[linux-2.6-block.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
#define MIN_IOS		16
#define MIN_BIOS	16

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;
	void *vma_invalidate_address;
	unsigned long vma_invalidate_size;
} __attribute__((aligned(DM_IO_MAX_REGIONS)));

static struct kmem_cache *_dm_io_cache;

/*
 * Create a client with mempool and bioset.
 */
struct dm_io_client *dm_io_client_create(void)
{
	struct dm_io_client *client;

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

	client->pool = mempool_create_slab_pool(MIN_IOS, _dm_io_cache);
	if (!client->pool)
		goto bad;

	client->bios = bioset_create(MIN_BIOS, 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);

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->vma_invalidate_size)
			invalidate_kernel_vmap_range(io->vma_invalidate_address,
						     io->vma_invalidate_size);

		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;

	void *vma_invalidate_address;
	unsigned long vma_invalidate_size;
};

/*
 * 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;

	io->vma_invalidate_address = dp->vma_invalidate_address;
	io->vma_invalidate_size = dp->vma_invalidate_size;

	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;

	io->vma_invalidate_address = dp->vma_invalidate_address;
	io->vma_invalidate_size = dp->vma_invalidate_size;

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

static int dp_init(struct dm_io_request *io_req, struct dpages *dp,
		   unsigned long size)
{
	/* Set up dpages based on memory type */

	dp->vma_invalidate_address = NULL;
	dp->vma_invalidate_size = 0;

