231
232	err = -EBUSY;
 
233	/* Hash already allocated, make sure that we still meet the
234	 * requirements for the allocated hash.
235	 */
236	if (cp.perfect) {
237		if (!valid_perfect_hash(&cp) ||
238		    cp.hash > cp.alloc_hash)
239			goto errout;
240	} else if (cp.h && cp.hash != cp.alloc_hash)
241		goto errout;
 
242
243	err = -EINVAL;
244	if (tb[TCA_TCINDEX_FALL_THROUGH])
245		cp.fall_through = nla_get_u32(tb[TCA_TCINDEX_FALL_THROUGH]);
246
247	if (!cp.hash) {
248		/* Hash not specified, use perfect hash if the upper limit
249		 * of the hashing index is below the threshold.
250		 */
251		if ((cp.mask >> cp.shift) < PERFECT_HASH_THRESHOLD)
252			cp.hash = (cp.mask >> cp.shift) + 1;
253		else
254			cp.hash = DEFAULT_HASH_SIZE;
255	}
256
257	if (!cp.perfect && !cp.h)
258		cp.alloc_hash = cp.hash;
259
260	/* Note: this could be as restrictive as if (handle & ~(mask >> shift))
261	 * but then, we'd fail handles that may become valid after some future
262	 * mask change. While this is extremely unlikely to ever matter,
263	 * the check below is safer (and also more backwards-compatible).
264	 */
265	if (cp.perfect || valid_perfect_hash(&cp))
266		if (handle >= cp.alloc_hash)
267			goto errout;
268
269
270	err = -ENOMEM;
271	if (!cp.perfect && !cp.h) {
272		if (valid_perfect_hash(&cp)) {
273			cp.perfect = kcalloc(cp.hash, sizeof(*r), GFP_KERNEL);
274			if (!cp.perfect)
275				goto errout;
276			balloc = 1;
277		} else {
278			cp.h = kcalloc(cp.hash, sizeof(f), GFP_KERNEL);
279			if (!cp.h)
280				goto errout;
 
 
 
 
 
 
 
281			balloc = 2;
282		}
283	}
284
285	if (cp.perfect)
286		r = cp.perfect + handle;
287	else
288		r = tcindex_lookup(&cp, handle) ? : &new_filter_result;
289
290	if (r == &new_filter_result) {
291		f = kzalloc(sizeof(*f), GFP_KERNEL);
292		if (!f)
293			goto errout_alloc;
 
 
 
 
 
 
 
294	}
295
296	if (tb[TCA_TCINDEX_CLASSID]) {
297		cr.res.classid = nla_get_u32(tb[TCA_TCINDEX_CLASSID]);
298		tcf_bind_filter(tp, &cr.res, base);
299	}
300
301	tcf_exts_change(tp, &cr.exts, &e);
 
 
 
 
 
 
302
303	tcf_tree_lock(tp);
304	if (old_r && old_r != r)
305		memset(old_r, 0, sizeof(*old_r));
306
307	memcpy(p, &cp, sizeof(cp));
308	memcpy(r, &cr, sizeof(cr));
309
310	if (r == &new_filter_result) {
311		struct tcindex_filter **fp;
 
312
313		f->key = handle;
314		f->result = new_filter_result;
315		f->next = NULL;
316		for (fp = p->h+(handle % p->hash); *fp; fp = &(*fp)->next)
317			/* nothing */;
318		*fp = f;
 
 
 
 
 
 
319	}
320	tcf_tree_unlock(tp);
321
 
 
322	return 0;
323
324errout_alloc:
325	if (balloc == 1)
326		kfree(cp.perfect);
327	else if (balloc == 2)
328		kfree(cp.h);
 
329errout:
330	tcf_exts_destroy(tp, &e);
 
331	return err;
332}
333
334static int
335tcindex_change(struct tcf_proto *tp, unsigned long base, u32 handle,
336	       struct nlattr **tca, unsigned long *arg)
 
 
337{
338	struct nlattr *opt = tca[TCA_OPTIONS];
339	struct nlattr *tb[TCA_TCINDEX_MAX + 1];
340	struct tcindex_data *p = PRIV(tp);
341	struct tcindex_filter_result *r = (struct tcindex_filter_result *) *arg;
342	int err;
343
344	pr_debug("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
345	    "p %p,r %p,*arg 0x%lx\n",
346	    tp, handle, tca, arg, opt, p, r, arg ? *arg : 0L);
347
348	if (!opt)
349		return 0;
350
351	err = nla_parse_nested(tb, TCA_TCINDEX_MAX, opt, tcindex_policy);
 
