[net-next-2.6.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 "dm.h"

#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;
	unsigned long eopnotsupp_bits;
	atomic_t count;
	struct task_struct *sleeper;
	struct dm_io_client *client;
	io_notify_fn callback;
	void *context;
};

static struct kmem_cache *_dm_io_cache;

/*
 * 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_slab_pool(ios, _dm_io_cache);
	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 (error == -EOPNOTSUPP)
			set_bit(region, &io->eopnotsupp_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;
	}

retry:
	io.error_bits = 0;
	io.eopnotsupp_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 (io.eopnotsupp_bits && (rw & (1 << BIO_RW_BARRIER))) {
		rw &= ~(1 << BIO_RW_BARRIER);
		goto retry;
	}

	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;
	io->eopnotsupp_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);

int __init dm_io_init(void)
{
	_dm_io_cache = KMEM_CACHE(io, 0);
	if (!_dm_io_cache)
		return -ENOMEM;

	return 0;
}

void dm_io_exit(void)
{
	kmem_cache_destroy(_dm_io_cache);
	_dm_io_cache = NULL;
}
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
952b3557 MP	8	#include "dm.h"
952b3557 MP	9
586e80e6	10	#include <linux/device-mapper.h>
1da177e4 LT	11
	12	#include <linux/bio.h>
	13	#include <linux/mempool.h>
	14	#include <linux/module.h>
	15	#include <linux/sched.h>
	16	#include <linux/slab.h>
a765e20e	17	#include <linux/dm-io.h>
1da177e4	18
891ce207 HM	19	struct dm_io_client {
	20	mempool_t *pool;
	21	struct bio_set *bios;
	22	};
	23
1da177e4 LT	24	/* FIXME: can we shrink this ? */
1da177e4 LT	25	struct io {
e01fd7ee	26	unsigned long error_bits;
5af443a7	27	unsigned long eopnotsupp_bits;
1da177e4 LT	28	atomic_t count;
1da177e4 LT	29	struct task_struct *sleeper;
891ce207	30	struct dm_io_client *client;
1da177e4 LT	31	io_notify_fn callback;
	32	void *context;
	33	};
	34
952b3557 MP	35	static struct kmem_cache *_dm_io_cache;
952b3557 MP	36
1da177e4 LT	37	/*
	38	* io contexts are only dynamically allocated for asynchronous
	39	* io. Since async io is likely to be the majority of io we'll
891ce207	40	* have the same number of io contexts as bios! (FIXME: must reduce this).
1da177e4	41	*/
891ce207	42
1da177e4 LT	43	static unsigned int pages_to_ios(unsigned int pages)
	44	{
	45	return 4 * pages; /* too many ? */
	46	}
	47
c8b03afe HM	48	/*
	49	* Create a client with mempool and bioset.
	50	*/
	51	struct dm_io_client *dm_io_client_create(unsigned num_pages)
	52	{
	53	unsigned ios = pages_to_ios(num_pages);
	54	struct dm_io_client *client;
	55
	56	client = kmalloc(sizeof(*client), GFP_KERNEL);
	57	if (!client)
	58	return ERR_PTR(-ENOMEM);
	59
952b3557	60	client->pool = mempool_create_slab_pool(ios, _dm_io_cache);
c8b03afe HM	61	if (!client->pool)
	62	goto bad;
	63
bb799ca0	64	client->bios = bioset_create(16, 0);
c8b03afe HM	65	if (!client->bios)
	66	goto bad;
	67
	68	return client;
	69
	70	bad:
	71	if (client->pool)
	72	mempool_destroy(client->pool);
	73	kfree(client);
	74	return ERR_PTR(-ENOMEM);
	75	}
	76	EXPORT_SYMBOL(dm_io_client_create);
	77
	78	int dm_io_client_resize(unsigned num_pages, struct dm_io_client *client)
	79	{
	80	return mempool_resize(client->pool, pages_to_ios(num_pages),
	81	GFP_KERNEL);
	82	}
	83	EXPORT_SYMBOL(dm_io_client_resize);
	84
	85	void dm_io_client_destroy(struct dm_io_client *client)
	86	{
	87	mempool_destroy(client->pool);
	88	bioset_free(client->bios);
	89	kfree(client);
	90	}
	91	EXPORT_SYMBOL(dm_io_client_destroy);
	92
1da177e4 LT	93	/*-----------------------------------------------------------------
	94	* We need to keep track of which region a bio is doing io for.
