[GitHub/exynos8895/android_kernel_samsung_universal8895.git] / drivers / md / bcache / writeback.c

/*
 * background writeback - scan btree for dirty data and write it to the backing
 * device
 *
 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
 * Copyright 2012 Google, Inc.
 */

#include "bcache.h"
#include "btree.h"
#include "debug.h"
#include "writeback.h"

#include <linux/delay.h>
#include <linux/freezer.h>
#include <linux/kthread.h>
#include <trace/events/bcache.h>

/* Rate limiting */

static void __update_writeback_rate(struct cached_dev *dc)
{
	struct cache_set *c = dc->disk.c;
	uint64_t cache_sectors = c->nbuckets * c->sb.bucket_size;
	uint64_t cache_dirty_target =
		div_u64(cache_sectors * dc->writeback_percent, 100);

	int64_t target = div64_u64(cache_dirty_target * bdev_sectors(dc->bdev),
				   c->cached_dev_sectors);

	/* PD controller */

	int change = 0;
	int64_t error;
	int64_t dirty = bcache_dev_sectors_dirty(&dc->disk);
	int64_t derivative = dirty - dc->disk.sectors_dirty_last;

	dc->disk.sectors_dirty_last = dirty;

	derivative *= dc->writeback_rate_d_term;
	derivative = clamp(derivative, -dirty, dirty);

	derivative = ewma_add(dc->disk.sectors_dirty_derivative, derivative,
			      dc->writeback_rate_d_smooth, 0);

	/* Avoid divide by zero */
	if (!target)
		goto out;

	error = div64_s64((dirty + derivative - target) << 8, target);

	change = div_s64((dc->writeback_rate.rate * error) >> 8,
			 dc->writeback_rate_p_term_inverse);

	/* Don't increase writeback rate if the device isn't keeping up */
	if (change > 0 &&
	    time_after64(local_clock(),
			 dc->writeback_rate.next + 10 * NSEC_PER_MSEC))
		change = 0;

	dc->writeback_rate.rate =
		clamp_t(int64_t, dc->writeback_rate.rate + change,
			1, NSEC_PER_MSEC);
out:
	dc->writeback_rate_derivative = derivative;
	dc->writeback_rate_change = change;
	dc->writeback_rate_target = target;
}

static void update_writeback_rate(struct work_struct *work)
{
	struct cached_dev *dc = container_of(to_delayed_work(work),
					     struct cached_dev,
					     writeback_rate_update);

	down_read(&dc->writeback_lock);

	if (atomic_read(&dc->has_dirty) &&
	    dc->writeback_percent)
		__update_writeback_rate(dc);

	up_read(&dc->writeback_lock);

	schedule_delayed_work(&dc->writeback_rate_update,
			      dc->writeback_rate_update_seconds * HZ);
}

static unsigned writeback_delay(struct cached_dev *dc, unsigned sectors)
{
	uint64_t ret;

	if (atomic_read(&dc->disk.detaching) ||
	    !dc->writeback_percent)
		return 0;

	ret = bch_next_delay(&dc->writeback_rate, sectors * 10000000ULL);

	return min_t(uint64_t, ret, HZ);
}

struct dirty_io {
	struct closure		cl;
	struct cached_dev	*dc;
	struct bio		bio;
};

static void dirty_init(struct keybuf_key *w)
{
	struct dirty_io *io = w->private;
	struct bio *bio = &io->bio;

	bio_init(bio);
	if (!io->dc->writeback_percent)
		bio_set_prio(bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));

	bio->bi_size		= KEY_SIZE(&w->key) << 9;
	bio->bi_max_vecs	= DIV_ROUND_UP(KEY_SIZE(&w->key), PAGE_SECTORS);
	bio->bi_private		= w;
	bio->bi_io_vec		= bio->bi_inline_vecs;
	bch_bio_map(bio, NULL);
}

static void dirty_io_destructor(struct closure *cl)
{
	struct dirty_io *io = container_of(cl, struct dirty_io, cl);
	kfree(io);
}

static void write_dirty_finish(struct closure *cl)
{
	struct dirty_io *io = container_of(cl, struct dirty_io, cl);
	struct keybuf_key *w = io->bio.bi_private;
	struct cached_dev *dc = io->dc;
	struct bio_vec *bv;
	int i;

	bio_for_each_segment_all(bv, &io->bio, i)
		__free_page(bv->bv_page);