352	if (err < 0)
353		return err;
354
355	return tcindex_set_parms(tp, base, handle, p, r, tb, tca[TCA_RATE]);
 
356}
357
358
359static void tcindex_walk(struct tcf_proto *tp, struct tcf_walker *walker)
360{
361	struct tcindex_data *p = PRIV(tp);
362	struct tcindex_filter *f, *next;
363	int i;
364
365	pr_debug("tcindex_walk(tp %p,walker %p),p %p\n", tp, walker, p);
366	if (p->perfect) {
367		for (i = 0; i < p->hash; i++) {
368			if (!p->perfect[i].res.class)
369				continue;
370			if (walker->count >= walker->skip) {
371				if (walker->fn(tp,
372				    (unsigned long) (p->perfect+i), walker)
373				     < 0) {
374					walker->stop = 1;
375					return;
376				}
377			}
378			walker->count++;
379		}
380	}
381	if (!p->h)
382		return;
383	for (i = 0; i < p->hash; i++) {
384		for (f = p->h[i]; f; f = next) {
385			next = f->next;
386			if (walker->count >= walker->skip) {
387				if (walker->fn(tp, (unsigned long) &f->result,
388				    walker) < 0) {
389					walker->stop = 1;
390					return;
391				}
392			}
393			walker->count++;
394		}
395	}
396}
397
398
399static int tcindex_destroy_element(struct tcf_proto *tp,
400    unsigned long arg, struct tcf_walker *walker)
401{
402	return __tcindex_delete(tp, arg, 0);
403}
404
 
405
406static void tcindex_destroy(struct tcf_proto *tp)
407{
408	struct tcindex_data *p = PRIV(tp);
409	struct tcf_walker walker;
410
411	pr_debug("tcindex_destroy(tp %p),p %p\n", tp, p);
412	walker.count = 0;
413	walker.skip = 0;
414	walker.fn = &tcindex_destroy_element;
415	tcindex_walk(tp, &walker);
416	kfree(p->perfect);
417	kfree(p->h);
418	kfree(p);
419	tp->root = NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
420}
421
422
423static int tcindex_dump(struct tcf_proto *tp, unsigned long fh,
424    struct sk_buff *skb, struct tcmsg *t)
425{
426	struct tcindex_data *p = PRIV(tp);
427	struct tcindex_filter_result *r = (struct tcindex_filter_result *) fh;
428	unsigned char *b = skb_tail_pointer(skb);
429	struct nlattr *nest;
430
431	pr_debug("tcindex_dump(tp %p,fh 0x%lx,skb %p,t %p),p %p,r %p,b %p\n",
432		 tp, fh, skb, t, p, r, b);
433	pr_debug("p->perfect %p p->h %p\n", p->perfect, p->h);
434
435	nest = nla_nest_start(skb, TCA_OPTIONS);
436	if (nest == NULL)
437		goto nla_put_failure;
438
439	if (!fh) {
440		t->tcm_handle = ~0; /* whatever ... */
441		NLA_PUT_U32(skb, TCA_TCINDEX_HASH, p->hash);
442		NLA_PUT_U16(skb, TCA_TCINDEX_MASK, p->mask);
443		NLA_PUT_U32(skb, TCA_TCINDEX_SHIFT, p->shift);
444		NLA_PUT_U32(skb, TCA_TCINDEX_FALL_THROUGH, p->fall_through);
 
445		nla_nest_end(skb, nest);
446	} else {
447		if (p->perfect) {
448			t->tcm_handle = r-p->perfect;
449		} else {
450			struct tcindex_filter *f;
 
451			int i;
452
453			t->tcm_handle = 0;
454			for (i = 0; !t->tcm_handle && i < p->hash; i++) {
455				for (f = p->h[i]; !t->tcm_handle && f;
456				     f = f->next) {
 
 
457					if (&f->result == r)
458						t->tcm_handle = f->key;
459				}
460			}
461		}
462		pr_debug("handle = %d\n", t->tcm_handle);
463		if (r->res.class)
464			NLA_PUT_U32(skb, TCA_TCINDEX_CLASSID, r->res.classid);
 
465
466		if (tcf_exts_dump(skb, &r->exts, &tcindex_ext_map) < 0)
467			goto nla_put_failure;
468		nla_nest_end(skb, nest);
469
470		if (tcf_exts_dump_stats(skb, &r->exts, &tcindex_ext_map) < 0)
471			goto nla_put_failure;
472	}
473
474	return skb->len;
475
476nla_put_failure:
477	nlmsg_trim(skb, b);
478	return -1;
479}
480
 