	95	* In order to save a memory allocation we store this the last
	96	* bvec which we know is unused (blech).
	97	* XXX This is ugly and can OOPS with some configs... find another way.
	98	---------------------------------------------------------------/
	99	static inline void bio_set_region(struct bio *bio, unsigned region)
	100	{
f00b16ad	101	bio->bi_io_vec[bio->bi_max_vecs].bv_len = region;
1da177e4 LT	102	}
	103
	104	static inline unsigned bio_get_region(struct bio *bio)
	105	{
f00b16ad	106	return bio->bi_io_vec[bio->bi_max_vecs].bv_len;
1da177e4 LT	107	}
	108
	109	/*-----------------------------------------------------------------
	110	* We need an io object to keep track of the number of bios that
	111	* have been dispatched for a particular io.
	112	---------------------------------------------------------------/
	113	static void dec_count(struct io *io, unsigned int region, int error)
	114	{
5af443a7	115	if (error) {
e01fd7ee	116	set_bit(region, &io->error_bits);
5af443a7 MP	117	if (error == -EOPNOTSUPP)
	118	set_bit(region, &io->eopnotsupp_bits);
	119	}
1da177e4 LT	120
	121	if (atomic_dec_and_test(&io->count)) {
	122	if (io->sleeper)
	123	wake_up_process(io->sleeper);
	124
	125	else {
e01fd7ee	126	unsigned long r = io->error_bits;
1da177e4 LT	127	io_notify_fn fn = io->callback;
	128	void *context = io->context;
	129
bf17ce3a	130	mempool_free(io, io->client->pool);
1da177e4 LT	131	fn(r, context);
	132	}
	133	}
	134	}
	135
6712ecf8	136	static void endio(struct bio *bio, int error)
1da177e4	137	{
c897feb3 HM	138	struct io *io;
c897feb3 HM	139	unsigned region;
1da177e4	140
1da177e4 LT	141	if (error && bio_data_dir(bio) == READ)
	142	zero_fill_bio(bio);
	143
c897feb3 HM	144	/*
	145	* The bio destructor in bio_put() may use the io object.
	146	*/
	147	io = bio->bi_private;
	148	region = bio_get_region(bio);
	149
f00b16ad	150	bio->bi_max_vecs++;
1da177e4 LT	151	bio_put(bio);
1da177e4 LT	152
c897feb3	153	dec_count(io, region, error);
1da177e4 LT	154	}
	155
	156	/*-----------------------------------------------------------------
	157	* These little objects provide an abstraction for getting a new
	158	* destination page for io.
	159	---------------------------------------------------------------/
	160	struct dpages {
	161	void (get_page)(struct dpages dp,
	162	struct page *p, unsigned long len, unsigned *offset);
	163	void (next_page)(struct dpages dp);
	164
	165	unsigned context_u;
	166	void *context_ptr;
	167	};
	168
	169	/*
	170	* Functions for getting the pages from a list.
	171	*/
	172	static void list_get_page(struct dpages *dp,
	173	struct page *p, unsigned long len, unsigned *offset)
	174	{
	175	unsigned o = dp->context_u;
	176	struct page_list pl = (struct page_list ) dp->context_ptr;
	177
	178	*p = pl->page;
	179	*len = PAGE_SIZE - o;
	180	*offset = o;
	181	}
	182
	183	static void list_next_page(struct dpages *dp)
	184	{
	185	struct page_list pl = (struct page_list ) dp->context_ptr;
	186	dp->context_ptr = pl->next;
	187	dp->context_u = 0;
	188	}
	189
	190	static void list_dp_init(struct dpages dp, struct page_list pl, unsigned offset)
	191	{
	192	dp->get_page = list_get_page;
	193	dp->next_page = list_next_page;
	194	dp->context_u = offset;
	195	dp->context_ptr = pl;
	196	}
	197
	198	/*
	199	* Functions for getting the pages from a bvec.
