[mirror_ubuntu-artful-kernel.git] / drivers / md / dm-io.c

/*
 * Copyright (C) 2003 Sistina Software
 * Copyright (C) 2006 Red Hat GmbH
 *
 * This file is released under the GPL.
 */

#include <linux/device-mapper.h>

#include <linux/bio.h>
#include <linux/mempool.h>
#include <linux/module.h>
#include <linux/sched.h>
#include <linux/slab.h>
#include <linux/dm-io.h>

struct dm_io_client {
	mempool_t *pool;
	struct bio_set *bios;
};

/* FIXME: can we shrink this ? */
struct io {
	unsigned long error_bits;
	atomic_t count;
	struct task_struct *sleeper;
	struct dm_io_client *client;
	io_notify_fn callback;
	void *context;
};

/*
 * io contexts are only dynamically allocated for asynchronous
 * io.  Since async io is likely to be the majority of io we'll
 * have the same number of io contexts as bios! (FIXME: must reduce this).
 */

static unsigned int pages_to_ios(unsigned int pages)
{
	return 4 * pages;	/* too many ? */
}

/*
 * Create a client with mempool and bioset.
 */
struct dm_io_client *dm_io_client_create(unsigned num_pages)
{
	unsigned ios = pages_to_ios(num_pages);
	struct dm_io_client *client;

	client = kmalloc(sizeof(*client), GFP_KERNEL);
	if (!client)
		return ERR_PTR(-ENOMEM);

	client->pool = mempool_create_kmalloc_pool(ios, sizeof(struct io));
	if (!client->pool)
		goto bad;

	client->bios = bioset_create(16, 0);
	if (!client->bios)
		goto bad;

	return client;

   bad:
	if (client->pool)
		mempool_destroy(client->pool);
	kfree(client);
	return ERR_PTR(-ENOMEM);
}
EXPORT_SYMBOL(dm_io_client_create);

int dm_io_client_resize(unsigned num_pages, struct dm_io_client *client)
{
	return mempool_resize(client->pool, pages_to_ios(num_pages),
			      GFP_KERNEL);
}
EXPORT_SYMBOL(dm_io_client_resize);

void dm_io_client_destroy(struct dm_io_client *client)
{
	mempool_destroy(client->pool);
	bioset_free(client->bios);
	kfree(client);
}
EXPORT_SYMBOL(dm_io_client_destroy);

/*-----------------------------------------------------------------
 * We need to keep track of which region a bio is doing io for.
 * In order to save a memory allocation we store this the last
 * bvec which we know is unused (blech).
 * XXX This is ugly and can OOPS with some configs... find another way.
 *---------------------------------------------------------------*/
static inline void bio_set_region(struct bio *bio, unsigned region)
{
	bio->bi_io_vec[bio->bi_max_vecs].bv_len = region;
}

static inline unsigned bio_get_region(struct bio *bio)
{
	return bio->bi_io_vec[bio->bi_max_vecs].bv_len;
}

/*-----------------------------------------------------------------
 * We need an io object to keep track of the number of bios that
 * have been dispatched for a particular io.
 *---------------------------------------------------------------*/
static void dec_count(struct io *io, unsigned int region, int error)
{
	if (error)
		set_bit(region, &io->error_bits);

	if (atomic_dec_and_test(&io->count)) {
		if (io->sleeper)
			wake_up_process(io->sleeper);

		else {
			unsigned long r = io->error_bits;
			io_notify_fn fn = io->callback;
			void *context = io->context;

			mempool_free(io, io->client->pool);
			fn(r, context);
		}
	}
}

static void endio(struct bio *bio, int error)
{
	struct io *io;
	unsigned region;

	if (error && bio_data_dir(bio) == READ)
		zero_fill_bio(bio);