 
 
 
 
 
 
 
 
 
 
 
 
481static struct tcf_proto_ops cls_tcindex_ops __read_mostly = {
482	.kind		=	"tcindex",
483	.classify	=	tcindex_classify,
484	.init		=	tcindex_init,
485	.destroy	=	tcindex_destroy,
486	.get		=	tcindex_get,
487	.put		=	tcindex_put,
488	.change		=	tcindex_change,
489	.delete		=	tcindex_delete,
490	.walk		=	tcindex_walk,
491	.dump		=	tcindex_dump,
 
492	.owner		=	THIS_MODULE,
493};
494
495static int __init init_tcindex(void)
496{
497	return register_tcf_proto_ops(&cls_tcindex_ops);
498}
499
500static void __exit exit_tcindex(void)
501{
502	unregister_tcf_proto_ops(&cls_tcindex_ops);
503}
504
505module_init(init_tcindex)
506module_exit(exit_tcindex)
507MODULE_LICENSE("GPL");
v5.9
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * net/sched/cls_tcindex.c	Packet classifier for skb->tc_index
  4 *
  5 * Written 1998,1999 by Werner Almesberger, EPFL ICA
  6 */
  7
  8#include <linux/module.h>
  9#include <linux/types.h>
 10#include <linux/kernel.h>
 11#include <linux/skbuff.h>
 12#include <linux/errno.h>
 13#include <linux/slab.h>
 14#include <linux/refcount.h>
 15#include <net/act_api.h>
 16#include <net/netlink.h>
 17#include <net/pkt_cls.h>
 18#include <net/sch_generic.h>
 19
 20/*
 21 * Passing parameters to the root seems to be done more awkwardly than really
 22 * necessary. At least, u32 doesn't seem to use such dirty hacks. To be
 23 * verified. FIXME.
 24 */
 25
 26#define PERFECT_HASH_THRESHOLD	64	/* use perfect hash if not bigger */
 27#define DEFAULT_HASH_SIZE	64	/* optimized for diffserv */
 28
 29
 30struct tcindex_data;
 
 31
 32struct tcindex_filter_result {
 33	struct tcf_exts		exts;
 34	struct tcf_result	res;
 35	struct tcindex_data	*p;
 36	struct rcu_work		rwork;
 37};
 38
 39struct tcindex_filter {
 40	u16 key;
 41	struct tcindex_filter_result result;
 42	struct tcindex_filter __rcu *next;
 43	struct rcu_work rwork;
 44};
 45
 46
 47struct tcindex_data {
 48	struct tcindex_filter_result *perfect; /* perfect hash; NULL if none */
 49	struct tcindex_filter __rcu **h; /* imperfect hash; */
 50	struct tcf_proto *tp;
 51	u16 mask;		/* AND key with mask */
 52	u32 shift;		/* shift ANDed key to the right */
 53	u32 hash;		/* hash table size; 0 if undefined */
 54	u32 alloc_hash;		/* allocated size */
 55	u32 fall_through;	/* 0: only classify if explicit match */
 56	refcount_t refcnt;	/* a temporary refcnt for perfect hash */
 57	struct rcu_work rwork;
 58};
 59
 60static inline int tcindex_filter_is_set(struct tcindex_filter_result *r)
 61{
 62	return tcf_exts_has_actions(&r->exts) || r->res.classid;
 63}
 64
 65static void tcindex_data_get(struct tcindex_data *p)
 
 66{
 67	refcount_inc(&p->refcnt);
 68}
 69
 70static void tcindex_data_put(struct tcindex_data *p)
 
