[net-next-2.6.git] / net / core / flow.c

/* flow.c: Generic flow cache.
 *
 * Copyright (C) 2003 Alexey N. Kuznetsov (kuznet@ms2.inr.ac.ru)
 * Copyright (C) 2003 David S. Miller (davem@redhat.com)
 */

#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/list.h>
#include <linux/jhash.h>
#include <linux/interrupt.h>
#include <linux/mm.h>
#include <linux/random.h>
#include <linux/init.h>
#include <linux/slab.h>
#include <linux/smp.h>
#include <linux/completion.h>
#include <linux/percpu.h>
#include <linux/bitops.h>
#include <linux/notifier.h>
#include <linux/cpu.h>
#include <linux/cpumask.h>
#include <linux/mutex.h>
#include <net/flow.h>
#include <asm/atomic.h>
#include <linux/security.h>

struct flow_cache_entry {
	struct flow_cache_entry	*next;
	u16			family;
	u8			dir;
	u32			genid;
	struct flowi		key;
	void			*object;
	atomic_t		*object_ref;
};

atomic_t flow_cache_genid = ATOMIC_INIT(0);

static u32 flow_hash_shift;
#define flow_hash_size	(1 << flow_hash_shift)
static DEFINE_PER_CPU(struct flow_cache_entry **, flow_tables) = { NULL };

#define flow_table(cpu) (per_cpu(flow_tables, cpu))

static struct kmem_cache *flow_cachep __read_mostly;

static int flow_lwm, flow_hwm;

struct flow_percpu_info {
	int hash_rnd_recalc;
	u32 hash_rnd;
	int count;
};
static DEFINE_PER_CPU(struct flow_percpu_info, flow_hash_info) = { 0 };

#define flow_hash_rnd_recalc(cpu) \
	(per_cpu(flow_hash_info, cpu).hash_rnd_recalc)
#define flow_hash_rnd(cpu) \
	(per_cpu(flow_hash_info, cpu).hash_rnd)
#define flow_count(cpu) \
	(per_cpu(flow_hash_info, cpu).count)

static struct timer_list flow_hash_rnd_timer;

#define FLOW_HASH_RND_PERIOD	(10 * 60 * HZ)

struct flow_flush_info {
	atomic_t cpuleft;
	struct completion completion;
};
static DEFINE_PER_CPU(struct tasklet_struct, flow_flush_tasklets) = { NULL };

#define flow_flush_tasklet(cpu) (&per_cpu(flow_flush_tasklets, cpu))

static void flow_cache_new_hashrnd(unsigned long arg)
{
	int i;

	for_each_possible_cpu(i)
		flow_hash_rnd_recalc(i) = 1;

	flow_hash_rnd_timer.expires = jiffies + FLOW_HASH_RND_PERIOD;
	add_timer(&flow_hash_rnd_timer);
}

static void flow_entry_kill(int cpu, struct flow_cache_entry *fle)
{
	if (fle->object)
		atomic_dec(fle->object_ref);
	kmem_cache_free(flow_cachep, fle);
	flow_count(cpu)--;
}

static void __flow_cache_shrink(int cpu, int shrink_to)
{
	struct flow_cache_entry *fle, **flp;
	int i;

	for (i = 0; i < flow_hash_size; i++) {
		int k = 0;

		flp = &flow_table(cpu)[i];
		while ((fle = *flp) != NULL && k < shrink_to) {
			k++;
			flp = &fle->next;
		}
		while ((fle = *flp) != NULL) {
			*flp = fle->next;
			flow_entry_kill(cpu, fle);
		}
	}
}

static void flow_cache_shrink(int cpu)
{
	int shrink_to = flow_lwm / flow_hash_size;

	__flow_cache_shrink(cpu, shrink_to);
}

static void flow_new_hash_rnd(int cpu)
{
	get_random_bytes(&flow_hash_rnd(cpu), sizeof(u32));
	flow_hash_rnd_recalc(cpu) = 0;

	__flow_cache_shrink(cpu, 0);
}

static u32 flow_hash_code(struct flowi *key, int cpu)
{
	u32 *k = (u32 *) key;

	return (jhash2(k, (sizeof(*key) / sizeof(u32)), flow_hash_rnd(cpu)) &
		(flow_hash_size - 1));
}