	/*
	 * The bio destructor in bio_put() may use the io object.
	 */
	io = bio->bi_private;
	region = bio_get_region(bio);

	bio->bi_max_vecs++;
	bio_put(bio);

	dec_count(io, region, error);
}

/*-----------------------------------------------------------------
 * These little objects provide an abstraction for getting a new
 * destination page for io.
 *---------------------------------------------------------------*/
struct dpages {
	void (*get_page)(struct dpages *dp,
			 struct page **p, unsigned long *len, unsigned *offset);
	void (*next_page)(struct dpages *dp);

	unsigned context_u;
	void *context_ptr;
};

/*
 * Functions for getting the pages from a list.
 */
static void list_get_page(struct dpages *dp,
		  struct page **p, unsigned long *len, unsigned *offset)
{
	unsigned o = dp->context_u;
	struct page_list *pl = (struct page_list *) dp->context_ptr;

	*p = pl->page;
	*len = PAGE_SIZE - o;
	*offset = o;
}

static void list_next_page(struct dpages *dp)
{
	struct page_list *pl = (struct page_list *) dp->context_ptr;
	dp->context_ptr = pl->next;
	dp->context_u = 0;
}

static void list_dp_init(struct dpages *dp, struct page_list *pl, unsigned offset)
{
	dp->get_page = list_get_page;
	dp->next_page = list_next_page;
	dp->context_u = offset;
	dp->context_ptr = pl;
}

/*
 * Functions for getting the pages from a bvec.
 */
static void bvec_get_page(struct dpages *dp,
		  struct page **p, unsigned long *len, unsigned *offset)
{
	struct bio_vec *bvec = (struct bio_vec *) dp->context_ptr;
	*p = bvec->bv_page;
	*len = bvec->bv_len;
	*offset = bvec->bv_offset;
}

static void bvec_next_page(struct dpages *dp)
{
	struct bio_vec *bvec = (struct bio_vec *) dp->context_ptr;
	dp->context_ptr = bvec + 1;
}

static void bvec_dp_init(struct dpages *dp, struct bio_vec *bvec)
{
	dp->get_page = bvec_get_page;
	dp->next_page = bvec_next_page;
	dp->context_ptr = bvec;
}

/*
 * Functions for getting the pages from a VMA.
 */
static void vm_get_page(struct dpages *dp,
		 struct page **p, unsigned long *len, unsigned *offset)
{
	*p = vmalloc_to_page(dp->context_ptr);
	*offset = dp->context_u;
	*len = PAGE_SIZE - dp->context_u;
}

static void vm_next_page(struct dpages *dp)
{
	dp->context_ptr += PAGE_SIZE - dp->context_u;
	dp->context_u = 0;
}

static void vm_dp_init(struct dpages *dp, void *data)
{
	dp->get_page = vm_get_page;
	dp->next_page = vm_next_page;
	dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1);
	dp->context_ptr = data;
}

static void dm_bio_destructor(struct bio *bio)
{
	struct io *io = bio->bi_private;

	bio_free(bio, io->client->bios);
}

/*
 * Functions for getting the pages from kernel memory.
 */
static void km_get_page(struct dpages *dp, struct page **p, unsigned long *len,
			unsigned *offset)
{
	*p = virt_to_page(dp->context_ptr);
	*offset = dp->context_u;
	*len = PAGE_SIZE - dp->context_u;
}

static void km_next_page(struct dpages *dp)
{
	dp->context_ptr += PAGE_SIZE - dp->context_u;
	dp->context_u = 0;
}

static void km_dp_init(struct dpages *dp, void *data)
{
	dp->get_page = km_get_page;
	dp->next_page = km_next_page;
	dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1);
	dp->context_ptr = data;
}