 71{
 72	if (refcount_dec_and_test(&p->refcnt)) {
 73		kfree(p->perfect);
 74		kfree(p->h);
 75		kfree(p);
 76	}
 77}
 78
 79static struct tcindex_filter_result *tcindex_lookup(struct tcindex_data *p,
 80						    u16 key)
 81{
 82	if (p->perfect) {
 83		struct tcindex_filter_result *f = p->perfect + key;
 84
 85		return tcindex_filter_is_set(f) ? f : NULL;
 86	} else if (p->h) {
 87		struct tcindex_filter __rcu **fp;
 88		struct tcindex_filter *f;
 89
 90		fp = &p->h[key % p->hash];
 91		for (f = rcu_dereference_bh_rtnl(*fp);
 92		     f;
 93		     fp = &f->next, f = rcu_dereference_bh_rtnl(*fp))
 94			if (f->key == key)
 95				return &f->result;
 96	}
 97
 98	return NULL;
 99}
100
101
102static int tcindex_classify(struct sk_buff *skb, const struct tcf_proto *tp,
103			    struct tcf_result *res)
104{
105	struct tcindex_data *p = rcu_dereference_bh(tp->root);
106	struct tcindex_filter_result *f;
107	int key = (skb->tc_index & p->mask) >> p->shift;
108
109	pr_debug("tcindex_classify(skb %p,tp %p,res %p),p %p\n",
110		 skb, tp, res, p);
111
112	f = tcindex_lookup(p, key);
113	if (!f) {
114		struct Qdisc *q = tcf_block_q(tp->chain->block);
115
116		if (!p->fall_through)
117			return -1;
118		res->classid = TC_H_MAKE(TC_H_MAJ(q->handle), key);
119		res->class = 0;
120		pr_debug("alg 0x%x\n", res->classid);
121		return 0;
122	}
123	*res = f->res;
124	pr_debug("map 0x%x\n", res->classid);
125
126	return tcf_exts_exec(skb, &f->exts, res);
127}
128
129
130static void *tcindex_get(struct tcf_proto *tp, u32 handle)
131{
132	struct tcindex_data *p = rtnl_dereference(tp->root);
133	struct tcindex_filter_result *r;
134
135	pr_debug("tcindex_get(tp %p,handle 0x%08x)\n", tp, handle);
136	if (p->perfect && handle >= p->alloc_hash)
137		return NULL;
138	r = tcindex_lookup(p, handle);
139	return r && tcindex_filter_is_set(r) ? r : NULL;
140}
141
 
 
 
 
 
 
 
142static int tcindex_init(struct tcf_proto *tp)
143{
144	struct tcindex_data *p;
145
146	pr_debug("tcindex_init(tp %p)\n", tp);
147	p = kzalloc(sizeof(struct tcindex_data), GFP_KERNEL);
148	if (!p)
149		return -ENOMEM;
150
151	p->mask = 0xffff;
152	p->hash = DEFAULT_HASH_SIZE;
153	p->fall_through = 1;
154	refcount_set(&p->refcnt, 1); /* Paired with tcindex_destroy_work() */
155
156	rcu_assign_pointer(tp->root, p);
157	return 0;
158}
159
160static void __tcindex_destroy_rexts(struct tcindex_filter_result *r)
161{
162	tcf_exts_destroy(&r->exts);
163	tcf_exts_put_net(&r->exts);
164	tcindex_data_put(r->p);
165}
166
167static void tcindex_destroy_rexts_work(struct work_struct *work)
168{
169	struct tcindex_filter_result *r;
170
171	r = container_of(to_rcu_work(work),
172			 struct tcindex_filter_result,
173			 rwork);
174	rtnl_lock();
175	__tcindex_destroy_rexts(r);
176	rtnl_unlock();
177}
178
179static void __tcindex_destroy_fexts(struct tcindex_filter *f)
180{
181	tcf_exts_destroy(&f->result.exts);
182	tcf_exts_put_net(&f->result.exts);
183	kfree(f);
184}
185
186static void tcindex_destroy_fexts_work(struct work_struct *work)
187{
188	struct tcindex_filter *f = container_of(to_rcu_work(work),
189						struct tcindex_filter,
190						rwork);
191
192	rtnl_lock();
193	__tcindex_destroy_fexts(f);
194	rtnl_unlock();
195}
196
197static int tcindex_delete(struct tcf_proto *tp, void *arg, bool *last,
198			  bool rtnl_held, struct netlink_ext_ack *extack)
199{
200	struct tcindex_data *p = rtnl_dereference(tp->root);
201	struct tcindex_filter_result *r = arg;
202	struct tcindex_filter __rcu **walk;
203	struct tcindex_filter *f = NULL;
204
205	pr_debug("tcindex_delete(tp %p,arg %p),p %p\n", tp, arg, p);
206	if (p->perfect) {
207		if (!r->res.class)
208			return -ENOENT;
209	} else {
210		int i;
 
211
212		for (i = 0; i < p->hash; i++) {
213			walk = p->h + i;
214			for (f = rtnl_dereference(*walk); f;
215			     walk = &f->next, f = rtnl_dereference(*walk)) {
216				if (&f->result == r)
217					goto found;
218			}
219		}
220		return -ENOENT;
221
222found:
223		rcu_assign_pointer(*walk, rtnl_dereference(f->next));
 
 
 
 
 
