[mirror_ubuntu-artful-kernel.git] / drivers / md / bcache / movinggc.c

/*
 * Moving/copying garbage collector
 *
 * Copyright 2012 Google, Inc.
 */

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

#include <trace/events/bcache.h>

struct moving_io {
	struct closure		cl;
	struct keybuf_key	*w;
	struct data_insert_op	op;
	struct bbio		bio;
};

static bool moving_pred(struct keybuf *buf, struct bkey *k)
{
	struct cache_set *c = container_of(buf, struct cache_set,
					   moving_gc_keys);
	unsigned i;

	for (i = 0; i < KEY_PTRS(k); i++)
		if (ptr_available(c, k, i) &&
		    GC_MOVE(PTR_BUCKET(c, k, i)))
			return true;

	return false;
}

/* Moving GC - IO loop */

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

static void write_moving_finish(struct closure *cl)
{
	struct moving_io *io = container_of(cl, struct moving_io, cl);
	struct bio *bio = &io->bio.bio;

	bio_free_pages(bio);

	if (io->op.replace_collision)
		trace_bcache_gc_copy_collision(&io->w->key);

	bch_keybuf_del(&io->op.c->moving_gc_keys, io->w);

	up(&io->op.c->moving_in_flight);

	closure_return_with_destructor(cl, moving_io_destructor);
}

static void read_moving_endio(struct bio *bio)
{
	struct bbio *b = container_of(bio, struct bbio, bio);
	struct moving_io *io = container_of(bio->bi_private,
					    struct moving_io, cl);

	if (bio->bi_status)
		io->op.status = bio->bi_status;
	else if (!KEY_DIRTY(&b->key) &&
		 ptr_stale(io->op.c, &b->key, 0)) {
		io->op.status = BLK_STS_IOERR;
	}

	bch_bbio_endio(io->op.c, bio, bio->bi_status, "reading data to move");
}

static void moving_init(struct moving_io *io)
{
	struct bio *bio = &io->bio.bio;

	bio_init(bio, bio->bi_inline_vecs,
		 DIV_ROUND_UP(KEY_SIZE(&io->w->key), PAGE_SECTORS));
	bio_get(bio);
	bio_set_prio(bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));

	bio->bi_iter.bi_size	= KEY_SIZE(&io->w->key) << 9;
	bio->bi_private		= &io->cl;
	bch_bio_map(bio, NULL);
}

static void write_moving(struct closure *cl)
{
	struct moving_io *io = container_of(cl, struct moving_io, cl);
	struct data_insert_op *op = &io->op;

	if (!op->status) {
		moving_init(io);

		io->bio.bio.bi_iter.bi_sector = KEY_START(&io->w->key);
		op->write_prio		= 1;
		op->bio			= &io->bio.bio;

		op->writeback		= KEY_DIRTY(&io->w->key);
		op->csum		= KEY_CSUM(&io->w->key);

		bkey_copy(&op->replace_key, &io->w->key);
		op->replace		= true;

		closure_call(&op->cl, bch_data_insert, NULL, cl);
	}

	continue_at(cl, write_moving_finish, op->wq);
}

static void read_moving_submit(struct closure *cl)
{
	struct moving_io *io = container_of(cl, struct moving_io, cl);
	struct bio *bio = &io->bio.bio;

	bch_submit_bbio(bio, io->op.c, &io->w->key, 0);

	continue_at(cl, write_moving, io->op.wq);
}

static void read_moving(struct cache_set *c)
{
	struct keybuf_key *w;
	struct moving_io *io;
	struct bio *bio;
	struct closure cl;

	closure_init_stack(&cl);

	/* XXX: if we error, background writeback could stall indefinitely */

	while (!test_bit(CACHE_SET_STOPPING, &c->flags)) {
		w = bch_keybuf_next_rescan(c, &c->moving_gc_keys,
					   &MAX_KEY, moving_pred);
		if (!w)
			break;

		if (ptr_stale(c, &w->key, 0)) {
			bch_keybuf_del(&c->moving_gc_keys, w);
			continue;
		}

