blob: 314a0e2faf79d6c097e7c1201bf25af8600baaf6 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
Milan Broz784aae72009-01-06 03:05:12 +00003 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
5 * This file is released under the GPL.
6 */
7
8#include "dm.h"
Mike Anderson51e5b2b2007-10-19 22:48:00 +01009#include "dm-uevent.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070010
11#include <linux/init.h>
12#include <linux/module.h>
Arjan van de Ven48c9c272006-03-27 01:18:20 -080013#include <linux/mutex.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/moduleparam.h>
15#include <linux/blkpg.h>
16#include <linux/bio.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/mempool.h>
18#include <linux/slab.h>
19#include <linux/idr.h>
Darrick J. Wong3ac51e72006-03-27 01:17:54 -080020#include <linux/hdreg.h>
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +010021#include <linux/delay.h>
Li Zefan55782132009-06-09 13:43:05 +080022
23#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024
Alasdair G Kergon72d94862006-06-26 00:27:35 -070025#define DM_MSG_PREFIX "core"
26
Namhyung Kim71a16732011-10-31 20:18:54 +000027#ifdef CONFIG_PRINTK
28/*
29 * ratelimit state to be used in DMXXX_LIMIT().
30 */
31DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
32 DEFAULT_RATELIMIT_INTERVAL,
33 DEFAULT_RATELIMIT_BURST);
34EXPORT_SYMBOL(dm_ratelimit_state);
35#endif
36
Milan Broz60935eb2009-06-22 10:12:30 +010037/*
38 * Cookies are numeric values sent with CHANGE and REMOVE
39 * uevents while resuming, removing or renaming the device.
40 */
41#define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
42#define DM_COOKIE_LENGTH 24
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044static const char *_name = DM_NAME;
45
46static unsigned int major = 0;
47static unsigned int _major = 0;
48
Alasdair G Kergond15b7742011-08-02 12:32:01 +010049static DEFINE_IDR(_minor_idr);
50
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -070051static DEFINE_SPINLOCK(_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070052/*
Kiyoshi Ueda8fbf26a2009-01-06 03:05:06 +000053 * For bio-based dm.
Linus Torvalds1da177e2005-04-16 15:20:36 -070054 * One of these is allocated per bio.
55 */
56struct dm_io {
57 struct mapped_device *md;
58 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 atomic_t io_count;
Richard Kennedy6ae2fa62008-07-21 12:00:28 +010060 struct bio *bio;
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -080061 unsigned long start_time;
Kiyoshi Uedaf88fb982009-10-16 23:18:15 +010062 spinlock_t endio_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -070063};
64
65/*
Kiyoshi Ueda8fbf26a2009-01-06 03:05:06 +000066 * For request-based dm.
67 * One of these is allocated per request.
68 */
69struct dm_rq_target_io {
70 struct mapped_device *md;
71 struct dm_target *ti;
72 struct request *orig, clone;
73 int error;
74 union map_info info;
75};
76
77/*
Kent Overstreet94818742012-09-07 13:44:01 -070078 * For request-based dm - the bio clones we allocate are embedded in these
79 * structs.
80 *
81 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
82 * the bioset is created - this means the bio has to come at the end of the
83 * struct.
Kiyoshi Ueda8fbf26a2009-01-06 03:05:06 +000084 */
85struct dm_rq_clone_bio_info {
86 struct bio *orig;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +010087 struct dm_rq_target_io *tio;
Kent Overstreet94818742012-09-07 13:44:01 -070088 struct bio clone;
Kiyoshi Ueda8fbf26a2009-01-06 03:05:06 +000089};
90
Linus Torvalds1da177e2005-04-16 15:20:36 -070091union map_info *dm_get_mapinfo(struct bio *bio)
92{
Alasdair G Kergon17b2f662006-06-26 00:27:33 -070093 if (bio && bio->bi_private)
Alasdair G Kergon028867a2007-07-12 17:26:32 +010094 return &((struct dm_target_io *)bio->bi_private)->info;
Alasdair G Kergon17b2f662006-06-26 00:27:33 -070095 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070096}
97
Kiyoshi Uedacec47e32009-06-22 10:12:35 +010098union map_info *dm_get_rq_mapinfo(struct request *rq)
99{
100 if (rq && rq->end_io_data)
101 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
102 return NULL;
103}
104EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
105
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -0700106#define MINOR_ALLOCED ((void *)-1)
107
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108/*
109 * Bits for the md->flags field.
110 */
Alasdair G Kergon1eb787e2009-04-09 00:27:14 +0100111#define DMF_BLOCK_IO_FOR_SUSPEND 0
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112#define DMF_SUSPENDED 1
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -0800113#define DMF_FROZEN 2
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700114#define DMF_FREEING 3
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700115#define DMF_DELETING 4
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800116#define DMF_NOFLUSH_SUSPENDING 5
Mikulas Patockad5b9dd02011-08-02 12:32:04 +0100117#define DMF_MERGE_IS_OPTIONAL 6
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118
Milan Broz304f3f62008-02-08 02:11:17 +0000119/*
120 * Work processed by per-device workqueue.
121 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122struct mapped_device {
Alasdair G Kergon2ca33102005-07-28 21:16:00 -0700123 struct rw_semaphore io_lock;
Daniel Walkere61290a2008-02-08 02:10:08 +0000124 struct mutex suspend_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 rwlock_t map_lock;
126 atomic_t holders;
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700127 atomic_t open_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
129 unsigned long flags;
130
Jens Axboe165125e2007-07-24 09:28:11 +0200131 struct request_queue *queue;
Mike Snitzera5664da2010-08-12 04:14:01 +0100132 unsigned type;
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +0100133 /* Protect queue and type against concurrent access. */
Mike Snitzera5664da2010-08-12 04:14:01 +0100134 struct mutex type_lock;
135
Alasdair G Kergon36a04562011-10-31 20:19:04 +0000136 struct target_type *immutable_target_type;
137
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 struct gendisk *disk;
Mike Anderson7e51f252006-03-27 01:17:52 -0800139 char name[16];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
141 void *interface_ptr;
142
143 /*
144 * A list of ios that arrived while we were suspended.
145 */
Nikanth Karthikesan316d3152009-10-06 20:16:55 +0200146 atomic_t pending[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 wait_queue_head_t wait;
Mikulas Patocka53d59142009-04-02 19:55:37 +0100148 struct work_struct work;
Kiyoshi Ueda74859362006-12-08 02:41:02 -0800149 struct bio_list deferred;
Mikulas Patocka022c2612009-04-02 19:55:39 +0100150 spinlock_t deferred_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
152 /*
Tejun Heo29e40132010-09-08 18:07:00 +0200153 * Processing queue (flush)
Milan Broz304f3f62008-02-08 02:11:17 +0000154 */
155 struct workqueue_struct *wq;
156
157 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 * The current mapping.
159 */
160 struct dm_table *map;
161
162 /*
163 * io objects are allocated from here.
164 */
165 mempool_t *io_pool;
166 mempool_t *tio_pool;
167
Stefan Bader9faf4002006-10-03 01:15:41 -0700168 struct bio_set *bs;
169
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 /*
171 * Event handling.
172 */
173 atomic_t event_nr;
174 wait_queue_head_t eventq;
Mike Anderson7a8c3d32007-10-19 22:48:01 +0100175 atomic_t uevent_seq;
176 struct list_head uevent_list;
177 spinlock_t uevent_lock; /* Protect access to uevent_list */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
179 /*
180 * freeze/thaw support require holding onto a super block
181 */
182 struct super_block *frozen_sb;
Mikulas Patockadb8fef42009-06-22 10:12:15 +0100183 struct block_device *bdev;
Darrick J. Wong3ac51e72006-03-27 01:17:54 -0800184
185 /* forced geometry settings */
186 struct hd_geometry geometry;
Milan Broz784aae72009-01-06 03:05:12 +0000187
188 /* sysfs handle */
189 struct kobject kobj;
Mikulas Patocka52b1fd52009-06-22 10:12:21 +0100190
Tejun Heod87f4c12010-09-03 11:56:19 +0200191 /* zero-length flush that will be cloned and submitted to targets */
192 struct bio flush_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193};
194
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100195/*
196 * For mempools pre-allocation at the table loading time.
197 */
198struct dm_md_mempools {
199 mempool_t *io_pool;
200 mempool_t *tio_pool;
201 struct bio_set *bs;
202};
203
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204#define MIN_IOS 256
Christoph Lametere18b8902006-12-06 20:33:20 -0800205static struct kmem_cache *_io_cache;
Kiyoshi Ueda8fbf26a2009-01-06 03:05:06 +0000206static struct kmem_cache *_rq_tio_cache;
Kent Overstreet94818742012-09-07 13:44:01 -0700207
208/*
209 * Unused now, and needs to be deleted. But since io_pool is overloaded and it's
210 * still used for _io_cache, I'm leaving this for a later cleanup
211 */
Kiyoshi Ueda8fbf26a2009-01-06 03:05:06 +0000212static struct kmem_cache *_rq_bio_info_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214static int __init local_init(void)
215{
Kiyoshi Ueda51157b42008-10-21 17:45:08 +0100216 int r = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 /* allocate a slab for the dm_ios */
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100219 _io_cache = KMEM_CACHE(dm_io, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 if (!_io_cache)
Kiyoshi Ueda51157b42008-10-21 17:45:08 +0100221 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
Kiyoshi Ueda8fbf26a2009-01-06 03:05:06 +0000223 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
224 if (!_rq_tio_cache)
Mikulas Patockadba14162012-10-12 21:02:15 +0100225 goto out_free_io_cache;
Kiyoshi Ueda8fbf26a2009-01-06 03:05:06 +0000226
227 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
228 if (!_rq_bio_info_cache)
229 goto out_free_rq_tio_cache;
230
Mike Anderson51e5b2b2007-10-19 22:48:00 +0100231 r = dm_uevent_init();
Kiyoshi Ueda51157b42008-10-21 17:45:08 +0100232 if (r)
Kiyoshi Ueda8fbf26a2009-01-06 03:05:06 +0000233 goto out_free_rq_bio_info_cache;
Mike Anderson51e5b2b2007-10-19 22:48:00 +0100234
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 _major = major;
236 r = register_blkdev(_major, _name);
Kiyoshi Ueda51157b42008-10-21 17:45:08 +0100237 if (r < 0)
238 goto out_uevent_exit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
240 if (!_major)
241 _major = r;
242
243 return 0;
Kiyoshi Ueda51157b42008-10-21 17:45:08 +0100244
245out_uevent_exit:
246 dm_uevent_exit();
Kiyoshi Ueda8fbf26a2009-01-06 03:05:06 +0000247out_free_rq_bio_info_cache:
248 kmem_cache_destroy(_rq_bio_info_cache);
249out_free_rq_tio_cache:
250 kmem_cache_destroy(_rq_tio_cache);
Kiyoshi Ueda51157b42008-10-21 17:45:08 +0100251out_free_io_cache:
252 kmem_cache_destroy(_io_cache);
253
254 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255}
256
257static void local_exit(void)
258{
Kiyoshi Ueda8fbf26a2009-01-06 03:05:06 +0000259 kmem_cache_destroy(_rq_bio_info_cache);
260 kmem_cache_destroy(_rq_tio_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 kmem_cache_destroy(_io_cache);
Akinobu Mita00d59402007-07-17 04:03:46 -0700262 unregister_blkdev(_major, _name);
Mike Anderson51e5b2b2007-10-19 22:48:00 +0100263 dm_uevent_exit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264
265 _major = 0;
266
267 DMINFO("cleaned up");
268}
269
Alasdair G Kergonb9249e52008-02-08 02:09:51 +0000270static int (*_inits[])(void) __initdata = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 local_init,
272 dm_target_init,
273 dm_linear_init,
274 dm_stripe_init,
Mikulas Patocka952b3552009-12-10 23:51:57 +0000275 dm_io_init,
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100276 dm_kcopyd_init,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 dm_interface_init,
278};
279
Alasdair G Kergonb9249e52008-02-08 02:09:51 +0000280static void (*_exits[])(void) = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 local_exit,
282 dm_target_exit,
283 dm_linear_exit,
284 dm_stripe_exit,
Mikulas Patocka952b3552009-12-10 23:51:57 +0000285 dm_io_exit,
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100286 dm_kcopyd_exit,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 dm_interface_exit,
288};
289
290static int __init dm_init(void)
291{
292 const int count = ARRAY_SIZE(_inits);
293
294 int r, i;
295
296 for (i = 0; i < count; i++) {
297 r = _inits[i]();
298 if (r)
299 goto bad;
300 }
301
302 return 0;
303
304 bad:
305 while (i--)
306 _exits[i]();
307
308 return r;
309}
310
311static void __exit dm_exit(void)
312{
313 int i = ARRAY_SIZE(_exits);
314
315 while (i--)
316 _exits[i]();
Alasdair G Kergond15b7742011-08-02 12:32:01 +0100317
318 /*
319 * Should be empty by this point.
320 */
321 idr_remove_all(&_minor_idr);
322 idr_destroy(&_minor_idr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323}
324
325/*
326 * Block device functions
327 */
Mike Anderson432a2122009-12-10 23:52:20 +0000328int dm_deleting_md(struct mapped_device *md)
329{
330 return test_bit(DMF_DELETING, &md->flags);
331}
332
Al Virofe5f9f22008-03-02 10:29:31 -0500333static int dm_blk_open(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334{
335 struct mapped_device *md;
336
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700337 spin_lock(&_minor_lock);
338
Al Virofe5f9f22008-03-02 10:29:31 -0500339 md = bdev->bd_disk->private_data;
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700340 if (!md)
341 goto out;
342
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700343 if (test_bit(DMF_FREEING, &md->flags) ||
Mike Anderson432a2122009-12-10 23:52:20 +0000344 dm_deleting_md(md)) {
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700345 md = NULL;
346 goto out;
347 }
348
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 dm_get(md);
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700350 atomic_inc(&md->open_count);
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700351
352out:
353 spin_unlock(&_minor_lock);
354
355 return md ? 0 : -ENXIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356}
357
Al Virofe5f9f22008-03-02 10:29:31 -0500358static int dm_blk_close(struct gendisk *disk, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359{
Al Virofe5f9f22008-03-02 10:29:31 -0500360 struct mapped_device *md = disk->private_data;
Arnd Bergmann6e9624b2010-08-07 18:25:34 +0200361
Milan Broz4a1aeb92011-01-13 19:59:48 +0000362 spin_lock(&_minor_lock);
363
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700364 atomic_dec(&md->open_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 dm_put(md);
Milan Broz4a1aeb92011-01-13 19:59:48 +0000366
367 spin_unlock(&_minor_lock);
Arnd Bergmann6e9624b2010-08-07 18:25:34 +0200368
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 return 0;
370}
371
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700372int dm_open_count(struct mapped_device *md)
373{
374 return atomic_read(&md->open_count);
375}
376
377/*
378 * Guarantees nothing is using the device before it's deleted.