224	}
225	tcf_unbind_filter(tp, &r->res);
226	/* all classifiers are required to call tcf_exts_destroy() after rcu
227	 * grace period, since converted-to-rcu actions are relying on that
228	 * in cleanup() callback
229	 */
230	if (f) {
231		if (tcf_exts_get_net(&f->result.exts))
232			tcf_queue_work(&f->rwork, tcindex_destroy_fexts_work);
233		else
234			__tcindex_destroy_fexts(f);
235	} else {
236		tcindex_data_get(p);
237
238		if (tcf_exts_get_net(&r->exts))
239			tcf_queue_work(&r->rwork, tcindex_destroy_rexts_work);
240		else
241			__tcindex_destroy_rexts(r);
242	}
243
244	*last = false;
245	return 0;
246}
247
248static void tcindex_destroy_work(struct work_struct *work)
249{
250	struct tcindex_data *p = container_of(to_rcu_work(work),
251					      struct tcindex_data,
252					      rwork);
253
254	tcindex_data_put(p);
255}
256
257static inline int
258valid_perfect_hash(struct tcindex_data *p)
259{
260	return  p->hash > (p->mask >> p->shift);
261}
262
263static const struct nla_policy tcindex_policy[TCA_TCINDEX_MAX + 1] = {
264	[TCA_TCINDEX_HASH]		= { .type = NLA_U32 },
265	[TCA_TCINDEX_MASK]		= { .type = NLA_U16 },
266	[TCA_TCINDEX_SHIFT]		= { .type = NLA_U32 },
267	[TCA_TCINDEX_FALL_THROUGH]	= { .type = NLA_U32 },
268	[TCA_TCINDEX_CLASSID]		= { .type = NLA_U32 },
269};
270
271static int tcindex_filter_result_init(struct tcindex_filter_result *r,
272				      struct tcindex_data *p,
273				      struct net *net)
274{
275	memset(r, 0, sizeof(*r));
276	r->p = p;
277	return tcf_exts_init(&r->exts, net, TCA_TCINDEX_ACT,
278			     TCA_TCINDEX_POLICE);
279}
280
281static void tcindex_partial_destroy_work(struct work_struct *work)
282{
283	struct tcindex_data *p = container_of(to_rcu_work(work),
284					      struct tcindex_data,
285					      rwork);
286
287	rtnl_lock();
288	kfree(p->perfect);
289	kfree(p);
290	rtnl_unlock();
291}
292
293static void tcindex_free_perfect_hash(struct tcindex_data *cp)
294{
295	int i;
296
297	for (i = 0; i < cp->hash; i++)
298		tcf_exts_destroy(&cp->perfect[i].exts);
299	kfree(cp->perfect);
300}
301
302static int tcindex_alloc_perfect_hash(struct net *net, struct tcindex_data *cp)
303{
304	int i, err = 0;
305
306	cp->perfect = kcalloc(cp->hash, sizeof(struct tcindex_filter_result),
307			      GFP_KERNEL);
308	if (!cp->perfect)
309		return -ENOMEM;
310
311	for (i = 0; i < cp->hash; i++) {
312		err = tcf_exts_init(&cp->perfect[i].exts, net,
313				    TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
314		if (err < 0)
315			goto errout;
316		cp->perfect[i].p = cp;
317	}
318
319	return 0;
320
321errout:
322	tcindex_free_perfect_hash(cp);
323	return err;
324}
325
326static int
327tcindex_set_parms(struct net *net, struct tcf_proto *tp, unsigned long base,
328		  u32 handle, struct tcindex_data *p,
329		  struct tcindex_filter_result *r, struct nlattr **tb,
330		  struct nlattr *est, bool ovr, struct netlink_ext_ack *extack)
331{
 
332	struct tcindex_filter_result new_filter_result, *old_r = r;
333	struct tcindex_data *cp = NULL, *oldp;
 