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

		w->private	= io;
		io->w		= w;
		io->op.inode	= KEY_INODE(&w->key);
		io->op.c	= c;
		io->op.wq	= c->moving_gc_wq;

		moving_init(io);
		bio = &io->bio.bio;

		bio_set_op_attrs(bio, REQ_OP_READ, 0);
		bio->bi_end_io	= read_moving_endio;

		if (bio_alloc_pages(bio, GFP_KERNEL))
			goto err;

		trace_bcache_gc_copy(&w->key);

		down(&c->moving_in_flight);
		closure_call(&io->cl, read_moving_submit, NULL, &cl);
	}

	if (0) {
err:		if (!IS_ERR_OR_NULL(w->private))
			kfree(w->private);

		bch_keybuf_del(&c->moving_gc_keys, w);
	}

	closure_sync(&cl);
}

static bool bucket_cmp(struct bucket *l, struct bucket *r)
{
	return GC_SECTORS_USED(l) < GC_SECTORS_USED(r);
}

static unsigned bucket_heap_top(struct cache *ca)
{
	struct bucket *b;
	return (b = heap_peek(&ca->heap)) ? GC_SECTORS_USED(b) : 0;
}

void bch_moving_gc(struct cache_set *c)
{
	struct cache *ca;
	struct bucket *b;
	unsigned i;

	if (!c->copy_gc_enabled)
		return;

	mutex_lock(&c->bucket_lock);

	for_each_cache(ca, c, i) {
		unsigned sectors_to_move = 0;
		unsigned reserve_sectors = ca->sb.bucket_size *
			fifo_used(&ca->free[RESERVE_MOVINGGC]);

		ca->heap.used = 0;

		for_each_bucket(b, ca) {
			if (GC_MARK(b) == GC_MARK_METADATA ||
			    !GC_SECTORS_USED(b) ||
			    GC_SECTORS_USED(b) == ca->sb.bucket_size ||
			    atomic_read(&b->pin))
				continue;

			if (!heap_full(&ca->heap)) {
				sectors_to_move += GC_SECTORS_USED(b);
				heap_add(&ca->heap, b, bucket_cmp);
			} else if (bucket_cmp(b, heap_peek(&ca->heap))) {
				sectors_to_move -= bucket_heap_top(ca);
				sectors_to_move += GC_SECTORS_USED(b);

				ca->heap.data[0] = b;
				heap_sift(&ca->heap, 0, bucket_cmp);
			}
		}

		while (sectors_to_move > reserve_sectors) {
			heap_pop(&ca->heap, b, bucket_cmp);
			sectors_to_move -= GC_SECTORS_USED(b);
		}

		while (heap_pop(&ca->heap, b, bucket_cmp))
			SET_GC_MOVE(b, 1);
	}

	mutex_unlock(&c->bucket_lock);

	c->moving_gc_keys.last_scanned = ZERO_KEY;

