[linux.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>

#define DM_MSG_PREFIX "io"

#define DM_IO_MAX_REGIONS	BITS_PER_LONG

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

/*
 * Aligning 'struct io' reduces the number of bits required to store
 * its address.  Refer to store_io_and_region_in_bio() below.
 */
struct io {
	unsigned long error_bits;
	atomic_t count;
	struct task_struct *sleeper;
	struct dm_io_client *client;
	io_notify_fn callback;
	void *context;
} __attribute__((aligned(DM_IO_MAX_REGIONS)));

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.
 * To avoid a memory allocation to store just 5 or 6 bits, we
 * ensure the 'struct io' pointer is aligned so enough low bits are
 * always zero and then combine it with the region number directly in
 * bi_private.
 *---------------------------------------------------------------*/
static void store_io_and_region_in_bio(struct bio *bio, struct io *io,
				       unsigned region)
{
	if (unlikely(!IS_ALIGNED((unsigned long)io, DM_IO_MAX_REGIONS))) {
		DMCRIT("Unaligned struct io pointer %p", io);
		BUG();
	}

	bio->bi_private = (void *)((unsigned long)io | region);
}

static void retrieve_io_and_region_from_bio(struct bio *bio, struct io **io,
				       unsigned *region)
{
	unsigned long val = (unsigned long)bio->bi_private;

	*io = (void *)(val & -(unsigned long)DM_IO_MAX_REGIONS);
	*region = val & (DM_IO_MAX_REGIONS - 1);
}

/*-----------------------------------------------------------------
 * 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.
	 */
	retrieve_io_and_region_from_bio(bio, &io, &region);

	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)
{
	unsigned region;
	struct io *io;

	retrieve_io_and_region_from_bio(bio, &io, &region);

	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;

	/*
	 * where->count may be zero if rw holds a flush and we need to
	 * send a zero-sized flush.
	 */
	do {
		/*
		 * Allocate a suitably sized-bio.
		 */
		num_bvecs = dm_sector_div_up(remaining,
					     (PAGE_SIZE >> SECTOR_SHIFT));
		num_bvecs = min_t(int, bio_get_nr_vecs(where->bdev), num_bvecs);
		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_destructor = dm_bio_destructor;
		store_io_and_region_in_bio(bio, io, 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);
	} while (remaining);
}

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;

	BUG_ON(num_regions > DM_IO_MAX_REGIONS);

	if (sync)
		rw |= REQ_SYNC;