334	struct tcindex_filter *f = NULL; /* make gcc behave */
335	struct tcf_result cr = {};
336	int err, balloc = 0;
337	struct tcf_exts e;
338
339	err = tcf_exts_init(&e, net, TCA_TCINDEX_ACT, TCA_TCINDEX_POLICE);
340	if (err < 0)
341		return err;
342	err = tcf_exts_validate(net, tp, tb, est, &e, ovr, true, extack);
343	if (err < 0)
344		goto errout;
345
346	err = -ENOMEM;
347	/* tcindex_data attributes must look atomic to classifier/lookup so
348	 * allocate new tcindex data and RCU assign it onto root. Keeping
349	 * perfect hash and hash pointers from old data.
350	 */
351	cp = kzalloc(sizeof(*cp), GFP_KERNEL);
352	if (!cp)
353		goto errout;
354
355	cp->mask = p->mask;
356	cp->shift = p->shift;
357	cp->hash = p->hash;
358	cp->alloc_hash = p->alloc_hash;
359	cp->fall_through = p->fall_through;
360	cp->tp = tp;
361	refcount_set(&cp->refcnt, 1); /* Paired with tcindex_destroy_work() */
362
363	if (tb[TCA_TCINDEX_HASH])
364		cp->hash = nla_get_u32(tb[TCA_TCINDEX_HASH]);
365
366	if (tb[TCA_TCINDEX_MASK])
367		cp->mask = nla_get_u16(tb[TCA_TCINDEX_MASK]);
368
369	if (tb[TCA_TCINDEX_SHIFT])
370		cp->shift = nla_get_u32(tb[TCA_TCINDEX_SHIFT]);
371
372	if (!cp->hash) {
373		/* Hash not specified, use perfect hash if the upper limit
374		 * of the hashing index is below the threshold.
375		 */
376		if ((cp->mask >> cp->shift) < PERFECT_HASH_THRESHOLD)
377			cp->hash = (cp->mask >> cp->shift) + 1;
378		else
379			cp->hash = DEFAULT_HASH_SIZE;
380	}
381
382	if (p->perfect) {
383		int i;
384
385		if (tcindex_alloc_perfect_hash(net, cp) < 0)
386			goto errout;
387		cp->alloc_hash = cp->hash;
388		for (i = 0; i < min(cp->hash, p->hash); i++)
389			cp->perfect[i].res = p->perfect[i].res;
390		balloc = 1;
391	}
392	cp->h = p->h;
393
394	err = tcindex_filter_result_init(&new_filter_result, cp, net);
395	if (err < 0)
396		goto errout_alloc;
397	if (old_r)
398		cr = r->res;
399
400	err = -EBUSY;
401
402	/* Hash already allocated, make sure that we still meet the
403	 * requirements for the allocated hash.
404	 */
405	if (cp->perfect) {
406		if (!valid_perfect_hash(cp) ||
407		    cp->hash > cp->alloc_hash)
408			goto errout_alloc;
409	} else if (cp->h && cp->hash != cp->alloc_hash) {
410		goto errout_alloc;
411	}
412
413	err = -EINVAL;
414	if (tb[TCA_TCINDEX_FALL_THROUGH])
415		cp->fall_through = nla_get_u32(tb[TCA_TCINDEX_FALL_THROUGH]);
 
 
 
 
 
 
 
 
 
 
416
417	if (!cp->perfect && !cp->h)
418		cp->alloc_hash = cp->hash;
419
420	/* Note: this could be as restrictive as if (handle & ~(mask >> shift))
421	 * but then, we'd fail handles that may become valid after some future
422	 * mask change. While this is extremely unlikely to ever matter,
423	 * the check below is safer (and also more backwards-compatible).
424	 */
425	if (cp->perfect || valid_perfect_hash(cp))
426		if (handle >= cp->alloc_hash)
427			goto errout_alloc;
428
429
430	err = -ENOMEM;
431	if (!cp->perfect && !cp->h) {
432		if (valid_perfect_hash(cp)) {
433			if (tcindex_alloc_perfect_hash(net, cp) < 0)
434				goto errout_alloc;
 
435			balloc = 1;
436		} else {
437			struct tcindex_filter __rcu **hash;
438
439			hash = kcalloc(cp->hash,
440				       sizeof(struct tcindex_filter *),
441				       GFP_KERNEL);
442
443			if (!hash)
444				goto errout_alloc;
445
446			cp->h = hash;
447			balloc = 2;
448		}
449	}
450
451	if (cp->perfect)
452		r = cp->perfect + handle;
453	else
454		r = tcindex_lookup(cp, handle) ? : &new_filter_result;
455
456	if (r == &new_filter_result) {
457		f = kzalloc(sizeof(*f), GFP_KERNEL);
458		if (!f)
459			goto errout_alloc;
460		f->key = handle;
461		f->next = NULL;
462		err = tcindex_filter_result_init(&f->result, cp, net);
463		if (err < 0) {
464			kfree(f);
465			goto errout_alloc;
466		}
467	}
468
469	if (tb[TCA_TCINDEX_CLASSID]) {
470		cr.classid = nla_get_u32(tb[TCA_TCINDEX_CLASSID]);
471		tcf_bind_filter(tp, &cr, base);
472	}
473
474	if (old_r && old_r != r) {
475		err = tcindex_filter_result_init(old_r, cp, net);
476		if (err < 0) {
477			kfree(f);
478			goto errout_alloc;
479		}
480	}
481
482	oldp = p;
483	r->res = cr;
484	tcf_exts_change(&r->exts, &e);
485
486	rcu_assign_pointer(tp->root, cp);
 