379 */
380int dm_lock_for_deletion(struct mapped_device *md)
381{
382 int r = 0;
383
384 spin_lock(&_minor_lock);
385
386 if (dm_open_count(md))
387 r = -EBUSY;
388 else
389 set_bit(DMF_DELETING, &md->flags);
390
391 spin_unlock(&_minor_lock);
392
393 return r;
394}
395
Darrick J. Wong3ac51e72006-03-27 01:17:54 -0800396static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
397{
398 struct mapped_device *md = bdev->bd_disk->private_data;
399
400 return dm_get_geometry(md, geo);
401}
402
Al Virofe5f9f22008-03-02 10:29:31 -0500403static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
Milan Brozaa129a22006-10-03 01:15:15 -0700404 unsigned int cmd, unsigned long arg)
405{
Al Virofe5f9f22008-03-02 10:29:31 -0500406 struct mapped_device *md = bdev->bd_disk->private_data;
Alasdair G Kergon7c666412009-12-10 23:52:19 +0000407 struct dm_table *map = dm_get_live_table(md);
Milan Brozaa129a22006-10-03 01:15:15 -0700408 struct dm_target *tgt;
409 int r = -ENOTTY;
410
Milan Brozaa129a22006-10-03 01:15:15 -0700411 if (!map || !dm_table_get_size(map))
412 goto out;
413
414 /* We only support devices that have a single target */
415 if (dm_table_get_num_targets(map) != 1)
416 goto out;
417
418 tgt = dm_table_get_target(map, 0);
419
Kiyoshi Ueda4f186f82009-12-10 23:52:26 +0000420 if (dm_suspended_md(md)) {
Milan Brozaa129a22006-10-03 01:15:15 -0700421 r = -EAGAIN;
422 goto out;
423 }
424
425 if (tgt->type->ioctl)
Al Viro647b3d02007-08-28 22:15:59 -0400426 r = tgt->type->ioctl(tgt, cmd, arg);
Milan Brozaa129a22006-10-03 01:15:15 -0700427
428out:
429 dm_table_put(map);
430
Milan Brozaa129a22006-10-03 01:15:15 -0700431 return r;
432}
433
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100434static struct dm_io *alloc_io(struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435{
436 return mempool_alloc(md->io_pool, GFP_NOIO);
437}
438
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100439static void free_io(struct mapped_device *md, struct dm_io *io)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440{
441 mempool_free(io, md->io_pool);
442}
443
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100444static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445{
Mikulas Patockadba14162012-10-12 21:02:15 +0100446 bio_put(&tio->clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447}
448
Kiyoshi Ueda08885642009-12-10 23:52:15 +0000449static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
450 gfp_t gfp_mask)
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100451{
Kiyoshi Ueda08885642009-12-10 23:52:15 +0000452 return mempool_alloc(md->tio_pool, gfp_mask);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100453}
454
455static void free_rq_tio(struct dm_rq_target_io *tio)
456{
457 mempool_free(tio, tio->md->tio_pool);
458}
459
Kiyoshi Ueda90abb8c2009-12-10 23:52:13 +0000460static int md_in_flight(struct mapped_device *md)
461{
462 return atomic_read(&md->pending[READ]) +
463 atomic_read(&md->pending[WRITE]);
464}
465
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800466static void start_io_acct(struct dm_io *io)
467{
468 struct mapped_device *md = io->md;
Tejun Heoc9959052008-08-25 19:47:21 +0900469 int cpu;
Nikanth Karthikesan316d3152009-10-06 20:16:55 +0200470 int rw = bio_data_dir(io->bio);
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800471
472 io->start_time = jiffies;
473
Tejun Heo074a7ac2008-08-25 19:56:14 +0900474 cpu = part_stat_lock();
475 part_round_stats(cpu, &dm_disk(md)->part0);
476 part_stat_unlock();
Shaohua Li1e9bb882011-03-22 08:35:35 +0100477 atomic_set(&dm_disk(md)->part0.in_flight[rw],
478 atomic_inc_return(&md->pending[rw]));
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800479}
480
Mikulas Patockad221d2e2008-11-13 23:39:10 +0000481static void end_io_acct(struct dm_io *io)
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800482{
483 struct mapped_device *md = io->md;
484 struct bio *bio = io->bio;
485 unsigned long duration = jiffies - io->start_time;
Tejun Heoc9959052008-08-25 19:47:21 +0900486 int pending, cpu;
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800487 int rw = bio_data_dir(bio);
488
Tejun Heo074a7ac2008-08-25 19:56:14 +0900489 cpu = part_stat_lock();
490 part_round_stats(cpu, &dm_disk(md)->part0);
491 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
492 part_stat_unlock();
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800493
Mikulas Patockaaf7e4662009-04-09 00:27:16 +0100494 /*
495 * After this is decremented the bio must not be touched if it is
Tejun Heod87f4c12010-09-03 11:56:19 +0200496 * a flush.
Mikulas Patockaaf7e4662009-04-09 00:27:16 +0100497 */
Shaohua Li1e9bb882011-03-22 08:35:35 +0100498 pending = atomic_dec_return(&md->pending[rw]);
499 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +0200500 pending += atomic_read(&md->pending[rw^0x1]);
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800501
Mikulas Patockad221d2e2008-11-13 23:39:10 +0000502 /* nudge anyone waiting on suspend queue */
503 if (!pending)
504 wake_up(&md->wait);
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800505}
506
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507/*
508 * Add the bio to the list of deferred io.
509 */
Mikulas Patocka92c63902009-04-09 00:27:15 +0100510static void queue_io(struct mapped_device *md, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511{
Kiyoshi Ueda054474202010-09-08 18:07:01 +0200512 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Kiyoshi Ueda054474202010-09-08 18:07:01 +0200514 spin_lock_irqsave(&md->deferred_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 bio_list_add(&md->deferred, bio);
Kiyoshi Ueda054474202010-09-08 18:07:01 +0200516 spin_unlock_irqrestore(&md->deferred_lock, flags);
Tejun Heo6a8736d2010-09-08 18:07:00 +0200517 queue_work(md->wq, &md->work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518}
519
520/*
521 * Everyone (including functions in this file), should use this
522 * function to access the md->map field, and make sure they call
523 * dm_table_put() when finished.
524 */
Alasdair G Kergon7c666412009-12-10 23:52:19 +0000525struct dm_table *dm_get_live_table(struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526{
527 struct dm_table *t;
Kiyoshi Ueda523d9292009-06-22 10:12:37 +0100528 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529
Kiyoshi Ueda523d9292009-06-22 10:12:37 +0100530 read_lock_irqsave(&md->map_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 t = md->map;
532 if (t)
533 dm_table_get(t);
Kiyoshi Ueda523d9292009-06-22 10:12:37 +0100534 read_unlock_irqrestore(&md->map_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535
536 return t;
537}
538
Darrick J. Wong3ac51e72006-03-27 01:17:54 -0800539/*
540 * Get the geometry associated with a dm device
541 */
542int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
543{
544 *geo = md->geometry;
545
546 return 0;
547}
548
549/*
550 * Set the geometry of a device.
551 */
552int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
553{
554 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
555
556 if (geo->start > sz) {
557 DMWARN("Start sector is beyond the geometry limits.");
558 return -EINVAL;
559 }
560
561 md->geometry = *geo;
562
563 return 0;
564}
565
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566/*-----------------------------------------------------------------
567 * CRUD START:
568 * A more elegant soln is in the works that uses the queue
569 * merge fn, unfortunately there are a couple of changes to
570 * the block layer that I want to make for this. So in the
571 * interests of getting something for people to use I give
572 * you this clearly demarcated crap.
573 *---------------------------------------------------------------*/
574
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800575static int __noflush_suspending(struct mapped_device *md)
576{
577 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
578}
579
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580/*
581 * Decrements the number of outstanding ios that a bio has been
582 * cloned into, completing the original io if necc.
583 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800584static void dec_pending(struct dm_io *io, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585{
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800586 unsigned long flags;
Milan Brozb35f8ca2009-03-16 17:44:36 +0000587 int io_error;
588 struct bio *bio;
589 struct mapped_device *md = io->md;
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800590
591 /* Push-back supersedes any I/O errors */
Kiyoshi Uedaf88fb982009-10-16 23:18:15 +0100592 if (unlikely(error)) {
593 spin_lock_irqsave(&io->endio_lock, flags);
594 if (!(io->error > 0 && __noflush_suspending(md)))
595 io->error = error;
596 spin_unlock_irqrestore(&io->endio_lock, flags);
597 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
599 if (atomic_dec_and_test(&io->io_count)) {
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800600 if (io->error == DM_ENDIO_REQUEUE) {
601 /*
602 * Target requested pushing back the I/O.
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800603 */
Mikulas Patocka022c2612009-04-02 19:55:39 +0100604 spin_lock_irqsave(&md->deferred_lock, flags);
Tejun Heo6a8736d2010-09-08 18:07:00 +0200605 if (__noflush_suspending(md))
606 bio_list_add_head(&md->deferred, io->bio);
607 else
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800608 /* noflush suspend was interrupted. */
609 io->error = -EIO;
Mikulas Patocka022c2612009-04-02 19:55:39 +0100610 spin_unlock_irqrestore(&md->deferred_lock, flags);
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800611 }
612
Milan Brozb35f8ca2009-03-16 17:44:36 +0000613 io_error = io->error;
614 bio = io->bio;
Tejun Heo6a8736d2010-09-08 18:07:00 +0200615 end_io_acct(io);
616 free_io(md, io);
Jens Axboe2056a782006-03-23 20:00:26 +0100617
Tejun Heo6a8736d2010-09-08 18:07:00 +0200618 if (io_error == DM_ENDIO_REQUEUE)
619 return;
620
Mike Snitzerb372d362010-09-08 18:07:01 +0200621 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) {
Mikulas Patockaaf7e4662009-04-09 00:27:16 +0100622 /*
Tejun Heo6a8736d2010-09-08 18:07:00 +0200623 * Preflush done for flush with data, reissue
624 * without REQ_FLUSH.
Mikulas Patockaaf7e4662009-04-09 00:27:16 +0100625 */
Tejun Heo6a8736d2010-09-08 18:07:00 +0200626 bio->bi_rw &= ~REQ_FLUSH;
627 queue_io(md, bio);
Mikulas Patockaaf7e4662009-04-09 00:27:16 +0100628 } else {
Mike Snitzerb372d362010-09-08 18:07:01 +0200629 /* done with normal IO or empty flush */
Jeff Moyerb7908c12011-01-06 20:41:42 +0100630 trace_block_bio_complete(md->queue, bio, io_error);
Mike Snitzerb372d362010-09-08 18:07:01 +0200631 bio_endio(bio, io_error);
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800632 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 }
634}
635
NeilBrown6712ecf2007-09-27 12:47:43 +0200636static void clone_endio(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637{
638 int r = 0;
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100639 struct dm_target_io *tio = bio->bi_private;
Milan Brozb35f8ca2009-03-16 17:44:36 +0000640 struct dm_io *io = tio->io;
Stefan Bader9faf4002006-10-03 01:15:41 -0700641 struct mapped_device *md = tio->io->md;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 dm_endio_fn endio = tio->ti->type->end_io;
643
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
645 error = -EIO;
646
647 if (endio) {
Mikulas Patocka7de3ee52012-12-21 20:23:41 +0000648 r = endio(tio->ti, bio, error);
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800649 if (r < 0 || r == DM_ENDIO_REQUEUE)
650 /*
651 * error and requeue request are handled
652 * in dec_pending().
653 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 error = r;
Kiyoshi Ueda45cbcd72006-12-08 02:41:05 -0800655 else if (r == DM_ENDIO_INCOMPLETE)
656 /* The target will handle the io */
NeilBrown6712ecf2007-09-27 12:47:43 +0200657 return;
Kiyoshi Ueda45cbcd72006-12-08 02:41:05 -0800658 else if (r) {
659 DMWARN("unimplemented target endio return value: %d", r);
660 BUG();
661 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 }
663
Stefan Bader9faf4002006-10-03 01:15:41 -0700664 free_tio(md, tio);
Milan Brozb35f8ca2009-03-16 17:44:36 +0000665 dec_pending(io, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666}
667
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100668/*
669 * Partial completion handling for request-based dm
670 */
671static void end_clone_bio(struct bio *clone, int error)
672{
673 struct dm_rq_clone_bio_info *info = clone->bi_private;
674 struct dm_rq_target_io *tio = info->tio;
675 struct bio *bio = info->orig;
676 unsigned int nr_bytes = info->orig->bi_size;
677
678 bio_put(clone);
679
680 if (tio->error)
681 /*
682 * An error has already been detected on the request.
683 * Once error occurred, just let clone->end_io() handle
684 * the remainder.
685 */
686 return;
687 else if (error) {
688 /*
689 * Don't notice the error to the upper layer yet.
690 * The error handling decision is made by the target driver,
691 * when the request is completed.
692 */
693 tio->error = error;
694 return;
695 }
696
697 /*
698 * I/O for the bio successfully completed.
699 * Notice the data completion to the upper layer.
700 */
701
702 /*
703 * bios are processed from the head of the list.
704 * So the completing bio should always be rq->bio.
705 * If it's not, something wrong is happening.
706 */
707 if (tio->orig->bio != bio)
708 DMERR("bio completion is going in the middle of the request");
709
710 /*
711 * Update the original request.
712 * Do not use blk_end_request() here, because it may complete
713 * the original request before the clone, and break the ordering.
714 */
715 blk_update_request(tio->orig, 0, nr_bytes);
716}
717
718/*
719 * Don't touch any member of the md after calling this function because
720 * the md may be freed in dm_put() at the end of this function.
721 * Or do dm_get() before calling this function and dm_put() later.
722 */
Kiyoshi Uedab4324fe2009-12-10 23:52:16 +0000723static void rq_completed(struct mapped_device *md, int rw, int run_queue)
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100724{
Kiyoshi Uedab4324fe2009-12-10 23:52:16 +0000725 atomic_dec(&md->pending[rw]);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100726
727 /* nudge anyone waiting on suspend queue */
Kiyoshi Uedab4324fe2009-12-10 23:52:16 +0000728 if (!md_in_flight(md))
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100729 wake_up(&md->wait);
730
Jens Axboea8c32a52012-11-06 12:24:26 +0100731 /*
732 * Run this off this callpath, as drivers could invoke end_io while
733 * inside their request_fn (and holding the queue lock). Calling
734 * back into ->request_fn() could deadlock attempting to grab the
735 * queue lock again.