	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:
		flush_kernel_vmap_range(io_req->mem.ptr.vma, size);
		if ((io_req->bi_rw & RW_MASK) == READ) {
			dp->vma_invalidate_address = io_req->mem.ptr.vma;
			dp->vma_invalidate_size = size;
		}
		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, (unsigned long)where->count << SECTOR_SHIFT);
	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
bda8efec MP	22	#define MIN_IOS 16
bda8efec MP	23	#define MIN_BIOS 16
f1e53987	24
891ce207 HM	25	struct dm_io_client {
	26	mempool_t *pool;
	27	struct bio_set *bios;
	28	};
	29
f1e53987 MP	30	/*
	31	* Aligning 'struct io' reduces the number of bits required to store
	32	* its address. Refer to store_io_and_region_in_bio() below.
	33	*/
1da177e4	34	struct io {
e01fd7ee	35	unsigned long error_bits;
1da177e4 LT	36	atomic_t count;
1da177e4 LT	37	struct task_struct *sleeper;
891ce207	38	struct dm_io_client *client;
1da177e4 LT	39	io_notify_fn callback;
1da177e4 LT	40	void *context;
bb91bc7b MP	41	void *vma_invalidate_address;
bb91bc7b MP	42	unsigned long vma_invalidate_size;
f1e53987	43	} __attribute__((aligned(DM_IO_MAX_REGIONS)));
1da177e4	44
952b3557 MP	45	static struct kmem_cache *_dm_io_cache;
952b3557 MP	46
c8b03afe HM	47	/*
	48	* Create a client with mempool and bioset.
	49	*/
bda8efec	50	struct dm_io_client *dm_io_client_create(void)
c8b03afe	51	{
c8b03afe HM	52	struct dm_io_client *client;
	53
	54	client = kmalloc(sizeof(*client), GFP_KERNEL);
	55	if (!client)
	56	return ERR_PTR(-ENOMEM);
	57
bda8efec	58	client->pool = mempool_create_slab_pool(MIN_IOS, _dm_io_cache);
c8b03afe HM	59	if (!client->pool)
	60	goto bad;
	61
bda8efec	62	client->bios = bioset_create(MIN_BIOS, 0);
c8b03afe HM	63	if (!client->bios)
	64	goto bad;
	65
	66	return client;
	67
	68	bad:
	69	if (client->pool)
	70	mempool_destroy(client->pool);
	71	kfree(client);
	72	return ERR_PTR(-ENOMEM);
	73	}
	74	EXPORT_SYMBOL(dm_io_client_create);
	75
c8b03afe HM	76	void dm_io_client_destroy(struct dm_io_client *client)
	77	{
	78	mempool_destroy(client->pool);
	79	bioset_free(client->bios);
	80	kfree(client);
	81	}
	82	EXPORT_SYMBOL(dm_io_client_destroy);
	83
1da177e4 LT	84	/*-----------------------------------------------------------------
1da177e4 LT	85	* We need to keep track of which region a bio is doing io for.
f1e53987 MP	86	* To avoid a memory allocation to store just 5 or 6 bits, we
	87	* ensure the 'struct io' pointer is aligned so enough low bits are
	88	* always zero and then combine it with the region number directly in
	89	* bi_private.
1da177e4	90	---------------------------------------------------------------/
f1e53987 MP	91	static void store_io_and_region_in_bio(struct bio bio, struct io io,
f1e53987 MP	92	unsigned region)
1da177e4	93	{
f1e53987 MP	94	if (unlikely(!IS_ALIGNED((unsigned long)io, DM_IO_MAX_REGIONS))) {
	95	DMCRIT("Unaligned struct io pointer %p", io);
	96	BUG();
	97	}
	98
	99	bio->bi_private = (void *)((unsigned long)io \| region);
1da177e4 LT	100	}
1da177e4 LT	101
f1e53987 MP	102	static void retrieve_io_and_region_from_bio(struct bio bio, struct io *io,
f1e53987 MP	103	unsigned *region)
1da177e4	104	{
f1e53987 MP	105	unsigned long val = (unsigned long)bio->bi_private;
	106
	107	io = (void )(val & -(unsigned long)DM_IO_MAX_REGIONS);
	108	*region = val & (DM_IO_MAX_REGIONS - 1);
1da177e4 LT	109	}
	110
	111	/*-----------------------------------------------------------------
	112	* We need an io object to keep track of the number of bios that
	113	* have been dispatched for a particular io.
	114	---------------------------------------------------------------/
	115	static void dec_count(struct io *io, unsigned int region, int error)
	116	{
d87f4c14	117	if (error)
e01fd7ee	118	set_bit(region, &io->error_bits);
1da177e4 LT	119
1da177e4 LT	120	if (atomic_dec_and_test(&io->count)) {
bb91bc7b MP	121	if (io->vma_invalidate_size)
	122	invalidate_kernel_vmap_range(io->vma_invalidate_address,
	123	io->vma_invalidate_size);
