[deliverable/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/completion.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 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;
	unsigned min_ios = dm_get_reserved_bio_based_ios();

	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_ios, 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 complete_io(struct io *io)
{
	unsigned long error_bits = io->error_bits;
	io_notify_fn fn = io->callback;
	void *context = io->context;

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

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

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))
		complete_io(io);
}

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 bio_get_page(struct dpages *dp, struct page **p,
			 unsigned long *len, unsigned *offset)
{
	struct bio_vec *bvec = dp->context_ptr;
	*p = bvec->bv_page;
	*len = bvec->bv_len - dp->context_u;
	*offset = bvec->bv_offset + dp->context_u;
}

static void bio_next_page(struct dpages *dp)
{
	struct bio_vec *bvec = dp->context_ptr;
	dp->context_ptr = bvec + 1;
	dp->context_u = 0;
}

static void bio_dp_init(struct dpages *dp, struct bio *bio)
{
	dp->get_page = bio_get_page;
	dp->next_page = bio_next_page;
	dp->context_ptr = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
	dp->context_u = bio->bi_iter.bi_bvec_done;
}

/*
 * 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;
}

/*
 * 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;
	struct request_queue *q = bdev_get_queue(where->bdev);
	unsigned short logical_block_size = queue_logical_block_size(q);
	sector_t num_sectors;

	/* Reject unsupported discard requests */
	if ((rw & REQ_DISCARD) && !blk_queue_discard(q)) {
		dec_count(io, region, -EOPNOTSUPP);
		return;
	}

	/*
	 * 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.
		 */
		if ((rw & REQ_DISCARD) || (rw & REQ_WRITE_SAME))
			num_bvecs = 1;
		else
			num_bvecs = min_t(int, bio_get_nr_vecs(where->bdev),
					  dm_sector_div_up(remaining, (PAGE_SIZE >> SECTOR_SHIFT)));

		bio = bio_alloc_bioset(GFP_NOIO, num_bvecs, io->client->bios);
		bio->bi_iter.bi_sector = where->sector + (where->count - remaining);
		bio->bi_bdev = where->bdev;
		bio->bi_end_io = endio;
		store_io_and_region_in_bio(bio, io, region);

		if (rw & REQ_DISCARD) {
			num_sectors = min_t(sector_t, q->limits.max_discard_sectors, remaining);
			bio->bi_iter.bi_size = num_sectors << SECTOR_SHIFT;
			remaining -= num_sectors;
		} else if (rw & REQ_WRITE_SAME) {
			/*
			 * WRITE SAME only uses a single page.
			 */
			dp->get_page(dp, &page, &len, &offset);
			bio_add_page(bio, page, logical_block_size, offset);
			num_sectors = min_t(sector_t, q->limits.max_write_same_sectors, remaining);
			bio->bi_iter.bi_size = num_sectors << SECTOR_SHIFT;

			offset = 0;
			remaining -= num_sectors;
			dp->next_page(dp);
		} else while (remaining) {
			/*
			 * Try and add as many pages as possible.
			 */
			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);
}

struct sync_io {
	unsigned long error_bits;
	struct completion wait;
};

static void sync_io_complete(unsigned long error, void *context)
{
	struct sync_io *sio = context;

	sio->error_bits = error;
	complete(&sio->wait);
}

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;
	struct sync_io sio;

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

	init_completion(&sio.wait);

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

	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);

	wait_for_completion_io(&sio.wait);

	if (error_bits)
		*error_bits = sio.error_bits;