	/* This is kind of a dumb way of signalling errors. */
	if (KEY_DIRTY(&w->key)) {
		unsigned i;
		struct btree_op op;
		struct keylist keys;
		int ret;

		bch_btree_op_init(&op, -1);
		bch_keylist_init(&keys);

		bkey_copy(keys.top, &w->key);
		SET_KEY_DIRTY(keys.top, false);
		bch_keylist_push(&keys);

		for (i = 0; i < KEY_PTRS(&w->key); i++)
			atomic_inc(&PTR_BUCKET(dc->disk.c, &w->key, i)->pin);

		ret = bch_btree_insert(&op, dc->disk.c, &keys, NULL, &w->key);

		if (ret)
			trace_bcache_writeback_collision(&w->key);

		atomic_long_inc(ret
				? &dc->disk.c->writeback_keys_failed
				: &dc->disk.c->writeback_keys_done);
	}

	bch_keybuf_del(&dc->writeback_keys, w);
	up(&dc->in_flight);

	closure_return_with_destructor(cl, dirty_io_destructor);
}

static void dirty_endio(struct bio *bio, int error)
{
	struct keybuf_key *w = bio->bi_private;
	struct dirty_io *io = w->private;

	if (error)
		SET_KEY_DIRTY(&w->key, false);

	closure_put(&io->cl);
}

static void write_dirty(struct closure *cl)
{
	struct dirty_io *io = container_of(cl, struct dirty_io, cl);
	struct keybuf_key *w = io->bio.bi_private;

	dirty_init(w);
	io->bio.bi_rw		= WRITE;
	io->bio.bi_sector	= KEY_START(&w->key);
	io->bio.bi_bdev		= io->dc->bdev;
	io->bio.bi_end_io	= dirty_endio;

	closure_bio_submit(&io->bio, cl, &io->dc->disk);

	continue_at(cl, write_dirty_finish, system_wq);
}

static void read_dirty_endio(struct bio *bio, int error)
{
	struct keybuf_key *w = bio->bi_private;
	struct dirty_io *io = w->private;

	bch_count_io_errors(PTR_CACHE(io->dc->disk.c, &w->key, 0),
			    error, "reading dirty data from cache");

	dirty_endio(bio, error);
}

static void read_dirty_submit(struct closure *cl)
{
	struct dirty_io *io = container_of(cl, struct dirty_io, cl);

	closure_bio_submit(&io->bio, cl, &io->dc->disk);

	continue_at(cl, write_dirty, system_wq);
}

static void read_dirty(struct cached_dev *dc)
{
	unsigned delay = 0;
	struct keybuf_key *w;
	struct dirty_io *io;
	struct closure cl;

	closure_init_stack(&cl);

	/*
	 * XXX: if we error, background writeback just spins. Should use some
	 * mempools.
	 */

	while (!kthread_should_stop()) {
		try_to_freeze();

		w = bch_keybuf_next(&dc->writeback_keys);
		if (!w)
			break;

		BUG_ON(ptr_stale(dc->disk.c, &w->key, 0));

		if (KEY_START(&w->key) != dc->last_read ||
		    jiffies_to_msecs(delay) > 50)
			while (!kthread_should_stop() && delay)
				delay = schedule_timeout_interruptible(delay);

		dc->last_read	= KEY_OFFSET(&w->key);

		io = kzalloc(sizeof(struct dirty_io) + sizeof(struct bio_vec)
			     * DIV_ROUND_UP(KEY_SIZE(&w->key), PAGE_SECTORS),
			     GFP_KERNEL);
		if (!io)
			goto err;

		w->private	= io;
		io->dc		= dc;

		dirty_init(w);
		io->bio.bi_sector	= PTR_OFFSET(&w->key, 0);
		io->bio.bi_bdev		= PTR_CACHE(dc->disk.c,
						    &w->key, 0)->bdev;
		io->bio.bi_rw		= READ;
		io->bio.bi_end_io	= read_dirty_endio;

		if (bio_alloc_pages(&io->bio, GFP_KERNEL))
			goto err_free;

		trace_bcache_writeback(&w->key);

		down(&dc->in_flight);
		closure_call(&io->cl, read_dirty_submit, NULL, &cl);

		delay = writeback_delay(dc, KEY_SIZE(&w->key));
	}

	if (0) {
err_free:
		kfree(w->private);
err:
		bch_keybuf_del(&dc->writeback_keys, w);
	}