487
488	if (r == &new_filter_result) {
489		struct tcindex_filter *nfp;
490		struct tcindex_filter __rcu **fp;
491
492		f->result.res = r->res;
493		tcf_exts_change(&f->result.exts, &r->exts);
494
495		fp = cp->h + (handle % cp->hash);
496		for (nfp = rtnl_dereference(*fp);
497		     nfp;
498		     fp = &nfp->next, nfp = rtnl_dereference(*fp))
499				; /* nothing */
500
501		rcu_assign_pointer(*fp, f);
502	} else {
503		tcf_exts_destroy(&new_filter_result.exts);
504	}
 
505
506	if (oldp)
507		tcf_queue_work(&oldp->rwork, tcindex_partial_destroy_work);
508	return 0;
509
510errout_alloc:
511	if (balloc == 1)
512		tcindex_free_perfect_hash(cp);
513	else if (balloc == 2)
514		kfree(cp->h);
515	tcf_exts_destroy(&new_filter_result.exts);
516errout:
517	kfree(cp);
518	tcf_exts_destroy(&e);
519	return err;
520}
521
522static int
523tcindex_change(struct net *net, struct sk_buff *in_skb,
524	       struct tcf_proto *tp, unsigned long base, u32 handle,
525	       struct nlattr **tca, void **arg, bool ovr,
526	       bool rtnl_held, struct netlink_ext_ack *extack)
527{
528	struct nlattr *opt = tca[TCA_OPTIONS];
529	struct nlattr *tb[TCA_TCINDEX_MAX + 1];
530	struct tcindex_data *p = rtnl_dereference(tp->root);
531	struct tcindex_filter_result *r = *arg;
532	int err;
533
534	pr_debug("tcindex_change(tp %p,handle 0x%08x,tca %p,arg %p),opt %p,"
535	    "p %p,r %p,*arg %p\n",
536	    tp, handle, tca, arg, opt, p, r, *arg);
537
538	if (!opt)
539		return 0;
540
541	err = nla_parse_nested_deprecated(tb, TCA_TCINDEX_MAX, opt,
542					  tcindex_policy, NULL);
543	if (err < 0)
544		return err;
545
546	return tcindex_set_parms(net, tp, base, handle, p, r, tb,
547				 tca[TCA_RATE], ovr, extack);
548}
549
550static void tcindex_walk(struct tcf_proto *tp, struct tcf_walker *walker,
551			 bool rtnl_held)
552{
553	struct tcindex_data *p = rtnl_dereference(tp->root);
554	struct tcindex_filter *f, *next;
555	int i;
556
557	pr_debug("tcindex_walk(tp %p,walker %p),p %p\n", tp, walker, p);
558	if (p->perfect) {
559		for (i = 0; i < p->hash; i++) {
560			if (!p->perfect[i].res.class)
561				continue;
562			if (walker->count >= walker->skip) {
563				if (walker->fn(tp, p->perfect + i, walker) < 0) {
 
 
564					walker->stop = 1;
565					return;
566				}
567			}
568			walker->count++;
569		}
570	}
571	if (!p->h)
572		return;
573	for (i = 0; i < p->hash; i++) {
574		for (f = rtnl_dereference(p->h[i]); f; f = next) {
575			next = rtnl_dereference(f->next);
576			if (walker->count >= walker->skip) {
577				if (walker->fn(tp, &f->result, walker) < 0) {
 
578					walker->stop = 1;
579					return;
580				}
581			}
582			walker->count++;
583		}
584	}
585}
586
587static void tcindex_destroy(struct tcf_proto *tp, bool rtnl_held,
588			    struct netlink_ext_ack *extack)
 
589{
590	struct tcindex_data *p = rtnl_dereference(tp->root);
591	int i;
592
593	pr_debug("tcindex_destroy(tp %p),p %p\n", tp, p);
594
595	if (p->perfect) {
596		for (i = 0; i < p->hash; i++) {
597			struct tcindex_filter_result *r = p->perfect + i;
 