/*-----------------------------------------------------------------
 * IO routines that accept a list of pages.
 *---------------------------------------------------------------*/
static void do_region(int rw, unsigned region, struct dm_io_region *where,
		      struct dpages *dp, struct io *io)
{
	struct bio *bio;
	struct page *page;
	unsigned long len;
	unsigned offset;
	unsigned num_bvecs;
	sector_t remaining = where->count;

	while (remaining) {
		/*
		 * Allocate a suitably sized-bio: we add an extra
		 * bvec for bio_get/set_region() and decrement bi_max_vecs
		 * to hide it from bio_add_page().
		 */
		num_bvecs = dm_sector_div_up(remaining,
					     (PAGE_SIZE >> SECTOR_SHIFT));
		num_bvecs = 1 + min_t(int, bio_get_nr_vecs(where->bdev),
				      num_bvecs);
		if (unlikely(num_bvecs > BIO_MAX_PAGES))
			num_bvecs = BIO_MAX_PAGES;
		bio = bio_alloc_bioset(GFP_NOIO, num_bvecs, io->client->bios);
		bio->bi_sector = where->sector + (where->count - remaining);
		bio->bi_bdev = where->bdev;
		bio->bi_end_io = endio;
		bio->bi_private = io;
		bio->bi_destructor = dm_bio_destructor;
		bio->bi_max_vecs--;
		bio_set_region(bio, region);

		/*
		 * Try and add as many pages as possible.
		 */
		while (remaining) {
			dp->get_page(dp, &page, &len, &offset);
			len = min(len, to_bytes(remaining));
			if (!bio_add_page(bio, page, len, offset))
				break;

			offset = 0;
			remaining -= to_sector(len);
			dp->next_page(dp);
		}

		atomic_inc(&io->count);
		submit_bio(rw, bio);
	}
}

static void dispatch_io(int rw, unsigned int num_regions,
			struct dm_io_region *where, struct dpages *dp,
			struct io *io, int sync)
{
	int i;
	struct dpages old_pages = *dp;

	if (sync)
		rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);

	/*
	 * For multiple regions we need to be careful to rewind
	 * the dp object for each call to do_region.
	 */
	for (i = 0; i < num_regions; i++) {
		*dp = old_pages;
		if (where[i].count)
			do_region(rw, i, where + i, dp, io);
	}

	/*
	 * Drop the extra reference that we were holding to avoid
	 * the io being completed too early.
	 */
	dec_count(io, 0, 0);
}

static int sync_io(struct dm_io_client *client, unsigned int num_regions,
		   struct dm_io_region *where, int rw, struct dpages *dp,
		   unsigned long *error_bits)
{
	struct io io;

	if (num_regions > 1 && (rw & RW_MASK) != WRITE) {
		WARN_ON(1);
		return -EIO;
	}

	io.error_bits = 0;
	atomic_set(&io.count, 1); /* see dispatch_io() */
	io.sleeper = current;
	io.client = client;

	dispatch_io(rw, num_regions, where, dp, &io, 1);

	while (1) {
		set_current_state(TASK_UNINTERRUPTIBLE);

		if (!atomic_read(&io.count))
			break;

		io_schedule();
	}
	set_current_state(TASK_RUNNING);

	if (error_bits)
		*error_bits = io.error_bits;

	return io.error_bits ? -EIO : 0;
}

static int async_io(struct dm_io_client *client, unsigned int num_regions,
		    struct dm_io_region *where, int rw, struct dpages *dp,
		    io_notify_fn fn, void *context)
{
	struct io *io;

	if (num_regions > 1 && (rw & RW_MASK) != WRITE) {
		WARN_ON(1);
		fn(1, context);
		return -EIO;
	}

	io = mempool_alloc(client->pool, GFP_NOIO);
	io->error_bits = 0;
	atomic_set(&io->count, 1); /* see dispatch_io() */
	io->sleeper = NULL;
	io->client = client;
	io->callback = fn;
	io->context = context;

	dispatch_io(rw, num_regions, where, dp, io, 0);
	return 0;
}

static int dp_init(struct dm_io_request *io_req, struct dpages *dp)
{
	/* Set up dpages based on memory type */
	switch (io_req->mem.type) {
	case DM_IO_PAGE_LIST:
		list_dp_init(dp, io_req->mem.ptr.pl, io_req->mem.offset);
		break;