	return sio.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->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_BIO:
		bio_dp_init(dp, io_req->mem.ptr.bio);
		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
1da177e4 LT	12	#include <linux/bio.h>
10f1d5d1	13	#include <linux/completion.h>
1da177e4 LT	14	#include <linux/mempool.h>
	15	#include <linux/module.h>
	16	#include <linux/sched.h>
	17	#include <linux/slab.h>
a765e20e	18	#include <linux/dm-io.h>
1da177e4	19
f1e53987 MP	20	#define DM_MSG_PREFIX "io"
	21
	22	#define DM_IO_MAX_REGIONS BITS_PER_LONG
	23
891ce207 HM	24	struct dm_io_client {
	25	mempool_t *pool;
	26	struct bio_set *bios;
	27	};
	28
f1e53987 MP	29	/*
	30	* Aligning 'struct io' reduces the number of bits required to store
	31	* its address. Refer to store_io_and_region_in_bio() below.
	32	*/
1da177e4	33	struct io {
e01fd7ee	34	unsigned long error_bits;
1da177e4	35	atomic_t count;
891ce207	36	struct dm_io_client *client;
1da177e4 LT	37	io_notify_fn callback;
1da177e4 LT	38	void *context;
bb91bc7b MP	39	void *vma_invalidate_address;
bb91bc7b MP	40	unsigned long vma_invalidate_size;
f1e53987	41	} __attribute__((aligned(DM_IO_MAX_REGIONS)));
1da177e4	42
952b3557 MP	43	static struct kmem_cache *_dm_io_cache;
952b3557 MP	44
c8b03afe HM	45	/*
	46	* Create a client with mempool and bioset.
	47	*/
bda8efec	48	struct dm_io_client *dm_io_client_create(void)
c8b03afe	49	{
c8b03afe	50	struct dm_io_client *client;
e8603136	51	unsigned min_ios = dm_get_reserved_bio_based_ios();
c8b03afe HM	52
	53	client = kmalloc(sizeof(*client), GFP_KERNEL);
	54	if (!client)
	55	return ERR_PTR(-ENOMEM);
	56
e8603136	57	client->pool = mempool_create_slab_pool(min_ios, _dm_io_cache);
c8b03afe HM	58	if (!client->pool)
	59	goto bad;
	60
e8603136	61	client->bios = bioset_create(min_ios, 0);
c8b03afe HM	62	if (!client->bios)
	63	goto bad;
	64
	65	return client;
	66
	67	bad:
	68	if (client->pool)
	69	mempool_destroy(client->pool);
	70	kfree(client);
	71	return ERR_PTR(-ENOMEM);
	72	}
	73	EXPORT_SYMBOL(dm_io_client_create);
	74
c8b03afe HM	75	void dm_io_client_destroy(struct dm_io_client *client)
	76	{
	77	mempool_destroy(client->pool);
	78	bioset_free(client->bios);
	79	kfree(client);
	80	}
	81	EXPORT_SYMBOL(dm_io_client_destroy);
	82
1da177e4 LT	83	/*-----------------------------------------------------------------
1da177e4 LT	84	* We need to keep track of which region a bio is doing io for.
f1e53987 MP	85	* To avoid a memory allocation to store just 5 or 6 bits, we
	86	* ensure the 'struct io' pointer is aligned so enough low bits are
	87	* always zero and then combine it with the region number directly in
	88	* bi_private.
1da177e4	89	---------------------------------------------------------------/
f1e53987 MP	90	static void store_io_and_region_in_bio(struct bio bio, struct io io,
f1e53987 MP	91	unsigned region)
1da177e4	92	{
f1e53987 MP	93	if (unlikely(!IS_ALIGNED((unsigned long)io, DM_IO_MAX_REGIONS))) {
	94	DMCRIT("Unaligned struct io pointer %p", io);
	95	BUG();
	96	}
	97
	98	bio->bi_private = (void *)((unsigned long)io \| region);
1da177e4 LT	99	}
1da177e4 LT	100
f1e53987 MP	101	static void retrieve_io_and_region_from_bio(struct bio bio, struct io *io,
f1e53987 MP	102	unsigned *region)
1da177e4	103	{
