dm-io.c 12.9 KB
Newer Older
L
Linus Torvalds 已提交
1 2
/*
 * Copyright (C) 2003 Sistina Software
3
 * Copyright (C) 2006 Red Hat GmbH
L
Linus Torvalds 已提交
4 5 6 7
 *
 * This file is released under the GPL.
 */

M
Mikulas Patocka 已提交
8 9
#include "dm.h"

10
#include <linux/device-mapper.h>
L
Linus Torvalds 已提交
11 12 13 14 15 16

#include <linux/bio.h>
#include <linux/mempool.h>
#include <linux/module.h>
#include <linux/sched.h>
#include <linux/slab.h>
A
Alasdair G Kergon 已提交
17
#include <linux/dm-io.h>
L
Linus Torvalds 已提交
18

19 20 21 22
#define DM_MSG_PREFIX "io"

#define DM_IO_MAX_REGIONS	BITS_PER_LONG

23 24 25 26 27
struct dm_io_client {
	mempool_t *pool;
	struct bio_set *bios;
};

28 29 30 31
/*
 * Aligning 'struct io' reduces the number of bits required to store
 * its address.  Refer to store_io_and_region_in_bio() below.
 */
L
Linus Torvalds 已提交
32
struct io {
33
	unsigned long error_bits;
L
Linus Torvalds 已提交
34 35
	atomic_t count;
	struct task_struct *sleeper;
36
	struct dm_io_client *client;
L
Linus Torvalds 已提交
37 38
	io_notify_fn callback;
	void *context;
39 40
	void *vma_invalidate_address;
	unsigned long vma_invalidate_size;
41
} __attribute__((aligned(DM_IO_MAX_REGIONS)));
L
Linus Torvalds 已提交
42

M
Mikulas Patocka 已提交
43 44
static struct kmem_cache *_dm_io_cache;

H
Heinz Mauelshagen 已提交
45 46 47
/*
 * Create a client with mempool and bioset.
 */
48
struct dm_io_client *dm_io_client_create(void)
H
Heinz Mauelshagen 已提交
49 50
{
	struct dm_io_client *client;
51
	unsigned min_ios = dm_get_reserved_bio_based_ios();
H
Heinz Mauelshagen 已提交
52 53 54 55 56

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

57
	client->pool = mempool_create_slab_pool(min_ios, _dm_io_cache);
H
Heinz Mauelshagen 已提交
58 59 60
	if (!client->pool)
		goto bad;

61
	client->bios = bioset_create(min_ios, 0);
H
Heinz Mauelshagen 已提交
62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82
	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);

L
Linus Torvalds 已提交
83 84
/*-----------------------------------------------------------------
 * We need to keep track of which region a bio is doing io for.
85 86 87 88
 * 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.
L
Linus Torvalds 已提交
89
 *---------------------------------------------------------------*/
90 91
static void store_io_and_region_in_bio(struct bio *bio, struct io *io,
				       unsigned region)
L
Linus Torvalds 已提交
92
{
93 94 95 96 97 98
	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);
L
Linus Torvalds 已提交
99 100
}

101 102
static void retrieve_io_and_region_from_bio(struct bio *bio, struct io **io,
				       unsigned *region)
L
Linus Torvalds 已提交
103
{
104 105 106 107
	unsigned long val = (unsigned long)bio->bi_private;

	*io = (void *)(val & -(unsigned long)DM_IO_MAX_REGIONS);
	*region = val & (DM_IO_MAX_REGIONS - 1);
L
Linus Torvalds 已提交
108 109 110 111 112 113 114 115
}

/*-----------------------------------------------------------------
 * 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)
{
116
	if (error)
117
		set_bit(region, &io->error_bits);
L
Linus Torvalds 已提交
118 119

	if (atomic_dec_and_test(&io->count)) {
120 121 122 123
		if (io->vma_invalidate_size)
			invalidate_kernel_vmap_range(io->vma_invalidate_address,
						     io->vma_invalidate_size);

L
Linus Torvalds 已提交
124 125 126 127
		if (io->sleeper)
			wake_up_process(io->sleeper);

		else {
128
			unsigned long r = io->error_bits;
L
Linus Torvalds 已提交
129 130 131
			io_notify_fn fn = io->callback;
			void *context = io->context;

M
Milan Broz 已提交
132
			mempool_free(io, io->client->pool);
L
Linus Torvalds 已提交
133 134 135 136 137
			fn(r, context);
		}
	}
}

138
static void endio(struct bio *bio, int error)
L
Linus Torvalds 已提交
139
{
H
Heinz Mauelshagen 已提交
140 141
	struct io *io;
	unsigned region;
L
Linus Torvalds 已提交
142 143 144 145

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

H
Heinz Mauelshagen 已提交
146 147 148
	/*
	 * The bio destructor in bio_put() may use the io object.
	 */
149
	retrieve_io_and_region_from_bio(bio, &io, &region);
H
Heinz Mauelshagen 已提交
150

L
Linus Torvalds 已提交
151 152
	bio_put(bio);

