blob: e24143cc2040b7f3f8f90a63c3b1b2cfe2b21c24 [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 bio-based dm.
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 * One of these is allocated per target within a bio. Hopefully
68 * this will be simplified out one day.
69 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +010070struct dm_target_io {
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 struct dm_io *io;
72 struct dm_target *ti;
73 union map_info info;
74};
75
Kiyoshi Ueda8fbf26a2009-01-06 03:05:06 +000076/*
77 * For request-based dm.
78 * One of these is allocated per request.
79 */
80struct dm_rq_target_io {
81 struct mapped_device *md;
82 struct dm_target *ti;
83 struct request *orig, clone;
84 int error;
85 union map_info info;
86};
87
88/*
89 * For request-based dm.
90 * One of these is allocated per bio.
91 */
92struct dm_rq_clone_bio_info {
93 struct bio *orig;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +010094 struct dm_rq_target_io *tio;
Kiyoshi Ueda8fbf26a2009-01-06 03:05:06 +000095};
96
Linus Torvalds1da177e2005-04-16 15:20:36 -070097union map_info *dm_get_mapinfo(struct bio *bio)
98{
Alasdair G Kergon17b2f662006-06-26 00:27:33 -070099 if (bio && bio->bi_private)
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100100 return &((struct dm_target_io *)bio->bi_private)->info;
Alasdair G Kergon17b2f662006-06-26 00:27:33 -0700101 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102}
103
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100104union map_info *dm_get_rq_mapinfo(struct request *rq)
105{
106 if (rq && rq->end_io_data)
107 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
108 return NULL;
109}
110EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
111
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -0700112#define MINOR_ALLOCED ((void *)-1)
113
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114/*
115 * Bits for the md->flags field.
116 */
Alasdair G Kergon1eb787e2009-04-09 00:27:14 +0100117#define DMF_BLOCK_IO_FOR_SUSPEND 0
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118#define DMF_SUSPENDED 1
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -0800119#define DMF_FROZEN 2
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700120#define DMF_FREEING 3
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700121#define DMF_DELETING 4
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800122#define DMF_NOFLUSH_SUSPENDING 5
Mikulas Patockad5b9dd02011-08-02 12:32:04 +0100123#define DMF_MERGE_IS_OPTIONAL 6
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
Milan Broz304f3f62008-02-08 02:11:17 +0000125/*
126 * Work processed by per-device workqueue.
127 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128struct mapped_device {
Alasdair G Kergon2ca33102005-07-28 21:16:00 -0700129 struct rw_semaphore io_lock;
Daniel Walkere61290a2008-02-08 02:10:08 +0000130 struct mutex suspend_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 rwlock_t map_lock;
132 atomic_t holders;
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700133 atomic_t open_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134
135 unsigned long flags;
136
Jens Axboe165125e2007-07-24 09:28:11 +0200137 struct request_queue *queue;
Mike Snitzera5664da2010-08-12 04:14:01 +0100138 unsigned type;
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +0100139 /* Protect queue and type against concurrent access. */
Mike Snitzera5664da2010-08-12 04:14:01 +0100140 struct mutex type_lock;
141
Alasdair G Kergon36a04562011-10-31 20:19:04 +0000142 struct target_type *immutable_target_type;
143
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 struct gendisk *disk;
Mike Anderson7e51f252006-03-27 01:17:52 -0800145 char name[16];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
147 void *interface_ptr;
148
149 /*
150 * A list of ios that arrived while we were suspended.
151 */
Nikanth Karthikesan316d3152009-10-06 20:16:55 +0200152 atomic_t pending[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 wait_queue_head_t wait;
Mikulas Patocka53d59142009-04-02 19:55:37 +0100154 struct work_struct work;
Kiyoshi Ueda74859362006-12-08 02:41:02 -0800155 struct bio_list deferred;
Mikulas Patocka022c2612009-04-02 19:55:39 +0100156 spinlock_t deferred_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
158 /*
Tejun Heo29e40132010-09-08 18:07:00 +0200159 * Processing queue (flush)
Milan Broz304f3f62008-02-08 02:11:17 +0000160 */
161 struct workqueue_struct *wq;
162
163 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 * The current mapping.