f1e53987 MP	104	unsigned long val = (unsigned long)bio->bi_private;
	105
	106	io = (void )(val & -(unsigned long)DM_IO_MAX_REGIONS);
	107	*region = val & (DM_IO_MAX_REGIONS - 1);
1da177e4 LT	108	}
	109
	110	/*-----------------------------------------------------------------
	111	* We need an io object to keep track of the number of bios that
	112	* have been dispatched for a particular io.
	113	---------------------------------------------------------------/
97e7cdf1	114	static void complete_io(struct io *io)
1da177e4	115	{
97e7cdf1 JT	116	unsigned long error_bits = io->error_bits;
	117	io_notify_fn fn = io->callback;
	118	void *context = io->context;
1da177e4	119
97e7cdf1 JT	120	if (io->vma_invalidate_size)
	121	invalidate_kernel_vmap_range(io->vma_invalidate_address,
	122	io->vma_invalidate_size);
bb91bc7b	123
97e7cdf1 JT	124	mempool_free(io, io->client->pool);
	125	fn(error_bits, context);
	126	}
1da177e4	127
97e7cdf1 JT	128	static void dec_count(struct io *io, unsigned int region, int error)
	129	{
	130	if (error)
	131	set_bit(region, &io->error_bits);
1da177e4	132
97e7cdf1 JT	133	if (atomic_dec_and_test(&io->count))
97e7cdf1 JT	134	complete_io(io);
1da177e4 LT	135	}
1da177e4 LT	136
6712ecf8	137	static void endio(struct bio *bio, int error)
1da177e4	138	{
c897feb3 HM	139	struct io *io;
c897feb3 HM	140	unsigned region;
1da177e4	141
1da177e4 LT	142	if (error && bio_data_dir(bio) == READ)
	143	zero_fill_bio(bio);
	144
c897feb3 HM	145	/*
	146	* The bio destructor in bio_put() may use the io object.
	147	*/
f1e53987	148	retrieve_io_and_region_from_bio(bio, &io, &region);
c897feb3	149
1da177e4 LT	150	bio_put(bio);
1da177e4 LT	151
c897feb3	152	dec_count(io, region, error);
1da177e4 LT	153	}
	154
	155	/*-----------------------------------------------------------------
	156	* These little objects provide an abstraction for getting a new
	157	* destination page for io.
	158	---------------------------------------------------------------/
	159	struct dpages {
	160	void (get_page)(struct dpages dp,
	161	struct page *p, unsigned long len, unsigned *offset);
	162	void (next_page)(struct dpages dp);
	163
	164	unsigned context_u;
	165	void *context_ptr;
bb91bc7b MP	166
	167	void *vma_invalidate_address;
	168	unsigned long vma_invalidate_size;
1da177e4 LT	169	};
	170
	171	/*
	172	* Functions for getting the pages from a list.
	173	*/
	174	static void list_get_page(struct dpages *dp,
	175	struct page *p, unsigned long len, unsigned *offset)
	176	{
	177	unsigned o = dp->context_u;
	178	struct page_list pl = (struct page_list ) dp->context_ptr;
	179
	180	*p = pl->page;
	181	*len = PAGE_SIZE - o;
	182	*offset = o;
	183	}
	184
	185	static void list_next_page(struct dpages *dp)
	186	{
	187	struct page_list pl = (struct page_list ) dp->context_ptr;
	188	dp->context_ptr = pl->next;
	189	dp->context_u = 0;
	190	}
	191
	192	static void list_dp_init(struct dpages dp, struct page_list pl, unsigned offset)
	193	{
	194	dp->get_page = list_get_page;
	195	dp->next_page = list_next_page;
	196	dp->context_u = offset;
	197	dp->context_ptr = pl;
	198	}
	199
	200	/*
	201	* Functions for getting the pages from a bvec.
	202	*/
d73f9907 MP	203	static void bio_get_page(struct dpages dp, struct page *p,
d73f9907 MP	204	unsigned long len, unsigned offset)
1da177e4	205	{