	/*
	 * 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 || (rw & REQ_FLUSH))
			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)
{
	/*
	 * gcc <= 4.3 can't do the alignment for stack variables, so we must
	 * align it on our own.
	 * volatile prevents the optimizer from removing or reusing
	 * "io_" field from the stack frame (allowed in ANSI C).
	 */
	volatile char io_[sizeof(struct io) + __alignof__(struct io) - 1];
	struct io *io = (struct io *)PTR_ALIGN(&io_, __alignof__(struct 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 REQ_SYNC 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
f1e53987 MP	19	#define DM_MSG_PREFIX "io"
	20
	21	#define DM_IO_MAX_REGIONS BITS_PER_LONG
	22
891ce207 HM	23	struct dm_io_client {
	24	mempool_t *pool;
	25	struct bio_set *bios;
	26	};
	27
f1e53987 MP	28	/*
	29	* Aligning 'struct io' reduces the number of bits required to store
	30	* its address. Refer to store_io_and_region_in_bio() below.
	31	*/
1da177e4	32	struct io {
e01fd7ee	33	unsigned long error_bits;
1da177e4 LT	34	atomic_t count;
1da177e4 LT	35	struct task_struct *sleeper;
891ce207	36	struct dm_io_client *client;
1da177e4 LT	37	io_notify_fn callback;
1da177e4 LT	38	void *context;
f1e53987	39	} __attribute__((aligned(DM_IO_MAX_REGIONS)));
1da177e4	40
952b3557 MP	41	static struct kmem_cache *_dm_io_cache;
952b3557 MP	42
1da177e4 LT	43	/*
	44	* io contexts are only dynamically allocated for asynchronous
	45	* io. Since async io is likely to be the majority of io we'll
891ce207	46	* have the same number of io contexts as bios! (FIXME: must reduce this).
1da177e4	47	*/
891ce207	48
1da177e4 LT	49	static unsigned int pages_to_ios(unsigned int pages)
	50	{
	51	return 4 * pages; /* too many ? */
	52	}
	53
c8b03afe HM	54	/*
	55	* Create a client with mempool and bioset.
	56	*/
	57	struct dm_io_client *dm_io_client_create(unsigned num_pages)
	58	{
	59	unsigned ios = pages_to_ios(num_pages);
	60	struct dm_io_client *client;
	61
	62	client = kmalloc(sizeof(*client), GFP_KERNEL);
	63	if (!client)
	64	return ERR_PTR(-ENOMEM);
	65
952b3557	66	client->pool = mempool_create_slab_pool(ios, _dm_io_cache);
c8b03afe HM	67	if (!client->pool)
	68	goto bad;
	69
bb799ca0	70	client->bios = bioset_create(16, 0);
c8b03afe HM	71	if (!client->bios)
	72	goto bad;
	73
	74	return client;
	75
	76	bad:
	77	if (client->pool)
	78	mempool_destroy(client->pool);
	79	kfree(client);
	80	return ERR_PTR(-ENOMEM);
	81	}
	82	EXPORT_SYMBOL(dm_io_client_create);
	83
	84	int dm_io_client_resize(unsigned num_pages, struct dm_io_client *client)
	85	{
	86	return mempool_resize(client->pool, pages_to_ios(num_pages),
	87	GFP_KERNEL);
	88	}
	89	EXPORT_SYMBOL(dm_io_client_resize);
	90
	91	void dm_io_client_destroy(struct dm_io_client *client)
	92	{
	93	mempool_destroy(client->pool);
	94	bioset_free(client->bios);
	95	kfree(client);
	96	}
	97	EXPORT_SYMBOL(dm_io_client_destroy);
	98
1da177e4 LT	99	/*-----------------------------------------------------------------
1da177e4 LT	100	* We need to keep track of which region a bio is doing io for.
f1e53987 MP	101	* To avoid a memory allocation to store just 5 or 6 bits, we
	102	* ensure the 'struct io' pointer is aligned so enough low bits are
	103	* always zero and then combine it with the region number directly in
	104	* bi_private.
1da177e4	105	---------------------------------------------------------------/
f1e53987 MP	106	static void store_io_and_region_in_bio(struct bio bio, struct io io,
f1e53987 MP	107	unsigned region)
1da177e4	108	{
f1e53987 MP	109	if (unlikely(!IS_ALIGNED((unsigned long)io, DM_IO_MAX_REGIONS))) {
	110	DMCRIT("Unaligned struct io pointer %p", io);
	111	BUG();
	112	}
	113
	114	bio->bi_private = (void *)((unsigned long)io \| region);
1da177e4 LT	115	}
1da177e4 LT	116
f1e53987 MP	117	static void retrieve_io_and_region_from_bio(struct bio bio, struct io *io,
f1e53987 MP	118	unsigned *region)
1da177e4	119	{