H
Heinz Mauelshagen 已提交
153
	dec_count(io, region, error);
L
Linus Torvalds 已提交
154 155 156 157 158 159 160 161 162 163 164 165 166
}

/*-----------------------------------------------------------------
 * 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;
167 168 169

	void *vma_invalidate_address;
	unsigned long vma_invalidate_size;
L
Linus Torvalds 已提交
170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203
};

/*
 * 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.
 */
204 205
static void bio_get_page(struct dpages *dp, struct page **p,
			 unsigned long *len, unsigned *offset)
L
Linus Torvalds 已提交
206
{
207 208 209 210
	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;
L
Linus Torvalds 已提交
211 212
}

213
static void bio_next_page(struct dpages *dp)
L
Linus Torvalds 已提交
214
{
215 216 217
	struct bio_vec *bvec = dp->context_ptr;
	dp->context_ptr = bvec + 1;
	dp->context_u = 0;
L
Linus Torvalds 已提交
218 219
}

220
static void bio_dp_init(struct dpages *dp, struct bio *bio)
L
Linus Torvalds 已提交
221
{
222 223
	dp->get_page = bio_get_page;
	dp->next_page = bio_next_page;
224 225
	dp->context_ptr = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
	dp->context_u = bio->bi_iter.bi_bvec_done;
L
Linus Torvalds 已提交
226 227
}

H
Heinz Mauelshagen 已提交
228 229 230
/*
 * Functions for getting the pages from a VMA.
 */
L
Linus Torvalds 已提交
231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252
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;
}

H
Heinz Mauelshagen 已提交
253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277
/*
 * 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;
}

L
Linus Torvalds 已提交
278 279 280
/*-----------------------------------------------------------------
 * IO routines that accept a list of pages.
 *---------------------------------------------------------------*/
H
Heinz Mauelshagen 已提交
281
static void do_region(int rw, unsigned region, struct dm_io_region *where,
L
Linus Torvalds 已提交
282 283 284 285 286 287 288 289
		      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;
M
Milan Broz 已提交
290
	struct request_queue *q = bdev_get_queue(where->bdev);
291 292
	unsigned short logical_block_size = queue_logical_block_size(q);
	sector_t num_sectors;
L
Linus Torvalds 已提交
293

M
Mikulas Patocka 已提交
294
	/*
295 296
	 * where->count may be zero if rw holds a flush and we need to
	 * send a zero-sized flush.
M
Mikulas Patocka 已提交
297 298
	 */
	do {
L
Linus Torvalds 已提交
299
		/*
300
		 * Allocate a suitably sized-bio.
L
Linus Torvalds 已提交
301
		 */
302
		if ((rw & REQ_DISCARD) || (rw & REQ_WRITE_SAME))
M
Milan Broz 已提交
303 304 305 306 307
			num_bvecs = 1;
		else
			num_bvecs = min_t(int, bio_get_nr_vecs(where->bdev),
					  dm_sector_div_up(remaining, (PAGE_SIZE >> SECTOR_SHIFT)));

M
Milan Broz 已提交
308
		bio = bio_alloc_bioset(GFP_NOIO, num_bvecs, io->client->bios);
309
		bio->bi_iter.bi_sector = where->sector + (where->count - remaining);
L
Linus Torvalds 已提交
310 311
		bio->bi_bdev = where->bdev;
		bio->bi_end_io = endio;
312
		store_io_and_region_in_bio(bio, io, region);
L
Linus Torvalds 已提交
313

M
Milan Broz 已提交
314
		if (rw & REQ_DISCARD) {
315
			num_sectors = min_t(sector_t, q->limits.max_discard_sectors, remaining);
316
			bio->bi_iter.bi_size = num_sectors << SECTOR_SHIFT;
317 318 319 320 321 322 323 324
			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);
325
			bio->bi_iter.bi_size = num_sectors << SECTOR_SHIFT;
326 327 328 329

			offset = 0;
			remaining -= num_sectors;
			dp->next_page(dp);
M
Milan Broz 已提交
330 331 332 333
		} else while (remaining) {
			/*
			 * Try and add as many pages as possible.
			 */
L
Linus Torvalds 已提交
334 335 336 337 338 339 340 341 342 343 344 345
			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);
M
Mikulas Patocka 已提交
346
	} while (remaining);
L
Linus Torvalds 已提交
347 348 349
}

static void dispatch_io(int rw, unsigned int num_regions,
H
Heinz Mauelshagen 已提交
350
			struct dm_io_region *where, struct dpages *dp,
L
Linus Torvalds 已提交
351 352 353 354 355
			struct io *io, int sync)
{
	int i;
	struct dpages old_pages = *dp;

356 357
	BUG_ON(num_regions > DM_IO_MAX_REGIONS);

L
Linus Torvalds 已提交
358
	if (sync)
J
Jens Axboe 已提交
359
		rw |= REQ_SYNC;
L
Linus Torvalds 已提交
360 361 362 363 364 365 366

	/*
	 * 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;
367
		if (where[i].count || (rw & REQ_FLUSH))
L
Linus Torvalds 已提交
368 369 370 371
			do_region(rw, i, where + i, dp, io);
	}