736 */
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100737 if (run_queue)
Jens Axboea8c32a52012-11-06 12:24:26 +0100738 blk_run_queue_async(md->queue);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100739
740 /*
741 * dm_put() must be at the end of this function. See the comment above
742 */
743 dm_put(md);
744}
745
Kiyoshi Uedaa77e28c2009-09-04 20:40:16 +0100746static void free_rq_clone(struct request *clone)
747{
748 struct dm_rq_target_io *tio = clone->end_io_data;
749
750 blk_rq_unprep_clone(clone);
751 free_rq_tio(tio);
752}
753
Kiyoshi Ueda980691e2009-12-10 23:52:17 +0000754/*
755 * Complete the clone and the original request.
756 * Must be called without queue lock.
757 */
758static void dm_end_request(struct request *clone, int error)
759{
760 int rw = rq_data_dir(clone);
761 struct dm_rq_target_io *tio = clone->end_io_data;
762 struct mapped_device *md = tio->md;
763 struct request *rq = tio->orig;
764
Tejun Heo29e40132010-09-08 18:07:00 +0200765 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Kiyoshi Ueda980691e2009-12-10 23:52:17 +0000766 rq->errors = clone->errors;
767 rq->resid_len = clone->resid_len;
768
769 if (rq->sense)
770 /*
771 * We are using the sense buffer of the original
772 * request.
773 * So setting the length of the sense data is enough.
774 */
775 rq->sense_len = clone->sense_len;
776 }
777
778 free_rq_clone(clone);
Tejun Heo29e40132010-09-08 18:07:00 +0200779 blk_end_request_all(rq, error);
780 rq_completed(md, rw, true);
Kiyoshi Ueda980691e2009-12-10 23:52:17 +0000781}
782
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100783static void dm_unprep_request(struct request *rq)
784{
785 struct request *clone = rq->special;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100786
787 rq->special = NULL;
788 rq->cmd_flags &= ~REQ_DONTPREP;
789
Kiyoshi Uedaa77e28c2009-09-04 20:40:16 +0100790 free_rq_clone(clone);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100791}
792
793/*
794 * Requeue the original request of a clone.
795 */
796void dm_requeue_unmapped_request(struct request *clone)
797{
Kiyoshi Uedab4324fe2009-12-10 23:52:16 +0000798 int rw = rq_data_dir(clone);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100799 struct dm_rq_target_io *tio = clone->end_io_data;
800 struct mapped_device *md = tio->md;
801 struct request *rq = tio->orig;
802 struct request_queue *q = rq->q;
803 unsigned long flags;
804
805 dm_unprep_request(rq);
806
807 spin_lock_irqsave(q->queue_lock, flags);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100808 blk_requeue_request(q, rq);
809 spin_unlock_irqrestore(q->queue_lock, flags);
810
Kiyoshi Uedab4324fe2009-12-10 23:52:16 +0000811 rq_completed(md, rw, 0);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100812}
813EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
814
815static void __stop_queue(struct request_queue *q)
816{
817 blk_stop_queue(q);
818}
819
820static void stop_queue(struct request_queue *q)
821{
822 unsigned long flags;
823
824 spin_lock_irqsave(q->queue_lock, flags);
825 __stop_queue(q);
826 spin_unlock_irqrestore(q->queue_lock, flags);
827}
828
829static void __start_queue(struct request_queue *q)
830{
831 if (blk_queue_stopped(q))
832 blk_start_queue(q);
833}
834
835static void start_queue(struct request_queue *q)
836{
837 unsigned long flags;
838
839 spin_lock_irqsave(q->queue_lock, flags);
840 __start_queue(q);
841 spin_unlock_irqrestore(q->queue_lock, flags);
842}
843
Kiyoshi Ueda11a68242009-12-10 23:52:17 +0000844static void dm_done(struct request *clone, int error, bool mapped)
845{
846 int r = error;
847 struct dm_rq_target_io *tio = clone->end_io_data;
Mike Snitzerba1cbad2012-09-26 23:45:42 +0100848 dm_request_endio_fn rq_end_io = NULL;
Kiyoshi Ueda11a68242009-12-10 23:52:17 +0000849
Mike Snitzerba1cbad2012-09-26 23:45:42 +0100850 if (tio->ti) {
851 rq_end_io = tio->ti->type->rq_end_io;
852
853 if (mapped && rq_end_io)
854 r = rq_end_io(tio->ti, clone, error, &tio->info);
855 }
Kiyoshi Ueda11a68242009-12-10 23:52:17 +0000856
857 if (r <= 0)
858 /* The target wants to complete the I/O */
859 dm_end_request(clone, r);
860 else if (r == DM_ENDIO_INCOMPLETE)
861 /* The target will handle the I/O */
862 return;
863 else if (r == DM_ENDIO_REQUEUE)
864 /* The target wants to requeue the I/O */
865 dm_requeue_unmapped_request(clone);
866 else {
867 DMWARN("unimplemented target endio return value: %d", r);
868 BUG();
869 }
870}
871
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100872/*
873 * Request completion handler for request-based dm
874 */
875static void dm_softirq_done(struct request *rq)
876{
Kiyoshi Ueda11a68242009-12-10 23:52:17 +0000877 bool mapped = true;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100878 struct request *clone = rq->completion_data;
879 struct dm_rq_target_io *tio = clone->end_io_data;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100880
Kiyoshi Ueda11a68242009-12-10 23:52:17 +0000881 if (rq->cmd_flags & REQ_FAILED)
882 mapped = false;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100883
Kiyoshi Ueda11a68242009-12-10 23:52:17 +0000884 dm_done(clone, tio->error, mapped);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100885}
886
887/*
888 * Complete the clone and the original request with the error status
889 * through softirq context.
890 */
891static void dm_complete_request(struct request *clone, int error)
892{
893 struct dm_rq_target_io *tio = clone->end_io_data;
894 struct request *rq = tio->orig;
895
896 tio->error = error;
897 rq->completion_data = clone;
898 blk_complete_request(rq);
899}
900
901/*
902 * Complete the not-mapped clone and the original request with the error status
903 * through softirq context.
904 * Target's rq_end_io() function isn't called.
905 * This may be used when the target's map_rq() function fails.
906 */
907void dm_kill_unmapped_request(struct request *clone, int error)
908{
909 struct dm_rq_target_io *tio = clone->end_io_data;
910 struct request *rq = tio->orig;
911
912 rq->cmd_flags |= REQ_FAILED;
913 dm_complete_request(clone, error);
914}
915EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
916
917/*
918 * Called with the queue lock held
919 */
920static void end_clone_request(struct request *clone, int error)
921{
922 /*
923 * For just cleaning up the information of the queue in which
924 * the clone was dispatched.
925 * The clone is *NOT* freed actually here because it is alloced from
926 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
927 */
928 __blk_put_request(clone->q, clone);
929
930 /*
931 * Actual request completion is done in a softirq context which doesn't
932 * hold the queue lock. Otherwise, deadlock could occur because:
933 * - another request may be submitted by the upper level driver
934 * of the stacking during the completion
935 * - the submission which requires queue lock may be done
936 * against this queue
937 */
938 dm_complete_request(clone, error);
939}
940
Mike Snitzer56a67df2010-08-12 04:14:10 +0100941/*
942 * Return maximum size of I/O possible at the supplied sector up to the current
943 * target boundary.
944 */
945static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946{
Mike Snitzer56a67df2010-08-12 04:14:10 +0100947 sector_t target_offset = dm_target_offset(ti, sector);
948
949 return ti->len - target_offset;
950}
951
952static sector_t max_io_len(sector_t sector, struct dm_target *ti)
953{
954 sector_t len = max_io_len_target_boundary(sector, ti);
Mike Snitzer542f9032012-07-27 15:08:00 +0100955 sector_t offset, max_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956
957 /*
Mike Snitzer542f9032012-07-27 15:08:00 +0100958 * Does the target need to split even further?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 */
Mike Snitzer542f9032012-07-27 15:08:00 +0100960 if (ti->max_io_len) {
961 offset = dm_target_offset(ti, sector);
962 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
963 max_len = sector_div(offset, ti->max_io_len);
964 else
965 max_len = offset & (ti->max_io_len - 1);
966 max_len = ti->max_io_len - max_len;
967
968 if (len > max_len)
969 len = max_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 }
971
972 return len;
973}
974
Mike Snitzer542f9032012-07-27 15:08:00 +0100975int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
976{
977 if (len > UINT_MAX) {
978 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
979 (unsigned long long)len, UINT_MAX);
980 ti->error = "Maximum size of target IO is too large";
981 return -EINVAL;
982 }
983
984 ti->max_io_len = (uint32_t) len;
985
986 return 0;
987}
988EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
989
Mikulas Patockadba14162012-10-12 21:02:15 +0100990static void __map_bio(struct dm_target *ti, struct dm_target_io *tio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991{
992 int r;
Jens Axboe2056a782006-03-23 20:00:26 +0100993 sector_t sector;
Stefan Bader9faf4002006-10-03 01:15:41 -0700994 struct mapped_device *md;
Mikulas Patockadba14162012-10-12 21:02:15 +0100995 struct bio *clone = &tio->clone;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 clone->bi_end_io = clone_endio;
998 clone->bi_private = tio;
999
1000 /*
1001 * Map the clone. If r == 0 we don't need to do
1002 * anything, the target has assumed ownership of
1003 * this io.
1004 */
1005 atomic_inc(&tio->io->io_count);
Jens Axboe2056a782006-03-23 20:00:26 +01001006 sector = clone->bi_sector;
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00001007 r = ti->type->map(ti, clone);
Kiyoshi Ueda45cbcd72006-12-08 02:41:05 -08001008 if (r == DM_MAPIO_REMAPPED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 /* the bio has been remapped so dispatch it */
Jens Axboe2056a782006-03-23 20:00:26 +01001010
Mike Snitzerd07335e2010-11-16 12:52:38 +01001011 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1012 tio->io->bio->bi_bdev->bd_dev, sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001013
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 generic_make_request(clone);
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -08001015 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1016 /* error the io and bail out, or requeue it if needed */
Stefan Bader9faf4002006-10-03 01:15:41 -07001017 md = tio->io->md;
1018 dec_pending(tio->io, r);
Stefan Bader9faf4002006-10-03 01:15:41 -07001019 free_tio(md, tio);
Kiyoshi Ueda45cbcd72006-12-08 02:41:05 -08001020 } else if (r) {
1021 DMWARN("unimplemented target map return value: %d", r);
1022 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 }
1024}
1025
1026struct clone_info {
1027 struct mapped_device *md;
1028 struct dm_table *map;
1029 struct bio *bio;
1030 struct dm_io *io;
1031 sector_t sector;
1032 sector_t sector_count;
1033 unsigned short idx;
1034};
1035
1036/*
Tejun Heod87f4c12010-09-03 11:56:19 +02001037 * Creates a little bio that just does part of a bvec.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 */
Mikulas Patockadba14162012-10-12 21:02:15 +01001039static void split_bvec(struct dm_target_io *tio, struct bio *bio,
1040 sector_t sector, unsigned short idx, unsigned int offset,
1041 unsigned int len, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042{
Mikulas Patockadba14162012-10-12 21:02:15 +01001043 struct bio *clone = &tio->clone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 struct bio_vec *bv = bio->bi_io_vec + idx;
1045
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 *clone->bi_io_vec = *bv;
1047
1048 clone->bi_sector = sector;
1049 clone->bi_bdev = bio->bi_bdev;
Tejun Heod87f4c12010-09-03 11:56:19 +02001050 clone->bi_rw = bio->bi_rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 clone->bi_vcnt = 1;
1052 clone->bi_size = to_bytes(len);
1053 clone->bi_io_vec->bv_offset = offset;
1054 clone->bi_io_vec->bv_len = clone->bi_size;
Martin K. Petersenf3e1d262008-10-21 17:45:04 +01001055 clone->bi_flags |= 1 << BIO_CLONED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056
Martin K. Petersen9c470082009-04-09 00:27:12 +01001057 if (bio_integrity(bio)) {
Kent Overstreet1e2a410f2012-09-06 15:34:56 -07001058 bio_integrity_clone(clone, bio, GFP_NOIO);
Martin K. Petersen9c470082009-04-09 00:27:12 +01001059 bio_integrity_trim(clone,
1060 bio_sector_offset(bio, idx, offset), len);
1061 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062}
1063
1064/*
1065 * Creates a bio that consists of range of complete bvecs.
1066 */
Mikulas Patockadba14162012-10-12 21:02:15 +01001067static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1068 sector_t sector, unsigned short idx,
1069 unsigned short bv_count, unsigned int len,
1070 struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071{
Mikulas Patockadba14162012-10-12 21:02:15 +01001072 struct bio *clone = &tio->clone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073
Stefan Bader9faf4002006-10-03 01:15:41 -07001074 __bio_clone(clone, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 clone->bi_sector = sector;
1076 clone->bi_idx = idx;
1077 clone->bi_vcnt = idx + bv_count;
1078 clone->bi_size = to_bytes(len);
1079 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1080
Martin K. Petersen9c470082009-04-09 00:27:12 +01001081 if (bio_integrity(bio)) {
Kent Overstreet1e2a410f2012-09-06 15:34:56 -07001082 bio_integrity_clone(clone, bio, GFP_NOIO);
Martin K. Petersen9c470082009-04-09 00:27:12 +01001083
1084 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1085 bio_integrity_trim(clone,
1086 bio_sector_offset(bio, idx, 0), len);
1087 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088}
1089
Alasdair G Kergon9015df22009-06-22 10:12:21 +01001090static struct dm_target_io *alloc_tio(struct clone_info *ci,
Mikulas Patockadba14162012-10-12 21:02:15 +01001091 struct dm_target *ti, int nr_iovecs)
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001092{
Mikulas Patockadba14162012-10-12 21:02:15 +01001093 struct dm_target_io *tio;
1094 struct bio *clone;
1095
1096 clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs);
1097 tio = container_of(clone, struct dm_target_io, clone);
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001098
1099 tio->io = ci->io;
1100 tio->ti = ti;
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001101 memset(&tio->info, 0, sizeof(tio->info));
Mikulas Patockaddbd6582012-12-21 20:23:39 +00001102 tio->target_request_nr = 0;
Alasdair G Kergon9015df22009-06-22 10:12:21 +01001103
1104 return tio;
1105}
1106
Mike Snitzer06a426c2010-08-12 04:14:09 +01001107static void __issue_target_request(struct clone_info *ci, struct dm_target *ti,
Mike Snitzera79245b2010-08-12 04:14:24 +01001108 unsigned request_nr, sector_t len)
Alasdair G Kergon9015df22009-06-22 10:12:21 +01001109{
Mikulas Patockadba14162012-10-12 21:02:15 +01001110 struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs);
1111 struct bio *clone = &tio->clone;
Alasdair G Kergon9015df22009-06-22 10:12:21 +01001112
Mikulas Patockaddbd6582012-12-21 20:23:39 +00001113 tio->target_request_nr = request_nr;
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001114
Mike Snitzer06a426c2010-08-12 04:14:09 +01001115 /*
1116 * Discard requests require the bio's inline iovecs be initialized.
