[linux.git] / net / core / flow.c

/* flow.c: Generic flow cache.
 *
 * Copyright (C) 2003 Alexey N. Kuznetsov ([email protected])
 * Copyright (C) 2003 David S. Miller ([email protected])
 */

#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 <asm/semaphore.h>
#include <linux/security.h>

struct flow_cache_entry {
	struct flow_cache_entry	*next;
	u16			family;
	u8			dir;
	struct flowi		key;
	u32			genid;
	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 kmem_cache_t *flow_cachep __read_mostly;

static int flow_lwm, flow_hwm;

struct flow_percpu_info {
	int hash_rnd_recalc;
	u32 hash_rnd;
	int count;
} ____cacheline_aligned;
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_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;
			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 = 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

extern void flowi_is_missized(void);

/* 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);

	if (sizeof(struct flowi) % sizeof(flow_compare_t))
		flowi_is_missized();

	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 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, SLAB_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:
	{
		void *obj;
		atomic_t *obj_ref;

		resolver(key, family, dir, &obj, &obj_ref);

		if (fle) {
			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();

		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. */
	lock_cpu_hotplug();
	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, 1, 0);
	flow_cache_flush_tasklet((unsigned long)&info);
	local_bh_enable();

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

static void __devinit 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);
}

#ifdef CONFIG_HOTPLUG_CPU
static int flow_cache_cpu(struct notifier_block *nfb,
			  unsigned long action,
			  void *hcpu)
{
	if (action == CPU_DEAD)
		__flow_cache_shrink((unsigned long)hcpu, 0);
	return NOTIFY_OK;
}
#endif /* CONFIG_HOTPLUG_CPU */

static int __init flow_cache_init(void)
{
	int i;