	/*
	 * Wait for outstanding writeback IOs to finish (and keybuf slots to be
	 * freed) before refilling again
	 */
	closure_sync(&cl);
}

/* Scan for dirty data */

void bcache_dev_sectors_dirty_add(struct cache_set *c, unsigned inode,
				  uint64_t offset, int nr_sectors)
{
	struct bcache_device *d = c->devices[inode];
	unsigned stripe_offset;
	uint64_t stripe = offset;

	if (!d)
		return;

	do_div(stripe, d->stripe_size);

	stripe_offset = offset & (d->stripe_size - 1);

	while (nr_sectors) {
		int s = min_t(unsigned, abs(nr_sectors),
			      d->stripe_size - stripe_offset);

		if (nr_sectors < 0)
			s = -s;

		atomic_add(s, d->stripe_sectors_dirty + stripe);
		nr_sectors -= s;
		stripe_offset = 0;
		stripe++;
	}
}

static bool dirty_pred(struct keybuf *buf, struct bkey *k)
{
	return KEY_DIRTY(k);
}

static bool dirty_full_stripe_pred(struct keybuf *buf, struct bkey *k)
{
	uint64_t stripe = KEY_START(k);
	unsigned nr_sectors = KEY_SIZE(k);
	struct cached_dev *dc = container_of(buf, struct cached_dev,
					     writeback_keys);

	if (!KEY_DIRTY(k))
		return false;

	do_div(stripe, dc->disk.stripe_size);

	while (1) {
		if (atomic_read(dc->disk.stripe_sectors_dirty + stripe) ==
		    dc->disk.stripe_size)
			return true;

		if (nr_sectors <= dc->disk.stripe_size)
			return false;

		nr_sectors -= dc->disk.stripe_size;
		stripe++;
	}
}

static bool refill_dirty(struct cached_dev *dc)
{
	struct keybuf *buf = &dc->writeback_keys;
	bool searched_from_start = false;
	struct bkey end = KEY(dc->disk.id, MAX_KEY_OFFSET, 0);

	if (bkey_cmp(&buf->last_scanned, &end) >= 0) {
		buf->last_scanned = KEY(dc->disk.id, 0, 0);
		searched_from_start = true;
	}

	if (dc->partial_stripes_expensive) {
		uint64_t i;

		for (i = 0; i < dc->disk.nr_stripes; i++)
			if (atomic_read(dc->disk.stripe_sectors_dirty + i) ==
			    dc->disk.stripe_size)
				goto full_stripes;

		goto normal_refill;
full_stripes:
		searched_from_start = false;	/* not searching entire btree */
		bch_refill_keybuf(dc->disk.c, buf, &end,
				  dirty_full_stripe_pred);
	} else {
normal_refill:
		bch_refill_keybuf(dc->disk.c, buf, &end, dirty_pred);
	}

	return bkey_cmp(&buf->last_scanned, &end) >= 0 && searched_from_start;
}

static int bch_writeback_thread(void *arg)
{
	struct cached_dev *dc = arg;
	bool searched_full_index;

	while (!kthread_should_stop()) {
		down_write(&dc->writeback_lock);
		if (!atomic_read(&dc->has_dirty) ||
		    (!atomic_read(&dc->disk.detaching) &&
		     !dc->writeback_running)) {
			up_write(&dc->writeback_lock);
			set_current_state(TASK_INTERRUPTIBLE);

			if (kthread_should_stop())
				return 0;

			try_to_freeze();
			schedule();
			continue;
		}

		searched_full_index = refill_dirty(dc);

		if (searched_full_index &&
		    RB_EMPTY_ROOT(&dc->writeback_keys.keys)) {
			atomic_set(&dc->has_dirty, 0);
			cached_dev_put(dc);
			SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
			bch_write_bdev_super(dc, NULL);
		}

		up_write(&dc->writeback_lock);

		bch_ratelimit_reset(&dc->writeback_rate);
		read_dirty(dc);

		if (searched_full_index) {
			unsigned delay = dc->writeback_delay * HZ;

			while (delay &&
			       !kthread_should_stop() &&
			       !atomic_read(&dc->disk.detaching))
				delay = schedule_timeout_interruptible(delay);
		}
	}

	return 0;
}

/* Init */

struct sectors_dirty_init {
	struct btree_op	op;
	unsigned	inode;
};