f1e53987 MP	120	unsigned long val = (unsigned long)bio->bi_private;
	121
	122	io = (void )(val & -(unsigned long)DM_IO_MAX_REGIONS);
	123	*region = val & (DM_IO_MAX_REGIONS - 1);
1da177e4 LT	124	}
	125
	126	/*-----------------------------------------------------------------
	127	* We need an io object to keep track of the number of bios that
	128	* have been dispatched for a particular io.
	129	---------------------------------------------------------------/
	130	static void dec_count(struct io *io, unsigned int region, int error)
	131	{
d87f4c14	132	if (error)
e01fd7ee	133	set_bit(region, &io->error_bits);
1da177e4 LT	134
	135	if (atomic_dec_and_test(&io->count)) {
	136	if (io->sleeper)
	137	wake_up_process(io->sleeper);
	138
	139	else {
e01fd7ee	140	unsigned long r = io->error_bits;
1da177e4 LT	141	io_notify_fn fn = io->callback;
	142	void *context = io->context;
	143
bf17ce3a	144	mempool_free(io, io->client->pool);
1da177e4 LT	145	fn(r, context);
	146	}
	147	}
	148	}
	149
6712ecf8	150	static void endio(struct bio *bio, int error)
1da177e4	151	{
c897feb3 HM	152	struct io *io;
c897feb3 HM	153	unsigned region;
1da177e4	154
1da177e4 LT	155	if (error && bio_data_dir(bio) == READ)
	156	zero_fill_bio(bio);
	157
c897feb3 HM	158	/*
	159	* The bio destructor in bio_put() may use the io object.
	160	*/
f1e53987	161	retrieve_io_and_region_from_bio(bio, &io, &region);
c897feb3	162
1da177e4 LT	163	bio_put(bio);
1da177e4 LT	164
c897feb3	165	dec_count(io, region, error);
1da177e4 LT	166	}
	167
	168	/*-----------------------------------------------------------------
	169	* These little objects provide an abstraction for getting a new
	170	* destination page for io.
	171	---------------------------------------------------------------/
	172	struct dpages {
	173	void (get_page)(struct dpages dp,
	174	struct page *p, unsigned long len, unsigned *offset);
	175	void (next_page)(struct dpages dp);
	176
	177	unsigned context_u;
	178	void *context_ptr;
	179	};
	180
	181	/*
	182	* Functions for getting the pages from a list.
	183	*/
	184	static void list_get_page(struct dpages *dp,
	185	struct page *p, unsigned long len, unsigned *offset)
	186	{
	187	unsigned o = dp->context_u;
	188	struct page_list pl = (struct page_list ) dp->context_ptr;
	189
	190	*p = pl->page;
	191	*len = PAGE_SIZE - o;
	192	*offset = o;
	193	}
	194
	195	static void list_next_page(struct dpages *dp)
	196	{
	197	struct page_list pl = (struct page_list ) dp->context_ptr;
	198	dp->context_ptr = pl->next;
	199	dp->context_u = 0;
	200	}
	201
	202	static void list_dp_init(struct dpages dp, struct page_list pl, unsigned offset)
	203	{
	204	dp->get_page = list_get_page;
	205	dp->next_page = list_next_page;
	206	dp->context_u = offset;
	207	dp->context_ptr = pl;
	208	}
	209
	210	/*
	211	* Functions for getting the pages from a bvec.
	212	*/
	213	static void bvec_get_page(struct dpages *dp,
	214	struct page *p, unsigned long len, unsigned *offset)
	215	{
	216	struct bio_vec bvec = (struct bio_vec ) dp->context_ptr;
	217	*p = bvec->bv_page;
	218	*len = bvec->bv_len;
	219	*offset = bvec->bv_offset;
	220	}
	221
	222	static void bvec_next_page(struct dpages *dp)
	223	{
	224	struct bio_vec bvec = (struct bio_vec ) dp->context_ptr;
	225	dp->context_ptr = bvec + 1;
	226	}
	227
	228	static void bvec_dp_init(struct dpages dp, struct bio_vec bvec)
	229	{
230	dp->get_page = bvec_get_page;
231	dp->next_page = bvec_next_page;
232	dp->context_ptr = bvec;
233	}
234
c8b03afe HM	235	/*
	236	* Functions for getting the pages from a VMA.
	237	*/
1da177e4 LT	238	static void vm_get_page(struct dpages *dp,
	239	struct page *p, unsigned long len, unsigned *offset)
	240	{
	241	*p = vmalloc_to_page(dp->context_ptr);
	242	*offset = dp->context_u;
	243	*len = PAGE_SIZE - dp->context_u;
	244	}
	245
	246	static void vm_next_page(struct dpages *dp)
	247	{
	248	dp->context_ptr += PAGE_SIZE - dp->context_u;
	249	dp->context_u = 0;