d73f9907 MP	206	struct bio_vec *bvec = dp->context_ptr;
	207	*p = bvec->bv_page;
	208	*len = bvec->bv_len - dp->context_u;
	209	*offset = bvec->bv_offset + dp->context_u;
1da177e4 LT	210	}
1da177e4 LT	211
003b5c57	212	static void bio_next_page(struct dpages *dp)
1da177e4	213	{
d73f9907 MP	214	struct bio_vec *bvec = dp->context_ptr;
	215	dp->context_ptr = bvec + 1;
	216	dp->context_u = 0;
1da177e4 LT	217	}
1da177e4 LT	218
003b5c57	219	static void bio_dp_init(struct dpages dp, struct bio bio)
1da177e4	220	{
003b5c57 KO	221	dp->get_page = bio_get_page;
003b5c57 KO	222	dp->next_page = bio_next_page;
d73f9907 MP	223	dp->context_ptr = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
d73f9907 MP	224	dp->context_u = bio->bi_iter.bi_bvec_done;
1da177e4 LT	225	}
1da177e4 LT	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
c8b03afe HM	252	/*
	253	* Functions for getting the pages from kernel memory.
	254	*/
	255	static void km_get_page(struct dpages dp, struct page p, unsigned long len,
	256	unsigned *offset)
	257	{
	258	*p = virt_to_page(dp->context_ptr);
	259	*offset = dp->context_u;
	260	*len = PAGE_SIZE - dp->context_u;
	261	}
	262
	263	static void km_next_page(struct dpages *dp)
	264	{
	265	dp->context_ptr += PAGE_SIZE - dp->context_u;
	266	dp->context_u = 0;
	267	}
	268
	269	static void km_dp_init(struct dpages dp, void data)
	270	{
	271	dp->get_page = km_get_page;
	272	dp->next_page = km_next_page;
	273	dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1);
	274	dp->context_ptr = data;
	275	}
	276
1da177e4 LT	277	/*-----------------------------------------------------------------
	278	* IO routines that accept a list of pages.
	279	---------------------------------------------------------------/
22a1ceb1	280	static void do_region(int rw, unsigned region, struct dm_io_region *where,
1da177e4 LT	281	struct dpages dp, struct io io)
	282	{
	283	struct bio *bio;
	284	struct page *page;
	285	unsigned long len;
	286	unsigned offset;
	287	unsigned num_bvecs;
	288	sector_t remaining = where->count;
0c535e0d	289	struct request_queue *q = bdev_get_queue(where->bdev);
70d6c400 MS	290	unsigned short logical_block_size = queue_logical_block_size(q);
70d6c400 MS	291	sector_t num_sectors;
1da177e4	292
37527b86 DW	293	/* Reject unsupported discard requests */
	294	if ((rw & REQ_DISCARD) && !blk_queue_discard(q)) {
	295	dec_count(io, region, -EOPNOTSUPP);
	296	return;
	297	}
	298
12fc0f49	299	/*
d87f4c14 TH	300	* where->count may be zero if rw holds a flush and we need to
d87f4c14 TH	301	* send a zero-sized flush.
12fc0f49 MP	302	*/
12fc0f49 MP	303	do {
1da177e4	304	/*
f1e53987	305	* Allocate a suitably sized-bio.
1da177e4	306	*/
70d6c400	307	if ((rw & REQ_DISCARD) \|\| (rw & REQ_WRITE_SAME))
0c535e0d MB	308	num_bvecs = 1;
	309	else
	310	num_bvecs = min_t(int, bio_get_nr_vecs(where->bdev),
	311	dm_sector_div_up(remaining, (PAGE_SIZE >> SECTOR_SHIFT)));
	312
bf17ce3a	313	bio = bio_alloc_bioset(GFP_NOIO, num_bvecs, io->client->bios);
4f024f37	314	bio->bi_iter.bi_sector = where->sector + (where->count - remaining);
1da177e4 LT	315	bio->bi_bdev = where->bdev;
1da177e4 LT	316	bio->bi_end_io = endio;
f1e53987	317	store_io_and_region_in_bio(bio, io, region);
1da177e4	318
0c535e0d	319	if (rw & REQ_DISCARD) {
70d6c400	320	num_sectors = min_t(sector_t, q->limits.max_discard_sectors, remaining);