	read_moving(c);
}

void bch_moving_init_cache_set(struct cache_set *c)
{
	bch_keybuf_init(&c->moving_gc_keys);
	sema_init(&c->moving_in_flight, 64);
}
Commit	Line	Data
cafe5635 KO	1	/*
	2	* Moving/copying garbage collector
	3	*
	4	* Copyright 2012 Google, Inc.
	5	*/
	6
	7	#include "bcache.h"
	8	#include "btree.h"
	9	#include "debug.h"
	10	#include "request.h"
	11
c37511b8 KO	12	#include <trace/events/bcache.h>
c37511b8 KO	13
cafe5635	14	struct moving_io {
220bb38c	15	struct closure cl;
cafe5635	16	struct keybuf_key *w;
220bb38c	17	struct data_insert_op op;
cafe5635 KO	18	struct bbio bio;
	19	};
	20
	21	static bool moving_pred(struct keybuf buf, struct bkey k)
	22	{
	23	struct cache_set *c = container_of(buf, struct cache_set,
	24	moving_gc_keys);
	25	unsigned i;
	26
10d9dcf6 KO	27	for (i = 0; i < KEY_PTRS(k); i++)
	28	if (ptr_available(c, k, i) &&
	29	GC_MOVE(PTR_BUCKET(c, k, i)))
cafe5635	30	return true;
cafe5635 KO	31
	32	return false;
	33	}
	34
	35	/* Moving GC - IO loop */
	36
	37	static void moving_io_destructor(struct closure *cl)
	38	{
220bb38c	39	struct moving_io *io = container_of(cl, struct moving_io, cl);
cafe5635 KO	40	kfree(io);
	41	}
	42
	43	static void write_moving_finish(struct closure *cl)
	44	{
220bb38c	45	struct moving_io *io = container_of(cl, struct moving_io, cl);
cafe5635	46	struct bio *bio = &io->bio.bio;
cafe5635	47
491221f8	48	bio_free_pages(bio);
cafe5635	49
220bb38c	50	if (io->op.replace_collision)
c37511b8	51	trace_bcache_gc_copy_collision(&io->w->key);
cafe5635	52
220bb38c	53	bch_keybuf_del(&io->op.c->moving_gc_keys, io->w);
cafe5635	54
220bb38c	55	up(&io->op.c->moving_in_flight);
cafe5635 KO	56
	57	closure_return_with_destructor(cl, moving_io_destructor);
	58	}
	59
4246a0b6	60	static void read_moving_endio(struct bio *bio)
cafe5635	61	{
6d3d1a9c	62	struct bbio *b = container_of(bio, struct bbio, bio);
cafe5635	63	struct moving_io *io = container_of(bio->bi_private,
220bb38c	64	struct moving_io, cl);
cafe5635	65
4e4cbee9 CH	66	if (bio->bi_status)
4e4cbee9 CH	67	io->op.status = bio->bi_status;
6d3d1a9c KO	68	else if (!KEY_DIRTY(&b->key) &&
6d3d1a9c KO	69	ptr_stale(io->op.c, &b->key, 0)) {
4e4cbee9	70	io->op.status = BLK_STS_IOERR;
6d3d1a9c	71	}
cafe5635	72
4e4cbee9	73	bch_bbio_endio(io->op.c, bio, bio->bi_status, "reading data to move");
cafe5635 KO	74	}
	75
	76	static void moving_init(struct moving_io *io)
	77	{
	78	struct bio *bio = &io->bio.bio;
	79
3a83f467 ML	80	bio_init(bio, bio->bi_inline_vecs,
3a83f467 ML	81	DIV_ROUND_UP(KEY_SIZE(&io->w->key), PAGE_SECTORS));
cafe5635 KO	82	bio_get(bio);
	83	bio_set_prio(bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
	84
4f024f37	85	bio->bi_iter.bi_size = KEY_SIZE(&io->w->key) << 9;
220bb38c	86	bio->bi_private = &io->cl;
169ef1cf	87	bch_bio_map(bio, NULL);
cafe5635 KO	88	}
	89
	90	static void write_moving(struct closure *cl)
	91	{
220bb38c KO	92	struct moving_io *io = container_of(cl, struct moving_io, cl);
220bb38c KO	93	struct data_insert_op *op = &io->op;
cafe5635	94
4e4cbee9	95	if (!op->status) {
cafe5635 KO	96	moving_init(io);
cafe5635 KO	97
4f024f37	98	io->bio.bio.bi_iter.bi_sector = KEY_START(&io->w->key);
220bb38c KO	99	op->write_prio = 1;
220bb38c KO	100	op->bio = &io->bio.bio;
cafe5635	101
220bb38c KO	102	op->writeback = KEY_DIRTY(&io->w->key);
220bb38c KO	103	op->csum = KEY_CSUM(&io->w->key);
cafe5635	104
220bb38c KO	105	bkey_copy(&op->replace_key, &io->w->key);
220bb38c KO	106	op->replace = true;
cafe5635	107
220bb38c	108	closure_call(&op->cl, bch_data_insert, NULL, cl);
cafe5635 KO	109	}
cafe5635 KO	110
da415a09	111	continue_at(cl, write_moving_finish, op->wq);
cafe5635 KO	112	}
	113
	114	static void read_moving_submit(struct closure *cl)
	115	{
220bb38c	116	struct moving_io *io = container_of(cl, struct moving_io, cl);
cafe5635 KO	117	struct bio *bio = &io->bio.bio;
cafe5635 KO	118