	124
1da177e4 LT	125	if (io->sleeper)
	126	wake_up_process(io->sleeper);
	127
	128	else {
e01fd7ee	129	unsigned long r = io->error_bits;
1da177e4 LT	130	io_notify_fn fn = io->callback;
	131	void *context = io->context;
	132
bf17ce3a	133	mempool_free(io, io->client->pool);
1da177e4 LT	134	fn(r, context);
	135	}
	136	}
	137	}
	138
6712ecf8	139	static void endio(struct bio *bio, int error)
1da177e4	140	{
c897feb3 HM	141	struct io *io;
c897feb3 HM	142	unsigned region;
1da177e4	143
1da177e4 LT	144	if (error && bio_data_dir(bio) == READ)
	145	zero_fill_bio(bio);
	146
c897feb3 HM	147	/*
	148	* The bio destructor in bio_put() may use the io object.
	149	*/
f1e53987	150	retrieve_io_and_region_from_bio(bio, &io, &region);
c897feb3	151
1da177e4 LT	152	bio_put(bio);
1da177e4 LT	153
c897feb3	154	dec_count(io, region, error);
1da177e4 LT	155	}
	156
	157	/*-----------------------------------------------------------------
	158	* These little objects provide an abstraction for getting a new
	159	* destination page for io.
	160	---------------------------------------------------------------/
	161	struct dpages {
	162	void (get_page)(struct dpages dp,
	163	struct page *p, unsigned long len, unsigned *offset);
	164	void (next_page)(struct dpages dp);
	165
	166	unsigned context_u;
	167	void *context_ptr;
bb91bc7b MP	168
	169	void *vma_invalidate_address;
	170	unsigned long vma_invalidate_size;
1da177e4 LT	171	};
	172
	173	/*
	174	* Functions for getting the pages from a list.
	175	*/
	176	static void list_get_page(struct dpages *dp,
	177	struct page *p, unsigned long len, unsigned *offset)
	178	{
	179	unsigned o = dp->context_u;
	180	struct page_list pl = (struct page_list ) dp->context_ptr;
	181
	182	*p = pl->page;
	183	*len = PAGE_SIZE - o;
	184	*offset = o;
	185	}
	186
	187	static void list_next_page(struct dpages *dp)
	188	{
	189	struct page_list pl = (struct page_list ) dp->context_ptr;
	190	dp->context_ptr = pl->next;
	191	dp->context_u = 0;
	192	}
	193
	194	static void list_dp_init(struct dpages dp, struct page_list pl, unsigned offset)
	195	{
	196	dp->get_page = list_get_page;
	197	dp->next_page = list_next_page;
	198	dp->context_u = offset;
	199	dp->context_ptr = pl;
	200	}
	201
	202	/*
	203	* Functions for getting the pages from a bvec.
	204	*/
	205	static void bvec_get_page(struct dpages *dp,
	206	struct page *p, unsigned long len, unsigned *offset)
	207	{
	208	struct bio_vec bvec = (struct bio_vec ) dp->context_ptr;
	209	*p = bvec->bv_page;
	210	*len = bvec->bv_len;
	211	*offset = bvec->bv_offset;
	212	}
	213
	214	static void bvec_next_page(struct dpages *dp)
	215	{
	216	struct bio_vec bvec = (struct bio_vec ) dp->context_ptr;
	217	dp->context_ptr = bvec + 1;
	218	}
	219
	220	static void bvec_dp_init(struct dpages dp, struct bio_vec bvec)
	221	{
	222	dp->get_page = bvec_get_page;
	223	dp->next_page = bvec_next_page;
	224	dp->context_ptr = bvec;
	225	}
	226
c8b03afe HM	227	/*
	228	* Functions for getting the pages from a VMA.
	229	*/
1da177e4 LT	230	static void vm_get_page(struct dpages *dp,
	231	struct page *p, unsigned long len, unsigned *offset)
	232	{
	233	*p = vmalloc_to_page(dp->context_ptr);
	234	*offset = dp->context_u;
	235	*len = PAGE_SIZE - dp->context_u;
	236	}
	237
	238	static void vm_next_page(struct dpages *dp)
	239	{
	240	dp->context_ptr += PAGE_SIZE - dp->context_u;
	241	dp->context_u = 0;
	242	}
	243
	244	static void vm_dp_init(struct dpages dp, void data)
	245	{
	246	dp->get_page = vm_get_page;
	247	dp->next_page = vm_next_page;
	248	dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1);
	249	dp->context_ptr = data;
	250	}
	251
3676347a PO	252	static void dm_bio_destructor(struct bio *bio)
3676347a PO	253	{
f1e53987 MP	254	unsigned region;
	255	struct io *io;
	256
	257	retrieve_io_and_region_from_bio(bio, &io, &region);
891ce207	258
bf17ce3a	259	bio_free(bio, io->client->bios);
3676347a PO	260	}
3676347a PO	261
c8b03afe HM	262	/*
	263	* Functions for getting the pages from kernel memory.