#if (BITS_PER_LONG == 64)
typedef u64 flow_compare_t;
#else
typedef u32 flow_compare_t;
#endif

/* I hear what you're saying, use memcmp.  But memcmp cannot make
 * important assumptions that we can here, such as alignment and
 * constant size.
 */
static int flow_key_compare(struct flowi *key1, struct flowi *key2)
{
	flow_compare_t *k1, *k1_lim, *k2;
	const int n_elem = sizeof(struct flowi) / sizeof(flow_compare_t);

	BUILD_BUG_ON(sizeof(struct flowi) % sizeof(flow_compare_t));

	k1 = (flow_compare_t *) key1;
	k1_lim = k1 + n_elem;

	k2 = (flow_compare_t *) key2;

	do {
		if (*k1++ != *k2++)
			return 1;
	} while (k1 < k1_lim);

	return 0;
}

void *flow_cache_lookup(struct net *net, struct flowi *key, u16 family, u8 dir,
			flow_resolve_t resolver)
{
	struct flow_cache_entry *fle, **head;
	unsigned int hash;
	int cpu;

	local_bh_disable();
	cpu = smp_processor_id();

	fle = NULL;
	/* Packet really early in init?  Making flow_cache_init a
	 * pre-smp initcall would solve this.  --RR */
	if (!flow_table(cpu))
		goto nocache;

	if (flow_hash_rnd_recalc(cpu))
		flow_new_hash_rnd(cpu);
	hash = flow_hash_code(key, cpu);

	head = &flow_table(cpu)[hash];
	for (fle = *head; fle; fle = fle->next) {
		if (fle->family == family &&
		    fle->dir == dir &&
		    flow_key_compare(key, &fle->key) == 0) {
			if (fle->genid == atomic_read(&flow_cache_genid)) {
				void *ret = fle->object;

				if (ret)
					atomic_inc(fle->object_ref);
				local_bh_enable();

				return ret;
			}
			break;
		}
	}

	if (!fle) {
		if (flow_count(cpu) > flow_hwm)
			flow_cache_shrink(cpu);

		fle = kmem_cache_alloc(flow_cachep, GFP_ATOMIC);
		if (fle) {
			fle->next = *head;
			*head = fle;
			fle->family = family;
			fle->dir = dir;
			memcpy(&fle->key, key, sizeof(*key));
			fle->object = NULL;
			flow_count(cpu)++;
		}
	}

nocache:
	{
		int err;
		void *obj;
		atomic_t *obj_ref;

		err = resolver(net, key, family, dir, &obj, &obj_ref);

		if (fle && !err) {
			fle->genid = atomic_read(&flow_cache_genid);

			if (fle->object)
				atomic_dec(fle->object_ref);

			fle->object = obj;
			fle->object_ref = obj_ref;
			if (obj)
				atomic_inc(fle->object_ref);
		}
		local_bh_enable();

		if (err)
			obj = ERR_PTR(err);
		return obj;
	}
}

static void flow_cache_flush_tasklet(unsigned long data)
{
	struct flow_flush_info *info = (void *)data;
	int i;
	int cpu;

	cpu = smp_processor_id();
	for (i = 0; i < flow_hash_size; i++) {
		struct flow_cache_entry *fle;

		fle = flow_table(cpu)[i];
		for (; fle; fle = fle->next) {
			unsigned genid = atomic_read(&flow_cache_genid);

			if (!fle->object || fle->genid == genid)
				continue;

			fle->object = NULL;
			atomic_dec(fle->object_ref);
		}
	}

	if (atomic_dec_and_test(&info->cpuleft))
		complete(&info->completion);
}

static void flow_cache_flush_per_cpu(void *) __attribute__((__unused__));
static void flow_cache_flush_per_cpu(void *data)
{
	struct flow_flush_info *info = data;
	int cpu;
	struct tasklet_struct *tasklet;

	cpu = smp_processor_id();

	tasklet = flow_flush_tasklet(cpu);
	tasklet->data = (unsigned long)info;
	tasklet_schedule(tasklet);
}

void flow_cache_flush(void)
{
	struct flow_flush_info info;
	static DEFINE_MUTEX(flow_flush_sem);