1117 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1118 * and discard, so no need for concern about wasted bvec allocations.
1119 */
Kent Overstreetbf800ef2012-09-06 15:35:02 -07001120
Mikulas Patockadba14162012-10-12 21:02:15 +01001121 __bio_clone(clone, ci->bio);
Mike Snitzera79245b2010-08-12 04:14:24 +01001122 if (len) {
1123 clone->bi_sector = ci->sector;
1124 clone->bi_size = to_bytes(len);
1125 }
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001126
Mikulas Patockadba14162012-10-12 21:02:15 +01001127 __map_bio(ti, tio);
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001128}
1129
Mike Snitzer06a426c2010-08-12 04:14:09 +01001130static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti,
Mike Snitzera79245b2010-08-12 04:14:24 +01001131 unsigned num_requests, sector_t len)
Mike Snitzer06a426c2010-08-12 04:14:09 +01001132{
1133 unsigned request_nr;
1134
1135 for (request_nr = 0; request_nr < num_requests; request_nr++)
Mike Snitzera79245b2010-08-12 04:14:24 +01001136 __issue_target_request(ci, ti, request_nr, len);
Mike Snitzer06a426c2010-08-12 04:14:09 +01001137}
1138
Mike Snitzerb372d362010-09-08 18:07:01 +02001139static int __clone_and_map_empty_flush(struct clone_info *ci)
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001140{
Mike Snitzer06a426c2010-08-12 04:14:09 +01001141 unsigned target_nr = 0;
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001142 struct dm_target *ti;
1143
Mike Snitzerb372d362010-09-08 18:07:01 +02001144 BUG_ON(bio_has_data(ci->bio));
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001145 while ((ti = dm_table_get_target(ci->map, target_nr++)))
Mike Snitzera79245b2010-08-12 04:14:24 +01001146 __issue_target_requests(ci, ti, ti->num_flush_requests, 0);
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001147
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001148 return 0;
1149}
1150
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001151/*
1152 * Perform all io with a single clone.
1153 */
1154static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti)
1155{
Mikulas Patockadba14162012-10-12 21:02:15 +01001156 struct bio *bio = ci->bio;
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001157 struct dm_target_io *tio;
1158
Mikulas Patockadba14162012-10-12 21:02:15 +01001159 tio = alloc_tio(ci, ti, bio->bi_max_vecs);
1160 clone_bio(tio, bio, ci->sector, ci->idx, bio->bi_vcnt - ci->idx,
1161 ci->sector_count, ci->md->bs);
1162 __map_bio(ti, tio);
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001163 ci->sector_count = 0;
1164}
1165
Mike Snitzer23508a92012-12-21 20:23:37 +00001166typedef unsigned (*get_num_requests_fn)(struct dm_target *ti);
1167
1168static unsigned get_num_discard_requests(struct dm_target *ti)
1169{
1170 return ti->num_discard_requests;
1171}
1172
1173static unsigned get_num_write_same_requests(struct dm_target *ti)
1174{
1175 return ti->num_write_same_requests;
1176}
1177
1178typedef bool (*is_split_required_fn)(struct dm_target *ti);
1179
1180static bool is_split_required_for_discard(struct dm_target *ti)
1181{
1182 return ti->split_discard_requests;
1183}
1184
1185static int __clone_and_map_changing_extent_only(struct clone_info *ci,
1186 get_num_requests_fn get_num_requests,
1187 is_split_required_fn is_split_required)
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001188{
1189 struct dm_target *ti;
Mike Snitzera79245b2010-08-12 04:14:24 +01001190 sector_t len;
Alasdair G Kergonfe7af2d2013-01-31 14:23:36 +00001191 unsigned num_requests;
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001192
Mike Snitzera79245b2010-08-12 04:14:24 +01001193 do {
1194 ti = dm_table_find_target(ci->map, ci->sector);
1195 if (!dm_target_is_valid(ti))
1196 return -EIO;
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001197
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001198 /*
Mike Snitzer23508a92012-12-21 20:23:37 +00001199 * Even though the device advertised support for this type of
1200 * request, that does not mean every target supports it, and
Mike Snitzer936688d2011-08-02 12:32:01 +01001201 * reconfiguration might also have changed that since the
Mike Snitzera79245b2010-08-12 04:14:24 +01001202 * check was performed.
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001203 */
Alasdair G Kergonfe7af2d2013-01-31 14:23:36 +00001204 num_requests = get_num_requests ? get_num_requests(ti) : 0;
1205 if (!num_requests)
Mike Snitzera79245b2010-08-12 04:14:24 +01001206 return -EOPNOTSUPP;
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001207
Mike Snitzer23508a92012-12-21 20:23:37 +00001208 if (is_split_required && !is_split_required(ti))
Mikulas Patocka7acf0272012-07-27 15:08:03 +01001209 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1210 else
1211 len = min(ci->sector_count, max_io_len(ci->sector, ti));
Mike Snitzer06a426c2010-08-12 04:14:09 +01001212
Alasdair G Kergonfe7af2d2013-01-31 14:23:36 +00001213 __issue_target_requests(ci, ti, num_requests, len);
Mike Snitzera79245b2010-08-12 04:14:24 +01001214
1215 ci->sector += len;
1216 } while (ci->sector_count -= len);
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001217
1218 return 0;
1219}
1220
Mike Snitzer23508a92012-12-21 20:23:37 +00001221static int __clone_and_map_discard(struct clone_info *ci)
1222{
1223 return __clone_and_map_changing_extent_only(ci, get_num_discard_requests,
1224 is_split_required_for_discard);
1225}
1226
1227static int __clone_and_map_write_same(struct clone_info *ci)
1228{
1229 return __clone_and_map_changing_extent_only(ci, get_num_write_same_requests, NULL);
1230}
1231
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001232static int __clone_and_map(struct clone_info *ci)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233{
Mikulas Patockadba14162012-10-12 21:02:15 +01001234 struct bio *bio = ci->bio;
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001235 struct dm_target *ti;
1236 sector_t len = 0, max;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001237 struct dm_target_io *tio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001239 if (unlikely(bio->bi_rw & REQ_DISCARD))
1240 return __clone_and_map_discard(ci);
Mike Snitzer23508a92012-12-21 20:23:37 +00001241 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1242 return __clone_and_map_write_same(ci);
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001243
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001244 ti = dm_table_find_target(ci->map, ci->sector);
1245 if (!dm_target_is_valid(ti))
1246 return -EIO;
1247
Mike Snitzer56a67df2010-08-12 04:14:10 +01001248 max = max_io_len(ci->sector, ti);
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001249
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 if (ci->sector_count <= max) {
1251 /*
1252 * Optimise for the simple case where we can do all of
1253 * the remaining io with a single clone.
1254 */
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001255 __clone_and_map_simple(ci, ti);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256
1257 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1258 /*
1259 * There are some bvecs that don't span targets.
1260 * Do as many of these as possible.
1261 */
1262 int i;
1263 sector_t remaining = max;
1264 sector_t bv_len;
1265
1266 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1267 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1268
1269 if (bv_len > remaining)
1270 break;
1271
1272 remaining -= bv_len;
1273 len += bv_len;
1274 }
1275
Mikulas Patockadba14162012-10-12 21:02:15 +01001276 tio = alloc_tio(ci, ti, bio->bi_max_vecs);
1277 clone_bio(tio, bio, ci->sector, ci->idx, i - ci->idx, len,
1278 ci->md->bs);
1279 __map_bio(ti, tio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280
1281 ci->sector += len;
1282 ci->sector_count -= len;
1283 ci->idx = i;
1284
1285 } else {
1286 /*
Alasdair G Kergond2044a92006-03-22 00:07:42 -08001287 * Handle a bvec that must be split between two or more targets.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 */
1289 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
Alasdair G Kergond2044a92006-03-22 00:07:42 -08001290 sector_t remaining = to_sector(bv->bv_len);
1291 unsigned int offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292
Alasdair G Kergond2044a92006-03-22 00:07:42 -08001293 do {
1294 if (offset) {
1295 ti = dm_table_find_target(ci->map, ci->sector);
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001296 if (!dm_target_is_valid(ti))
1297 return -EIO;
1298
Mike Snitzer56a67df2010-08-12 04:14:10 +01001299 max = max_io_len(ci->sector, ti);
Alasdair G Kergond2044a92006-03-22 00:07:42 -08001300 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301
Alasdair G Kergond2044a92006-03-22 00:07:42 -08001302 len = min(remaining, max);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
Mikulas Patockadba14162012-10-12 21:02:15 +01001304 tio = alloc_tio(ci, ti, 1);
1305 split_bvec(tio, bio, ci->sector, ci->idx,
1306 bv->bv_offset + offset, len, ci->md->bs);
Alasdair G Kergond2044a92006-03-22 00:07:42 -08001307
Mikulas Patockadba14162012-10-12 21:02:15 +01001308 __map_bio(ti, tio);
Alasdair G Kergond2044a92006-03-22 00:07:42 -08001309
1310 ci->sector += len;
1311 ci->sector_count -= len;
1312 offset += to_bytes(len);
1313 } while (remaining -= len);
1314
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 ci->idx++;
1316 }
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001317
1318 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319}
1320
1321/*
Mikulas Patocka8a53c282009-04-02 19:55:37 +01001322 * Split the bio into several clones and submit it to targets.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 */
Mikulas Patockaf0b9a452009-04-02 19:55:38 +01001324static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325{
1326 struct clone_info ci;
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001327 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328
Alasdair G Kergon7c666412009-12-10 23:52:19 +00001329 ci.map = dm_get_live_table(md);
Mikulas Patockaf0b9a452009-04-02 19:55:38 +01001330 if (unlikely(!ci.map)) {
Tejun Heo6a8736d2010-09-08 18:07:00 +02001331 bio_io_error(bio);
Mikulas Patockaf0b9a452009-04-02 19:55:38 +01001332 return;
1333 }
Mikulas Patocka692d0eb2009-04-09 00:27:13 +01001334
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 ci.md = md;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 ci.io = alloc_io(md);
1337 ci.io->error = 0;
1338 atomic_set(&ci.io->io_count, 1);
1339 ci.io->bio = bio;
1340 ci.io->md = md;
Kiyoshi Uedaf88fb982009-10-16 23:18:15 +01001341 spin_lock_init(&ci.io->endio_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 ci.sector = bio->bi_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 ci.idx = bio->bi_idx;
1344
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -08001345 start_io_acct(ci.io);
Mike Snitzerb372d362010-09-08 18:07:01 +02001346 if (bio->bi_rw & REQ_FLUSH) {
1347 ci.bio = &ci.md->flush_bio;
1348 ci.sector_count = 0;
1349 error = __clone_and_map_empty_flush(&ci);
1350 /* dec_pending submits any data associated with flush */
1351 } else {
Tejun Heo6a8736d2010-09-08 18:07:00 +02001352 ci.bio = bio;
Tejun Heod87f4c12010-09-03 11:56:19 +02001353 ci.sector_count = bio_sectors(bio);
Mike Snitzerb372d362010-09-08 18:07:01 +02001354 while (ci.sector_count && !error)
Tejun Heod87f4c12010-09-03 11:56:19 +02001355 error = __clone_and_map(&ci);
Tejun Heod87f4c12010-09-03 11:56:19 +02001356 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
1358 /* drop the extra reference count */
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001359 dec_pending(ci.io, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 dm_table_put(ci.map);
1361}
1362/*-----------------------------------------------------------------
1363 * CRUD END
1364 *---------------------------------------------------------------*/
1365
Milan Brozf6fccb12008-07-21 12:00:37 +01001366static int dm_merge_bvec(struct request_queue *q,
1367 struct bvec_merge_data *bvm,
1368 struct bio_vec *biovec)
1369{
1370 struct mapped_device *md = q->queuedata;
Alasdair G Kergon7c666412009-12-10 23:52:19 +00001371 struct dm_table *map = dm_get_live_table(md);
Milan Brozf6fccb12008-07-21 12:00:37 +01001372 struct dm_target *ti;
1373 sector_t max_sectors;
Mikulas Patocka50371082008-10-01 14:39:17 +01001374 int max_size = 0;
Milan Brozf6fccb12008-07-21 12:00:37 +01001375
1376 if (unlikely(!map))
Mikulas Patocka50371082008-10-01 14:39:17 +01001377 goto out;
Milan Brozf6fccb12008-07-21 12:00:37 +01001378
1379 ti = dm_table_find_target(map, bvm->bi_sector);
Mikulas Patockab01cd5a2008-10-01 14:39:24 +01001380 if (!dm_target_is_valid(ti))
1381 goto out_table;
Milan Brozf6fccb12008-07-21 12:00:37 +01001382
1383 /*
1384 * Find maximum amount of I/O that won't need splitting
1385 */
Mike Snitzer56a67df2010-08-12 04:14:10 +01001386 max_sectors = min(max_io_len(bvm->bi_sector, ti),
Milan Brozf6fccb12008-07-21 12:00:37 +01001387 (sector_t) BIO_MAX_SECTORS);
1388 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1389 if (max_size < 0)
1390 max_size = 0;
1391
1392 /*
1393 * merge_bvec_fn() returns number of bytes
1394 * it can accept at this offset
1395 * max is precomputed maximal io size
1396 */
1397 if (max_size && ti->type->merge)
1398 max_size = ti->type->merge(ti, bvm, biovec, max_size);
Mikulas Patocka8cbeb672009-06-22 10:12:14 +01001399 /*
1400 * If the target doesn't support merge method and some of the devices
1401 * provided their merge_bvec method (we know this by looking at
1402 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1403 * entries. So always set max_size to 0, and the code below allows
1404 * just one page.