	case DM_IO_BVEC:
		bvec_dp_init(dp, io_req->mem.ptr.bvec);
		break;

	case DM_IO_VMA:
		vm_dp_init(dp, io_req->mem.ptr.vma);
		break;

	case DM_IO_KMEM:
		km_dp_init(dp, io_req->mem.ptr.addr);
		break;

	default:
		return -EINVAL;
	}

	return 0;
}

/*
 * New collapsed (a)synchronous interface.
 *
 * If the IO is asynchronous (i.e. it has notify.fn), you must either unplug
 * the queue with blk_unplug() some time later or set the BIO_RW_SYNC bit in
 * io_req->bi_rw. If you fail to do one of these, the IO will be submitted to
 * the disk after q->unplug_delay, which defaults to 3ms in blk-settings.c.
 */
int dm_io(struct dm_io_request *io_req, unsigned num_regions,
	  struct dm_io_region *where, unsigned long *sync_error_bits)
{
	int r;
	struct dpages dp;

	r = dp_init(io_req, &dp);
	if (r)
		return r;

	if (!io_req->notify.fn)
		return sync_io(io_req->client, num_regions, where,
			       io_req->bi_rw, &dp, sync_error_bits);

	return async_io(io_req->client, num_regions, where, io_req->bi_rw,
			&dp, io_req->notify.fn, io_req->notify.context);
}
EXPORT_SYMBOL(dm_io);
Commit	Line	Data
1da177e4 LT	1	/*
1da177e4 LT	2	* Copyright (C) 2003 Sistina Software
891ce207	3	* Copyright (C) 2006 Red Hat GmbH
1da177e4 LT	4	*
	5	* This file is released under the GPL.
	6	*/
	7
586e80e6	8	#include <linux/device-mapper.h>
1da177e4 LT	9
	10	#include <linux/bio.h>
	11	#include <linux/mempool.h>
	12	#include <linux/module.h>
	13	#include <linux/sched.h>
	14	#include <linux/slab.h>
a765e20e	15	#include <linux/dm-io.h>
1da177e4	16
891ce207 HM	17	struct dm_io_client {
	18	mempool_t *pool;
	19	struct bio_set *bios;
	20	};
	21
1da177e4 LT	22	/* FIXME: can we shrink this ? */
1da177e4 LT	23	struct io {
e01fd7ee	24	unsigned long error_bits;
1da177e4 LT	25	atomic_t count;
1da177e4 LT	26	struct task_struct *sleeper;
891ce207	27	struct dm_io_client *client;
1da177e4 LT	28	io_notify_fn callback;
	29	void *context;
	30	};
	31
	32	/*
	33	* io contexts are only dynamically allocated for asynchronous
	34	* io. Since async io is likely to be the majority of io we'll
891ce207	35	* have the same number of io contexts as bios! (FIXME: must reduce this).
1da177e4	36	*/
891ce207	37
1da177e4 LT	38	static unsigned int pages_to_ios(unsigned int pages)
	39	{
	40	return 4 * pages; /* too many ? */
	41	}
	42
c8b03afe HM	43	/*
	44	* Create a client with mempool and bioset.
	45	*/
	46	struct dm_io_client *dm_io_client_create(unsigned num_pages)
	47	{
	48	unsigned ios = pages_to_ios(num_pages);
	49	struct dm_io_client *client;
	50
	51	client = kmalloc(sizeof(*client), GFP_KERNEL);
	52	if (!client)
	53	return ERR_PTR(-ENOMEM);
	54
	55	client->pool = mempool_create_kmalloc_pool(ios, sizeof(struct io));
	56	if (!client->pool)
	57	goto bad;
	58
bb799ca0	59	client->bios = bioset_create(16, 0);
c8b03afe HM	60	if (!client->bios)
	61	goto bad;
	62
	63	return client;
	64
	65	bad:
	66	if (client->pool)
	67	mempool_destroy(client->pool);
	68	kfree(client);
	69	return ERR_PTR(-ENOMEM);