	/* Don't want cpus going down or up during this. */
	get_online_cpus();
	mutex_lock(&flow_flush_sem);
	atomic_set(&info.cpuleft, num_online_cpus());
	init_completion(&info.completion);

	local_bh_disable();
	smp_call_function(flow_cache_flush_per_cpu, &info, 0);
	flow_cache_flush_tasklet((unsigned long)&info);
	local_bh_enable();

	wait_for_completion(&info.completion);
	mutex_unlock(&flow_flush_sem);
	put_online_cpus();
}

static void __init flow_cache_cpu_prepare(int cpu)
{
	struct tasklet_struct *tasklet;
	unsigned long order;

	for (order = 0;
	     (PAGE_SIZE << order) <
		     (sizeof(struct flow_cache_entry *)*flow_hash_size);
	     order++)
		/* NOTHING */;

	flow_table(cpu) = (struct flow_cache_entry **)
		__get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
	if (!flow_table(cpu))
		panic("NET: failed to allocate flow cache order %lu\n", order);

	flow_hash_rnd_recalc(cpu) = 1;
	flow_count(cpu) = 0;

	tasklet = flow_flush_tasklet(cpu);
	tasklet_init(tasklet, flow_cache_flush_tasklet, 0);
}

static int flow_cache_cpu(struct notifier_block *nfb,
			  unsigned long action,
			  void *hcpu)
{
	if (action == CPU_DEAD || action == CPU_DEAD_FROZEN)
		__flow_cache_shrink((unsigned long)hcpu, 0);
	return NOTIFY_OK;
}

static int __init flow_cache_init(void)
{
	int i;

	flow_cachep = kmem_cache_create("flow_cache",
					sizeof(struct flow_cache_entry),
					0, SLAB_PANIC,
					NULL);
	flow_hash_shift = 10;
	flow_lwm = 2 * flow_hash_size;
	flow_hwm = 4 * flow_hash_size;

	setup_timer(&flow_hash_rnd_timer, flow_cache_new_hashrnd, 0);
	flow_hash_rnd_timer.expires = jiffies + FLOW_HASH_RND_PERIOD;
	add_timer(&flow_hash_rnd_timer);

	for_each_possible_cpu(i)
		flow_cache_cpu_prepare(i);

	hotcpu_notifier(flow_cache_cpu, 0);
	return 0;
}

module_init(flow_cache_init);