	264	*/
	265	static void km_get_page(struct dpages dp, struct page p, unsigned long len,
	266	unsigned *offset)
	267	{
	268	*p = virt_to_page(dp->context_ptr);
	269	*offset = dp->context_u;
	270	*len = PAGE_SIZE - dp->context_u;
	271	}
	272
	273	static void km_next_page(struct dpages *dp)
	274	{
	275	dp->context_ptr += PAGE_SIZE - dp->context_u;
	276	dp->context_u = 0;
	277	}
	278
	279	static void km_dp_init(struct dpages dp, void data)
	280	{
	281	dp->get_page = km_get_page;
	282	dp->next_page = km_next_page;
	283	dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1);
	284	dp->context_ptr = data;
	285	}
	286
1da177e4 LT	287	/*-----------------------------------------------------------------
	288	* IO routines that accept a list of pages.
	289	---------------------------------------------------------------/
22a1ceb1	290	static void do_region(int rw, unsigned region, struct dm_io_region *where,
1da177e4 LT	291	struct dpages dp, struct io io)
	292	{
	293	struct bio *bio;
	294	struct page *page;
	295	unsigned long len;
	296	unsigned offset;
	297	unsigned num_bvecs;
	298	sector_t remaining = where->count;
	299
12fc0f49	300	/*
d87f4c14 TH	301	* where->count may be zero if rw holds a flush and we need to
d87f4c14 TH	302	* send a zero-sized flush.
12fc0f49 MP	303	*/
12fc0f49 MP	304	do {
1da177e4	305	/*
f1e53987	306	* Allocate a suitably sized-bio.
1da177e4	307	*/
596f138e JN	308	num_bvecs = dm_sector_div_up(remaining,
596f138e JN	309	(PAGE_SIZE >> SECTOR_SHIFT));
f1e53987	310	num_bvecs = min_t(int, bio_get_nr_vecs(where->bdev), num_bvecs);
bf17ce3a	311	bio = bio_alloc_bioset(GFP_NOIO, num_bvecs, io->client->bios);
1da177e4 LT	312	bio->bi_sector = where->sector + (where->count - remaining);
	313	bio->bi_bdev = where->bdev;
	314	bio->bi_end_io = endio;
3676347a	315	bio->bi_destructor = dm_bio_destructor;
f1e53987	316	store_io_and_region_in_bio(bio, io, region);
1da177e4 LT	317
	318	/*
	319	* Try and add as many pages as possible.
	320	*/
	321	while (remaining) {
	322	dp->get_page(dp, &page, &len, &offset);
	323	len = min(len, to_bytes(remaining));
	324	if (!bio_add_page(bio, page, len, offset))
	325	break;
	326
	327	offset = 0;
	328	remaining -= to_sector(len);
	329	dp->next_page(dp);
	330	}
	331
	332	atomic_inc(&io->count);
	333	submit_bio(rw, bio);
12fc0f49	334	} while (remaining);
1da177e4 LT	335	}
	336
	337	static void dispatch_io(int rw, unsigned int num_regions,
22a1ceb1	338	struct dm_io_region where, struct dpages dp,
1da177e4 LT	339	struct io *io, int sync)
	340	{
	341	int i;
	342	struct dpages old_pages = *dp;
	343
f1e53987 MP	344	BUG_ON(num_regions > DM_IO_MAX_REGIONS);
f1e53987 MP	345
1da177e4	346	if (sync)
721a9602	347	rw \|= REQ_SYNC;
1da177e4 LT	348
	349	/*
	350	* For multiple regions we need to be careful to rewind
	351	* the dp object for each call to do_region.
	352	*/
	353	for (i = 0; i < num_regions; i++) {
	354	*dp = old_pages;
d87f4c14	355	if (where[i].count \|\| (rw & REQ_FLUSH))
1da177e4 LT	356	do_region(rw, i, where + i, dp, io);
	357	}
	358
	359	/*
f00b16ad	360	* Drop the extra reference that we were holding to avoid
1da177e4 LT	361	* the io being completed too early.
	362	*/
	363	dec_count(io, 0, 0);
	364	}
	365
891ce207	366	static int sync_io(struct dm_io_client *client, unsigned int num_regions,
22a1ceb1	367	struct dm_io_region where, int rw, struct dpages dp,
891ce207	368	unsigned long *error_bits)
1da177e4	369	{
f1e53987 MP	370	/*
	371	* gcc <= 4.3 can't do the alignment for stack variables, so we must
	372	* align it on our own.
	373	* volatile prevents the optimizer from removing or reusing
	374	* "io_" field from the stack frame (allowed in ANSI C).
	375	*/
	376	volatile char io_[sizeof(struct io) + __alignof__(struct io) - 1];
	377	struct io io = (struct io )PTR_ALIGN(&io_, __alignof__(struct io));
1da177e4	378
7ff14a36	379	if (num_regions > 1 && (rw & RW_MASK) != WRITE) {
1da177e4 LT	380	WARN_ON(1);
	381	return -EIO;
	382	}
	383