	200	*/
	201	static void bvec_get_page(struct dpages *dp,
	202	struct page *p, unsigned long len, unsigned *offset)
	203	{
	204	struct bio_vec bvec = (struct bio_vec ) dp->context_ptr;
	205	*p = bvec->bv_page;
	206	*len = bvec->bv_len;
	207	*offset = bvec->bv_offset;
	208	}
	209
	210	static void bvec_next_page(struct dpages *dp)
	211	{
	212	struct bio_vec bvec = (struct bio_vec ) dp->context_ptr;
	213	dp->context_ptr = bvec + 1;
	214	}
	215
	216	static void bvec_dp_init(struct dpages dp, struct bio_vec bvec)
	217	{
218	dp->get_page = bvec_get_page;
219	dp->next_page = bvec_next_page;
220	dp->context_ptr = bvec;
221	}
222
c8b03afe HM	223	/*
	224	* Functions for getting the pages from a VMA.
	225	*/
1da177e4 LT	226	static void vm_get_page(struct dpages *dp,
	227	struct page *p, unsigned long len, unsigned *offset)
	228	{
	229	*p = vmalloc_to_page(dp->context_ptr);
	230	*offset = dp->context_u;
	231	*len = PAGE_SIZE - dp->context_u;
	232	}
	233
	234	static void vm_next_page(struct dpages *dp)
	235	{
	236	dp->context_ptr += PAGE_SIZE - dp->context_u;
	237	dp->context_u = 0;
	238	}
	239
	240	static void vm_dp_init(struct dpages dp, void data)
	241	{
	242	dp->get_page = vm_get_page;
	243	dp->next_page = vm_next_page;
	244	dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1);
	245	dp->context_ptr = data;
	246	}
	247
3676347a PO	248	static void dm_bio_destructor(struct bio *bio)
3676347a PO	249	{
891ce207 HM	250	struct io *io = bio->bi_private;
891ce207 HM	251
bf17ce3a	252	bio_free(bio, io->client->bios);
3676347a PO	253	}
3676347a PO	254
c8b03afe HM	255	/*
	256	* Functions for getting the pages from kernel memory.
	257	*/
	258	static void km_get_page(struct dpages dp, struct page p, unsigned long len,
	259	unsigned *offset)
	260	{
	261	*p = virt_to_page(dp->context_ptr);
	262	*offset = dp->context_u;
	263	*len = PAGE_SIZE - dp->context_u;
	264	}
	265
	266	static void km_next_page(struct dpages *dp)
	267	{
	268	dp->context_ptr += PAGE_SIZE - dp->context_u;
	269	dp->context_u = 0;
	270	}
	271
	272	static void km_dp_init(struct dpages dp, void data)
	273	{
	274	dp->get_page = km_get_page;
	275	dp->next_page = km_next_page;
	276	dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1);
	277	dp->context_ptr = data;
	278	}
	279
1da177e4 LT	280	/*-----------------------------------------------------------------
	281	* IO routines that accept a list of pages.
	282	---------------------------------------------------------------/
22a1ceb1	283	static void do_region(int rw, unsigned region, struct dm_io_region *where,
1da177e4 LT	284	struct dpages dp, struct io io)
	285	{
	286	struct bio *bio;
	287	struct page *page;
	288	unsigned long len;
	289	unsigned offset;
	290	unsigned num_bvecs;
	291	sector_t remaining = where->count;
	292
	293	while (remaining) {
	294	/*
f00b16ad HM	295	* Allocate a suitably sized-bio: we add an extra
	296	* bvec for bio_get/set_region() and decrement bi_max_vecs
	297	* to hide it from bio_add_page().