EXPORT_SYMBOL(flow_cache_genid);
EXPORT_SYMBOL(flow_cache_lookup);
Commit	Line	Data
1da177e4 LT	1	/* flow.c: Generic flow cache.
	2	*
	3	* Copyright (C) 2003 Alexey N. Kuznetsov (kuznet@ms2.inr.ac.ru)
	4	* Copyright (C) 2003 David S. Miller (davem@redhat.com)
	5	*/
	6
	7	#include <linux/kernel.h>
	8	#include <linux/module.h>
	9	#include <linux/list.h>
	10	#include <linux/jhash.h>
	11	#include <linux/interrupt.h>
	12	#include <linux/mm.h>
	13	#include <linux/random.h>
	14	#include <linux/init.h>
	15	#include <linux/slab.h>
	16	#include <linux/smp.h>
	17	#include <linux/completion.h>
	18	#include <linux/percpu.h>
	19	#include <linux/bitops.h>
	20	#include <linux/notifier.h>
	21	#include <linux/cpu.h>
	22	#include <linux/cpumask.h>
4a3e2f71	23	#include <linux/mutex.h>
1da177e4 LT	24	#include <net/flow.h>
1da177e4 LT	25	#include <asm/atomic.h>
df71837d	26	#include <linux/security.h>
1da177e4 LT	27
	28	struct flow_cache_entry {
	29	struct flow_cache_entry *next;
	30	u16 family;
	31	u8 dir;
1da177e4	32	u32 genid;
dd5a1843	33	struct flowi key;
1da177e4 LT	34	void *object;
	35	atomic_t *object_ref;
	36	};
	37
	38	atomic_t flow_cache_genid = ATOMIC_INIT(0);
	39
	40	static u32 flow_hash_shift;
	41	#define flow_hash_size (1 << flow_hash_shift)
	42	static DEFINE_PER_CPU(struct flow_cache_entry **, flow_tables) = { NULL };
	43
	44	#define flow_table(cpu) (per_cpu(flow_tables, cpu))
	45
e18b890b	46	static struct kmem_cache *flow_cachep __read_mostly;
1da177e4 LT	47
	48	static int flow_lwm, flow_hwm;
	49
	50	struct flow_percpu_info {
	51	int hash_rnd_recalc;
	52	u32 hash_rnd;
	53	int count;
5f58a5c8	54	};
1da177e4 LT	55	static DEFINE_PER_CPU(struct flow_percpu_info, flow_hash_info) = { 0 };
	56
	57	#define flow_hash_rnd_recalc(cpu) \
	58	(per_cpu(flow_hash_info, cpu).hash_rnd_recalc)
	59	#define flow_hash_rnd(cpu) \
	60	(per_cpu(flow_hash_info, cpu).hash_rnd)
	61	#define flow_count(cpu) \
	62	(per_cpu(flow_hash_info, cpu).count)
	63
	64	static struct timer_list flow_hash_rnd_timer;
	65
	66	#define FLOW_HASH_RND_PERIOD (10 * 60 * HZ)
	67
	68	struct flow_flush_info {
	69	atomic_t cpuleft;
	70	struct completion completion;
	71	};
	72	static DEFINE_PER_CPU(struct tasklet_struct, flow_flush_tasklets) = { NULL };
	73
	74	#define flow_flush_tasklet(cpu) (&per_cpu(flow_flush_tasklets, cpu))
	75
	76	static void flow_cache_new_hashrnd(unsigned long arg)
	77	{
	78	int i;
	79
6f912042	80	for_each_possible_cpu(i)
1da177e4 LT	81	flow_hash_rnd_recalc(i) = 1;
	82
	83	flow_hash_rnd_timer.expires = jiffies + FLOW_HASH_RND_PERIOD;
	84	add_timer(&flow_hash_rnd_timer);
	85	}
	86
134b0fc5 JM	87	static void flow_entry_kill(int cpu, struct flow_cache_entry *fle)
	88	{
	89	if (fle->object)
	90	atomic_dec(fle->object_ref);
	91	kmem_cache_free(flow_cachep, fle);
	92	flow_count(cpu)--;
	93	}
	94
1da177e4 LT	95	static void __flow_cache_shrink(int cpu, int shrink_to)
	96	{
	97	struct flow_cache_entry fle, *flp;
	98	int i;
	99
	100	for (i = 0; i < flow_hash_size; i++) {
	101	int k = 0;
	102
	103	flp = &flow_table(cpu)[i];
	104	while ((fle = *flp) != NULL && k < shrink_to) {
	105	k++;
	106	flp = &fle->next;
	107	}
	108	while ((fle = *flp) != NULL) {
	109	*flp = fle->next;
134b0fc5	110	flow_entry_kill(cpu, fle);
1da177e4 LT	111	}
	112	}
	113	}
	114
	115	static void flow_cache_shrink(int cpu)
	116	{
	117	int shrink_to = flow_lwm / flow_hash_size;
	118