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

	init_timer(&flow_hash_rnd_timer);
	flow_hash_rnd_timer.function = flow_cache_new_hashrnd;
	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 ([email protected])
	4	* Copyright (C) 2003 David S. Miller ([email protected])
	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>
	25	#include <asm/atomic.h>
	26	#include <asm/semaphore.h>
df71837d	27	#include <linux/security.h>
1da177e4 LT	28
	29	struct flow_cache_entry {
	30	struct flow_cache_entry *next;
	31	u16 family;
	32	u8 dir;
	33	struct flowi key;
	34	u32 genid;
	35	void *object;
	36	atomic_t *object_ref;
	37	};
	38
	39	atomic_t flow_cache_genid = ATOMIC_INIT(0);
	40
	41	static u32 flow_hash_shift;
	42	#define flow_hash_size (1 << flow_hash_shift)
	43	static DEFINE_PER_CPU(struct flow_cache_entry **, flow_tables) = { NULL };
	44
	45	#define flow_table(cpu) (per_cpu(flow_tables, cpu))
	46
ba89966c	47	static kmem_cache_t *flow_cachep __read_mostly;
1da177e4 LT	48
	49	static int flow_lwm, flow_hwm;
	50
	51	struct flow_percpu_info {
	52	int hash_rnd_recalc;
	53	u32 hash_rnd;
	54	int count;
	55	} ____cacheline_aligned;
	56	static DEFINE_PER_CPU(struct flow_percpu_info, flow_hash_info) = { 0 };
	57
	58	#define flow_hash_rnd_recalc(cpu) \
	59	(per_cpu(flow_hash_info, cpu).hash_rnd_recalc)
	60	#define flow_hash_rnd(cpu) \
	61	(per_cpu(flow_hash_info, cpu).hash_rnd)
	62	#define flow_count(cpu) \
	63	(per_cpu(flow_hash_info, cpu).count)
	64
	65	static struct timer_list flow_hash_rnd_timer;
	66
	67	#define FLOW_HASH_RND_PERIOD (10 * 60 * HZ)
	68
	69	struct flow_flush_info {
	70	atomic_t cpuleft;
	71	struct completion completion;
	72	};
	73	static DEFINE_PER_CPU(struct tasklet_struct, flow_flush_tasklets) = { NULL };
	74
	75	#define flow_flush_tasklet(cpu) (&per_cpu(flow_flush_tasklets, cpu))
	76
	77	static void flow_cache_new_hashrnd(unsigned long arg)
	78	{
	79	int i;
	80
6f912042	81	for_each_possible_cpu(i)
1da177e4 LT	82	flow_hash_rnd_recalc(i) = 1;
	83
	84	flow_hash_rnd_timer.expires = jiffies + FLOW_HASH_RND_PERIOD;
	85	add_timer(&flow_hash_rnd_timer);
	86	}
	87
	88	static void __flow_cache_shrink(int cpu, int shrink_to)
	89	{
	90	struct flow_cache_entry fle, *flp;
	91	int i;
	92
	93	for (i = 0; i < flow_hash_size; i++) {
	94	int k = 0;
	95
	96	flp = &flow_table(cpu)[i];
	97	while ((fle = *flp) != NULL && k < shrink_to) {
	98	k++;
	99	flp = &fle->next;
	100	}
	101	while ((fle = *flp) != NULL) {
	102	*flp = fle->next;
	103	if (fle->object)
	104	atomic_dec(fle->object_ref);
	105	kmem_cache_free(flow_cachep, fle);
	106	flow_count(cpu)--;
	107	}
	108	}
	109	}
	110
	111	static void flow_cache_shrink(int cpu)
	112	{
	113	int shrink_to = flow_lwm / flow_hash_size;
	114
	115	__flow_cache_shrink(cpu, shrink_to);
	116	}
	117
	118	static void flow_new_hash_rnd(int cpu)
	119	{
	120	get_random_bytes(&flow_hash_rnd(cpu), sizeof(u32));
	121	flow_hash_rnd_recalc(cpu) = 0;
	122
	123	__flow_cache_shrink(cpu, 0);
	124	}
	125
	126	static u32 flow_hash_code(struct flowi *key, int cpu)
	127	{
	128	u32 k = (u32 ) key;
	129
	130	return (jhash2(k, (sizeof(*key) / sizeof(u32)), flow_hash_rnd(cpu)) &
	131	(flow_hash_size - 1));
	132	}
	133
	134	#if (BITS_PER_LONG == 64)
	135	typedef u64 flow_compare_t;
	136	#else
	137	typedef u32 flow_compare_t;
	138	#endif
	139
	140	extern void flowi_is_missized(void);
	141
	142	/* I hear what you're saying, use memcmp. But memcmp cannot make
	143	* important assumptions that we can here, such as alignment and
	144	* constant size.
	145	*/
146	static int flow_key_compare(struct flowi key1, struct flowi key2)
147	{
148	flow_compare_t k1, k1_lim, *k2;
149	const int n_elem = sizeof(struct flowi) / sizeof(flow_compare_t);
150
151	if (sizeof(struct flowi) % sizeof(flow_compare_t))
152	flowi_is_missized();
153
154	k1 = (flow_compare_t *) key1;
155	k1_lim = k1 + n_elem;
156
157	k2 = (flow_compare_t *) key2;
158
159	do {
160	if (k1++ != k2++)
161	return 1;
162	} while (k1 < k1_lim);
163
164	return 0;
165	}
166
e0d1caa7	167	void flow_cache_lookup(struct flowi key, u16 family, u8 dir,
1da177e4 LT	168	flow_resolve_t resolver)
	169	{
	170	struct flow_cache_entry fle, *head;
	171	unsigned int hash;
	172	int cpu;
	173
	174	local_bh_disable();
	175	cpu = smp_processor_id();
	176
	177	fle = NULL;
	178	/* Packet really early in init? Making flow_cache_init a
	179	* pre-smp initcall would solve this. --RR */
	180	if (!flow_table(cpu))
	181	goto nocache;
	182
	183	if (flow_hash_rnd_recalc(cpu))
	184	flow_new_hash_rnd(cpu);
	185	hash = flow_hash_code(key, cpu);
	186
	187	head = &flow_table(cpu)[hash];
	188	for (fle = *head; fle; fle = fle->next) {