	70	}
	71	EXPORT_SYMBOL(dm_io_client_create);
	72
	73	int dm_io_client_resize(unsigned num_pages, struct dm_io_client *client)
	74	{
	75	return mempool_resize(client->pool, pages_to_ios(num_pages),
	76	GFP_KERNEL);
	77	}
	78	EXPORT_SYMBOL(dm_io_client_resize);
	79
	80	void dm_io_client_destroy(struct dm_io_client *client)
	81	{
	82	mempool_destroy(client->pool);
	83	bioset_free(client->bios);
	84	kfree(client);
	85	}
	86	EXPORT_SYMBOL(dm_io_client_destroy);
	87
1da177e4 LT	88	/*-----------------------------------------------------------------
	89	* We need to keep track of which region a bio is doing io for.
	90	* In order to save a memory allocation we store this the last
	91	* bvec which we know is unused (blech).
	92	* XXX This is ugly and can OOPS with some configs... find another way.
	93	---------------------------------------------------------------/
	94	static inline void bio_set_region(struct bio *bio, unsigned region)
	95	{
f00b16ad	96	bio->bi_io_vec[bio->bi_max_vecs].bv_len = region;
1da177e4 LT	97	}
	98
	99	static inline unsigned bio_get_region(struct bio *bio)
	100	{
f00b16ad	101	return bio->bi_io_vec[bio->bi_max_vecs].bv_len;
1da177e4 LT	102	}
	103
	104	/*-----------------------------------------------------------------
	105	* We need an io object to keep track of the number of bios that
	106	* have been dispatched for a particular io.
	107	---------------------------------------------------------------/
	108	static void dec_count(struct io *io, unsigned int region, int error)
	109	{
	110	if (error)
e01fd7ee	111	set_bit(region, &io->error_bits);
1da177e4 LT	112
	113	if (atomic_dec_and_test(&io->count)) {
	114	if (io->sleeper)
	115	wake_up_process(io->sleeper);
	116
	117	else {
e01fd7ee	118	unsigned long r = io->error_bits;
1da177e4 LT	119	io_notify_fn fn = io->callback;
	120	void *context = io->context;
	121
bf17ce3a	122	mempool_free(io, io->client->pool);
1da177e4 LT	123	fn(r, context);
	124	}
	125	}
	126	}
	127
6712ecf8	128	static void endio(struct bio *bio, int error)
1da177e4	129	{
c897feb3 HM	130	struct io *io;
c897feb3 HM	131	unsigned region;
1da177e4	132
1da177e4 LT	133	if (error && bio_data_dir(bio) == READ)
	134	zero_fill_bio(bio);
	135
c897feb3 HM	136	/*
	137	* The bio destructor in bio_put() may use the io object.
	138	*/
	139	io = bio->bi_private;
	140	region = bio_get_region(bio);
	141
f00b16ad	142	bio->bi_max_vecs++;
1da177e4 LT	143	bio_put(bio);
1da177e4 LT	144
c897feb3	145	dec_count(io, region, error);
1da177e4 LT	146	}
	147
	148	/*-----------------------------------------------------------------
	149	* These little objects provide an abstraction for getting a new
	150	* destination page for io.
	151	---------------------------------------------------------------/
	152	struct dpages {
	153	void (get_page)(struct dpages dp,
	154	struct page *p, unsigned long len, unsigned *offset);
	155	void (next_page)(struct dpages dp);
	156
	157	unsigned context_u;
	158	void *context_ptr;
	159	};
	160
	161	/*