	250	}
	251
	252	static void vm_dp_init(struct dpages dp, void data)
	253	{
	254	dp->get_page = vm_get_page;
	255	dp->next_page = vm_next_page;
	256	dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1);
	257	dp->context_ptr = data;
	258	}
	259
3676347a PO	260	static void dm_bio_destructor(struct bio *bio)
3676347a PO	261	{
f1e53987 MP	262	unsigned region;
	263	struct io *io;
	264
	265	retrieve_io_and_region_from_bio(bio, &io, &region);
891ce207	266
bf17ce3a	267	bio_free(bio, io->client->bios);
3676347a PO	268	}
3676347a PO	269
c8b03afe HM	270	/*
	271	* Functions for getting the pages from kernel memory.
	272	*/
	273	static void km_get_page(struct dpages dp, struct page p, unsigned long len,
	274	unsigned *offset)
	275	{
	276	*p = virt_to_page(dp->context_ptr);
	277	*offset = dp->context_u;
	278	*len = PAGE_SIZE - dp->context_u;
	279	}
	280
	281	static void km_next_page(struct dpages *dp)
	282	{
	283	dp->context_ptr += PAGE_SIZE - dp->context_u;
	284	dp->context_u = 0;
	285	}
	286
	287	static void km_dp_init(struct dpages dp, void data)
	288	{
	289	dp->get_page = km_get_page;
	290	dp->next_page = km_next_page;
	291	dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1);
	292	dp->context_ptr = data;
	293	}
	294
1da177e4 LT	295	/*-----------------------------------------------------------------
	296	* IO routines that accept a list of pages.
	297	---------------------------------------------------------------/
22a1ceb1	298	static void do_region(int rw, unsigned region, struct dm_io_region *where,
1da177e4 LT	299	struct dpages dp, struct io io)
	300	{
	301	struct bio *bio;
	302	struct page *page;
	303	unsigned long len;
	304	unsigned offset;
	305	unsigned num_bvecs;
	306	sector_t remaining = where->count;
	307
12fc0f49	308	/*
d87f4c14 TH	309	* where->count may be zero if rw holds a flush and we need to
d87f4c14 TH	310	* send a zero-sized flush.
12fc0f49 MP	311	*/
12fc0f49 MP	312	do {
1da177e4	313	/*
f1e53987	314	* Allocate a suitably sized-bio.
1da177e4	315	*/
596f138e JN	316	num_bvecs = dm_sector_div_up(remaining,
596f138e JN	317	(PAGE_SIZE >> SECTOR_SHIFT));
f1e53987	318	num_bvecs = min_t(int, bio_get_nr_vecs(where->bdev), num_bvecs);
bf17ce3a	319	bio = bio_alloc_bioset(GFP_NOIO, num_bvecs, io->client->bios);
1da177e4 LT	320	bio->bi_sector = where->sector + (where->count - remaining);
	321	bio->bi_bdev = where->bdev;
	322	bio->bi_end_io = endio;
3676347a	323	bio->bi_destructor = dm_bio_destructor;
f1e53987	324	store_io_and_region_in_bio(bio, io, region);
1da177e4 LT	325
	326	/*
	327	* Try and add as many pages as possible.
	328	*/
	329	while (remaining) {
	330	dp->get_page(dp, &page, &len, &offset);
	331	len = min(len, to_bytes(remaining));
	332	if (!bio_add_page(bio, page, len, offset))
	333	break;
	334
	335	offset = 0;
	336	remaining -= to_sector(len);
	337	dp->next_page(dp);
	338	}
	339
	340	atomic_inc(&io->count);
	341	submit_bio(rw, bio);
12fc0f49	342	} while (remaining);
1da177e4 LT	343	}
	344
	345	static void dispatch_io(int rw, unsigned int num_regions,
22a1ceb1	346	struct dm_io_region where, struct dpages dp,
1da177e4 LT	347	struct io *io, int sync)
	348	{
	349	int i;
	350	struct dpages old_pages = *dp;
	351
f1e53987 MP	352	BUG_ON(num_regions > DM_IO_MAX_REGIONS);
f1e53987 MP	353
1da177e4	354	if (sync)
721a9602	355	rw \|= REQ_SYNC;
1da177e4 LT	356
	357	/*
	358	* For multiple regions we need to be careful to rewind
	359	* the dp object for each call to do_region.
	360	*/
	361	for (i = 0; i < num_regions; i++) {
	362	*dp = old_pages;
d87f4c14	363	if (where[i].count \|\| (rw & REQ_FLUSH))
1da177e4 LT	364	do_region(rw, i, where + i, dp, io);
	365	}
	366
	367	/*
f00b16ad	368	* Drop the extra reference that we were holding to avoid
1da177e4 LT	369	* the io being completed too early.
	370	*/
	371	dec_count(io, 0, 0);
	372	}
	373
891ce207	374	static int sync_io(struct dm_io_client *client, unsigned int num_regions,