static int sectors_dirty_init_fn(struct btree_op *_op, struct btree *b,
				 struct bkey *k)
{
	struct sectors_dirty_init *op = container_of(_op,
						struct sectors_dirty_init, op);
	if (KEY_INODE(k) > op->inode)
		return MAP_DONE;

	if (KEY_DIRTY(k))
		bcache_dev_sectors_dirty_add(b->c, KEY_INODE(k),
					     KEY_START(k), KEY_SIZE(k));

	return MAP_CONTINUE;
}

void bch_sectors_dirty_init(struct cached_dev *dc)
{
	struct sectors_dirty_init op;

	bch_btree_op_init(&op.op, -1);
	op.inode = dc->disk.id;

	bch_btree_map_keys(&op.op, dc->disk.c, &KEY(op.inode, 0, 0),
			   sectors_dirty_init_fn, 0);
}

int bch_cached_dev_writeback_init(struct cached_dev *dc)
{
	sema_init(&dc->in_flight, 64);
	init_rwsem(&dc->writeback_lock);
	bch_keybuf_init(&dc->writeback_keys);

	dc->writeback_metadata		= true;
	dc->writeback_running		= true;
	dc->writeback_percent		= 10;
	dc->writeback_delay		= 30;
	dc->writeback_rate.rate		= 1024;

	dc->writeback_rate_update_seconds = 30;
	dc->writeback_rate_d_term	= 16;
	dc->writeback_rate_p_term_inverse = 64;
	dc->writeback_rate_d_smooth	= 8;

	dc->writeback_thread = kthread_create(bch_writeback_thread, dc,
					      "bcache_writeback");
	if (IS_ERR(dc->writeback_thread))
		return PTR_ERR(dc->writeback_thread);

	set_task_state(dc->writeback_thread, TASK_INTERRUPTIBLE);

	INIT_DELAYED_WORK(&dc->writeback_rate_update, update_writeback_rate);
	schedule_delayed_work(&dc->writeback_rate_update,
			      dc->writeback_rate_update_seconds * HZ);