	162	* Functions for getting the pages from a list.
	163	*/
	164	static void list_get_page(struct dpages *dp,
	165	struct page *p, unsigned long len, unsigned *offset)
	166	{
	167	unsigned o = dp->context_u;
	168	struct page_list pl = (struct page_list ) dp->context_ptr;
	169
	170	*p = pl->page;
	171	*len = PAGE_SIZE - o;
	172	*offset = o;
	173	}
	174
	175	static void list_next_page(struct dpages *dp)
	176	{
	177	struct page_list pl = (struct page_list ) dp->context_ptr;
	178	dp->context_ptr = pl->next;
	179	dp->context_u = 0;
	180	}
	181
	182	static void list_dp_init(struct dpages dp, struct page_list pl, unsigned offset)
	183	{
	184	dp->get_page = list_get_page;
	185	dp->next_page = list_next_page;
	186	dp->context_u = offset;
	187	dp->context_ptr = pl;
	188	}
	189
	190	/*
	191	* Functions for getting the pages from a bvec.
	192	*/
	193	static void bvec_get_page(struct dpages *dp,
	194	struct page *p, unsigned long len, unsigned *offset)
	195	{
	196	struct bio_vec bvec = (struct bio_vec ) dp->context_ptr;
	197	*p = bvec->bv_page;
	198	*len = bvec->bv_len;
	199	*offset = bvec->bv_offset;
	200	}
	201
	202	static void bvec_next_page(struct dpages *dp)
	203	{
	204	struct bio_vec bvec = (struct bio_vec ) dp->context_ptr;
	205	dp->context_ptr = bvec + 1;
	206	}
	207
	208	static void bvec_dp_init(struct dpages dp, struct bio_vec bvec)
	209	{
210	dp->get_page = bvec_get_page;
211	dp->next_page = bvec_next_page;
212	dp->context_ptr = bvec;
213	}
214
c8b03afe HM	215	/*
	216	* Functions for getting the pages from a VMA.
	217	*/
1da177e4 LT	218	static void vm_get_page(struct dpages *dp,
	219	struct page *p, unsigned long len, unsigned *offset)
	220	{
	221	*p = vmalloc_to_page(dp->context_ptr);
	222	*offset = dp->context_u;
	223	*len = PAGE_SIZE - dp->context_u;
	224	}
	225
	226	static void vm_next_page(struct dpages *dp)
	227	{
	228	dp->context_ptr += PAGE_SIZE - dp->context_u;
	229	dp->context_u = 0;
	230	}
	231
	232	static void vm_dp_init(struct dpages dp, void data)
	233	{
	234	dp->get_page = vm_get_page;
	235	dp->next_page = vm_next_page;
	236	dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1);
	237	dp->context_ptr = data;
	238	}
	239
3676347a PO	240	static void dm_bio_destructor(struct bio *bio)
3676347a PO	241	{
891ce207 HM	242	struct io *io = bio->bi_private;
891ce207 HM	243
bf17ce3a	244	bio_free(bio, io->client->bios);
3676347a PO	245	}
3676347a PO	246
c8b03afe HM	247	/*
	248	* Functions for getting the pages from kernel memory.
	249	*/
	250	static void km_get_page(struct dpages dp, struct page p, unsigned long len,
	251	unsigned *offset)
	252	{
	253	*p = virt_to_page(dp->context_ptr);
	254	*offset = dp->context_u;
	255	*len = PAGE_SIZE - dp->context_u;
	256	}
	257
	258	static void km_next_page(struct dpages *dp)
	259	{
	260	dp->context_ptr += PAGE_SIZE - dp->context_u;
	261	dp->context_u = 0;
	262	}
	263
	264	static void km_dp_init(struct dpages dp, void data)
	265	{
	266	dp->get_page = km_get_page;
	267	dp->next_page = km_next_page;