22a1ceb1	375	struct dm_io_region where, int rw, struct dpages dp,
891ce207	376	unsigned long *error_bits)
1da177e4	377	{
f1e53987 MP	378	/*
	379	* gcc <= 4.3 can't do the alignment for stack variables, so we must
	380	* align it on our own.
	381	* volatile prevents the optimizer from removing or reusing
	382	* "io_" field from the stack frame (allowed in ANSI C).
	383	*/
	384	volatile char io_[sizeof(struct io) + __alignof__(struct io) - 1];
	385	struct io io = (struct io )PTR_ALIGN(&io_, __alignof__(struct io));
1da177e4	386
7ff14a36	387	if (num_regions > 1 && (rw & RW_MASK) != WRITE) {
1da177e4 LT	388	WARN_ON(1);
	389	return -EIO;
	390	}
	391
f1e53987	392	io->error_bits = 0;
f1e53987 MP	393	atomic_set(&io->count, 1); /* see dispatch_io() */
	394	io->sleeper = current;
	395	io->client = client;
1da177e4	396
f1e53987	397	dispatch_io(rw, num_regions, where, dp, io, 1);
1da177e4 LT	398
	399	while (1) {
	400	set_current_state(TASK_UNINTERRUPTIBLE);
	401
f1e53987	402	if (!atomic_read(&io->count))
1da177e4 LT	403	break;
	404
	405	io_schedule();
	406	}
	407	set_current_state(TASK_RUNNING);
	408
891ce207	409	if (error_bits)
f1e53987	410	*error_bits = io->error_bits;
891ce207	411
f1e53987	412	return io->error_bits ? -EIO : 0;
1da177e4 LT	413	}
1da177e4 LT	414
891ce207	415	static int async_io(struct dm_io_client *client, unsigned int num_regions,
22a1ceb1	416	struct dm_io_region where, int rw, struct dpages dp,
891ce207	417	io_notify_fn fn, void *context)
1da177e4 LT	418	{
	419	struct io *io;
	420
7ff14a36	421	if (num_regions > 1 && (rw & RW_MASK) != WRITE) {
1da177e4 LT	422	WARN_ON(1);
	423	fn(1, context);
	424	return -EIO;
	425	}
	426
bf17ce3a	427	io = mempool_alloc(client->pool, GFP_NOIO);
e01fd7ee	428	io->error_bits = 0;
1da177e4 LT	429	atomic_set(&io->count, 1); /* see dispatch_io() */
1da177e4 LT	430	io->sleeper = NULL;
891ce207	431	io->client = client;
1da177e4 LT	432	io->callback = fn;
	433	io->context = context;
	434
	435	dispatch_io(rw, num_regions, where, dp, io, 0);
	436	return 0;
	437	}
	438
c8b03afe HM	439	static int dp_init(struct dm_io_request io_req, struct dpages dp)
	440	{
	441	/* Set up dpages based on memory type */
	442	switch (io_req->mem.type) {
	443	case DM_IO_PAGE_LIST:
	444	list_dp_init(dp, io_req->mem.ptr.pl, io_req->mem.offset);
	445	break;
	446
	447	case DM_IO_BVEC:
	448	bvec_dp_init(dp, io_req->mem.ptr.bvec);
	449	break;
	450
	451	case DM_IO_VMA:
	452	vm_dp_init(dp, io_req->mem.ptr.vma);
	453	break;
	454
	455	case DM_IO_KMEM:
	456	km_dp_init(dp, io_req->mem.ptr.addr);
	457	break;
	458
	459	default:
	460	return -EINVAL;
	461	}
	462
	463	return 0;
	464	}
	465
	466	/*
7ff14a36 MP	467	* New collapsed (a)synchronous interface.
	468	*
	469	* If the IO is asynchronous (i.e. it has notify.fn), you must either unplug
7b6d91da CH	470	* the queue with blk_unplug() some time later or set REQ_SYNC in
7b6d91da CH	471	io_req->bi_rw. If you fail to do one of these, the IO will be submitted to
7ff14a36	472	* the disk after q->unplug_delay, which defaults to 3ms in blk-settings.c.
c8b03afe HM	473	*/
c8b03afe HM	474	int dm_io(struct dm_io_request *io_req, unsigned num_regions,
22a1ceb1	475	struct dm_io_region where, unsigned long sync_error_bits)
c8b03afe HM	476	{
	477	int r;
	478	struct dpages dp;
	479
	480	r = dp_init(io_req, &dp);
	481	if (r)
	482	return r;
	483
	484	if (!io_req->notify.fn)
	485	return sync_io(io_req->client, num_regions, where,
	486	io_req->bi_rw, &dp, sync_error_bits);
	487
	488	return async_io(io_req->client, num_regions, where, io_req->bi_rw,
	489	&dp, io_req->notify.fn, io_req->notify.context);
	490	}
	491	EXPORT_SYMBOL(dm_io);
952b3557 MP	492
	493	int __init dm_io_init(void)
	494	{
	495	_dm_io_cache = KMEM_CACHE(io, 0);
	496	if (!_dm_io_cache)
	497	return -ENOMEM;
	498
	499	return 0;
	500	}
	501
	502	void dm_io_exit(void)
	503	{
	504	kmem_cache_destroy(_dm_io_cache);
	505	_dm_io_cache = NULL;
	506	}