f1e53987	384	io->error_bits = 0;
f1e53987 MP	385	atomic_set(&io->count, 1); /* see dispatch_io() */
	386	io->sleeper = current;
	387	io->client = client;
1da177e4	388
bb91bc7b MP	389	io->vma_invalidate_address = dp->vma_invalidate_address;
	390	io->vma_invalidate_size = dp->vma_invalidate_size;
	391
f1e53987	392	dispatch_io(rw, num_regions, where, dp, io, 1);
1da177e4 LT	393
	394	while (1) {
	395	set_current_state(TASK_UNINTERRUPTIBLE);
	396
f1e53987	397	if (!atomic_read(&io->count))
1da177e4 LT	398	break;
	399
	400	io_schedule();
	401	}
	402	set_current_state(TASK_RUNNING);
	403
891ce207	404	if (error_bits)
f1e53987	405	*error_bits = io->error_bits;
891ce207	406
f1e53987	407	return io->error_bits ? -EIO : 0;
1da177e4 LT	408	}
1da177e4 LT	409
891ce207	410	static int async_io(struct dm_io_client *client, unsigned int num_regions,
22a1ceb1	411	struct dm_io_region where, int rw, struct dpages dp,
891ce207	412	io_notify_fn fn, void *context)
1da177e4 LT	413	{
	414	struct io *io;
	415
7ff14a36	416	if (num_regions > 1 && (rw & RW_MASK) != WRITE) {
1da177e4 LT	417	WARN_ON(1);
	418	fn(1, context);
	419	return -EIO;
	420	}
	421
bf17ce3a	422	io = mempool_alloc(client->pool, GFP_NOIO);
e01fd7ee	423	io->error_bits = 0;
1da177e4 LT	424	atomic_set(&io->count, 1); /* see dispatch_io() */
1da177e4 LT	425	io->sleeper = NULL;
891ce207	426	io->client = client;
1da177e4 LT	427	io->callback = fn;
	428	io->context = context;
	429
bb91bc7b MP	430	io->vma_invalidate_address = dp->vma_invalidate_address;
	431	io->vma_invalidate_size = dp->vma_invalidate_size;
	432
1da177e4 LT	433	dispatch_io(rw, num_regions, where, dp, io, 0);
	434	return 0;
	435	}
	436
bb91bc7b MP	437	static int dp_init(struct dm_io_request io_req, struct dpages dp,
bb91bc7b MP	438	unsigned long size)
c8b03afe HM	439	{
c8b03afe HM	440	/* Set up dpages based on memory type */
bb91bc7b MP	441
	442	dp->vma_invalidate_address = NULL;
	443	dp->vma_invalidate_size = 0;
	444
c8b03afe HM	445	switch (io_req->mem.type) {
	446	case DM_IO_PAGE_LIST:
	447	list_dp_init(dp, io_req->mem.ptr.pl, io_req->mem.offset);
	448	break;
	449
	450	case DM_IO_BVEC:
	451	bvec_dp_init(dp, io_req->mem.ptr.bvec);
	452	break;
	453
	454	case DM_IO_VMA:
bb91bc7b MP	455	flush_kernel_vmap_range(io_req->mem.ptr.vma, size);
	456	if ((io_req->bi_rw & RW_MASK) == READ) {
	457	dp->vma_invalidate_address = io_req->mem.ptr.vma;
	458	dp->vma_invalidate_size = size;
	459	}
c8b03afe HM	460	vm_dp_init(dp, io_req->mem.ptr.vma);
	461	break;
	462
	463	case DM_IO_KMEM:
	464	km_dp_init(dp, io_req->mem.ptr.addr);
	465	break;
	466
	467	default:
	468	return -EINVAL;
	469	}
	470
	471	return 0;
	472	}
	473
	474	/*
7ff14a36 MP	475	* New collapsed (a)synchronous interface.
	476	*
	477	* If the IO is asynchronous (i.e. it has notify.fn), you must either unplug
7b6d91da CH	478	* the queue with blk_unplug() some time later or set REQ_SYNC in
7b6d91da CH	479	io_req->bi_rw. If you fail to do one of these, the IO will be submitted to
7ff14a36	480	* the disk after q->unplug_delay, which defaults to 3ms in blk-settings.c.
c8b03afe HM	481	*/
c8b03afe HM	482	int dm_io(struct dm_io_request *io_req, unsigned num_regions,
22a1ceb1	483	struct dm_io_region where, unsigned long sync_error_bits)
c8b03afe HM	484	{
	485	int r;
	486	struct dpages dp;
	487
bb91bc7b	488	r = dp_init(io_req, &dp, (unsigned long)where->count << SECTOR_SHIFT);
c8b03afe HM	489	if (r)
	490	return r;
	491
	492	if (!io_req->notify.fn)
	493	return sync_io(io_req->client, num_regions, where,
	494	io_req->bi_rw, &dp, sync_error_bits);
	495
	496	return async_io(io_req->client, num_regions, where, io_req->bi_rw,
	497	&dp, io_req->notify.fn, io_req->notify.context);
	498	}
	499	EXPORT_SYMBOL(dm_io);
952b3557 MP	500
	501	int __init dm_io_init(void)
	502	{
	503	_dm_io_cache = KMEM_CACHE(io, 0);
	504	if (!_dm_io_cache)
	505	return -ENOMEM;
	506
	507	return 0;
	508	}
	509
	510	void dm_io_exit(void)
	511	{
	512	kmem_cache_destroy(_dm_io_cache);
	513	_dm_io_cache = NULL;
	514	}