220bb38c	119	bch_submit_bbio(bio, io->op.c, &io->w->key, 0);
cafe5635	120
da415a09	121	continue_at(cl, write_moving, io->op.wq);
cafe5635 KO	122	}
cafe5635 KO	123
72a44517	124	static void read_moving(struct cache_set *c)
cafe5635	125	{
cafe5635 KO	126	struct keybuf_key *w;
	127	struct moving_io *io;
	128	struct bio *bio;
72a44517 KO	129	struct closure cl;
	130
	131	closure_init_stack(&cl);
cafe5635 KO	132
	133	/* XXX: if we error, background writeback could stall indefinitely */
	134
	135	while (!test_bit(CACHE_SET_STOPPING, &c->flags)) {
72c27061 KO	136	w = bch_keybuf_next_rescan(c, &c->moving_gc_keys,
72c27061 KO	137	&MAX_KEY, moving_pred);
cafe5635 KO	138	if (!w)
	139	break;
	140
6d3d1a9c KO	141	if (ptr_stale(c, &w->key, 0)) {
	142	bch_keybuf_del(&c->moving_gc_keys, w);
	143	continue;
	144	}
	145
cafe5635 KO	146	io = kzalloc(sizeof(struct moving_io) + sizeof(struct bio_vec)
	147	* DIV_ROUND_UP(KEY_SIZE(&w->key), PAGE_SECTORS),
	148	GFP_KERNEL);
	149	if (!io)
	150	goto err;
	151
	152	w->private = io;
	153	io->w = w;
220bb38c KO	154	io->op.inode = KEY_INODE(&w->key);
220bb38c KO	155	io->op.c = c;
da415a09	156	io->op.wq = c->moving_gc_wq;
cafe5635 KO	157
	158	moving_init(io);
	159	bio = &io->bio.bio;
	160
ad0d9e76	161	bio_set_op_attrs(bio, REQ_OP_READ, 0);
cafe5635 KO	162	bio->bi_end_io = read_moving_endio;
cafe5635 KO	163
8e51e414	164	if (bio_alloc_pages(bio, GFP_KERNEL))
cafe5635 KO	165	goto err;
cafe5635 KO	166
c37511b8	167	trace_bcache_gc_copy(&w->key);
cafe5635	168
72a44517	169	down(&c->moving_in_flight);
220bb38c	170	closure_call(&io->cl, read_moving_submit, NULL, &cl);
cafe5635 KO	171	}
	172
	173	if (0) {
	174	err: if (!IS_ERR_OR_NULL(w->private))
	175	kfree(w->private);
	176
	177	bch_keybuf_del(&c->moving_gc_keys, w);
	178	}
	179
72a44517	180	closure_sync(&cl);
cafe5635 KO	181	}
cafe5635 KO	182
b1a67b0f KO	183	static bool bucket_cmp(struct bucket l, struct bucket r)
	184	{
	185	return GC_SECTORS_USED(l) < GC_SECTORS_USED(r);
	186	}
	187
	188	static unsigned bucket_heap_top(struct cache *ca)
	189	{
bee63f40 NS	190	struct bucket *b;
bee63f40 NS	191	return (b = heap_peek(&ca->heap)) ? GC_SECTORS_USED(b) : 0;
b1a67b0f KO	192	}
b1a67b0f KO	193
72a44517	194	void bch_moving_gc(struct cache_set *c)
cafe5635	195	{
cafe5635 KO	196	struct cache *ca;
	197	struct bucket *b;
	198	unsigned i;
	199
cafe5635	200	if (!c->copy_gc_enabled)
72a44517	201	return;
cafe5635 KO	202
	203	mutex_lock(&c->bucket_lock);
	204
	205	for_each_cache(ca, c, i) {
	206	unsigned sectors_to_move = 0;
	207	unsigned reserve_sectors = ca->sb.bucket_size *
78365411	208	fifo_used(&ca->free[RESERVE_MOVINGGC]);
cafe5635 KO	209
	210	ca->heap.used = 0;
	211
	212	for_each_bucket(b, ca) {
3f6ef381 NS	213	if (GC_MARK(b) == GC_MARK_METADATA \|\|
	214	!GC_SECTORS_USED(b) \|\|
	215	GC_SECTORS_USED(b) == ca->sb.bucket_size \|\|
	216	atomic_read(&b->pin))
cafe5635 KO	217	continue;
	218
	219	if (!heap_full(&ca->heap)) {
	220	sectors_to_move += GC_SECTORS_USED(b);
	221	heap_add(&ca->heap, b, bucket_cmp);
	222	} else if (bucket_cmp(b, heap_peek(&ca->heap))) {
b1a67b0f	223	sectors_to_move -= bucket_heap_top(ca);
cafe5635 KO	224	sectors_to_move += GC_SECTORS_USED(b);
	225
	226	ca->heap.data[0] = b;
	227	heap_sift(&ca->heap, 0, bucket_cmp);
	228	}
	229	}
	230
	231	while (sectors_to_move > reserve_sectors) {
	232	heap_pop(&ca->heap, b, bucket_cmp);
	233	sectors_to_move -= GC_SECTORS_USED(b);
	234	}
	235
981aa8c0 NS	236	while (heap_pop(&ca->heap, b, bucket_cmp))
981aa8c0 NS	237	SET_GC_MOVE(b, 1);
cafe5635 KO	238	}
	239
	240	mutex_unlock(&c->bucket_lock);
	241
	242	c->moving_gc_keys.last_scanned = ZERO_KEY;
	243
72a44517	244	read_moving(c);
cafe5635 KO	245	}
	246
	247	void bch_moving_init_cache_set(struct cache_set *c)
	248	{
72c27061	249	bch_keybuf_init(&c->moving_gc_keys);
72a44517	250	sema_init(&c->moving_in_flight, 64);
cafe5635	251	}