4f024f37	321	bio->bi_iter.bi_size = num_sectors << SECTOR_SHIFT;
70d6c400 MS	322	remaining -= num_sectors;
	323	} else if (rw & REQ_WRITE_SAME) {
	324	/*
	325	* WRITE SAME only uses a single page.
	326	*/
	327	dp->get_page(dp, &page, &len, &offset);
	328	bio_add_page(bio, page, logical_block_size, offset);
	329	num_sectors = min_t(sector_t, q->limits.max_write_same_sectors, remaining);
4f024f37	330	bio->bi_iter.bi_size = num_sectors << SECTOR_SHIFT;
70d6c400 MS	331
	332	offset = 0;
	333	remaining -= num_sectors;
	334	dp->next_page(dp);
0c535e0d MB	335	} else while (remaining) {
	336	/*
	337	* Try and add as many pages as possible.
	338	*/
1da177e4 LT	339	dp->get_page(dp, &page, &len, &offset);
	340	len = min(len, to_bytes(remaining));
	341	if (!bio_add_page(bio, page, len, offset))
	342	break;
	343
	344	offset = 0;
	345	remaining -= to_sector(len);
	346	dp->next_page(dp);
	347	}
	348
	349	atomic_inc(&io->count);
	350	submit_bio(rw, bio);
12fc0f49	351	} while (remaining);
1da177e4 LT	352	}
	353
	354	static void dispatch_io(int rw, unsigned int num_regions,
22a1ceb1	355	struct dm_io_region where, struct dpages dp,
1da177e4 LT	356	struct io *io, int sync)
	357	{
	358	int i;
	359	struct dpages old_pages = *dp;
	360
f1e53987 MP	361	BUG_ON(num_regions > DM_IO_MAX_REGIONS);
f1e53987 MP	362
1da177e4	363	if (sync)
721a9602	364	rw \|= REQ_SYNC;
1da177e4 LT	365
	366	/*
	367	* For multiple regions we need to be careful to rewind
	368	* the dp object for each call to do_region.
	369	*/
	370	for (i = 0; i < num_regions; i++) {
	371	*dp = old_pages;
d87f4c14	372	if (where[i].count \|\| (rw & REQ_FLUSH))
1da177e4 LT	373	do_region(rw, i, where + i, dp, io);
	374	}
	375
	376	/*
f00b16ad	377	* Drop the extra reference that we were holding to avoid
1da177e4 LT	378	* the io being completed too early.
	379	*/
	380	dec_count(io, 0, 0);
	381	}
	382
97e7cdf1 JT	383	struct sync_io {
	384	unsigned long error_bits;
	385	struct completion wait;
	386	};
	387
	388	static void sync_io_complete(unsigned long error, void *context)
	389	{
	390	struct sync_io *sio = context;
	391
	392	sio->error_bits = error;
	393	complete(&sio->wait);
	394	}
	395
891ce207	396	static int sync_io(struct dm_io_client *client, unsigned int num_regions,
22a1ceb1	397	struct dm_io_region where, int rw, struct dpages dp,
891ce207	398	unsigned long *error_bits)
1da177e4	399	{
97e7cdf1 JT	400	struct io *io;
97e7cdf1 JT	401	struct sync_io sio;
1da177e4	402
7ff14a36	403	if (num_regions > 1 && (rw & RW_MASK) != WRITE) {
1da177e4 LT	404	WARN_ON(1);
	405	return -EIO;
	406	}
	407
97e7cdf1 JT	408	init_completion(&sio.wait);
	409
	410	io = mempool_alloc(client->pool, GFP_NOIO);
f1e53987	411	io->error_bits = 0;
f1e53987	412	atomic_set(&io->count, 1); /* see dispatch_io() */
f1e53987	413	io->client = client;
97e7cdf1 JT	414	io->callback = sync_io_complete;
97e7cdf1 JT	415	io->context = &sio;
1da177e4	416
bb91bc7b MP	417	io->vma_invalidate_address = dp->vma_invalidate_address;
	418	io->vma_invalidate_size = dp->vma_invalidate_size;
	419
f1e53987	420	dispatch_io(rw, num_regions, where, dp, io, 1);
1da177e4	421
97e7cdf1	422	wait_for_completion_io(&sio.wait);
1da177e4	423
891ce207	424	if (error_bits)
97e7cdf1	425	*error_bits = sio.error_bits;
891ce207	426