1405 */
1406 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1407
1408 max_size = 0;
Milan Brozf6fccb12008-07-21 12:00:37 +01001409
Mikulas Patockab01cd5a2008-10-01 14:39:24 +01001410out_table:
Mikulas Patocka50371082008-10-01 14:39:17 +01001411 dm_table_put(map);
1412
1413out:
Milan Brozf6fccb12008-07-21 12:00:37 +01001414 /*
1415 * Always allow an entire first page
1416 */
1417 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1418 max_size = biovec->bv_len;
1419
Milan Brozf6fccb12008-07-21 12:00:37 +01001420 return max_size;
1421}
1422
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423/*
1424 * The request function that just remaps the bio built up by
1425 * dm_merge_bvec.
1426 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001427static void _dm_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428{
Kevin Corry12f03a42006-02-01 03:04:52 -08001429 int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 struct mapped_device *md = q->queuedata;
Tejun Heoc9959052008-08-25 19:47:21 +09001431 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001433 down_read(&md->io_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434
Tejun Heo074a7ac2008-08-25 19:56:14 +09001435 cpu = part_stat_lock();
1436 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1437 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1438 part_stat_unlock();
Kevin Corry12f03a42006-02-01 03:04:52 -08001439
Tejun Heo6a8736d2010-09-08 18:07:00 +02001440 /* if we're suspended, we have to queue this io for later */
1441 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001442 up_read(&md->io_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443
Tejun Heo6a8736d2010-09-08 18:07:00 +02001444 if (bio_rw(bio) != READA)
1445 queue_io(md, bio);
1446 else
Alasdair G Kergon54d9a1b2009-04-09 00:27:14 +01001447 bio_io_error(bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001448 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 }
1450
Mikulas Patockaf0b9a452009-04-02 19:55:38 +01001451 __split_and_process_bio(md, bio);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001452 up_read(&md->io_lock);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001453 return;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001454}
1455
1456static int dm_request_based(struct mapped_device *md)
1457{
1458 return blk_queue_stackable(md->queue);
1459}
1460
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001461static void dm_request(struct request_queue *q, struct bio *bio)
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001462{
1463 struct mapped_device *md = q->queuedata;
1464
1465 if (dm_request_based(md))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001466 blk_queue_bio(q, bio);
1467 else
1468 _dm_request(q, bio);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001469}
1470
1471void dm_dispatch_request(struct request *rq)
1472{
1473 int r;
1474
1475 if (blk_queue_io_stat(rq->q))
1476 rq->cmd_flags |= REQ_IO_STAT;
1477
1478 rq->start_time = jiffies;
1479 r = blk_insert_cloned_request(rq->q, rq);
1480 if (r)
1481 dm_complete_request(rq, r);
1482}
1483EXPORT_SYMBOL_GPL(dm_dispatch_request);
1484
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001485static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1486 void *data)
1487{
1488 struct dm_rq_target_io *tio = data;
Kent Overstreet94818742012-09-07 13:44:01 -07001489 struct dm_rq_clone_bio_info *info =
1490 container_of(bio, struct dm_rq_clone_bio_info, clone);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001491
1492 info->orig = bio_orig;
1493 info->tio = tio;
1494 bio->bi_end_io = end_clone_bio;
1495 bio->bi_private = info;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001496
1497 return 0;
1498}
1499
1500static int setup_clone(struct request *clone, struct request *rq,
1501 struct dm_rq_target_io *tio)
1502{
Kiyoshi Uedad0bcb872009-12-10 23:52:18 +00001503 int r;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001504
Tejun Heo29e40132010-09-08 18:07:00 +02001505 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1506 dm_rq_bio_constructor, tio);
1507 if (r)
1508 return r;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001509
Tejun Heo29e40132010-09-08 18:07:00 +02001510 clone->cmd = rq->cmd;
1511 clone->cmd_len = rq->cmd_len;
1512 clone->sense = rq->sense;
1513 clone->buffer = rq->buffer;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001514 clone->end_io = end_clone_request;
1515 clone->end_io_data = tio;
1516
1517 return 0;
1518}
1519
Kiyoshi Ueda6facdaf2009-12-10 23:52:15 +00001520static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1521 gfp_t gfp_mask)
1522{
1523 struct request *clone;
1524 struct dm_rq_target_io *tio;
1525
1526 tio = alloc_rq_tio(md, gfp_mask);
1527 if (!tio)
1528 return NULL;
1529
1530 tio->md = md;
1531 tio->ti = NULL;
1532 tio->orig = rq;
1533 tio->error = 0;
1534 memset(&tio->info, 0, sizeof(tio->info));
1535
1536 clone = &tio->clone;
1537 if (setup_clone(clone, rq, tio)) {
1538 /* -ENOMEM */
1539 free_rq_tio(tio);
1540 return NULL;
1541 }
1542
1543 return clone;
1544}
1545
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001546/*
1547 * Called with the queue lock held.
1548 */
1549static int dm_prep_fn(struct request_queue *q, struct request *rq)
1550{
1551 struct mapped_device *md = q->queuedata;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001552 struct request *clone;
1553
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001554 if (unlikely(rq->special)) {
1555 DMWARN("Already has something in rq->special.");
1556 return BLKPREP_KILL;
1557 }
1558
Kiyoshi Ueda6facdaf2009-12-10 23:52:15 +00001559 clone = clone_rq(rq, md, GFP_ATOMIC);
1560 if (!clone)
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001561 return BLKPREP_DEFER;
1562
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001563 rq->special = clone;
1564 rq->cmd_flags |= REQ_DONTPREP;
1565
1566 return BLKPREP_OK;
1567}
1568
Kiyoshi Ueda9eef87d2010-02-16 18:43:01 +00001569/*
1570 * Returns:
1571 * 0 : the request has been processed (not requeued)
1572 * !0 : the request has been requeued
1573 */
1574static int map_request(struct dm_target *ti, struct request *clone,
1575 struct mapped_device *md)
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001576{
Kiyoshi Ueda9eef87d2010-02-16 18:43:01 +00001577 int r, requeued = 0;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001578 struct dm_rq_target_io *tio = clone->end_io_data;
1579
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001580 tio->ti = ti;
1581 r = ti->type->map_rq(ti, clone, &tio->info);
1582 switch (r) {
1583 case DM_MAPIO_SUBMITTED:
1584 /* The target has taken the I/O to submit by itself later */
1585 break;
1586 case DM_MAPIO_REMAPPED:
1587 /* The target has remapped the I/O so dispatch it */
Jun'ichi Nomura6db4ccd2009-12-10 23:52:25 +00001588 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1589 blk_rq_pos(tio->orig));
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001590 dm_dispatch_request(clone);
1591 break;
1592 case DM_MAPIO_REQUEUE:
1593 /* The target wants to requeue the I/O */
1594 dm_requeue_unmapped_request(clone);
Kiyoshi Ueda9eef87d2010-02-16 18:43:01 +00001595 requeued = 1;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001596 break;
1597 default:
1598 if (r > 0) {
1599 DMWARN("unimplemented target map return value: %d", r);
1600 BUG();
1601 }
1602
1603 /* The target wants to complete the I/O */
1604 dm_kill_unmapped_request(clone, r);
1605 break;
1606 }
Kiyoshi Ueda9eef87d2010-02-16 18:43:01 +00001607
1608 return requeued;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001609}
1610
Mike Snitzerba1cbad2012-09-26 23:45:42 +01001611static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1612{
1613 struct request *clone;
1614
1615 blk_start_request(orig);
1616 clone = orig->special;
1617 atomic_inc(&md->pending[rq_data_dir(clone)]);
1618
1619 /*
1620 * Hold the md reference here for the in-flight I/O.
1621 * We can't rely on the reference count by device opener,
1622 * because the device may be closed during the request completion
1623 * when all bios are completed.
1624 * See the comment in rq_completed() too.
1625 */
1626 dm_get(md);
1627
1628 return clone;
1629}
1630
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001631/*
1632 * q->request_fn for request-based dm.
1633 * Called with the queue lock held.
1634 */
1635static void dm_request_fn(struct request_queue *q)
1636{
1637 struct mapped_device *md = q->queuedata;
Alasdair G Kergon7c666412009-12-10 23:52:19 +00001638 struct dm_table *map = dm_get_live_table(md);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001639 struct dm_target *ti;
Kiyoshi Uedab4324fe2009-12-10 23:52:16 +00001640 struct request *rq, *clone;
Tejun Heo29e40132010-09-08 18:07:00 +02001641 sector_t pos;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001642
1643 /*
Kiyoshi Uedab4324fe2009-12-10 23:52:16 +00001644 * For suspend, check blk_queue_stopped() and increment
1645 * ->pending within a single queue_lock not to increment the
1646 * number of in-flight I/Os after the queue is stopped in
1647 * dm_suspend().
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001648 */
Jens Axboe7eaceac2011-03-10 08:52:07 +01001649 while (!blk_queue_stopped(q)) {
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001650 rq = blk_peek_request(q);
1651 if (!rq)
Jens Axboe7eaceac2011-03-10 08:52:07 +01001652 goto delay_and_out;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001653
Tejun Heo29e40132010-09-08 18:07:00 +02001654 /* always use block 0 to find the target for flushes for now */
1655 pos = 0;
1656 if (!(rq->cmd_flags & REQ_FLUSH))
1657 pos = blk_rq_pos(rq);
Kiyoshi Uedad0bcb872009-12-10 23:52:18 +00001658
Tejun Heo29e40132010-09-08 18:07:00 +02001659 ti = dm_table_find_target(map, pos);
Mike Snitzerba1cbad2012-09-26 23:45:42 +01001660 if (!dm_target_is_valid(ti)) {
1661 /*
1662 * Must perform setup, that dm_done() requires,
1663 * before calling dm_kill_unmapped_request
1664 */
1665 DMERR_LIMIT("request attempted access beyond the end of device");
1666 clone = dm_start_request(md, rq);
1667 dm_kill_unmapped_request(clone, -EIO);
1668 continue;
1669 }
Tejun Heo29e40132010-09-08 18:07:00 +02001670
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001671 if (ti->type->busy && ti->type->busy(ti))
Jens Axboe7eaceac2011-03-10 08:52:07 +01001672 goto delay_and_out;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001673
Mike Snitzerba1cbad2012-09-26 23:45:42 +01001674 clone = dm_start_request(md, rq);
Kiyoshi Uedab4324fe2009-12-10 23:52:16 +00001675
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001676 spin_unlock(q->queue_lock);
Kiyoshi Ueda9eef87d2010-02-16 18:43:01 +00001677 if (map_request(ti, clone, md))
1678 goto requeued;
1679
Kiyoshi Ueda052189a2011-01-13 20:00:00 +00001680 BUG_ON(!irqs_disabled());
1681 spin_lock(q->queue_lock);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001682 }
1683
1684 goto out;
1685
Kiyoshi Ueda9eef87d2010-02-16 18:43:01 +00001686requeued:
Kiyoshi Ueda052189a2011-01-13 20:00:00 +00001687 BUG_ON(!irqs_disabled());
1688 spin_lock(q->queue_lock);
Kiyoshi Ueda9eef87d2010-02-16 18:43:01 +00001689
Jens Axboe7eaceac2011-03-10 08:52:07 +01001690delay_and_out:
1691 blk_delay_queue(q, HZ / 10);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001692out:
1693 dm_table_put(map);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001694}
1695
1696int dm_underlying_device_busy(struct request_queue *q)
1697{
1698 return blk_lld_busy(q);
1699}
1700EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1701
1702static int dm_lld_busy(struct request_queue *q)
1703{
1704 int r;
1705 struct mapped_device *md = q->queuedata;
Alasdair G Kergon7c666412009-12-10 23:52:19 +00001706 struct dm_table *map = dm_get_live_table(md);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001707
1708 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1709 r = 1;
1710 else
1711 r = dm_table_any_busy_target(map);
1712
1713 dm_table_put(map);
1714
1715 return r;
1716}
1717
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718static int dm_any_congested(void *congested_data, int bdi_bits)
1719{
Chandra Seetharaman8a57dfc2008-11-13 23:39:14 +00001720 int r = bdi_bits;
1721 struct mapped_device *md = congested_data;
1722 struct dm_table *map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723
Alasdair G Kergon1eb787e2009-04-09 00:27:14 +01001724 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
Alasdair G Kergon7c666412009-12-10 23:52:19 +00001725 map = dm_get_live_table(md);
Chandra Seetharaman8a57dfc2008-11-13 23:39:14 +00001726 if (map) {
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001727 /*
1728 * Request-based dm cares about only own queue for
1729 * the query about congestion status of request_queue
1730 */
1731 if (dm_request_based(md))
1732 r = md->queue->backing_dev_info.state &
1733 bdi_bits;
1734 else
1735 r = dm_table_any_congested(map, bdi_bits);
1736
Chandra Seetharaman8a57dfc2008-11-13 23:39:14 +00001737 dm_table_put(map);
1738 }
1739 }
1740
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 return r;
1742}
1743
1744/*-----------------------------------------------------------------
1745 * An IDR is used to keep track of allocated minor numbers.
1746 *---------------------------------------------------------------*/
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07001747static void free_minor(int minor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748{
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001749 spin_lock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 idr_remove(&_minor_idr, minor);
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001751 spin_unlock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752}
1753
1754/*
1755 * See if the device with a specific minor # is free.
1756 */
Frederik Deweerdtcf13ab82008-04-24 22:10:59 +01001757static int specific_minor(int minor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758{
1759 int r, m;
1760
1761 if (minor >= (1 << MINORBITS))
1762 return -EINVAL;
1763
Jeff Mahoney62f75c22006-06-26 00:27:21 -07001764 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1765 if (!r)
1766 return -ENOMEM;
1767
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001768 spin_lock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
1770 if (idr_find(&_minor_idr, minor)) {
1771 r = -EBUSY;
1772 goto out;
1773 }
1774
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -07001775 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
Jeff Mahoney62f75c22006-06-26 00:27:21 -07001776 if (r)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778
1779 if (m != minor) {
1780 idr_remove(&_minor_idr, m);
1781 r = -EBUSY;
1782 goto out;
1783 }
1784
1785out:
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001786 spin_unlock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 return r;
1788}
1789
Frederik Deweerdtcf13ab82008-04-24 22:10:59 +01001790static int next_free_minor(int *minor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791{
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07001792 int r, m;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
Jeff Mahoney62f75c22006-06-26 00:27:21 -07001795 if (!r)
1796 return -ENOMEM;
1797
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001798 spin_lock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -07001800 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
Frederik Deweerdtcf13ab82008-04-24 22:10:59 +01001801 if (r)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803
1804 if (m >= (1 << MINORBITS)) {
1805 idr_remove(&_minor_idr, m);
1806 r = -ENOSPC;
1807 goto out;
1808 }
1809
1810 *minor = m;
1811
1812out:
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001813 spin_unlock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 return r;
1815}
1816
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001817static const struct block_device_operations dm_blk_dops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818
Mikulas Patocka53d59142009-04-02 19:55:37 +01001819static void dm_wq_work(struct work_struct *work);
1820
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +01001821static void dm_init_md_queue(struct mapped_device *md)
1822{
1823 /*
1824 * Request-based dm devices cannot be stacked on top of bio-based dm
1825 * devices. The type of this dm device has not been decided yet.
