[mirror_ubuntu-zesty-kernel.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 <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 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;
} ____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_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 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(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, 1, 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 __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);
}

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_HWCACHE_ALIGN|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>
	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
e18b890b	47	static struct kmem_cache *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
134b0fc5 JM	88	static void flow_entry_kill(int cpu, struct flow_cache_entry *fle)
	89	{
	90	if (fle->object)
	91	atomic_dec(fle->object_ref);
	92	kmem_cache_free(flow_cachep, fle);
	93	flow_count(cpu)--;
	94	}
	95
1da177e4 LT	96	static void __flow_cache_shrink(int cpu, int shrink_to)
	97	{
	98	struct flow_cache_entry fle, *flp;
	99	int i;
	100
	101	for (i = 0; i < flow_hash_size; i++) {
	102	int k = 0;
	103
	104	flp = &flow_table(cpu)[i];
	105	while ((fle = *flp) != NULL && k < shrink_to) {
	106	k++;
	107	flp = &fle->next;
	108	}
	109	while ((fle = *flp) != NULL) {
	110	*flp = fle->next;
134b0fc5	111	flow_entry_kill(cpu, fle);
1da177e4 LT	112	}
	113	}
	114	}
	115
	116	static void flow_cache_shrink(int cpu)
	117	{
	118	int shrink_to = flow_lwm / flow_hash_size;
	119
	120	__flow_cache_shrink(cpu, shrink_to);
	121	}
	122
	123	static void flow_new_hash_rnd(int cpu)
	124	{
	125	get_random_bytes(&flow_hash_rnd(cpu), sizeof(u32));
	126	flow_hash_rnd_recalc(cpu) = 0;
	127
	128	__flow_cache_shrink(cpu, 0);
	129	}
	130
	131	static u32 flow_hash_code(struct flowi *key, int cpu)
	132	{
	133	u32 k = (u32 ) key;
	134
	135	return (jhash2(k, (sizeof(*key) / sizeof(u32)), flow_hash_rnd(cpu)) &
	136	(flow_hash_size - 1));
	137	}
	138
	139	#if (BITS_PER_LONG == 64)
	140	typedef u64 flow_compare_t;
	141	#else
	142	typedef u32 flow_compare_t;
	143	#endif
	144
1da177e4 LT	145	/* I hear what you're saying, use memcmp. But memcmp cannot make
	146	* important assumptions that we can here, such as alignment and
	147	* constant size.
	148	*/
	149	static int flow_key_compare(struct flowi key1, struct flowi key2)
	150	{
	151	flow_compare_t k1, k1_lim, *k2;
	152	const int n_elem = sizeof(struct flowi) / sizeof(flow_compare_t);
	153
f0fe91de	154	BUILD_BUG_ON(sizeof(struct flowi) % sizeof(flow_compare_t));
1da177e4 LT	155
	156	k1 = (flow_compare_t *) key1;
	157	k1_lim = k1 + n_elem;
	158
	159	k2 = (flow_compare_t *) key2;
	160
	161	do {
	162	if (k1++ != k2++)
	163	return 1;
	164	} while (k1 < k1_lim);
	165
	166	return 0;
	167	}
	168
e0d1caa7	169	void flow_cache_lookup(struct flowi key, u16 family, u8 dir,
1da177e4 LT	170	flow_resolve_t resolver)
	171	{
	172	struct flow_cache_entry fle, *head;
	173	unsigned int hash;
	174	int cpu;
	175
	176	local_bh_disable();
	177	cpu = smp_processor_id();
	178
	179	fle = NULL;
	180	/* Packet really early in init? Making flow_cache_init a
	181	* pre-smp initcall would solve this. --RR */
	182	if (!flow_table(cpu))
	183	goto nocache;
	184
	185	if (flow_hash_rnd_recalc(cpu))
	186	flow_new_hash_rnd(cpu);
	187	hash = flow_hash_code(key, cpu);
	188
	189	head = &flow_table(cpu)[hash];
	190	for (fle = *head; fle; fle = fle->next) {
	191	if (fle->family == family &&
	192	fle->dir == dir &&
	193	flow_key_compare(key, &fle->key) == 0) {
	194	if (fle->genid == atomic_read(&flow_cache_genid)) {
	195	void *ret = fle->object;
	196
	197	if (ret)
	198	atomic_inc(fle->object_ref);
	199	local_bh_enable();
	200
	201	return ret;
	202	}
	203	break;
	204	}
	205	}
	206
	207	if (!fle) {
	208	if (flow_count(cpu) > flow_hwm)
	209	flow_cache_shrink(cpu);
	210
54e6ecb2	211	fle = kmem_cache_alloc(flow_cachep, GFP_ATOMIC);
1da177e4 LT	212	if (fle) {
	213	fle->next = *head;
	214	*head = fle;
	215	fle->family = family;
	216	fle->dir = dir;
	217	memcpy(&fle->key, key, sizeof(*key));
	218	fle->object = NULL;