	268	dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1);
	269	dp->context_ptr = data;
	270	}
	271
1da177e4 LT	272	/*-----------------------------------------------------------------
	273	* IO routines that accept a list of pages.
	274	---------------------------------------------------------------/
22a1ceb1	275	static void do_region(int rw, unsigned region, struct dm_io_region *where,
1da177e4 LT	276	struct dpages dp, struct io io)
	277	{
	278	struct bio *bio;
	279	struct page *page;
	280	unsigned long len;
	281	unsigned offset;
	282	unsigned num_bvecs;
	283	sector_t remaining = where->count;
	284
	285	while (remaining) {
	286	/*
f00b16ad HM	287	* Allocate a suitably sized-bio: we add an extra
	288	* bvec for bio_get/set_region() and decrement bi_max_vecs
	289	* to hide it from bio_add_page().
1da177e4	290	*/
596f138e JN	291	num_bvecs = dm_sector_div_up(remaining,
	292	(PAGE_SIZE >> SECTOR_SHIFT));
	293	num_bvecs = 1 + min_t(int, bio_get_nr_vecs(where->bdev),
	294	num_bvecs);
d659e6cc MP	295	if (unlikely(num_bvecs > BIO_MAX_PAGES))
d659e6cc MP	296	num_bvecs = BIO_MAX_PAGES;
bf17ce3a	297	bio = bio_alloc_bioset(GFP_NOIO, num_bvecs, io->client->bios);
1da177e4 LT	298	bio->bi_sector = where->sector + (where->count - remaining);
	299	bio->bi_bdev = where->bdev;
	300	bio->bi_end_io = endio;
	301	bio->bi_private = io;
3676347a	302	bio->bi_destructor = dm_bio_destructor;
f00b16ad	303	bio->bi_max_vecs--;
1da177e4 LT	304	bio_set_region(bio, region);
	305
	306	/*
	307	* Try and add as many pages as possible.
	308	*/
	309	while (remaining) {
	310	dp->get_page(dp, &page, &len, &offset);
	311	len = min(len, to_bytes(remaining));
	312	if (!bio_add_page(bio, page, len, offset))
	313	break;
	314
	315	offset = 0;
	316	remaining -= to_sector(len);
	317	dp->next_page(dp);
	318	}
	319
	320	atomic_inc(&io->count);
	321	submit_bio(rw, bio);
	322	}
	323	}
	324
	325	static void dispatch_io(int rw, unsigned int num_regions,
22a1ceb1	326	struct dm_io_region where, struct dpages dp,
1da177e4 LT	327	struct io *io, int sync)
	328	{
	329	int i;
	330	struct dpages old_pages = *dp;
	331
	332	if (sync)
93dbb393	333	rw \|= (1 << BIO_RW_SYNCIO) \| (1 << BIO_RW_UNPLUG);
1da177e4 LT	334
	335	/*
	336	* For multiple regions we need to be careful to rewind
	337	* the dp object for each call to do_region.
	338	*/
	339	for (i = 0; i < num_regions; i++) {
	340	*dp = old_pages;
	341	if (where[i].count)
	342	do_region(rw, i, where + i, dp, io);
	343	}
	344
	345	/*
f00b16ad	346	* Drop the extra reference that we were holding to avoid
1da177e4 LT	347	* the io being completed too early.
	348	*/
	349	dec_count(io, 0, 0);
	350	}
	351
891ce207	352	static int sync_io(struct dm_io_client *client, unsigned int num_regions,
22a1ceb1	353	struct dm_io_region where, int rw, struct dpages dp,
891ce207	354	unsigned long *error_bits)
1da177e4 LT	355	{
	356	struct io io;
	357
7ff14a36	358	if (num_regions > 1 && (rw & RW_MASK) != WRITE) {
1da177e4 LT	359	WARN_ON(1);