	return 0;
}
Commit	Line	Data
cafe5635 KO	1	/*
	2	* background writeback - scan btree for dirty data and write it to the backing
	3	* device
	4	*
	5	* Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
	6	* Copyright 2012 Google, Inc.
	7	*/
	8
	9	#include "bcache.h"
	10	#include "btree.h"
	11	#include "debug.h"
279afbad	12	#include "writeback.h"
cafe5635	13
5e6926da KO	14	#include <linux/delay.h>
	15	#include <linux/freezer.h>
	16	#include <linux/kthread.h>
c37511b8 KO	17	#include <trace/events/bcache.h>
c37511b8 KO	18
cafe5635 KO	19	/* Rate limiting */
	20
	21	static void __update_writeback_rate(struct cached_dev *dc)
	22	{
	23	struct cache_set *c = dc->disk.c;
	24	uint64_t cache_sectors = c->nbuckets * c->sb.bucket_size;
	25	uint64_t cache_dirty_target =
	26	div_u64(cache_sectors * dc->writeback_percent, 100);
	27
	28	int64_t target = div64_u64(cache_dirty_target * bdev_sectors(dc->bdev),
	29	c->cached_dev_sectors);
	30
	31	/* PD controller */
	32
	33	int change = 0;
	34	int64_t error;
279afbad	35	int64_t dirty = bcache_dev_sectors_dirty(&dc->disk);
cafe5635 KO	36	int64_t derivative = dirty - dc->disk.sectors_dirty_last;
	37
	38	dc->disk.sectors_dirty_last = dirty;
	39
	40	derivative *= dc->writeback_rate_d_term;
	41	derivative = clamp(derivative, -dirty, dirty);
	42
	43	derivative = ewma_add(dc->disk.sectors_dirty_derivative, derivative,
	44	dc->writeback_rate_d_smooth, 0);
	45
	46	/* Avoid divide by zero */
	47	if (!target)
	48	goto out;
	49
	50	error = div64_s64((dirty + derivative - target) << 8, target);
	51
	52	change = div_s64((dc->writeback_rate.rate * error) >> 8,
	53	dc->writeback_rate_p_term_inverse);
	54
	55	/* Don't increase writeback rate if the device isn't keeping up */
	56	if (change > 0 &&
	57	time_after64(local_clock(),
	58	dc->writeback_rate.next + 10 * NSEC_PER_MSEC))
	59	change = 0;
	60
	61	dc->writeback_rate.rate =
	62	clamp_t(int64_t, dc->writeback_rate.rate + change,
	63	1, NSEC_PER_MSEC);
	64	out:
	65	dc->writeback_rate_derivative = derivative;
	66	dc->writeback_rate_change = change;
	67	dc->writeback_rate_target = target;
cafe5635 KO	68	}
	69
	70	static void update_writeback_rate(struct work_struct *work)
	71	{
	72	struct cached_dev *dc = container_of(to_delayed_work(work),
	73	struct cached_dev,
	74	writeback_rate_update);
	75
	76	down_read(&dc->writeback_lock);
	77
	78	if (atomic_read(&dc->has_dirty) &&
	79	dc->writeback_percent)
	80	__update_writeback_rate(dc);
	81
	82	up_read(&dc->writeback_lock);
5e6926da KO	83
	84	schedule_delayed_work(&dc->writeback_rate_update,
	85	dc->writeback_rate_update_seconds * HZ);
cafe5635 KO	86	}
	87
	88	static unsigned writeback_delay(struct cached_dev *dc, unsigned sectors)
	89	{
c2a4f318 KO	90	uint64_t ret;
c2a4f318 KO	91
cafe5635 KO	92	if (atomic_read(&dc->disk.detaching) \|\|
	93	!dc->writeback_percent)
	94	return 0;
	95
c2a4f318 KO	96	ret = bch_next_delay(&dc->writeback_rate, sectors * 10000000ULL);
	97
	98	return min_t(uint64_t, ret, HZ);
cafe5635 KO	99	}
cafe5635 KO	100
5e6926da KO	101	struct dirty_io {
	102	struct closure cl;
	103	struct cached_dev *dc;
	104	struct bio bio;
	105	};
72c27061	106
cafe5635 KO	107	static void dirty_init(struct keybuf_key *w)
	108	{
	109	struct dirty_io *io = w->private;
	110	struct bio *bio = &io->bio;
	111
	112	bio_init(bio);
	113	if (!io->dc->writeback_percent)
	114	bio_set_prio(bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
	115
	116	bio->bi_size = KEY_SIZE(&w->key) << 9;
	117	bio->bi_max_vecs = DIV_ROUND_UP(KEY_SIZE(&w->key), PAGE_SECTORS);
	118	bio->bi_private = w;