1826 * The type is decided at the first table loading time.
1827 * To prevent problematic device stacking, clear the queue flag
1828 * for request stacking support until then.
1829 *
1830 * This queue is new, so no concurrency on the queue_flags.
1831 */
1832 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1833
1834 md->queue->queuedata = md;
1835 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1836 md->queue->backing_dev_info.congested_data = md;
1837 blk_queue_make_request(md->queue, dm_request);
1838 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +01001839 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1840}
1841
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842/*
1843 * Allocate and initialise a blank device with a given minor.
1844 */
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07001845static struct mapped_device *alloc_dev(int minor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846{
1847 int r;
Frederik Deweerdtcf13ab82008-04-24 22:10:59 +01001848 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -07001849 void *old_md;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850
1851 if (!md) {
1852 DMWARN("unable to allocate device, out of memory.");
1853 return NULL;
1854 }
1855
Jeff Mahoney10da4f72006-06-26 00:27:25 -07001856 if (!try_module_get(THIS_MODULE))
Milan Broz6ed7ade2008-02-08 02:10:19 +00001857 goto bad_module_get;
Jeff Mahoney10da4f72006-06-26 00:27:25 -07001858
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 /* get a minor number for the dev */
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07001860 if (minor == DM_ANY_MINOR)
Frederik Deweerdtcf13ab82008-04-24 22:10:59 +01001861 r = next_free_minor(&minor);
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07001862 else
Frederik Deweerdtcf13ab82008-04-24 22:10:59 +01001863 r = specific_minor(minor);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 if (r < 0)
Milan Broz6ed7ade2008-02-08 02:10:19 +00001865 goto bad_minor;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866
Mike Snitzera5664da2010-08-12 04:14:01 +01001867 md->type = DM_TYPE_NONE;
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001868 init_rwsem(&md->io_lock);
Daniel Walkere61290a2008-02-08 02:10:08 +00001869 mutex_init(&md->suspend_lock);
Mike Snitzera5664da2010-08-12 04:14:01 +01001870 mutex_init(&md->type_lock);
Mikulas Patocka022c2612009-04-02 19:55:39 +01001871 spin_lock_init(&md->deferred_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 rwlock_init(&md->map_lock);
1873 atomic_set(&md->holders, 1);
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -07001874 atomic_set(&md->open_count, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 atomic_set(&md->event_nr, 0);
Mike Anderson7a8c3d32007-10-19 22:48:01 +01001876 atomic_set(&md->uevent_seq, 0);
1877 INIT_LIST_HEAD(&md->uevent_list);
1878 spin_lock_init(&md->uevent_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +01001880 md->queue = blk_alloc_queue(GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 if (!md->queue)
Milan Broz6ed7ade2008-02-08 02:10:19 +00001882 goto bad_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +01001884 dm_init_md_queue(md);
Stefan Bader9faf4002006-10-03 01:15:41 -07001885
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 md->disk = alloc_disk(1);
1887 if (!md->disk)
Milan Broz6ed7ade2008-02-08 02:10:19 +00001888 goto bad_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001890 atomic_set(&md->pending[0], 0);
1891 atomic_set(&md->pending[1], 0);
Jeff Mahoneyf0b04112006-06-26 00:27:25 -07001892 init_waitqueue_head(&md->wait);
Mikulas Patocka53d59142009-04-02 19:55:37 +01001893 INIT_WORK(&md->work, dm_wq_work);
Jeff Mahoneyf0b04112006-06-26 00:27:25 -07001894 init_waitqueue_head(&md->eventq);
1895
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 md->disk->major = _major;
1897 md->disk->first_minor = minor;
1898 md->disk->fops = &dm_blk_dops;
1899 md->disk->queue = md->queue;
1900 md->disk->private_data = md;
1901 sprintf(md->disk->disk_name, "dm-%d", minor);
1902 add_disk(md->disk);
Mike Anderson7e51f252006-03-27 01:17:52 -08001903 format_dev_t(md->name, MKDEV(_major, minor));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904
Tejun Heo9c4376d2011-01-13 19:59:58 +00001905 md->wq = alloc_workqueue("kdmflush",
1906 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
Milan Broz304f3f62008-02-08 02:11:17 +00001907 if (!md->wq)
1908 goto bad_thread;
1909
Mikulas Patocka32a926d2009-06-22 10:12:17 +01001910 md->bdev = bdget_disk(md->disk, 0);
1911 if (!md->bdev)
1912 goto bad_bdev;
1913
Tejun Heo6a8736d2010-09-08 18:07:00 +02001914 bio_init(&md->flush_bio);
1915 md->flush_bio.bi_bdev = md->bdev;
1916 md->flush_bio.bi_rw = WRITE_FLUSH;
1917
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -07001918 /* Populate the mapping, nobody knows we exist yet */
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001919 spin_lock(&_minor_lock);
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -07001920 old_md = idr_replace(&_minor_idr, md, minor);
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001921 spin_unlock(&_minor_lock);
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -07001922
1923 BUG_ON(old_md != MINOR_ALLOCED);
1924
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 return md;
1926
Mikulas Patocka32a926d2009-06-22 10:12:17 +01001927bad_bdev:
1928 destroy_workqueue(md->wq);
Milan Broz304f3f62008-02-08 02:11:17 +00001929bad_thread:
Zdenek Kabelac03022c52009-10-16 23:18:15 +01001930 del_gendisk(md->disk);
Milan Broz304f3f62008-02-08 02:11:17 +00001931 put_disk(md->disk);
Milan Broz6ed7ade2008-02-08 02:10:19 +00001932bad_disk:
Al Viro1312f402006-03-12 11:02:03 -05001933 blk_cleanup_queue(md->queue);
Milan Broz6ed7ade2008-02-08 02:10:19 +00001934bad_queue:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 free_minor(minor);
Milan Broz6ed7ade2008-02-08 02:10:19 +00001936bad_minor:
Jeff Mahoney10da4f72006-06-26 00:27:25 -07001937 module_put(THIS_MODULE);
Milan Broz6ed7ade2008-02-08 02:10:19 +00001938bad_module_get:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 kfree(md);
1940 return NULL;
1941}
1942
Jun'ichi Nomuraae9da832007-10-19 22:38:43 +01001943static void unlock_fs(struct mapped_device *md);
1944
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945static void free_dev(struct mapped_device *md)
1946{
Tejun Heof331c022008-09-03 09:01:48 +02001947 int minor = MINOR(disk_devt(md->disk));
Jun'ichi Nomura63d94e42006-02-24 13:04:25 -08001948
Mikulas Patocka32a926d2009-06-22 10:12:17 +01001949 unlock_fs(md);
1950 bdput(md->bdev);
Milan Broz304f3f62008-02-08 02:11:17 +00001951 destroy_workqueue(md->wq);
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001952 if (md->tio_pool)
1953 mempool_destroy(md->tio_pool);
1954 if (md->io_pool)
1955 mempool_destroy(md->io_pool);
1956 if (md->bs)
1957 bioset_free(md->bs);
Martin K. Petersen9c470082009-04-09 00:27:12 +01001958 blk_integrity_unregister(md->disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 del_gendisk(md->disk);
Jun'ichi Nomura63d94e42006-02-24 13:04:25 -08001960 free_minor(minor);
Jeff Mahoneyfba9f902006-06-26 00:27:23 -07001961
1962 spin_lock(&_minor_lock);
1963 md->disk->private_data = NULL;
1964 spin_unlock(&_minor_lock);
1965
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 put_disk(md->disk);
Al Viro1312f402006-03-12 11:02:03 -05001967 blk_cleanup_queue(md->queue);
Jeff Mahoney10da4f72006-06-26 00:27:25 -07001968 module_put(THIS_MODULE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 kfree(md);
1970}
1971
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001972static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1973{
Mikulas Patockac0820cf2012-12-21 20:23:38 +00001974 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001975
Mikulas Patockac0820cf2012-12-21 20:23:38 +00001976 if (md->io_pool && (md->tio_pool || dm_table_get_type(t) == DM_TYPE_BIO_BASED) && md->bs) {
1977 /*
1978 * The md already has necessary mempools. Reload just the
1979 * bioset because front_pad may have changed because
1980 * a different table was loaded.
1981 */
1982 bioset_free(md->bs);
1983 md->bs = p->bs;
1984 p->bs = NULL;
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001985 goto out;
Mikulas Patockac0820cf2012-12-21 20:23:38 +00001986 }
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001987
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001988 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
1989
1990 md->io_pool = p->io_pool;
1991 p->io_pool = NULL;
1992 md->tio_pool = p->tio_pool;
1993 p->tio_pool = NULL;
1994 md->bs = p->bs;
1995 p->bs = NULL;
1996
1997out:
1998 /* mempool bind completed, now no need any mempools in the table */
1999 dm_table_free_md_mempools(t);
2000}
2001
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002/*
2003 * Bind a table to the device.
2004 */
2005static void event_callback(void *context)
2006{
Mike Anderson7a8c3d32007-10-19 22:48:01 +01002007 unsigned long flags;
2008 LIST_HEAD(uevents);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 struct mapped_device *md = (struct mapped_device *) context;
2010
Mike Anderson7a8c3d32007-10-19 22:48:01 +01002011 spin_lock_irqsave(&md->uevent_lock, flags);
2012 list_splice_init(&md->uevent_list, &uevents);
2013 spin_unlock_irqrestore(&md->uevent_lock, flags);
2014
Tejun Heoed9e1982008-08-25 19:56:05 +09002015 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
Mike Anderson7a8c3d32007-10-19 22:48:01 +01002016
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 atomic_inc(&md->event_nr);
2018 wake_up(&md->eventq);
2019}
2020
Mike Snitzerc2176492011-01-13 19:53:46 +00002021/*
2022 * Protected by md->suspend_lock obtained by dm_swap_table().
2023 */
Alasdair G Kergon4e90188be2005-07-28 21:15:59 -07002024static void __set_size(struct mapped_device *md, sector_t size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025{
Alasdair G Kergon4e90188be2005-07-28 21:15:59 -07002026 set_capacity(md->disk, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027
Mikulas Patockadb8fef42009-06-22 10:12:15 +01002028 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029}
2030
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002031/*
Mikulas Patockad5b9dd02011-08-02 12:32:04 +01002032 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2033 *
2034 * If this function returns 0, then the device is either a non-dm
2035 * device without a merge_bvec_fn, or it is a dm device that is
2036 * able to split any bios it receives that are too big.
2037 */
2038int dm_queue_merge_is_compulsory(struct request_queue *q)
2039{
2040 struct mapped_device *dev_md;
2041
2042 if (!q->merge_bvec_fn)
2043 return 0;
2044
2045 if (q->make_request_fn == dm_request) {
2046 dev_md = q->queuedata;
2047 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2048 return 0;
2049 }
2050
2051 return 1;
2052}
2053
2054static int dm_device_merge_is_compulsory(struct dm_target *ti,
2055 struct dm_dev *dev, sector_t start,
2056 sector_t len, void *data)
2057{
2058 struct block_device *bdev = dev->bdev;
2059 struct request_queue *q = bdev_get_queue(bdev);
2060
2061 return dm_queue_merge_is_compulsory(q);
2062}
2063
2064/*
2065 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2066 * on the properties of the underlying devices.
2067 */
2068static int dm_table_merge_is_optional(struct dm_table *table)
2069{
2070 unsigned i = 0;
2071 struct dm_target *ti;
2072
2073 while (i < dm_table_get_num_targets(table)) {
2074 ti = dm_table_get_target(table, i++);
2075
2076 if (ti->type->iterate_devices &&
2077 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2078 return 0;
2079 }
2080
2081 return 1;
2082}
2083
2084/*
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002085 * Returns old map, which caller must destroy.
2086 */
2087static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2088 struct queue_limits *limits)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089{
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002090 struct dm_table *old_map;
Jens Axboe165125e2007-07-24 09:28:11 +02002091 struct request_queue *q = md->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 sector_t size;
Kiyoshi Ueda523d9292009-06-22 10:12:37 +01002093 unsigned long flags;
Mikulas Patockad5b9dd02011-08-02 12:32:04 +01002094 int merge_is_optional;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095
2096 size = dm_table_get_size(t);
Darrick J. Wong3ac51e72006-03-27 01:17:54 -08002097
2098 /*
2099 * Wipe any geometry if the size of the table changed.
2100 */
2101 if (size != get_capacity(md->disk))
2102 memset(&md->geometry, 0, sizeof(md->geometry));
2103
Mikulas Patocka32a926d2009-06-22 10:12:17 +01002104 __set_size(md, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002106 dm_table_event_callback(t, event_callback, md);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002107
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002108 /*
2109 * The queue hasn't been stopped yet, if the old table type wasn't
2110 * for request-based during suspension. So stop it to prevent
2111 * I/O mapping before resume.
2112 * This must be done before setting the queue restrictions,
2113 * because request-based dm may be run just after the setting.
2114 */
2115 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2116 stop_queue(q);
2117
2118 __bind_mempools(md, t);
2119
Mikulas Patockad5b9dd02011-08-02 12:32:04 +01002120 merge_is_optional = dm_table_merge_is_optional(t);
2121
Kiyoshi Ueda523d9292009-06-22 10:12:37 +01002122 write_lock_irqsave(&md->map_lock, flags);
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002123 old_map = md->map;
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002124 md->map = t;
Alasdair G Kergon36a04562011-10-31 20:19:04 +00002125 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2126
Mike Snitzer754c5fc2009-06-22 10:12:34 +01002127 dm_table_set_restrictions(t, q, limits);
Mikulas Patockad5b9dd02011-08-02 12:32:04 +01002128 if (merge_is_optional)
2129 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2130 else
2131 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
Kiyoshi Ueda523d9292009-06-22 10:12:37 +01002132 write_unlock_irqrestore(&md->map_lock, flags);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002133
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002134 return old_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135}
2136
Alasdair G Kergona7940152009-12-10 23:52:23 +00002137/*
2138 * Returns unbound table for the caller to free.