598
599			/* tcf_queue_work() does not guarantee the ordering we
600			 * want, so we have to take this refcnt temporarily to
601			 * ensure 'p' is freed after all tcindex_filter_result
602			 * here. Imperfect hash does not need this, because it
603			 * uses linked lists rather than an array.
604			 */
605			tcindex_data_get(p);
606
607			tcf_unbind_filter(tp, &r->res);
608			if (tcf_exts_get_net(&r->exts))
609				tcf_queue_work(&r->rwork,
610					       tcindex_destroy_rexts_work);
611			else
612				__tcindex_destroy_rexts(r);
613		}
614	}
615
616	for (i = 0; p->h && i < p->hash; i++) {
617		struct tcindex_filter *f, *next;
618		bool last;
619
620		for (f = rtnl_dereference(p->h[i]); f; f = next) {
621			next = rtnl_dereference(f->next);
622			tcindex_delete(tp, &f->result, &last, rtnl_held, NULL);
623		}
624	}
625
626	tcf_queue_work(&p->rwork, tcindex_destroy_work);
627}
628
629
630static int tcindex_dump(struct net *net, struct tcf_proto *tp, void *fh,
631			struct sk_buff *skb, struct tcmsg *t, bool rtnl_held)
632{
633	struct tcindex_data *p = rtnl_dereference(tp->root);
634	struct tcindex_filter_result *r = fh;
 
635	struct nlattr *nest;
636
637	pr_debug("tcindex_dump(tp %p,fh %p,skb %p,t %p),p %p,r %p\n",
638		 tp, fh, skb, t, p, r);
639	pr_debug("p->perfect %p p->h %p\n", p->perfect, p->h);
640
641	nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
642	if (nest == NULL)
643		goto nla_put_failure;
644
645	if (!fh) {
646		t->tcm_handle = ~0; /* whatever ... */
647		if (nla_put_u32(skb, TCA_TCINDEX_HASH, p->hash) ||
648		    nla_put_u16(skb, TCA_TCINDEX_MASK, p->mask) ||
649		    nla_put_u32(skb, TCA_TCINDEX_SHIFT, p->shift) ||
650		    nla_put_u32(skb, TCA_TCINDEX_FALL_THROUGH, p->fall_through))
651			goto nla_put_failure;
652		nla_nest_end(skb, nest);
653	} else {
654		if (p->perfect) {
655			t->tcm_handle = r - p->perfect;
656		} else {
657			struct tcindex_filter *f;
658			struct tcindex_filter __rcu **fp;
659			int i;
660
661			t->tcm_handle = 0;
662			for (i = 0; !t->tcm_handle && i < p->hash; i++) {
663				fp = &p->h[i];
664				for (f = rtnl_dereference(*fp);
665				     !t->tcm_handle && f;
666				     fp = &f->next, f = rtnl_dereference(*fp)) {
667					if (&f->result == r)
668						t->tcm_handle = f->key;
669				}
670			}
671		}
672		pr_debug("handle = %d\n", t->tcm_handle);
673		if (r->res.class &&
674		    nla_put_u32(skb, TCA_TCINDEX_CLASSID, r->res.classid))
675			goto nla_put_failure;
676
677		if (tcf_exts_dump(skb, &r->exts) < 0)
678			goto nla_put_failure;
679		nla_nest_end(skb, nest);
680
681		if (tcf_exts_dump_stats(skb, &r->exts) < 0)
682			goto nla_put_failure;
683	}
684
685	return skb->len;
686
687nla_put_failure:
688	nla_nest_cancel(skb, nest);
689	return -1;
690}
691
692static void tcindex_bind_class(void *fh, u32 classid, unsigned long cl,
693			       void *q, unsigned long base)
694{
695	struct tcindex_filter_result *r = fh;
696
697	if (r && r->res.classid == classid) {
698		if (cl)
699			__tcf_bind_filter(q, &r->res, base);
700		else
701			__tcf_unbind_filter(q, &r->res);
702	}
703}
704
705static struct tcf_proto_ops cls_tcindex_ops __read_mostly = {
706	.kind		=	"tcindex",
707	.classify	=	tcindex_classify,
708	.init		=	tcindex_init,
709	.destroy	=	tcindex_destroy,
710	.get		=	tcindex_get,
 
711	.change		=	tcindex_change,
712	.delete		=	tcindex_delete,
713	.walk		=	tcindex_walk,
714	.dump		=	tcindex_dump,
715	.bind_class	=	tcindex_bind_class,
716	.owner		=	THIS_MODULE,
717};
718
719static int __init init_tcindex(void)
720{
721	return register_tcf_proto_ops(&cls_tcindex_ops);
722}
723
724static void __exit exit_tcindex(void)
725{
726	unregister_tcf_proto_ops(&cls_tcindex_ops);
727}
728
729module_init(init_tcindex)
730module_exit(exit_tcindex)
731MODULE_LICENSE("GPL");