97e7cdf1	427	return sio.error_bits ? -EIO : 0;
1da177e4 LT	428	}
1da177e4 LT	429
891ce207	430	static int async_io(struct dm_io_client *client, unsigned int num_regions,
22a1ceb1	431	struct dm_io_region where, int rw, struct dpages dp,
891ce207	432	io_notify_fn fn, void *context)
1da177e4 LT	433	{
	434	struct io *io;
	435
7ff14a36	436	if (num_regions > 1 && (rw & RW_MASK) != WRITE) {
1da177e4 LT	437	WARN_ON(1);
	438	fn(1, context);
	439	return -EIO;
	440	}
	441
bf17ce3a	442	io = mempool_alloc(client->pool, GFP_NOIO);
e01fd7ee	443	io->error_bits = 0;
1da177e4	444	atomic_set(&io->count, 1); /* see dispatch_io() */
891ce207	445	io->client = client;
1da177e4 LT	446	io->callback = fn;
	447	io->context = context;
	448
bb91bc7b MP	449	io->vma_invalidate_address = dp->vma_invalidate_address;
	450	io->vma_invalidate_size = dp->vma_invalidate_size;
	451
1da177e4 LT	452	dispatch_io(rw, num_regions, where, dp, io, 0);
	453	return 0;
	454	}
	455
bb91bc7b MP	456	static int dp_init(struct dm_io_request io_req, struct dpages dp,
bb91bc7b MP	457	unsigned long size)
c8b03afe HM	458	{
c8b03afe HM	459	/* Set up dpages based on memory type */
bb91bc7b MP	460
	461	dp->vma_invalidate_address = NULL;
	462	dp->vma_invalidate_size = 0;
	463
c8b03afe HM	464	switch (io_req->mem.type) {
	465	case DM_IO_PAGE_LIST:
	466	list_dp_init(dp, io_req->mem.ptr.pl, io_req->mem.offset);
	467	break;
	468
003b5c57 KO	469	case DM_IO_BIO:
003b5c57 KO	470	bio_dp_init(dp, io_req->mem.ptr.bio);
c8b03afe HM	471	break;
	472
	473	case DM_IO_VMA:
bb91bc7b MP	474	flush_kernel_vmap_range(io_req->mem.ptr.vma, size);
	475	if ((io_req->bi_rw & RW_MASK) == READ) {
	476	dp->vma_invalidate_address = io_req->mem.ptr.vma;
	477	dp->vma_invalidate_size = size;
	478	}
c8b03afe HM	479	vm_dp_init(dp, io_req->mem.ptr.vma);
	480	break;
	481
	482	case DM_IO_KMEM:
	483	km_dp_init(dp, io_req->mem.ptr.addr);
	484	break;
	485
	486	default:
	487	return -EINVAL;
	488	}
	489
	490	return 0;
	491	}
	492
	493	/*
7ff14a36 MP	494	* New collapsed (a)synchronous interface.
	495	*
	496	* If the IO is asynchronous (i.e. it has notify.fn), you must either unplug
97e7cdf1 JT	497	* the queue with blk_unplug() some time later or set REQ_SYNC in io_req->bi_rw.
	498	* If you fail to do one of these, the IO will be submitted to the disk after
	499	* q->unplug_delay, which defaults to 3ms in blk-settings.c.
c8b03afe HM	500	*/
c8b03afe HM	501	int dm_io(struct dm_io_request *io_req, unsigned num_regions,
22a1ceb1	502	struct dm_io_region where, unsigned long sync_error_bits)
c8b03afe HM	503	{
	504	int r;
	505	struct dpages dp;
	506
bb91bc7b	507	r = dp_init(io_req, &dp, (unsigned long)where->count << SECTOR_SHIFT);
c8b03afe HM	508	if (r)
	509	return r;
	510
	511	if (!io_req->notify.fn)
	512	return sync_io(io_req->client, num_regions, where,
	513	io_req->bi_rw, &dp, sync_error_bits);
	514
	515	return async_io(io_req->client, num_regions, where, io_req->bi_rw,
	516	&dp, io_req->notify.fn, io_req->notify.context);
	517	}
	518	EXPORT_SYMBOL(dm_io);
952b3557 MP	519
	520	int __init dm_io_init(void)
	521	{
	522	_dm_io_cache = KMEM_CACHE(io, 0);
	523	if (!_dm_io_cache)
	524	return -ENOMEM;
	525
	526	return 0;
	527	}
	528
	529	void dm_io_exit(void)
	530	{
	531	kmem_cache_destroy(_dm_io_cache);
	532	_dm_io_cache = NULL;
	533	}