165 */
166 struct dm_table *map;
167
168 /*
169 * io objects are allocated from here.
170 */
171 mempool_t *io_pool;
172 mempool_t *tio_pool;
173
Stefan Bader9faf4002006-10-03 01:15:41 -0700174 struct bio_set *bs;
175
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 /*
177 * Event handling.
178 */
179 atomic_t event_nr;
180 wait_queue_head_t eventq;
Mike Anderson7a8c3d32007-10-19 22:48:01 +0100181 atomic_t uevent_seq;
182 struct list_head uevent_list;
183 spinlock_t uevent_lock; /* Protect access to uevent_list */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
185 /*
186 * freeze/thaw support require holding onto a super block
187 */
188 struct super_block *frozen_sb;
Mikulas Patockadb8fef42009-06-22 10:12:15 +0100189 struct block_device *bdev;
Darrick J. Wong3ac51e72006-03-27 01:17:54 -0800190
191 /* forced geometry settings */
192 struct hd_geometry geometry;
Milan Broz784aae72009-01-06 03:05:12 +0000193
194 /* sysfs handle */
195 struct kobject kobj;
Mikulas Patocka52b1fd52009-06-22 10:12:21 +0100196
Tejun Heod87f4c12010-09-03 11:56:19 +0200197 /* zero-length flush that will be cloned and submitted to targets */
198 struct bio flush_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199};
200
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100201/*
202 * For mempools pre-allocation at the table loading time.
203 */
204struct dm_md_mempools {
205 mempool_t *io_pool;
206 mempool_t *tio_pool;
207 struct bio_set *bs;
208};
209
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210#define MIN_IOS 256
Christoph Lametere18b8902006-12-06 20:33:20 -0800211static struct kmem_cache *_io_cache;
212static struct kmem_cache *_tio_cache;
Kiyoshi Ueda8fbf26a2009-01-06 03:05:06 +0000213static struct kmem_cache *_rq_tio_cache;
214static struct kmem_cache *_rq_bio_info_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216static int __init local_init(void)
217{
Kiyoshi Ueda51157b42008-10-21 17:45:08 +0100218 int r = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 /* allocate a slab for the dm_ios */
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100221 _io_cache = KMEM_CACHE(dm_io, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 if (!_io_cache)
Kiyoshi Ueda51157b42008-10-21 17:45:08 +0100223 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224
225 /* allocate a slab for the target ios */
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100226 _tio_cache = KMEM_CACHE(dm_target_io, 0);
Kiyoshi Ueda51157b42008-10-21 17:45:08 +0100227 if (!_tio_cache)
228 goto out_free_io_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
Kiyoshi Ueda8fbf26a2009-01-06 03:05:06 +0000230 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
231 if (!_rq_tio_cache)
232 goto out_free_tio_cache;
233
234 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
235 if (!_rq_bio_info_cache)
236 goto out_free_rq_tio_cache;
237
Mike Anderson51e5b2b2007-10-19 22:48:00 +0100238 r = dm_uevent_init();
Kiyoshi Ueda51157b42008-10-21 17:45:08 +0100239 if (r)
Kiyoshi Ueda8fbf26a2009-01-06 03:05:06 +0000240 goto out_free_rq_bio_info_cache;
Mike Anderson51e5b2b2007-10-19 22:48:00 +0100241
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 _major = major;
243 r = register_blkdev(_major, _name);
Kiyoshi Ueda51157b42008-10-21 17:45:08 +0100244 if (r < 0)
245 goto out_uevent_exit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
247 if (!_major)
248 _major = r;
249
250 return 0;
Kiyoshi Ueda51157b42008-10-21 17:45:08 +0100251
252out_uevent_exit:
253 dm_uevent_exit();
Kiyoshi Ueda8fbf26a2009-01-06 03:05:06 +0000254out_free_rq_bio_info_cache:
255 kmem_cache_destroy(_rq_bio_info_cache);
256out_free_rq_tio_cache:
257 kmem_cache_destroy(_rq_tio_cache);
Kiyoshi Ueda51157b42008-10-21 17:45:08 +0100258out_free_tio_cache:
259 kmem_cache_destroy(_tio_cache);
260out_free_io_cache:
261 kmem_cache_destroy(_io_cache);
262
263 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264}
265
266static void local_exit(void)
267{
Kiyoshi Ueda8fbf26a2009-01-06 03:05:06 +0000268 kmem_cache_destroy(_rq_bio_info_cache);