	119	bio->bi_io_vec = bio->bi_inline_vecs;
169ef1cf	120	bch_bio_map(bio, NULL);
cafe5635 KO	121	}
cafe5635 KO	122
cafe5635 KO	123	static void dirty_io_destructor(struct closure *cl)
	124	{
	125	struct dirty_io *io = container_of(cl, struct dirty_io, cl);
	126	kfree(io);
	127	}
	128
	129	static void write_dirty_finish(struct closure *cl)
	130	{
	131	struct dirty_io *io = container_of(cl, struct dirty_io, cl);
	132	struct keybuf_key *w = io->bio.bi_private;
	133	struct cached_dev *dc = io->dc;
8e51e414 KO	134	struct bio_vec *bv;
8e51e414 KO	135	int i;
cafe5635	136
8e51e414	137	bio_for_each_segment_all(bv, &io->bio, i)
cafe5635 KO	138	__free_page(bv->bv_page);
	139
	140	/* This is kind of a dumb way of signalling errors. */
	141	if (KEY_DIRTY(&w->key)) {
	142	unsigned i;
	143	struct btree_op op;
0b93207a	144	struct keylist keys;
6054c6d4	145	int ret;
0b93207a	146
b54d6934	147	bch_btree_op_init(&op, -1);
0b93207a	148	bch_keylist_init(&keys);
cafe5635	149
1b207d80 KO	150	bkey_copy(keys.top, &w->key);
	151	SET_KEY_DIRTY(keys.top, false);
	152	bch_keylist_push(&keys);
cafe5635 KO	153
	154	for (i = 0; i < KEY_PTRS(&w->key); i++)
	155	atomic_inc(&PTR_BUCKET(dc->disk.c, &w->key, i)->pin);
	156
6054c6d4	157	ret = bch_btree_insert(&op, dc->disk.c, &keys, NULL, &w->key);
cafe5635	158
6054c6d4	159	if (ret)
c37511b8 KO	160	trace_bcache_writeback_collision(&w->key);
c37511b8 KO	161
6054c6d4	162	atomic_long_inc(ret
cafe5635 KO	163	? &dc->disk.c->writeback_keys_failed
	164	: &dc->disk.c->writeback_keys_done);
	165	}
	166
	167	bch_keybuf_del(&dc->writeback_keys, w);
c2a4f318	168	up(&dc->in_flight);
cafe5635 KO	169
	170	closure_return_with_destructor(cl, dirty_io_destructor);
	171	}
	172
	173	static void dirty_endio(struct bio *bio, int error)
	174	{
	175	struct keybuf_key *w = bio->bi_private;
	176	struct dirty_io *io = w->private;
	177
	178	if (error)
	179	SET_KEY_DIRTY(&w->key, false);
	180
	181	closure_put(&io->cl);
	182	}
	183
	184	static void write_dirty(struct closure *cl)
	185	{
	186	struct dirty_io *io = container_of(cl, struct dirty_io, cl);
	187	struct keybuf_key *w = io->bio.bi_private;
	188
	189	dirty_init(w);
	190	io->bio.bi_rw = WRITE;
	191	io->bio.bi_sector = KEY_START(&w->key);
	192	io->bio.bi_bdev = io->dc->bdev;
	193	io->bio.bi_end_io = dirty_endio;
	194
cafe5635 KO	195	closure_bio_submit(&io->bio, cl, &io->dc->disk);
cafe5635 KO	196
c2a4f318	197	continue_at(cl, write_dirty_finish, system_wq);
cafe5635 KO	198	}
	199
	200	static void read_dirty_endio(struct bio *bio, int error)
	201	{
	202	struct keybuf_key *w = bio->bi_private;
	203	struct dirty_io *io = w->private;
	204
	205	bch_count_io_errors(PTR_CACHE(io->dc->disk.c, &w->key, 0),
	206	error, "reading dirty data from cache");
	207
	208	dirty_endio(bio, error);
	209	}
	210
	211	static void read_dirty_submit(struct closure *cl)
	212	{
	213	struct dirty_io *io = container_of(cl, struct dirty_io, cl);
	214
cafe5635 KO	215	closure_bio_submit(&io->bio, cl, &io->dc->disk);
cafe5635 KO	216
c2a4f318	217	continue_at(cl, write_dirty, system_wq);
cafe5635 KO	218	}
cafe5635 KO	219
5e6926da	220	static void read_dirty(struct cached_dev *dc)
cafe5635	221	{
5e6926da	222	unsigned delay = 0;
cafe5635 KO	223	struct keybuf_key *w;
cafe5635 KO	224	struct dirty_io *io;
5e6926da KO	225	struct closure cl;
	226
	227	closure_init_stack(&cl);
cafe5635 KO	228
	229	/*
	230	* XXX: if we error, background writeback just spins. Should use some
	231	* mempools.
	232	*/
	233
5e6926da KO	234	while (!kthread_should_stop()) {
	235	try_to_freeze();
	236