	219	flow_count(cpu)++;
	220	}
	221	}
	222
	223	nocache:
	224	{
134b0fc5	225	int err;
1da177e4 LT	226	void *obj;
	227	atomic_t *obj_ref;
	228
134b0fc5	229	err = resolver(key, family, dir, &obj, &obj_ref);
1da177e4	230
e0e8f1c8 HX	231	if (fle && !err) {
	232	fle->genid = atomic_read(&flow_cache_genid);
	233
	234	if (fle->object)
	235	atomic_dec(fle->object_ref);
	236
	237	fle->object = obj;
	238	fle->object_ref = obj_ref;
	239	if (obj)
	240	atomic_inc(fle->object_ref);
1da177e4 LT	241	}
	242	local_bh_enable();
	243
134b0fc5 JM	244	if (err)
134b0fc5 JM	245	obj = ERR_PTR(err);
1da177e4 LT	246	return obj;
	247	}
	248	}
	249
	250	static void flow_cache_flush_tasklet(unsigned long data)
	251	{
	252	struct flow_flush_info info = (void )data;
	253	int i;
	254	int cpu;
	255
	256	cpu = smp_processor_id();
	257	for (i = 0; i < flow_hash_size; i++) {
	258	struct flow_cache_entry *fle;
	259
	260	fle = flow_table(cpu)[i];
	261	for (; fle; fle = fle->next) {
	262	unsigned genid = atomic_read(&flow_cache_genid);
	263
	264	if (!fle->object \|\| fle->genid == genid)
	265	continue;
	266
	267	fle->object = NULL;
	268	atomic_dec(fle->object_ref);
	269	}
	270	}
	271
	272	if (atomic_dec_and_test(&info->cpuleft))
	273	complete(&info->completion);
	274	}
	275
	276	static void flow_cache_flush_per_cpu(void *) __attribute__((__unused__));
	277	static void flow_cache_flush_per_cpu(void *data)
	278	{
	279	struct flow_flush_info *info = data;
	280	int cpu;
	281	struct tasklet_struct *tasklet;
	282
	283	cpu = smp_processor_id();
	284
	285	tasklet = flow_flush_tasklet(cpu);
	286	tasklet->data = (unsigned long)info;
	287	tasklet_schedule(tasklet);
	288	}
	289
	290	void flow_cache_flush(void)
	291	{
	292	struct flow_flush_info info;
4a3e2f71	293	static DEFINE_MUTEX(flow_flush_sem);
1da177e4 LT	294
1da177e4 LT	295	/* Don't want cpus going down or up during this. */
86ef5c9a	296	get_online_cpus();
4a3e2f71	297	mutex_lock(&flow_flush_sem);
1da177e4 LT	298	atomic_set(&info.cpuleft, num_online_cpus());
	299	init_completion(&info.completion);
	300
	301	local_bh_disable();
	302	smp_call_function(flow_cache_flush_per_cpu, &info, 1, 0);
	303	flow_cache_flush_tasklet((unsigned long)&info);
	304	local_bh_enable();
	305
	306	wait_for_completion(&info.completion);
4a3e2f71	307	mutex_unlock(&flow_flush_sem);
86ef5c9a	308	put_online_cpus();
1da177e4 LT	309	}
	310
	311	static void __devinit flow_cache_cpu_prepare(int cpu)
	312	{
	313	struct tasklet_struct *tasklet;
	314	unsigned long order;
	315
	316	for (order = 0;
	317	(PAGE_SIZE << order) <
	318	(sizeof(struct flow_cache_entry )flow_hash_size);
	319	order++)
	320	/* NOTHING */;
	321
	322	flow_table(cpu) = (struct flow_cache_entry **)
77d04bd9	323	__get_free_pages(GFP_KERNEL\|__GFP_ZERO, order);
1da177e4 LT	324	if (!flow_table(cpu))
	325	panic("NET: failed to allocate flow cache order %lu\n", order);
	326
1da177e4 LT	327	flow_hash_rnd_recalc(cpu) = 1;
	328	flow_count(cpu) = 0;
	329
	330	tasklet = flow_flush_tasklet(cpu);
	331	tasklet_init(tasklet, flow_cache_flush_tasklet, 0);
	332	}
	333
1da177e4 LT	334	static int flow_cache_cpu(struct notifier_block *nfb,
	335	unsigned long action,
	336	void *hcpu)
	337	{
8bb78442	338	if (action == CPU_DEAD \|\| action == CPU_DEAD_FROZEN)
1da177e4 LT	339	__flow_cache_shrink((unsigned long)hcpu, 0);
	340	return NOTIFY_OK;
	341	}
1da177e4 LT	342
	343	static int __init flow_cache_init(void)
	344	{
	345	int i;
	346
	347	flow_cachep = kmem_cache_create("flow_cache",
	348	sizeof(struct flow_cache_entry),
e5d679f3	349	0, SLAB_HWCACHE_ALIGN\|SLAB_PANIC,
20c2df83	350	NULL);
1da177e4 LT	351	flow_hash_shift = 10;
	352	flow_lwm = 2 * flow_hash_size;
	353	flow_hwm = 4 * flow_hash_size;
	354
b24b8a24	355	setup_timer(&flow_hash_rnd_timer, flow_cache_new_hashrnd, 0);
1da177e4 LT	356	flow_hash_rnd_timer.expires = jiffies + FLOW_HASH_RND_PERIOD;
	357	add_timer(&flow_hash_rnd_timer);
	358
6f912042	359	for_each_possible_cpu(i)
1da177e4 LT	360	flow_cache_cpu_prepare(i);
	361
	362	hotcpu_notifier(flow_cache_cpu, 0);
	363	return 0;
	364	}
	365
	366	module_init(flow_cache_init);
	367
	368	EXPORT_SYMBOL(flow_cache_genid);
	369	EXPORT_SYMBOL(flow_cache_lookup);