1da177e4	298	*/
596f138e JN	299	num_bvecs = dm_sector_div_up(remaining,
	300	(PAGE_SIZE >> SECTOR_SHIFT));
	301	num_bvecs = 1 + min_t(int, bio_get_nr_vecs(where->bdev),
	302	num_bvecs);
d659e6cc MP	303	if (unlikely(num_bvecs > BIO_MAX_PAGES))
d659e6cc MP	304	num_bvecs = BIO_MAX_PAGES;
bf17ce3a	305	bio = bio_alloc_bioset(GFP_NOIO, num_bvecs, io->client->bios);
1da177e4 LT	306	bio->bi_sector = where->sector + (where->count - remaining);
	307	bio->bi_bdev = where->bdev;
	308	bio->bi_end_io = endio;
	309	bio->bi_private = io;
3676347a	310	bio->bi_destructor = dm_bio_destructor;
f00b16ad	311	bio->bi_max_vecs--;
1da177e4 LT	312	bio_set_region(bio, region);
	313
	314	/*
	315	* Try and add as many pages as possible.
	316	*/
	317	while (remaining) {
	318	dp->get_page(dp, &page, &len, &offset);
	319	len = min(len, to_bytes(remaining));
	320	if (!bio_add_page(bio, page, len, offset))
	321	break;
	322
	323	offset = 0;
	324	remaining -= to_sector(len);
	325	dp->next_page(dp);
	326	}
	327
	328	atomic_inc(&io->count);
	329	submit_bio(rw, bio);
	330	}
	331	}
	332
	333	static void dispatch_io(int rw, unsigned int num_regions,
22a1ceb1	334	struct dm_io_region where, struct dpages dp,
1da177e4 LT	335	struct io *io, int sync)
	336	{
	337	int i;
	338	struct dpages old_pages = *dp;
	339
	340	if (sync)
93dbb393	341	rw \|= (1 << BIO_RW_SYNCIO) \| (1 << BIO_RW_UNPLUG);
1da177e4 LT	342
	343	/*
	344	* For multiple regions we need to be careful to rewind
	345	* the dp object for each call to do_region.
	346	*/
	347	for (i = 0; i < num_regions; i++) {
	348	*dp = old_pages;
	349	if (where[i].count)
	350	do_region(rw, i, where + i, dp, io);
	351	}
	352
	353	/*
f00b16ad	354	* Drop the extra reference that we were holding to avoid
1da177e4 LT	355	* the io being completed too early.
	356	*/
	357	dec_count(io, 0, 0);
	358	}
	359
891ce207	360	static int sync_io(struct dm_io_client *client, unsigned int num_regions,
22a1ceb1	361	struct dm_io_region where, int rw, struct dpages dp,
891ce207	362	unsigned long *error_bits)
1da177e4 LT	363	{
	364	struct io io;
	365
7ff14a36	366	if (num_regions > 1 && (rw & RW_MASK) != WRITE) {
1da177e4 LT	367	WARN_ON(1);
	368	return -EIO;
	369	}
	370
51aa3228	371	retry:
e01fd7ee	372	io.error_bits = 0;
5af443a7	373	io.eopnotsupp_bits = 0;
1da177e4 LT	374	atomic_set(&io.count, 1); /* see dispatch_io() */
1da177e4 LT	375	io.sleeper = current;
891ce207	376	io.client = client;
1da177e4 LT	377
	378	dispatch_io(rw, num_regions, where, dp, &io, 1);
	379
	380	while (1) {
	381	set_current_state(TASK_UNINTERRUPTIBLE);
	382
b64b6bf4	383	if (!atomic_read(&io.count))
1da177e4 LT	384	break;
	385
	386	io_schedule();
	387	}
	388	set_current_state(TASK_RUNNING);
	389
51aa3228 MP	390	if (io.eopnotsupp_bits && (rw & (1 << BIO_RW_BARRIER))) {
	391	rw &= ~(1 << BIO_RW_BARRIER);
	392	goto retry;