cafe5635 KO	237	w = bch_keybuf_next(&dc->writeback_keys);
	238	if (!w)
	239	break;
	240
	241	BUG_ON(ptr_stale(dc->disk.c, &w->key, 0));
	242
5e6926da KO	243	if (KEY_START(&w->key) != dc->last_read \|\|
	244	jiffies_to_msecs(delay) > 50)
	245	while (!kthread_should_stop() && delay)
	246	delay = schedule_timeout_interruptible(delay);
cafe5635 KO	247
	248	dc->last_read = KEY_OFFSET(&w->key);
	249
	250	io = kzalloc(sizeof(struct dirty_io) + sizeof(struct bio_vec)
	251	* DIV_ROUND_UP(KEY_SIZE(&w->key), PAGE_SECTORS),
	252	GFP_KERNEL);
	253	if (!io)
	254	goto err;
	255
	256	w->private = io;
	257	io->dc = dc;
	258
	259	dirty_init(w);
	260	io->bio.bi_sector = PTR_OFFSET(&w->key, 0);
	261	io->bio.bi_bdev = PTR_CACHE(dc->disk.c,
	262	&w->key, 0)->bdev;
	263	io->bio.bi_rw = READ;
	264	io->bio.bi_end_io = read_dirty_endio;
	265
8e51e414	266	if (bio_alloc_pages(&io->bio, GFP_KERNEL))
cafe5635 KO	267	goto err_free;
cafe5635 KO	268
c37511b8	269	trace_bcache_writeback(&w->key);
cafe5635	270
c2a4f318	271	down(&dc->in_flight);
5e6926da	272	closure_call(&io->cl, read_dirty_submit, NULL, &cl);
cafe5635 KO	273
cafe5635 KO	274	delay = writeback_delay(dc, KEY_SIZE(&w->key));
cafe5635 KO	275	}
	276
	277	if (0) {
	278	err_free:
	279	kfree(w->private);
	280	err:
	281	bch_keybuf_del(&dc->writeback_keys, w);
	282	}
	283
c2a4f318 KO	284	/*
	285	* Wait for outstanding writeback IOs to finish (and keybuf slots to be
	286	* freed) before refilling again
	287	*/
5e6926da KO	288	closure_sync(&cl);
	289	}
	290
	291	/* Scan for dirty data */
	292
	293	void bcache_dev_sectors_dirty_add(struct cache_set *c, unsigned inode,
	294	uint64_t offset, int nr_sectors)
	295	{
	296	struct bcache_device *d = c->devices[inode];
	297	unsigned stripe_offset;
	298	uint64_t stripe = offset;
	299
	300	if (!d)
	301	return;
	302
	303	do_div(stripe, d->stripe_size);
	304
	305	stripe_offset = offset & (d->stripe_size - 1);
	306
	307	while (nr_sectors) {
	308	int s = min_t(unsigned, abs(nr_sectors),
	309	d->stripe_size - stripe_offset);
	310
	311	if (nr_sectors < 0)
	312	s = -s;
	313
	314	atomic_add(s, d->stripe_sectors_dirty + stripe);
	315	nr_sectors -= s;
	316	stripe_offset = 0;
	317	stripe++;
	318	}
	319	}
	320
	321	static bool dirty_pred(struct keybuf buf, struct bkey k)
	322	{
	323	return KEY_DIRTY(k);
	324	}
	325
	326	static bool dirty_full_stripe_pred(struct keybuf buf, struct bkey k)
	327	{
	328	uint64_t stripe = KEY_START(k);
	329	unsigned nr_sectors = KEY_SIZE(k);
	330	struct cached_dev *dc = container_of(buf, struct cached_dev,
	331	writeback_keys);
	332
	333	if (!KEY_DIRTY(k))
	334	return false;
	335
	336	do_div(stripe, dc->disk.stripe_size);
	337
	338	while (1) {
	339	if (atomic_read(dc->disk.stripe_sectors_dirty + stripe) ==
	340	dc->disk.stripe_size)
	341	return true;
	342
	343	if (nr_sectors <= dc->disk.stripe_size)
	344	return false;
	345
	346	nr_sectors -= dc->disk.stripe_size;
	347	stripe++;
	348	}
	349	}
	350
	351	static bool refill_dirty(struct cached_dev *dc)
352	{
353	struct keybuf *buf = &dc->writeback_keys;
354	bool searched_from_start = false;
355	struct bkey end = KEY(dc->disk.id, MAX_KEY_OFFSET, 0);
356
357	if (bkey_cmp(&buf->last_scanned, &end) >= 0) {
358	buf->last_scanned = KEY(dc->disk.id, 0, 0);
359	searched_from_start = true;
360	}
361
362	if (dc->partial_stripes_expensive) {
363	uint64_t i;
364
365	for (i = 0; i < dc->disk.nr_stripes; i++)
366	if (atomic_read(dc->disk.stripe_sectors_dirty + i) ==
367	dc->disk.stripe_size)
368	goto full_stripes;
369
370	goto normal_refill;
371	full_stripes:
372	searched_from_start = false; /* not searching entire btree */