2139 */
2140static struct dm_table *__unbind(struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141{
2142 struct dm_table *map = md->map;
Kiyoshi Ueda523d9292009-06-22 10:12:37 +01002143 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144
2145 if (!map)
Alasdair G Kergona7940152009-12-10 23:52:23 +00002146 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147
2148 dm_table_event_callback(map, NULL, NULL);
Kiyoshi Ueda523d9292009-06-22 10:12:37 +01002149 write_lock_irqsave(&md->map_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 md->map = NULL;
Kiyoshi Ueda523d9292009-06-22 10:12:37 +01002151 write_unlock_irqrestore(&md->map_lock, flags);
Alasdair G Kergona7940152009-12-10 23:52:23 +00002152
2153 return map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154}
2155
2156/*
2157 * Constructor for a new device.
2158 */
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07002159int dm_create(int minor, struct mapped_device **result)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160{
2161 struct mapped_device *md;
2162
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07002163 md = alloc_dev(minor);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 if (!md)
2165 return -ENXIO;
2166
Milan Broz784aae72009-01-06 03:05:12 +00002167 dm_sysfs_init(md);
2168
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 *result = md;
2170 return 0;
2171}
2172
Mike Snitzera5664da2010-08-12 04:14:01 +01002173/*
2174 * Functions to manage md->type.
2175 * All are required to hold md->type_lock.
2176 */
2177void dm_lock_md_type(struct mapped_device *md)
2178{
2179 mutex_lock(&md->type_lock);
2180}
2181
2182void dm_unlock_md_type(struct mapped_device *md)
2183{
2184 mutex_unlock(&md->type_lock);
2185}
2186
2187void dm_set_md_type(struct mapped_device *md, unsigned type)
2188{
2189 md->type = type;
2190}
2191
2192unsigned dm_get_md_type(struct mapped_device *md)
2193{
2194 return md->type;
2195}
2196
Alasdair G Kergon36a04562011-10-31 20:19:04 +00002197struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2198{
2199 return md->immutable_target_type;
2200}
2201
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +01002202/*
2203 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2204 */
2205static int dm_init_request_based_queue(struct mapped_device *md)
2206{
2207 struct request_queue *q = NULL;
2208
2209 if (md->queue->elevator)
2210 return 1;
2211
2212 /* Fully initialize the queue */
2213 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2214 if (!q)
2215 return 0;
2216
2217 md->queue = q;
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +01002218 dm_init_md_queue(md);
2219 blk_queue_softirq_done(md->queue, dm_softirq_done);
2220 blk_queue_prep_rq(md->queue, dm_prep_fn);
2221 blk_queue_lld_busy(md->queue, dm_lld_busy);
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +01002222
2223 elv_register_queue(md->queue);
2224
2225 return 1;
2226}
2227
2228/*
2229 * Setup the DM device's queue based on md's type
2230 */
2231int dm_setup_md_queue(struct mapped_device *md)
2232{
2233 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2234 !dm_init_request_based_queue(md)) {
2235 DMWARN("Cannot initialize queue for request-based mapped device");
2236 return -EINVAL;
2237 }
2238
2239 return 0;
2240}
2241
David Teigland637842c2006-01-06 00:20:00 -08002242static struct mapped_device *dm_find_md(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243{
2244 struct mapped_device *md;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 unsigned minor = MINOR(dev);
2246
2247 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2248 return NULL;
2249
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07002250 spin_lock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251
2252 md = idr_find(&_minor_idr, minor);
Jeff Mahoneyfba9f902006-06-26 00:27:23 -07002253 if (md && (md == MINOR_ALLOCED ||
Tejun Heof331c022008-09-03 09:01:48 +02002254 (MINOR(disk_devt(dm_disk(md))) != minor) ||
Kiyoshi Uedaabdc5682010-08-12 04:13:54 +01002255 dm_deleting_md(md) ||
Alasdair G Kergon17b2f662006-06-26 00:27:33 -07002256 test_bit(DMF_FREEING, &md->flags))) {
David Teigland637842c2006-01-06 00:20:00 -08002257 md = NULL;
Jeff Mahoneyfba9f902006-06-26 00:27:23 -07002258 goto out;
2259 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260
Jeff Mahoneyfba9f902006-06-26 00:27:23 -07002261out:
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07002262 spin_unlock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263
David Teigland637842c2006-01-06 00:20:00 -08002264 return md;
2265}
2266
David Teiglandd229a952006-01-06 00:20:01 -08002267struct mapped_device *dm_get_md(dev_t dev)
2268{
2269 struct mapped_device *md = dm_find_md(dev);
2270
2271 if (md)
2272 dm_get(md);
2273
2274 return md;
2275}
Alasdair G Kergon3cf2e4b2011-10-31 20:19:06 +00002276EXPORT_SYMBOL_GPL(dm_get_md);
David Teiglandd229a952006-01-06 00:20:01 -08002277
Alasdair G Kergon9ade92a2006-03-27 01:17:53 -08002278void *dm_get_mdptr(struct mapped_device *md)
David Teigland637842c2006-01-06 00:20:00 -08002279{
Alasdair G Kergon9ade92a2006-03-27 01:17:53 -08002280 return md->interface_ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281}
2282
2283void dm_set_mdptr(struct mapped_device *md, void *ptr)
2284{
2285 md->interface_ptr = ptr;
2286}
2287
2288void dm_get(struct mapped_device *md)
2289{
2290 atomic_inc(&md->holders);
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +01002291 BUG_ON(test_bit(DMF_FREEING, &md->flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292}
2293
Alasdair G Kergon72d94862006-06-26 00:27:35 -07002294const char *dm_device_name(struct mapped_device *md)
2295{
2296 return md->name;
2297}
2298EXPORT_SYMBOL_GPL(dm_device_name);
2299
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +01002300static void __dm_destroy(struct mapped_device *md, bool wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301{
Mike Anderson1134e5ae2006-03-27 01:17:54 -08002302 struct dm_table *map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +01002304 might_sleep();
Jeff Mahoneyfba9f902006-06-26 00:27:23 -07002305
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +01002306 spin_lock(&_minor_lock);
2307 map = dm_get_live_table(md);
2308 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2309 set_bit(DMF_FREEING, &md->flags);
2310 spin_unlock(&_minor_lock);
2311
2312 if (!dm_suspended_md(md)) {
2313 dm_table_presuspend_targets(map);
2314 dm_table_postsuspend_targets(map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 }
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +01002316
2317 /*
2318 * Rare, but there may be I/O requests still going to complete,
2319 * for example. Wait for all references to disappear.
2320 * No one should increment the reference count of the mapped_device,
2321 * after the mapped_device state becomes DMF_FREEING.
2322 */
2323 if (wait)
2324 while (atomic_read(&md->holders))
2325 msleep(1);
2326 else if (atomic_read(&md->holders))
2327 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2328 dm_device_name(md), atomic_read(&md->holders));
2329
2330 dm_sysfs_exit(md);
2331 dm_table_put(map);
2332 dm_table_destroy(__unbind(md));
2333 free_dev(md);
2334}
2335
2336void dm_destroy(struct mapped_device *md)
2337{
2338 __dm_destroy(md, true);
2339}
2340
2341void dm_destroy_immediate(struct mapped_device *md)
2342{
2343 __dm_destroy(md, false);
2344}
2345
2346void dm_put(struct mapped_device *md)
2347{
2348 atomic_dec(&md->holders);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349}
Edward Goggin79eb8852007-05-09 02:32:56 -07002350EXPORT_SYMBOL_GPL(dm_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351
Mikulas Patocka401600d2009-04-02 19:55:38 +01002352static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
Milan Broz46125c12008-02-08 02:10:30 +00002353{
2354 int r = 0;
Mikulas Patockab44ebeb2009-04-02 19:55:39 +01002355 DECLARE_WAITQUEUE(wait, current);
2356
Mikulas Patockab44ebeb2009-04-02 19:55:39 +01002357 add_wait_queue(&md->wait, &wait);
Milan Broz46125c12008-02-08 02:10:30 +00002358
2359 while (1) {
Mikulas Patocka401600d2009-04-02 19:55:38 +01002360 set_current_state(interruptible);
Milan Broz46125c12008-02-08 02:10:30 +00002361
Kiyoshi Uedab4324fe2009-12-10 23:52:16 +00002362 if (!md_in_flight(md))
Milan Broz46125c12008-02-08 02:10:30 +00002363 break;
2364
Mikulas Patocka401600d2009-04-02 19:55:38 +01002365 if (interruptible == TASK_INTERRUPTIBLE &&
2366 signal_pending(current)) {
Milan Broz46125c12008-02-08 02:10:30 +00002367 r = -EINTR;
2368 break;
2369 }
2370
2371 io_schedule();
2372 }
2373 set_current_state(TASK_RUNNING);
2374
Mikulas Patockab44ebeb2009-04-02 19:55:39 +01002375 remove_wait_queue(&md->wait, &wait);
2376
Milan Broz46125c12008-02-08 02:10:30 +00002377 return r;
2378}
2379
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380/*
2381 * Process the deferred bios
2382 */
Mikulas Patockaef208582009-04-02 19:55:38 +01002383static void dm_wq_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384{
Mikulas Patockaef208582009-04-02 19:55:38 +01002385 struct mapped_device *md = container_of(work, struct mapped_device,
2386 work);
Milan Broz6d6f10d2008-02-08 02:10:22 +00002387 struct bio *c;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388
Tejun Heo6a8736d2010-09-08 18:07:00 +02002389 down_read(&md->io_lock);
Mikulas Patockaef208582009-04-02 19:55:38 +01002390
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002391 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
Alasdair G Kergondf12ee92009-04-09 00:27:13 +01002392 spin_lock_irq(&md->deferred_lock);
2393 c = bio_list_pop(&md->deferred);
2394 spin_unlock_irq(&md->deferred_lock);
Mikulas Patocka022c2612009-04-02 19:55:39 +01002395
Tejun Heo6a8736d2010-09-08 18:07:00 +02002396 if (!c)
Alasdair G Kergondf12ee92009-04-09 00:27:13 +01002397 break;
Alasdair G Kergondf12ee92009-04-09 00:27:13 +01002398
Tejun Heo6a8736d2010-09-08 18:07:00 +02002399 up_read(&md->io_lock);
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002400
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002401 if (dm_request_based(md))
2402 generic_make_request(c);
Tejun Heo6a8736d2010-09-08 18:07:00 +02002403 else
2404 __split_and_process_bio(md, c);
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002405
Tejun Heo6a8736d2010-09-08 18:07:00 +02002406 down_read(&md->io_lock);
Mikulas Patocka022c2612009-04-02 19:55:39 +01002407 }
Milan Broz73d410c2008-02-08 02:10:25 +00002408
Tejun Heo6a8736d2010-09-08 18:07:00 +02002409 up_read(&md->io_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410}
2411
Mikulas Patocka9a1fb462009-04-02 19:55:36 +01002412static void dm_queue_flush(struct mapped_device *md)
Milan Broz304f3f62008-02-08 02:11:17 +00002413{
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002414 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2415 smp_mb__after_clear_bit();
Mikulas Patocka53d59142009-04-02 19:55:37 +01002416 queue_work(md->wq, &md->work);
Milan Broz304f3f62008-02-08 02:11:17 +00002417}
2418
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419/*
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002420 * Swap in a new table, returning the old one for the caller to destroy.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 */
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002422struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423{
Mike Snitzer3ae70652012-09-26 23:45:45 +01002424 struct dm_table *live_map, *map = ERR_PTR(-EINVAL);
Mike Snitzer754c5fc2009-06-22 10:12:34 +01002425 struct queue_limits limits;
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002426 int r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427
Daniel Walkere61290a2008-02-08 02:10:08 +00002428 mutex_lock(&md->suspend_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429
2430 /* device must be suspended */
Kiyoshi Ueda4f186f82009-12-10 23:52:26 +00002431 if (!dm_suspended_md(md))
Alasdair G Kergon93c534a2005-07-12 15:53:05 -07002432 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433
Mike Snitzer3ae70652012-09-26 23:45:45 +01002434 /*
2435 * If the new table has no data devices, retain the existing limits.
2436 * This helps multipath with queue_if_no_path if all paths disappear,
2437 * then new I/O is queued based on these limits, and then some paths
2438 * reappear.
2439 */
2440 if (dm_table_has_no_data_devices(table)) {
2441 live_map = dm_get_live_table(md);
2442 if (live_map)
2443 limits = md->queue->limits;
2444 dm_table_put(live_map);
2445 }
2446
Mike Snitzer754c5fc2009-06-22 10:12:34 +01002447 r = dm_calculate_queue_limits(table, &limits);
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002448 if (r) {
2449 map = ERR_PTR(r);
Mike Snitzer754c5fc2009-06-22 10:12:34 +01002450 goto out;
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002451 }
Mike Snitzer754c5fc2009-06-22 10:12:34 +01002452
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002453 map = __bind(md, table, &limits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454
Alasdair G Kergon93c534a2005-07-12 15:53:05 -07002455out:
Daniel Walkere61290a2008-02-08 02:10:08 +00002456 mutex_unlock(&md->suspend_lock);
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002457 return map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458}
2459
2460/*
2461 * Functions to lock and unlock any filesystem running on the
2462 * device.
2463 */
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002464static int lock_fs(struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465{
Alasdair G Kergone39e2e92006-01-06 00:20:05 -08002466 int r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467
2468 WARN_ON(md->frozen_sb);
Alasdair G Kergondfbe03f2005-05-05 16:16:04 -07002469
Mikulas Patockadb8fef42009-06-22 10:12:15 +01002470 md->frozen_sb = freeze_bdev(md->bdev);
Alasdair G Kergondfbe03f2005-05-05 16:16:04 -07002471 if (IS_ERR(md->frozen_sb)) {
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002472 r = PTR_ERR(md->frozen_sb);
Alasdair G Kergone39e2e92006-01-06 00:20:05 -08002473 md->frozen_sb = NULL;
2474 return r;
Alasdair G Kergondfbe03f2005-05-05 16:16:04 -07002475 }
2476
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -08002477 set_bit(DMF_FROZEN, &md->flags);
2478
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 return 0;
2480}
2481
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002482static void unlock_fs(struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483{
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -08002484 if (!test_bit(DMF_FROZEN, &md->flags))
2485 return;
2486
Mikulas Patockadb8fef42009-06-22 10:12:15 +01002487 thaw_bdev(md->bdev, md->frozen_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 md->frozen_sb = NULL;
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -08002489 clear_bit(DMF_FROZEN, &md->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490}
2491
2492/*
2493 * We need to be able to change a mapping table under a mounted
2494 * filesystem. For example we might want to move some data in
2495 * the background. Before the table can be swapped with
2496 * dm_bind_table, dm_suspend must be called to flush any in
2497 * flight bios and ensure that any further io gets deferred.