269 kmem_cache_destroy(_rq_tio_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 kmem_cache_destroy(_tio_cache);
271 kmem_cache_destroy(_io_cache);
Akinobu Mita00d59402007-07-17 04:03:46 -0700272 unregister_blkdev(_major, _name);
Mike Anderson51e5b2b2007-10-19 22:48:00 +0100273 dm_uevent_exit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
275 _major = 0;
276
277 DMINFO("cleaned up");
278}
279
Alasdair G Kergonb9249e52008-02-08 02:09:51 +0000280static int (*_inits[])(void) __initdata = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 local_init,
282 dm_target_init,
283 dm_linear_init,
284 dm_stripe_init,
Mikulas Patocka952b3552009-12-10 23:51:57 +0000285 dm_io_init,
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100286 dm_kcopyd_init,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 dm_interface_init,
288};
289
Alasdair G Kergonb9249e52008-02-08 02:09:51 +0000290static void (*_exits[])(void) = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 local_exit,
292 dm_target_exit,
293 dm_linear_exit,
294 dm_stripe_exit,
Mikulas Patocka952b3552009-12-10 23:51:57 +0000295 dm_io_exit,
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100296 dm_kcopyd_exit,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 dm_interface_exit,
298};
299
300static int __init dm_init(void)
301{
302 const int count = ARRAY_SIZE(_inits);
303
304 int r, i;
305
306 for (i = 0; i < count; i++) {
307 r = _inits[i]();
308 if (r)
309 goto bad;
310 }
311
312 return 0;
313
314 bad:
315 while (i--)
316 _exits[i]();
317
318 return r;
319}
320
321static void __exit dm_exit(void)
322{
323 int i = ARRAY_SIZE(_exits);
324
325 while (i--)
326 _exits[i]();
Alasdair G Kergond15b7742011-08-02 12:32:01 +0100327
328 /*
329 * Should be empty by this point.
330 */
331 idr_remove_all(&_minor_idr);
332 idr_destroy(&_minor_idr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333}
334
335/*
336 * Block device functions
337 */
Mike Anderson432a2122009-12-10 23:52:20 +0000338int dm_deleting_md(struct mapped_device *md)
339{
340 return test_bit(DMF_DELETING, &md->flags);
341}
342
Al Virofe5f9f22008-03-02 10:29:31 -0500343static int dm_blk_open(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344{
345 struct mapped_device *md;
346
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700347 spin_lock(&_minor_lock);
348
Al Virofe5f9f22008-03-02 10:29:31 -0500349 md = bdev->bd_disk->private_data;
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700350 if (!md)
351 goto out;
352
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700353 if (test_bit(DMF_FREEING, &md->flags) ||
Mike Anderson432a2122009-12-10 23:52:20 +0000354 dm_deleting_md(md)) {
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700355 md = NULL;
356 goto out;
357 }
358
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 dm_get(md);
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700360 atomic_inc(&md->open_count);
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700361
362out:
363 spin_unlock(&_minor_lock);
364
365 return md ? 0 : -ENXIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366}
367
Al Virofe5f9f22008-03-02 10:29:31 -0500368static int dm_blk_close(struct gendisk *disk, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369{
Al Virofe5f9f22008-03-02 10:29:31 -0500370 struct mapped_device *md = disk->private_data;
Arnd Bergmann6e9624b2010-08-07 18:25:34 +0200371
Milan Broz4a1aeb92011-01-13 19:59:48 +0000372 spin_lock(&_minor_lock);
373
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700374 atomic_dec(&md->open_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 dm_put(md);
Milan Broz4a1aeb92011-01-13 19:59:48 +0000376
377 spin_unlock(&_minor_lock);
Arnd Bergmann6e9624b2010-08-07 18:25:34 +0200378
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 return 0;
380}
381
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700382int dm_open_count(struct mapped_device *md)
383{
384 return atomic_read(&md->open_count);
385}
386
387/*
388 * Guarantees nothing is using the device before it's deleted.