	189	if (fle->family == family &&
	190	fle->dir == dir &&
	191	flow_key_compare(key, &fle->key) == 0) {
	192	if (fle->genid == atomic_read(&flow_cache_genid)) {
	193	void *ret = fle->object;
	194
	195	if (ret)
	196	atomic_inc(fle->object_ref);
	197	local_bh_enable();
	198
	199	return ret;
	200	}
	201	break;
	202	}
	203	}
	204
	205	if (!fle) {
	206	if (flow_count(cpu) > flow_hwm)
	207	flow_cache_shrink(cpu);
	208
	209	fle = kmem_cache_alloc(flow_cachep, SLAB_ATOMIC);
	210	if (fle) {
	211	fle->next = *head;
	212	*head = fle;
	213	fle->family = family;
	214	fle->dir = dir;
	215	memcpy(&fle->key, key, sizeof(*key));
	216	fle->object = NULL;
	217	flow_count(cpu)++;
	218	}
	219	}
	220
	221	nocache:
	222	{
	223	void *obj;
	224	atomic_t *obj_ref;
	225
e0d1caa7	226	resolver(key, family, dir, &obj, &obj_ref);
1da177e4 LT	227
	228	if (fle) {
	229	fle->genid = atomic_read(&flow_cache_genid);
	230
	231	if (fle->object)
	232	atomic_dec(fle->object_ref);
	233
	234	fle->object = obj;
	235	fle->object_ref = obj_ref;
	236	if (obj)
	237	atomic_inc(fle->object_ref);
	238	}
	239	local_bh_enable();
	240
	241	return obj;
	242	}
	243	}
	244
	245	static void flow_cache_flush_tasklet(unsigned long data)
	246	{
	247	struct flow_flush_info info = (void )data;
	248	int i;
	249	int cpu;
	250
	251	cpu = smp_processor_id();
	252	for (i = 0; i < flow_hash_size; i++) {
	253	struct flow_cache_entry *fle;
	254
	255	fle = flow_table(cpu)[i];
	256	for (; fle; fle = fle->next) {
	257	unsigned genid = atomic_read(&flow_cache_genid);
	258
	259	if (!fle->object \|\| fle->genid == genid)
	260	continue;
	261
	262	fle->object = NULL;
	263	atomic_dec(fle->object_ref);
	264	}
	265	}
	266
	267	if (atomic_dec_and_test(&info->cpuleft))
	268	complete(&info->completion);
	269	}
	270
	271	static void flow_cache_flush_per_cpu(void *) __attribute__((__unused__));
	272	static void flow_cache_flush_per_cpu(void *data)
	273	{
	274	struct flow_flush_info *info = data;
	275	int cpu;
	276	struct tasklet_struct *tasklet;
	277
	278	cpu = smp_processor_id();
	279
	280	tasklet = flow_flush_tasklet(cpu);
	281	tasklet->data = (unsigned long)info;
	282	tasklet_schedule(tasklet);
	283	}
	284
	285	void flow_cache_flush(void)
	286	{
	287	struct flow_flush_info info;
4a3e2f71	288	static DEFINE_MUTEX(flow_flush_sem);
1da177e4 LT	289
	290	/* Don't want cpus going down or up during this. */
	291	lock_cpu_hotplug();
4a3e2f71	292	mutex_lock(&flow_flush_sem);
1da177e4 LT	293	atomic_set(&info.cpuleft, num_online_cpus());
	294	init_completion(&info.completion);
	295
	296	local_bh_disable();
	297	smp_call_function(flow_cache_flush_per_cpu, &info, 1, 0);
	298	flow_cache_flush_tasklet((unsigned long)&info);
	299	local_bh_enable();
	300
	301	wait_for_completion(&info.completion);
4a3e2f71	302	mutex_unlock(&flow_flush_sem);
1da177e4 LT	303	unlock_cpu_hotplug();
	304	}
	305
	306	static void __devinit flow_cache_cpu_prepare(int cpu)
	307	{
	308	struct tasklet_struct *tasklet;
	309	unsigned long order;
	310
	311	for (order = 0;
	312	(PAGE_SIZE << order) <
	313	(sizeof(struct flow_cache_entry )flow_hash_size);
	314	order++)
	315	/* NOTHING */;
	316
	317	flow_table(cpu) = (struct flow_cache_entry **)
77d04bd9	318	__get_free_pages(GFP_KERNEL\|__GFP_ZERO, order);
1da177e4 LT	319	if (!flow_table(cpu))
	320	panic("NET: failed to allocate flow cache order %lu\n", order);
	321
1da177e4 LT	322	flow_hash_rnd_recalc(cpu) = 1;
	323	flow_count(cpu) = 0;
	324
	325	tasklet = flow_flush_tasklet(cpu);
	326	tasklet_init(tasklet, flow_cache_flush_tasklet, 0);
	327	}
	328
	329	#ifdef CONFIG_HOTPLUG_CPU
	330	static int flow_cache_cpu(struct notifier_block *nfb,
	331	unsigned long action,
	332	void *hcpu)
	333	{
	334	if (action == CPU_DEAD)
	335	__flow_cache_shrink((unsigned long)hcpu, 0);
	336	return NOTIFY_OK;
	337	}
	338	#endif /* CONFIG_HOTPLUG_CPU */
	339
	340	static int __init flow_cache_init(void)
	341	{
	342	int i;
	343
	344	flow_cachep = kmem_cache_create("flow_cache",
	345	sizeof(struct flow_cache_entry),
e5d679f3	346	0, SLAB_HWCACHE_ALIGN\|SLAB_PANIC,
1da177e4	347	NULL, NULL);
1da177e4 LT	348	flow_hash_shift = 10;
	349	flow_lwm = 2 * flow_hash_size;
	350	flow_hwm = 4 * flow_hash_size;
	351
	352	init_timer(&flow_hash_rnd_timer);
	353	flow_hash_rnd_timer.function = flow_cache_new_hashrnd;
	354	flow_hash_rnd_timer.expires = jiffies + FLOW_HASH_RND_PERIOD;
	355	add_timer(&flow_hash_rnd_timer);
	356
6f912042	357	for_each_possible_cpu(i)
1da177e4 LT	358	flow_cache_cpu_prepare(i);
	359
	360	hotcpu_notifier(flow_cache_cpu, 0);
	361	return 0;
	362	}
	363
	364	module_init(flow_cache_init);
	365
	366	EXPORT_SYMBOL(flow_cache_genid);
	367	EXPORT_SYMBOL(flow_cache_lookup);