	119	__flow_cache_shrink(cpu, shrink_to);
	120	}
	121
	122	static void flow_new_hash_rnd(int cpu)
	123	{
	124	get_random_bytes(&flow_hash_rnd(cpu), sizeof(u32));
	125	flow_hash_rnd_recalc(cpu) = 0;
	126
	127	__flow_cache_shrink(cpu, 0);
	128	}
	129
	130	static u32 flow_hash_code(struct flowi *key, int cpu)
	131	{
	132	u32 k = (u32 ) key;
	133
	134	return (jhash2(k, (sizeof(*key) / sizeof(u32)), flow_hash_rnd(cpu)) &
	135	(flow_hash_size - 1));
	136	}
	137
	138	#if (BITS_PER_LONG == 64)
	139	typedef u64 flow_compare_t;
	140	#else
	141	typedef u32 flow_compare_t;
	142	#endif
	143
1da177e4 LT	144	/* I hear what you're saying, use memcmp. But memcmp cannot make
	145	* important assumptions that we can here, such as alignment and
	146	* constant size.
	147	*/
	148	static int flow_key_compare(struct flowi key1, struct flowi key2)
	149	{
	150	flow_compare_t k1, k1_lim, *k2;
	151	const int n_elem = sizeof(struct flowi) / sizeof(flow_compare_t);
	152
f0fe91de	153	BUILD_BUG_ON(sizeof(struct flowi) % sizeof(flow_compare_t));
1da177e4 LT	154
	155	k1 = (flow_compare_t *) key1;
	156	k1_lim = k1 + n_elem;
	157
	158	k2 = (flow_compare_t *) key2;
	159
	160	do {
	161	if (k1++ != k2++)
	162	return 1;
	163	} while (k1 < k1_lim);
	164
	165	return 0;
	166	}
	167
52479b62	168	void flow_cache_lookup(struct net net, struct flowi *key, u16 family, u8 dir,
1da177e4 LT	169	flow_resolve_t resolver)
	170	{
	171	struct flow_cache_entry fle, *head;
	172	unsigned int hash;
	173	int cpu;
	174
	175	local_bh_disable();
	176	cpu = smp_processor_id();
	177
	178	fle = NULL;
	179	/* Packet really early in init? Making flow_cache_init a
	180	* pre-smp initcall would solve this. --RR */
	181	if (!flow_table(cpu))
	182	goto nocache;
	183
	184	if (flow_hash_rnd_recalc(cpu))
	185	flow_new_hash_rnd(cpu);
	186	hash = flow_hash_code(key, cpu);
	187
	188	head = &flow_table(cpu)[hash];
	189	for (fle = *head; fle; fle = fle->next) {
	190	if (fle->family == family &&
	191	fle->dir == dir &&
	192	flow_key_compare(key, &fle->key) == 0) {
	193	if (fle->genid == atomic_read(&flow_cache_genid)) {
	194	void *ret = fle->object;
	195
	196	if (ret)
	197	atomic_inc(fle->object_ref);
	198	local_bh_enable();
	199
	200	return ret;
	201	}
	202	break;
	203	}
	204	}
	205
	206	if (!fle) {
	207	if (flow_count(cpu) > flow_hwm)
	208	flow_cache_shrink(cpu);
	209
54e6ecb2	210	fle = kmem_cache_alloc(flow_cachep, GFP_ATOMIC);
1da177e4 LT	211	if (fle) {
	212	fle->next = *head;
	213	*head = fle;
	214	fle->family = family;
	215	fle->dir = dir;
	216	memcpy(&fle->key, key, sizeof(*key));
	217	fle->object = NULL;
	218	flow_count(cpu)++;
	219	}
	220	}
	221
	222	nocache:
	223	{
134b0fc5	224	int err;
1da177e4 LT	225	void *obj;
	226	atomic_t *obj_ref;
	227
52479b62	228	err = resolver(net, key, family, dir, &obj, &obj_ref);
1da177e4	229
e0e8f1c8 HX	230	if (fle && !err) {
	231	fle->genid = atomic_read(&flow_cache_genid);
	232
	233	if (fle->object)
	234	atomic_dec(fle->object_ref);
	235
	236	fle->object = obj;
	237	fle->object_ref = obj_ref;
	238	if (obj)
	239	atomic_inc(fle->object_ref);
1da177e4 LT	240	}
	241	local_bh_enable();
	242
134b0fc5 JM	243	if (err)
134b0fc5 JM	244	obj = ERR_PTR(err);
1da177e4 LT	245	return obj;
	246	}
	247	}
	248
	249	static void flow_cache_flush_tasklet(unsigned long data)
	250	{
	251	struct flow_flush_info info = (void )data;
	252	int i;
	253	int cpu;
	254
	255	cpu = smp_processor_id();