389 */
390int dm_lock_for_deletion(struct mapped_device *md)
391{
392 int r = 0;
393
394 spin_lock(&_minor_lock);
395
396 if (dm_open_count(md))
397 r = -EBUSY;
398 else
399 set_bit(DMF_DELETING, &md->flags);
400
401 spin_unlock(&_minor_lock);
402
403 return r;
404}
405
Darrick J. Wong3ac51e72006-03-27 01:17:54 -0800406static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
407{
408 struct mapped_device *md = bdev->bd_disk->private_data;
409
410 return dm_get_geometry(md, geo);
411}
412
Al Virofe5f9f22008-03-02 10:29:31 -0500413static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
Milan Brozaa129a22006-10-03 01:15:15 -0700414 unsigned int cmd, unsigned long arg)
415{
Al Virofe5f9f22008-03-02 10:29:31 -0500416 struct mapped_device *md = bdev->bd_disk->private_data;
Alasdair G Kergon7c666412009-12-10 23:52:19 +0000417 struct dm_table *map = dm_get_live_table(md);
Milan Brozaa129a22006-10-03 01:15:15 -0700418 struct dm_target *tgt;
419 int r = -ENOTTY;
420
Milan Brozaa129a22006-10-03 01:15:15 -0700421 if (!map || !dm_table_get_size(map))
422 goto out;
423
424 /* We only support devices that have a single target */
425 if (dm_table_get_num_targets(map) != 1)
426 goto out;
427
428 tgt = dm_table_get_target(map, 0);
429
Kiyoshi Ueda4f186f82009-12-10 23:52:26 +0000430 if (dm_suspended_md(md)) {
Milan Brozaa129a22006-10-03 01:15:15 -0700431 r = -EAGAIN;
432 goto out;
433 }
434
435 if (tgt->type->ioctl)
Al Viro647b3d02007-08-28 22:15:59 -0400436 r = tgt->type->ioctl(tgt, cmd, arg);
Milan Brozaa129a22006-10-03 01:15:15 -0700437
438out:
439 dm_table_put(map);
440
Milan Brozaa129a22006-10-03 01:15:15 -0700441 return r;
442}
443
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100444static struct dm_io *alloc_io(struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445{
446 return mempool_alloc(md->io_pool, GFP_NOIO);
447}
448
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100449static void free_io(struct mapped_device *md, struct dm_io *io)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450{
451 mempool_free(io, md->io_pool);
452}
453
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100454static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455{
456 mempool_free(tio, md->tio_pool);
457}
458
Kiyoshi Ueda08885642009-12-10 23:52:15 +0000459static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
460 gfp_t gfp_mask)
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100461{
Kiyoshi Ueda08885642009-12-10 23:52:15 +0000462 return mempool_alloc(md->tio_pool, gfp_mask);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100463}
464
465static void free_rq_tio(struct dm_rq_target_io *tio)
466{
467 mempool_free(tio, tio->md->tio_pool);
468}
469
470static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md)
471{
472 return mempool_alloc(md->io_pool, GFP_ATOMIC);
473}
474
475static void free_bio_info(struct dm_rq_clone_bio_info *info)
476{
477 mempool_free(info, info->tio->md->io_pool);
478}
479
Kiyoshi Ueda90abb8c2009-12-10 23:52:13 +0000480static int md_in_flight(struct mapped_device *md)
481{
482 return atomic_read(&md->pending[READ]) +
483 atomic_read(&md->pending[WRITE]);
484}
485
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800486static void start_io_acct(struct dm_io *io)
487{
488 struct mapped_device *md = io->md;
Tejun Heoc9959052008-08-25 19:47:21 +0900489 int cpu;
Nikanth Karthikesan316d3152009-10-06 20:16:55 +0200490 int rw = bio_data_dir(io->bio);
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800491
492 io->start_time = jiffies;
493