2498 */
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002499/*
2500 * Suspend mechanism in request-based dm.
2501 *
Kiyoshi Ueda9f518b22009-12-10 23:52:16 +00002502 * 1. Flush all I/Os by lock_fs() if needed.
2503 * 2. Stop dispatching any I/O by stopping the request_queue.
2504 * 3. Wait for all in-flight I/Os to be completed or requeued.
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002505 *
Kiyoshi Ueda9f518b22009-12-10 23:52:16 +00002506 * To abort suspend, start the request_queue.
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002507 */
Kiyoshi Uedaa3d77d32006-12-08 02:41:04 -08002508int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509{
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002510 struct dm_table *map = NULL;
Milan Broz46125c12008-02-08 02:10:30 +00002511 int r = 0;
Kiyoshi Uedaa3d77d32006-12-08 02:41:04 -08002512 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -08002513 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514
Daniel Walkere61290a2008-02-08 02:10:08 +00002515 mutex_lock(&md->suspend_lock);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002516
Kiyoshi Ueda4f186f82009-12-10 23:52:26 +00002517 if (dm_suspended_md(md)) {
Milan Broz73d410c2008-02-08 02:10:25 +00002518 r = -EINVAL;
Alasdair G Kergond2874832006-11-08 17:44:43 -08002519 goto out_unlock;
Milan Broz73d410c2008-02-08 02:10:25 +00002520 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521
Alasdair G Kergon7c666412009-12-10 23:52:19 +00002522 map = dm_get_live_table(md);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -08002524 /*
2525 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2526 * This flag is cleared before dm_suspend returns.
2527 */
2528 if (noflush)
2529 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2530
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002531 /* This does not get reverted if there's an error later. */
2532 dm_table_presuspend_targets(map);
2533
Mikulas Patocka32a926d2009-06-22 10:12:17 +01002534 /*
Kiyoshi Ueda9f518b22009-12-10 23:52:16 +00002535 * Flush I/O to the device.
2536 * Any I/O submitted after lock_fs() may not be flushed.
2537 * noflush takes precedence over do_lockfs.
2538 * (lock_fs() flushes I/Os and waits for them to complete.)
Mikulas Patocka32a926d2009-06-22 10:12:17 +01002539 */
2540 if (!noflush && do_lockfs) {
2541 r = lock_fs(md);
2542 if (r)
Kiyoshi Uedaf431d962008-10-21 17:45:07 +01002543 goto out;
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -08002544 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545
2546 /*
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002547 * Here we must make sure that no processes are submitting requests
2548 * to target drivers i.e. no one may be executing
2549 * __split_and_process_bio. This is called from dm_request and
2550 * dm_wq_work.
2551 *
2552 * To get all processes out of __split_and_process_bio in dm_request,
2553 * we take the write lock. To prevent any process from reentering
Tejun Heo6a8736d2010-09-08 18:07:00 +02002554 * __split_and_process_bio from dm_request and quiesce the thread
2555 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2556 * flush_workqueue(md->wq).
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 */
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002558 down_write(&md->io_lock);
Alasdair G Kergon1eb787e2009-04-09 00:27:14 +01002559 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002560 up_write(&md->io_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561
Kiyoshi Uedad0bcb872009-12-10 23:52:18 +00002562 /*
Tejun Heo29e40132010-09-08 18:07:00 +02002563 * Stop md->queue before flushing md->wq in case request-based
2564 * dm defers requests to md->wq from md->queue.
Kiyoshi Uedad0bcb872009-12-10 23:52:18 +00002565 */
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002566 if (dm_request_based(md))
Kiyoshi Ueda9f518b22009-12-10 23:52:16 +00002567 stop_queue(md->queue);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002568
Kiyoshi Uedad0bcb872009-12-10 23:52:18 +00002569 flush_workqueue(md->wq);
2570
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 /*
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002572 * At this point no more requests are entering target request routines.
2573 * We call dm_wait_for_completion to wait for all existing requests
2574 * to finish.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 */
Mikulas Patocka401600d2009-04-02 19:55:38 +01002576 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002578 down_write(&md->io_lock);
Milan Broz6d6f10d2008-02-08 02:10:22 +00002579 if (noflush)
Mikulas Patocka022c2612009-04-02 19:55:39 +01002580 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
Milan Broz94d63512008-02-08 02:10:27 +00002581 up_write(&md->io_lock);
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -08002582
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 /* were we interrupted ? */
Milan Broz46125c12008-02-08 02:10:30 +00002584 if (r < 0) {
Mikulas Patocka9a1fb462009-04-02 19:55:36 +01002585 dm_queue_flush(md);
Milan Broz73d410c2008-02-08 02:10:25 +00002586
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002587 if (dm_request_based(md))
Kiyoshi Ueda9f518b22009-12-10 23:52:16 +00002588 start_queue(md->queue);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002589
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002590 unlock_fs(md);
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -08002591 goto out; /* pushback list is already flushed, so skip flush */
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002592 }
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002593
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002594 /*
2595 * If dm_wait_for_completion returned 0, the device is completely
2596 * quiescent now. There is no request-processing activity. All new
2597 * requests are being added to md->deferred list.
2598 */
2599
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 set_bit(DMF_SUSPENDED, &md->flags);
2601
Kiyoshi Ueda4d4471c2009-12-10 23:52:26 +00002602 dm_table_postsuspend_targets(map);
2603
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002604out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605 dm_table_put(map);
Alasdair G Kergond2874832006-11-08 17:44:43 -08002606
2607out_unlock:
Daniel Walkere61290a2008-02-08 02:10:08 +00002608 mutex_unlock(&md->suspend_lock);
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002609 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610}
2611
2612int dm_resume(struct mapped_device *md)
2613{
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002614 int r = -EINVAL;
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002615 struct dm_table *map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
Daniel Walkere61290a2008-02-08 02:10:08 +00002617 mutex_lock(&md->suspend_lock);
Kiyoshi Ueda4f186f82009-12-10 23:52:26 +00002618 if (!dm_suspended_md(md))
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002619 goto out;
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002620
Alasdair G Kergon7c666412009-12-10 23:52:19 +00002621 map = dm_get_live_table(md);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002622 if (!map || !dm_table_get_size(map))
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002623 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624
Milan Broz8757b772006-10-03 01:15:36 -07002625 r = dm_table_resume_targets(map);
2626 if (r)
2627 goto out;
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002628
Mikulas Patocka9a1fb462009-04-02 19:55:36 +01002629 dm_queue_flush(md);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002630
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002631 /*
2632 * Flushing deferred I/Os must be done after targets are resumed
2633 * so that mapping of targets can work correctly.
2634 * Request-based dm is queueing the deferred I/Os in its request_queue.
2635 */
2636 if (dm_request_based(md))
2637 start_queue(md->queue);
2638
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002639 unlock_fs(md);
2640
2641 clear_bit(DMF_SUSPENDED, &md->flags);
2642
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002643 r = 0;
2644out:
2645 dm_table_put(map);
Daniel Walkere61290a2008-02-08 02:10:08 +00002646 mutex_unlock(&md->suspend_lock);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002647
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002648 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649}
2650
2651/*-----------------------------------------------------------------
2652 * Event notification.
2653 *---------------------------------------------------------------*/
Peter Rajnoha3abf85b2010-03-06 02:32:31 +00002654int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
Milan Broz60935eb2009-06-22 10:12:30 +01002655 unsigned cookie)
Alasdair G Kergon69267a32007-12-13 14:15:57 +00002656{
Milan Broz60935eb2009-06-22 10:12:30 +01002657 char udev_cookie[DM_COOKIE_LENGTH];
2658 char *envp[] = { udev_cookie, NULL };
2659
2660 if (!cookie)
Peter Rajnoha3abf85b2010-03-06 02:32:31 +00002661 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
Milan Broz60935eb2009-06-22 10:12:30 +01002662 else {
2663 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2664 DM_COOKIE_ENV_VAR_NAME, cookie);
Peter Rajnoha3abf85b2010-03-06 02:32:31 +00002665 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2666 action, envp);
Milan Broz60935eb2009-06-22 10:12:30 +01002667 }
Alasdair G Kergon69267a32007-12-13 14:15:57 +00002668}
2669
Mike Anderson7a8c3d32007-10-19 22:48:01 +01002670uint32_t dm_next_uevent_seq(struct mapped_device *md)
2671{
2672 return atomic_add_return(1, &md->uevent_seq);
2673}
2674
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675uint32_t dm_get_event_nr(struct mapped_device *md)
2676{
2677 return atomic_read(&md->event_nr);
2678}
2679
2680int dm_wait_event(struct mapped_device *md, int event_nr)
2681{
2682 return wait_event_interruptible(md->eventq,
2683 (event_nr != atomic_read(&md->event_nr)));
2684}
2685
Mike Anderson7a8c3d32007-10-19 22:48:01 +01002686void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2687{
2688 unsigned long flags;
2689
2690 spin_lock_irqsave(&md->uevent_lock, flags);
2691 list_add(elist, &md->uevent_list);
2692 spin_unlock_irqrestore(&md->uevent_lock, flags);
2693}
2694
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695/*
2696 * The gendisk is only valid as long as you have a reference
2697 * count on 'md'.
2698 */
2699struct gendisk *dm_disk(struct mapped_device *md)
2700{
2701 return md->disk;
2702}
2703
Milan Broz784aae72009-01-06 03:05:12 +00002704struct kobject *dm_kobject(struct mapped_device *md)
2705{
2706 return &md->kobj;
2707}
2708
2709/*
2710 * struct mapped_device should not be exported outside of dm.c
2711 * so use this check to verify that kobj is part of md structure
2712 */
2713struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2714{
2715 struct mapped_device *md;
2716
2717 md = container_of(kobj, struct mapped_device, kobj);
2718 if (&md->kobj != kobj)
2719 return NULL;
2720
Milan Broz4d89b7b2009-06-22 10:12:11 +01002721 if (test_bit(DMF_FREEING, &md->flags) ||
Mike Anderson432a2122009-12-10 23:52:20 +00002722 dm_deleting_md(md))
Milan Broz4d89b7b2009-06-22 10:12:11 +01002723 return NULL;
2724
Milan Broz784aae72009-01-06 03:05:12 +00002725 dm_get(md);
2726 return md;
2727}
2728
Kiyoshi Ueda4f186f82009-12-10 23:52:26 +00002729int dm_suspended_md(struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730{
2731 return test_bit(DMF_SUSPENDED, &md->flags);
2732}
2733
Kiyoshi Ueda64dbce52009-12-10 23:52:27 +00002734int dm_suspended(struct dm_target *ti)
2735{
Kiyoshi Uedaecdb2e22010-03-06 02:29:52 +00002736 return dm_suspended_md(dm_table_get_md(ti->table));
Kiyoshi Ueda64dbce52009-12-10 23:52:27 +00002737}
2738EXPORT_SYMBOL_GPL(dm_suspended);
2739
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -08002740int dm_noflush_suspending(struct dm_target *ti)
2741{
Kiyoshi Uedaecdb2e22010-03-06 02:29:52 +00002742 return __noflush_suspending(dm_table_get_md(ti->table));
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -08002743}
2744EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2745
Mikulas Patockac0820cf2012-12-21 20:23:38 +00002746struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size)
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002747{
2748 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
Martin K. Petersena91a2782011-03-17 11:11:05 +01002749 unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS;
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002750
2751 if (!pools)
2752 return NULL;
2753
Mikulas Patockac0820cf2012-12-21 20:23:38 +00002754 per_bio_data_size = roundup(per_bio_data_size, __alignof__(struct dm_target_io));
2755
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002756 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2757 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2758 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2759 if (!pools->io_pool)
2760 goto free_pools_and_out;
2761
Mikulas Patockadba14162012-10-12 21:02:15 +01002762 pools->tio_pool = NULL;
2763 if (type == DM_TYPE_REQUEST_BASED) {
2764 pools->tio_pool = mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2765 if (!pools->tio_pool)
2766 goto free_io_pool_and_out;
2767 }
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002768
Kent Overstreet94818742012-09-07 13:44:01 -07002769 pools->bs = (type == DM_TYPE_BIO_BASED) ?
Mikulas Patockadba14162012-10-12 21:02:15 +01002770 bioset_create(pool_size,
Mikulas Patockac0820cf2012-12-21 20:23:38 +00002771 per_bio_data_size + offsetof(struct dm_target_io, clone)) :
Kent Overstreet94818742012-09-07 13:44:01 -07002772 bioset_create(pool_size,
2773 offsetof(struct dm_rq_clone_bio_info, clone));
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002774 if (!pools->bs)
2775 goto free_tio_pool_and_out;
2776
Martin K. Petersena91a2782011-03-17 11:11:05 +01002777 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2778 goto free_bioset_and_out;
2779
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002780 return pools;
2781
Martin K. Petersena91a2782011-03-17 11:11:05 +01002782free_bioset_and_out:
2783 bioset_free(pools->bs);
2784
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002785free_tio_pool_and_out:
Mikulas Patockadba14162012-10-12 21:02:15 +01002786 if (pools->tio_pool)
2787 mempool_destroy(pools->tio_pool);
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002788
2789free_io_pool_and_out:
2790 mempool_destroy(pools->io_pool);
2791
2792free_pools_and_out:
2793 kfree(pools);
2794
2795 return NULL;
2796}
2797
2798void dm_free_md_mempools(struct dm_md_mempools *pools)
2799{
2800 if (!pools)
2801 return;
2802
2803 if (pools->io_pool)
2804 mempool_destroy(pools->io_pool);
2805
2806 if (pools->tio_pool)
2807 mempool_destroy(pools->tio_pool);
2808
2809 if (pools->bs)
2810 bioset_free(pools->bs);
2811
2812 kfree(pools);
2813}
2814
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07002815static const struct block_device_operations dm_blk_dops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 .open = dm_blk_open,
2817 .release = dm_blk_close,
Milan Brozaa129a22006-10-03 01:15:15 -07002818 .ioctl = dm_blk_ioctl,
Darrick J. Wong3ac51e72006-03-27 01:17:54 -08002819 .getgeo = dm_blk_getgeo,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 .owner = THIS_MODULE
2821};
2822
2823EXPORT_SYMBOL(dm_get_mapinfo);
2824
2825/*
2826 * module hooks
2827 */
2828module_init(dm_init);
2829module_exit(dm_exit);
2830
2831module_param(major, uint, 0);
2832MODULE_PARM_DESC(major, "The major number of the device mapper");
2833MODULE_DESCRIPTION(DM_NAME " driver");
2834MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2835MODULE_LICENSE("GPL");