	393	}
	394
891ce207	395	if (error_bits)
e01fd7ee	396	*error_bits = io.error_bits;
891ce207	397
e01fd7ee	398	return io.error_bits ? -EIO : 0;
1da177e4 LT	399	}
1da177e4 LT	400
891ce207	401	static int async_io(struct dm_io_client *client, unsigned int num_regions,
22a1ceb1	402	struct dm_io_region where, int rw, struct dpages dp,
891ce207	403	io_notify_fn fn, void *context)
1da177e4 LT	404	{
	405	struct io *io;
	406
7ff14a36	407	if (num_regions > 1 && (rw & RW_MASK) != WRITE) {
1da177e4 LT	408	WARN_ON(1);
	409	fn(1, context);
	410	return -EIO;
	411	}
	412
bf17ce3a	413	io = mempool_alloc(client->pool, GFP_NOIO);
e01fd7ee	414	io->error_bits = 0;
5af443a7	415	io->eopnotsupp_bits = 0;
1da177e4 LT	416	atomic_set(&io->count, 1); /* see dispatch_io() */
1da177e4 LT	417	io->sleeper = NULL;
891ce207	418	io->client = client;
1da177e4 LT	419	io->callback = fn;
	420	io->context = context;
	421
	422	dispatch_io(rw, num_regions, where, dp, io, 0);
	423	return 0;
	424	}
	425
c8b03afe HM	426	static int dp_init(struct dm_io_request io_req, struct dpages dp)
	427	{
	428	/* Set up dpages based on memory type */
	429	switch (io_req->mem.type) {
	430	case DM_IO_PAGE_LIST:
	431	list_dp_init(dp, io_req->mem.ptr.pl, io_req->mem.offset);
	432	break;
	433
	434	case DM_IO_BVEC:
	435	bvec_dp_init(dp, io_req->mem.ptr.bvec);
	436	break;
	437
	438	case DM_IO_VMA:
	439	vm_dp_init(dp, io_req->mem.ptr.vma);
	440	break;
	441
	442	case DM_IO_KMEM:
	443	km_dp_init(dp, io_req->mem.ptr.addr);
	444	break;
	445
	446	default:
	447	return -EINVAL;
	448	}
	449
	450	return 0;
	451	}
	452
	453	/*
7ff14a36 MP	454	* New collapsed (a)synchronous interface.
	455	*
	456	* If the IO is asynchronous (i.e. it has notify.fn), you must either unplug
	457	* the queue with blk_unplug() some time later or set the BIO_RW_SYNC bit in
	458	* io_req->bi_rw. If you fail to do one of these, the IO will be submitted to
	459	* the disk after q->unplug_delay, which defaults to 3ms in blk-settings.c.
c8b03afe HM	460	*/
c8b03afe HM	461	int dm_io(struct dm_io_request *io_req, unsigned num_regions,
22a1ceb1	462	struct dm_io_region where, unsigned long sync_error_bits)
c8b03afe HM	463	{
	464	int r;
	465	struct dpages dp;
	466
	467	r = dp_init(io_req, &dp);
	468	if (r)
	469	return r;
	470
	471	if (!io_req->notify.fn)
	472	return sync_io(io_req->client, num_regions, where,
	473	io_req->bi_rw, &dp, sync_error_bits);
	474
	475	return async_io(io_req->client, num_regions, where, io_req->bi_rw,
	476	&dp, io_req->notify.fn, io_req->notify.context);
	477	}
	478	EXPORT_SYMBOL(dm_io);
952b3557 MP	479
	480	int __init dm_io_init(void)
	481	{
	482	_dm_io_cache = KMEM_CACHE(io, 0);
	483	if (!_dm_io_cache)
	484	return -ENOMEM;
	485
	486	return 0;
	487	}
	488
	489	void dm_io_exit(void)
	490	{
	491	kmem_cache_destroy(_dm_io_cache);
	492	_dm_io_cache = NULL;
	493	}