Tejun Heo074a7ac2008-08-25 19:56:14 +0900494 cpu = part_stat_lock();
495 part_round_stats(cpu, &dm_disk(md)->part0);
496 part_stat_unlock();
Shaohua Li1e9bb882011-03-22 08:35:35 +0100497 atomic_set(&dm_disk(md)->part0.in_flight[rw],
498 atomic_inc_return(&md->pending[rw]));
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800499}
500
Mikulas Patockad221d2e2008-11-13 23:39:10 +0000501static void end_io_acct(struct dm_io *io)
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800502{
503 struct mapped_device *md = io->md;
504 struct bio *bio = io->bio;
505 unsigned long duration = jiffies - io->start_time;
Tejun Heoc9959052008-08-25 19:47:21 +0900506 int pending, cpu;
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800507 int rw = bio_data_dir(bio);
508
Tejun Heo074a7ac2008-08-25 19:56:14 +0900509 cpu = part_stat_lock();
510 part_round_stats(cpu, &dm_disk(md)->part0);
511 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
512 part_stat_unlock();
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800513
Mikulas Patockaaf7e4662009-04-09 00:27:16 +0100514 /*
515 * After this is decremented the bio must not be touched if it is
Tejun Heod87f4c12010-09-03 11:56:19 +0200516 * a flush.
Mikulas Patockaaf7e4662009-04-09 00:27:16 +0100517 */
Shaohua Li1e9bb882011-03-22 08:35:35 +0100518 pending = atomic_dec_return(&md->pending[rw]);
519 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +0200520 pending += atomic_read(&md->pending[rw^0x1]);
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800521
Mikulas Patockad221d2e2008-11-13 23:39:10 +0000522 /* nudge anyone waiting on suspend queue */
523 if (!pending)
524 wake_up(&md->wait);
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800525}
526
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527/*
528 * Add the bio to the list of deferred io.
529 */
Mikulas Patocka92c63902009-04-09 00:27:15 +0100530static void queue_io(struct mapped_device *md, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531{
Kiyoshi Ueda054474202010-09-08 18:07:01 +0200532 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533
Kiyoshi Ueda054474202010-09-08 18:07:01 +0200534 spin_lock_irqsave(&md->deferred_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 bio_list_add(&md->deferred, bio);
Kiyoshi Ueda054474202010-09-08 18:07:01 +0200536 spin_unlock_irqrestore(&md->deferred_lock, flags);
Tejun Heo6a8736d2010-09-08 18:07:00 +0200537 queue_work(md->wq, &md->work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538}
539
540/*
541 * Everyone (including functions in this file), should use this
542 * function to access the md->map field, and make sure they call
543 * dm_table_put() when finished.
544 */
Alasdair G Kergon7c666412009-12-10 23:52:19 +0000545struct dm_table *dm_get_live_table(struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546{
547 struct dm_table *t;
Kiyoshi Ueda523d9292009-06-22 10:12:37 +0100548 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
Kiyoshi Ueda523d9292009-06-22 10:12:37 +0100550 read_lock_irqsave(&md->map_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 t = md->map;
552 if (t)
553 dm_table_get(t);
Kiyoshi Ueda523d9292009-06-22 10:12:37 +0100554 read_unlock_irqrestore(&md->map_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
556 return t;
557}
558
Darrick J. Wong3ac51e72006-03-27 01:17:54 -0800559/*
560 * Get the geometry associated with a dm device
561 */
562int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
563{
564 *geo = md->geometry;
565
566 return 0;
567}
568
569/*
570 * Set the geometry of a device.