	/*
H
Heinz Mauelshagen 已提交
372
	 * Drop the extra reference that we were holding to avoid
L
Linus Torvalds 已提交
373 374 375 376 377
	 * the io being completed too early.
	 */
	dec_count(io, 0, 0);
}

378
static int sync_io(struct dm_io_client *client, unsigned int num_regions,
H
Heinz Mauelshagen 已提交
379
		   struct dm_io_region *where, int rw, struct dpages *dp,
380
		   unsigned long *error_bits)
L
Linus Torvalds 已提交
381
{
382 383 384 385 386 387 388 389
	/*
	 * 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));
L
Linus Torvalds 已提交
390

M
Mikulas Patocka 已提交
391
	if (num_regions > 1 && (rw & RW_MASK) != WRITE) {
L
Linus Torvalds 已提交
392 393 394 395
		WARN_ON(1);
		return -EIO;
	}

396 397 398 399
	io->error_bits = 0;
	atomic_set(&io->count, 1); /* see dispatch_io() */
	io->sleeper = current;
	io->client = client;
L
Linus Torvalds 已提交
400

401 402 403
	io->vma_invalidate_address = dp->vma_invalidate_address;
	io->vma_invalidate_size = dp->vma_invalidate_size;

404
	dispatch_io(rw, num_regions, where, dp, io, 1);
L
Linus Torvalds 已提交
405 406 407 408

	while (1) {
		set_current_state(TASK_UNINTERRUPTIBLE);

409
		if (!atomic_read(&io->count))
L
Linus Torvalds 已提交
410 411 412 413 414 415
			break;

		io_schedule();
	}
	set_current_state(TASK_RUNNING);

416
	if (error_bits)
417
		*error_bits = io->error_bits;
418

419
	return io->error_bits ? -EIO : 0;
L
Linus Torvalds 已提交
420 421
}

422
static int async_io(struct dm_io_client *client, unsigned int num_regions,
H
Heinz Mauelshagen 已提交
423
		    struct dm_io_region *where, int rw, struct dpages *dp,
424
		    io_notify_fn fn, void *context)
L
Linus Torvalds 已提交
425 426 427
{
	struct io *io;

M
Mikulas Patocka 已提交
428
	if (num_regions > 1 && (rw & RW_MASK) != WRITE) {
L
Linus Torvalds 已提交
429 430 431 432 433
		WARN_ON(1);
		fn(1, context);
		return -EIO;
	}

M
Milan Broz 已提交
434
	io = mempool_alloc(client->pool, GFP_NOIO);
435
	io->error_bits = 0;
L
Linus Torvalds 已提交
436 437
	atomic_set(&io->count, 1); /* see dispatch_io() */
	io->sleeper = NULL;
438
	io->client = client;
L
Linus Torvalds 已提交
439 440 441
	io->callback = fn;
	io->context = context;

442 443 444
	io->vma_invalidate_address = dp->vma_invalidate_address;
	io->vma_invalidate_size = dp->vma_invalidate_size;

L
Linus Torvalds 已提交
445 446 447 448
	dispatch_io(rw, num_regions, where, dp, io, 0);
	return 0;
}

449 450
static int dp_init(struct dm_io_request *io_req, struct dpages *dp,
		   unsigned long size)
H
Heinz Mauelshagen 已提交
451 452
{
	/* Set up dpages based on memory type */
453 454 455 456

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

H
Heinz Mauelshagen 已提交
457 458 459 460 461
	switch (io_req->mem.type) {
	case DM_IO_PAGE_LIST:
		list_dp_init(dp, io_req->mem.ptr.pl, io_req->mem.offset);
		break;

462 463
	case DM_IO_BIO:
		bio_dp_init(dp, io_req->mem.ptr.bio);
H
Heinz Mauelshagen 已提交
464 465 466
		break;

	case DM_IO_VMA:
467 468 469 470 471
		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;
		}
H
Heinz Mauelshagen 已提交
472 473 474 475 476 477 478 479 480 481 482 483 484 485 486
		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;
}

/*
M
Mikulas Patocka 已提交
487 488 489
 * New collapsed (a)synchronous interface.
 *
 * If the IO is asynchronous (i.e. it has notify.fn), you must either unplug
490 491
 * 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
M
Mikulas Patocka 已提交
492
 * the disk after q->unplug_delay, which defaults to 3ms in blk-settings.c.
H
Heinz Mauelshagen 已提交
493 494
 */
int dm_io(struct dm_io_request *io_req, unsigned num_regions,
H
Heinz Mauelshagen 已提交
495
	  struct dm_io_region *where, unsigned long *sync_error_bits)
H
Heinz Mauelshagen 已提交
496 497 498 499
{
	int r;
	struct dpages dp;

500
	r = dp_init(io_req, &dp, (unsigned long)where->count << SECTOR_SHIFT);
H
Heinz Mauelshagen 已提交
501 502 503 504 505 506 507 508 509 510 511
	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);
M
Mikulas Patocka 已提交
512 513 514 515 516 517 518 519 520 521 522 523 524 525 526

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;
}