	360	return -EIO;
	361	}
	362
e01fd7ee	363	io.error_bits = 0;
1da177e4 LT	364	atomic_set(&io.count, 1); /* see dispatch_io() */
1da177e4 LT	365	io.sleeper = current;
891ce207	366	io.client = client;
1da177e4 LT	367
	368	dispatch_io(rw, num_regions, where, dp, &io, 1);
	369
	370	while (1) {
	371	set_current_state(TASK_UNINTERRUPTIBLE);
	372
b64b6bf4	373	if (!atomic_read(&io.count))
1da177e4 LT	374	break;
	375
	376	io_schedule();
	377	}
	378	set_current_state(TASK_RUNNING);
	379
891ce207	380	if (error_bits)
e01fd7ee	381	*error_bits = io.error_bits;
891ce207	382
e01fd7ee	383	return io.error_bits ? -EIO : 0;
1da177e4 LT	384	}
1da177e4 LT	385
891ce207	386	static int async_io(struct dm_io_client *client, unsigned int num_regions,
22a1ceb1	387	struct dm_io_region where, int rw, struct dpages dp,
891ce207	388	io_notify_fn fn, void *context)
1da177e4 LT	389	{
	390	struct io *io;
	391
7ff14a36	392	if (num_regions > 1 && (rw & RW_MASK) != WRITE) {
1da177e4 LT	393	WARN_ON(1);
	394	fn(1, context);
	395	return -EIO;
	396	}
	397
bf17ce3a	398	io = mempool_alloc(client->pool, GFP_NOIO);
e01fd7ee	399	io->error_bits = 0;
1da177e4 LT	400	atomic_set(&io->count, 1); /* see dispatch_io() */
1da177e4 LT	401	io->sleeper = NULL;
891ce207	402	io->client = client;
1da177e4 LT	403	io->callback = fn;
	404	io->context = context;
	405
	406	dispatch_io(rw, num_regions, where, dp, io, 0);
	407	return 0;
	408	}
	409
c8b03afe HM	410	static int dp_init(struct dm_io_request io_req, struct dpages dp)
	411	{
	412	/* Set up dpages based on memory type */
	413	switch (io_req->mem.type) {
	414	case DM_IO_PAGE_LIST:
	415	list_dp_init(dp, io_req->mem.ptr.pl, io_req->mem.offset);
	416	break;
	417
	418	case DM_IO_BVEC:
	419	bvec_dp_init(dp, io_req->mem.ptr.bvec);
	420	break;
	421
	422	case DM_IO_VMA:
	423	vm_dp_init(dp, io_req->mem.ptr.vma);
	424	break;
	425
	426	case DM_IO_KMEM:
	427	km_dp_init(dp, io_req->mem.ptr.addr);
	428	break;
	429
	430	default:
	431	return -EINVAL;
	432	}
	433
	434	return 0;
	435	}
	436
	437	/*
7ff14a36 MP	438	* New collapsed (a)synchronous interface.
	439	*
	440	* If the IO is asynchronous (i.e. it has notify.fn), you must either unplug
	441	* the queue with blk_unplug() some time later or set the BIO_RW_SYNC bit in
	442	* io_req->bi_rw. If you fail to do one of these, the IO will be submitted to
	443	* the disk after q->unplug_delay, which defaults to 3ms in blk-settings.c.
c8b03afe HM	444	*/
c8b03afe HM	445	int dm_io(struct dm_io_request *io_req, unsigned num_regions,
22a1ceb1	446	struct dm_io_region where, unsigned long sync_error_bits)
c8b03afe HM	447	{
	448	int r;
	449	struct dpages dp;
	450
	451	r = dp_init(io_req, &dp);
	452	if (r)
	453	return r;
	454
	455	if (!io_req->notify.fn)
	456	return sync_io(io_req->client, num_regions, where,
	457	io_req->bi_rw, &dp, sync_error_bits);
	458
	459	return async_io(io_req->client, num_regions, where, io_req->bi_rw,
	460	&dp, io_req->notify.fn, io_req->notify.context);
	461	}
	462	EXPORT_SYMBOL(dm_io);