571 */
572int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
573{
574 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
575
576 if (geo->start > sz) {
577 DMWARN("Start sector is beyond the geometry limits.");
578 return -EINVAL;
579 }
580
581 md->geometry = *geo;
582
583 return 0;
584}
585
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586/*-----------------------------------------------------------------
587 * CRUD START:
588 * A more elegant soln is in the works that uses the queue
589 * merge fn, unfortunately there are a couple of changes to
590 * the block layer that I want to make for this. So in the
591 * interests of getting something for people to use I give
592 * you this clearly demarcated crap.
593 *---------------------------------------------------------------*/
594
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800595static int __noflush_suspending(struct mapped_device *md)
596{
597 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
598}
599
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600/*
601 * Decrements the number of outstanding ios that a bio has been
602 * cloned into, completing the original io if necc.
603 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800604static void dec_pending(struct dm_io *io, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605{
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800606 unsigned long flags;
Milan Brozb35f8ca2009-03-16 17:44:36 +0000607 int io_error;
608 struct bio *bio;
609 struct mapped_device *md = io->md;
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800610
611 /* Push-back supersedes any I/O errors */
Kiyoshi Uedaf88fb982009-10-16 23:18:15 +0100612 if (unlikely(error)) {
613 spin_lock_irqsave(&io->endio_lock, flags);
614 if (!(io->error > 0 && __noflush_suspending(md)))
615 io->error = error;
616 spin_unlock_irqrestore(&io->endio_lock, flags);
617 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
619 if (atomic_dec_and_test(&io->io_count)) {
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800620 if (io->error == DM_ENDIO_REQUEUE) {
621 /*
622 * Target requested pushing back the I/O.
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800623 */
Mikulas Patocka022c2612009-04-02 19:55:39 +0100624 spin_lock_irqsave(&md->deferred_lock, flags);
Tejun Heo6a8736d2010-09-08 18:07:00 +0200625 if (__noflush_suspending(md))
626 bio_list_add_head(&md->deferred, io->bio);
627 else
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800628 /* noflush suspend was interrupted. */
629 io->error = -EIO;
Mikulas Patocka022c2612009-04-02 19:55:39 +0100630 spin_unlock_irqrestore(&md->deferred_lock, flags);
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800631 }
632
Milan Brozb35f8ca2009-03-16 17:44:36 +0000633 io_error = io->error;
634 bio = io->bio;
Tejun Heo6a8736d2010-09-08 18:07:00 +0200635 end_io_acct(io);
636 free_io(md, io);
Jens Axboe2056a782006-03-23 20:00:26 +0100637
Tejun Heo6a8736d2010-09-08 18:07:00 +0200638 if (io_error == DM_ENDIO_REQUEUE)
639 return;
640
Mike Snitzerb372d362010-09-08 18:07:01 +0200641 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) {
Mikulas Patockaaf7e4662009-04-09 00:27:16 +0100642 /*
Tejun Heo6a8736d2010-09-08 18:07:00 +0200643 * Preflush done for flush with data, reissue
644 * without REQ_FLUSH.
Mikulas Patockaaf7e4662009-04-09 00:27:16 +0100645 */
Tejun Heo6a8736d2010-09-08 18:07:00 +0200646 bio->bi_rw &= ~REQ_FLUSH;
647 queue_io(md, bio);
Mikulas Patockaaf7e4662009-04-09 00:27:16 +0100648 } else {
Mike Snitzerb372d362010-09-08 18:07:01 +0200649 /* done with normal IO or empty flush */
Jeff Moyerb7908c12011-01-06 20:41:42 +0100650 trace_block_bio_complete(md->queue, bio, io_error);
Mike Snitzerb372d362010-09-08 18:07:01 +0200651 bio_endio(bio, io_error);
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800652 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 }
654}
655
NeilBrown6712ecf2007-09-27 12:47:43 +0200656static void clone_endio(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657{
658 int r = 0;
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100659 struct dm_target_io *tio = bio->bi_private;
Milan Brozb35f8ca2009-03-16 17:44:36 +0000660 struct dm_io *io = tio->io;
Stefan Bader9faf4002006-10-03 01:15:41 -0700661 struct mapped_device *md = tio->io->md;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 dm_endio_fn endio = tio->ti->type->end_io;
663
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
665 error = -EIO;
666
667 if (endio) {
668 r = endio(tio->ti, bio, error, &tio->info);
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800669 if (r < 0 || r == DM_ENDIO_REQUEUE)
670 /*
671 * error and requeue request are handled
672 * in dec_pending().