	256	for (i = 0; i < flow_hash_size; i++) {
	257	struct flow_cache_entry *fle;
	258
	259	fle = flow_table(cpu)[i];
	260	for (; fle; fle = fle->next) {
	261	unsigned genid = atomic_read(&flow_cache_genid);
	262
	263	if (!fle->object \|\| fle->genid == genid)
	264	continue;
	265
	266	fle->object = NULL;
	267	atomic_dec(fle->object_ref);
	268	}
	269	}
	270
	271	if (atomic_dec_and_test(&info->cpuleft))
	272	complete(&info->completion);
	273	}
	274
	275	static void flow_cache_flush_per_cpu(void *) __attribute__((__unused__));
	276	static void flow_cache_flush_per_cpu(void *data)
	277	{
	278	struct flow_flush_info *info = data;
	279	int cpu;
	280	struct tasklet_struct *tasklet;
	281
	282	cpu = smp_processor_id();
	283
	284	tasklet = flow_flush_tasklet(cpu);
	285	tasklet->data = (unsigned long)info;
	286	tasklet_schedule(tasklet);
	287	}
	288
	289	void flow_cache_flush(void)
	290	{
	291	struct flow_flush_info info;
4a3e2f71	292	static DEFINE_MUTEX(flow_flush_sem);
1da177e4 LT	293
1da177e4 LT	294	/* Don't want cpus going down or up during this. */
86ef5c9a	295	get_online_cpus();
4a3e2f71	296	mutex_lock(&flow_flush_sem);
1da177e4 LT	297	atomic_set(&info.cpuleft, num_online_cpus());
	298	init_completion(&info.completion);
	299
	300	local_bh_disable();
8691e5a8	301	smp_call_function(flow_cache_flush_per_cpu, &info, 0);
1da177e4 LT	302	flow_cache_flush_tasklet((unsigned long)&info);
	303	local_bh_enable();
	304
	305	wait_for_completion(&info.completion);
4a3e2f71	306	mutex_unlock(&flow_flush_sem);
86ef5c9a	307	put_online_cpus();
1da177e4 LT	308	}
1da177e4 LT	309
76acfdb9	310	static void __init flow_cache_cpu_prepare(int cpu)
1da177e4 LT	311	{
	312	struct tasklet_struct *tasklet;
	313	unsigned long order;
	314
	315	for (order = 0;
	316	(PAGE_SIZE << order) <
	317	(sizeof(struct flow_cache_entry )flow_hash_size);
	318	order++)
	319	/* NOTHING */;
	320
	321	flow_table(cpu) = (struct flow_cache_entry **)
77d04bd9	322	__get_free_pages(GFP_KERNEL\|__GFP_ZERO, order);
1da177e4 LT	323	if (!flow_table(cpu))
	324	panic("NET: failed to allocate flow cache order %lu\n", order);
	325
1da177e4 LT	326	flow_hash_rnd_recalc(cpu) = 1;
	327	flow_count(cpu) = 0;
	328
	329	tasklet = flow_flush_tasklet(cpu);
	330	tasklet_init(tasklet, flow_cache_flush_tasklet, 0);
	331	}
	332
1da177e4 LT	333	static int flow_cache_cpu(struct notifier_block *nfb,
	334	unsigned long action,
	335	void *hcpu)
	336	{
8bb78442	337	if (action == CPU_DEAD \|\| action == CPU_DEAD_FROZEN)
1da177e4 LT	338	__flow_cache_shrink((unsigned long)hcpu, 0);
	339	return NOTIFY_OK;
	340	}
1da177e4 LT	341
	342	static int __init flow_cache_init(void)
	343	{
	344	int i;
	345
	346	flow_cachep = kmem_cache_create("flow_cache",
	347	sizeof(struct flow_cache_entry),
dd5a1843	348	0, SLAB_PANIC,
20c2df83	349	NULL);
1da177e4 LT	350	flow_hash_shift = 10;
	351	flow_lwm = 2 * flow_hash_size;
	352	flow_hwm = 4 * flow_hash_size;
	353
b24b8a24	354	setup_timer(&flow_hash_rnd_timer, flow_cache_new_hashrnd, 0);
1da177e4 LT	355	flow_hash_rnd_timer.expires = jiffies + FLOW_HASH_RND_PERIOD;
	356	add_timer(&flow_hash_rnd_timer);
	357
6f912042	358	for_each_possible_cpu(i)
1da177e4 LT	359	flow_cache_cpu_prepare(i);
	360
	361	hotcpu_notifier(flow_cache_cpu, 0);
	362	return 0;
	363	}
	364
	365	module_init(flow_cache_init);
	366
	367	EXPORT_SYMBOL(flow_cache_genid);
	368	EXPORT_SYMBOL(flow_cache_lookup);