373	bch_refill_keybuf(dc->disk.c, buf, &end,
374	dirty_full_stripe_pred);
375	} else {
376	normal_refill:
377	bch_refill_keybuf(dc->disk.c, buf, &end, dirty_pred);
378	}
379
380	return bkey_cmp(&buf->last_scanned, &end) >= 0 && searched_from_start;
381	}
382
383	static int bch_writeback_thread(void *arg)
384	{
385	struct cached_dev *dc = arg;
386	bool searched_full_index;
387
388	while (!kthread_should_stop()) {
389	down_write(&dc->writeback_lock);
390	if (!atomic_read(&dc->has_dirty) \|\|
391	(!atomic_read(&dc->disk.detaching) &&
392	!dc->writeback_running)) {
393	up_write(&dc->writeback_lock);
394	set_current_state(TASK_INTERRUPTIBLE);
395
396	if (kthread_should_stop())
397	return 0;
398
399	try_to_freeze();
400	schedule();
401	continue;
402	}
403
404	searched_full_index = refill_dirty(dc);
405
406	if (searched_full_index &&
407	RB_EMPTY_ROOT(&dc->writeback_keys.keys)) {
408	atomic_set(&dc->has_dirty, 0);
409	cached_dev_put(dc);
410	SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
411	bch_write_bdev_super(dc, NULL);
412	}
413
414	up_write(&dc->writeback_lock);
415
416	bch_ratelimit_reset(&dc->writeback_rate);
417	read_dirty(dc);
418
419	if (searched_full_index) {
420	unsigned delay = dc->writeback_delay * HZ;
421
422	while (delay &&
423	!kthread_should_stop() &&
424	!atomic_read(&dc->disk.detaching))
425	delay = schedule_timeout_interruptible(delay);
426	}
427	}
428
429	return 0;
cafe5635 KO	430	}
cafe5635 KO	431
444fc0b6 KO	432	/* Init */
444fc0b6 KO	433
c18536a7 KO	434	struct sectors_dirty_init {
	435	struct btree_op op;
	436	unsigned inode;
	437	};
	438
	439	static int sectors_dirty_init_fn(struct btree_op _op, struct btree b,
48dad8ba	440	struct bkey *k)
444fc0b6	441	{
c18536a7 KO	442	struct sectors_dirty_init *op = container_of(_op,
c18536a7 KO	443	struct sectors_dirty_init, op);
48dad8ba KO	444	if (KEY_INODE(k) > op->inode)
48dad8ba KO	445	return MAP_DONE;
444fc0b6	446
48dad8ba KO	447	if (KEY_DIRTY(k))
	448	bcache_dev_sectors_dirty_add(b->c, KEY_INODE(k),
	449	KEY_START(k), KEY_SIZE(k));
	450
	451	return MAP_CONTINUE;
444fc0b6 KO	452	}
	453
	454	void bch_sectors_dirty_init(struct cached_dev *dc)
	455	{
c18536a7	456	struct sectors_dirty_init op;
444fc0b6	457
b54d6934	458	bch_btree_op_init(&op.op, -1);
48dad8ba KO	459	op.inode = dc->disk.id;
48dad8ba KO	460
c18536a7	461	bch_btree_map_keys(&op.op, dc->disk.c, &KEY(op.inode, 0, 0),
48dad8ba	462	sectors_dirty_init_fn, 0);
444fc0b6 KO	463	}
444fc0b6 KO	464
5e6926da	465	int bch_cached_dev_writeback_init(struct cached_dev *dc)
cafe5635	466	{
c2a4f318	467	sema_init(&dc->in_flight, 64);
cafe5635	468	init_rwsem(&dc->writeback_lock);
72c27061	469	bch_keybuf_init(&dc->writeback_keys);
cafe5635 KO	470
	471	dc->writeback_metadata = true;
	472	dc->writeback_running = true;
	473	dc->writeback_percent = 10;
	474	dc->writeback_delay = 30;
	475	dc->writeback_rate.rate = 1024;
	476
	477	dc->writeback_rate_update_seconds = 30;
	478	dc->writeback_rate_d_term = 16;
	479	dc->writeback_rate_p_term_inverse = 64;
	480	dc->writeback_rate_d_smooth = 8;
	481
5e6926da KO	482	dc->writeback_thread = kthread_create(bch_writeback_thread, dc,
	483	"bcache_writeback");
	484	if (IS_ERR(dc->writeback_thread))
	485	return PTR_ERR(dc->writeback_thread);
	486
	487	set_task_state(dc->writeback_thread, TASK_INTERRUPTIBLE);
	488
cafe5635 KO	489	INIT_DELAYED_WORK(&dc->writeback_rate_update, update_writeback_rate);
	490	schedule_delayed_work(&dc->writeback_rate_update,
	491	dc->writeback_rate_update_seconds * HZ);
cafe5635 KO	492
	493	return 0;
	494	}