673 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 error = r;
Kiyoshi Ueda45cbcd72006-12-08 02:41:05 -0800675 else if (r == DM_ENDIO_INCOMPLETE)
676 /* The target will handle the io */
NeilBrown6712ecf2007-09-27 12:47:43 +0200677 return;
Kiyoshi Ueda45cbcd72006-12-08 02:41:05 -0800678 else if (r) {
679 DMWARN("unimplemented target endio return value: %d", r);
680 BUG();
681 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 }
683
Stefan Bader9faf4002006-10-03 01:15:41 -0700684 /*
685 * Store md for cleanup instead of tio which is about to get freed.
686 */
687 bio->bi_private = md->bs;
688
Stefan Bader9faf4002006-10-03 01:15:41 -0700689 free_tio(md, tio);
Milan Brozb35f8ca2009-03-16 17:44:36 +0000690 bio_put(bio);
691 dec_pending(io, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692}
693
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100694/*
695 * Partial completion handling for request-based dm
696 */
697static void end_clone_bio(struct bio *clone, int error)
698{
699 struct dm_rq_clone_bio_info *info = clone->bi_private;
700 struct dm_rq_target_io *tio = info->tio;
701 struct bio *bio = info->orig;
702 unsigned int nr_bytes = info->orig->bi_size;
703
704 bio_put(clone);
705
706 if (tio->error)
707 /*
708 * An error has already been detected on the request.
709 * Once error occurred, just let clone->end_io() handle
710 * the remainder.
711 */
712 return;
713 else if (error) {
714 /*
715 * Don't notice the error to the upper layer yet.
716 * The error handling decision is made by the target driver,
717 * when the request is completed.
718 */
719 tio->error = error;
720 return;
721 }
722
723 /*
724 * I/O for the bio successfully completed.
725 * Notice the data completion to the upper layer.
726 */
727
728 /*
729 * bios are processed from the head of the list.
730 * So the completing bio should always be rq->bio.
731 * If it's not, something wrong is happening.
732 */
733 if (tio->orig->bio != bio)
734 DMERR("bio completion is going in the middle of the request");
735
736 /*
737 * Update the original request.
738 * Do not use blk_end_request() here, because it may complete
739 * the original request before the clone, and break the ordering.
740 */
741 blk_update_request(tio->orig, 0, nr_bytes);
742}
743
744/*
745 * Don't touch any member of the md after calling this function because
746 * the md may be freed in dm_put() at the end of this function.
747 * Or do dm_get() before calling this function and dm_put() later.
748 */
Kiyoshi Uedab4324fe2009-12-10 23:52:16 +0000749static void rq_completed(struct mapped_device *md, int rw, int run_queue)
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100750{
Kiyoshi Uedab4324fe2009-12-10 23:52:16 +0000751 atomic_dec(&md->pending[rw]);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100752
753 /* nudge anyone waiting on suspend queue */
Kiyoshi Uedab4324fe2009-12-10 23:52:16 +0000754 if (!md_in_flight(md))
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100755 wake_up(&md->wait);
756
757 if (run_queue)
Kiyoshi Uedab4324fe2009-12-10 23:52:16 +0000758 blk_run_queue(md->queue);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100759
760 /*
761 * dm_put() must be at the end of this function. See the comment above
762 */
763 dm_put(md);
764}
765
Kiyoshi Uedaa77e28c2009-09-04 20:40:16 +0100766static void free_rq_clone(struct request *clone)
767{
768 struct dm_rq_target_io *tio = clone->end_io_data;
769
770 blk_rq_unprep_clone(clone);
771 free_rq_tio(tio);
772}
773
Kiyoshi Ueda980691e2009-12-10 23:52:17 +0000774/*
775 * Complete the clone and the original request.
776 * Must be called without queue lock.
777 */
778static void dm_end_request(struct request *clone, int error)
779{
780 int rw = rq_data_dir(clone);
781 struct dm_rq_target_io *tio = clone->end_io_data;
782 struct mapped_device *md = tio->md;
783 struct request *rq = tio->orig;
784
Tejun Heo29e40132010-09-08 18:07:00 +0200785 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Kiyoshi Ueda980691e2009-12-10 23:52:17 +0000786 rq->errors = clone->errors;
787 rq->resid_len = clone->resid_len;
788
789 if (rq->sense)
790 /*
791 * We are using the sense buffer of the original
792 * request.
793 * So setting the length of the sense data is enough.
794 */
795 rq->sense_len = clone->sense_len;
796 }
797
798 free_rq_clone(clone);
Tejun Heo29e40132010-09-08 18:07:00 +0200799 blk_end_request_all(rq, error);
800 rq_completed(md, rw, true);
Kiyoshi Ueda980691e2009-12-10 23:52:17 +0000801}
802
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100803static void dm_unprep_request(struct request *rq)
804{
805 struct request *clone = rq->special;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100806
807 rq->special = NULL;
808 rq->cmd_flags &= ~REQ_DONTPREP;
809
Kiyoshi Uedaa77e28c2009-09-04 20:40:16 +0100810 free_rq_clone(clone);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100811}
812
813/*
814 * Requeue the original request of a clone.
815 */
816void dm_requeue_unmapped_request(struct request *clone)
817{
Kiyoshi Uedab4324fe2009-12-10 23:52:16 +0000818 int rw = rq_data_dir(clone);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100819 struct dm_rq_target_io *tio = clone->end_io_data;
820 struct mapped_device *md = tio->md;
821 struct request *rq = tio->orig;
822 struct request_queue *q = rq->q;
823 unsigned long flags;
824
825 dm_unprep_request(rq);
826
827 spin_lock_irqsave(q->queue_lock, flags);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100828 blk_requeue_request(q, rq);
829 spin_unlock_irqrestore(q->queue_lock, flags);
830
Kiyoshi Uedab4324fe2009-12-10 23:52:16 +0000831 rq_completed(md, rw, 0);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100832}
833EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
834
835static void __stop_queue(struct request_queue *q)
836{
837 blk_stop_queue(q);
838}
839
840static void stop_queue(struct request_queue *q)
841{
842 unsigned long flags;
843
844 spin_lock_irqsave(q->queue_lock, flags);
845 __stop_queue(q);
846 spin_unlock_irqrestore(q->queue_lock, flags);
847}
848
849static void __start_queue(struct request_queue *q)
850{
851 if (blk_queue_stopped(q))
852 blk_start_queue(q);
853}
854
855static void start_queue(struct request_queue *q)
856{
857 unsigned long flags;
858
859 spin_lock_irqsave(q->queue_lock, flags);
860 __start_queue(q);
861 spin_unlock_irqrestore(q->queue_lock, flags);
862}
863
Kiyoshi Ueda11a68242009-12-10 23:52:17 +0000864static void dm_done(struct request *clone, int error, bool mapped)
865{
866 int r = error;
867 struct dm_rq_target_io *tio = clone->end_io_data;
868 dm_request_endio_fn rq_end_io = tio->ti->type->rq_end_io;
869
870 if (mapped && rq_end_io)
871 r = rq_end_io(tio->ti, clone, error, &tio->info);
872
873 if (r <= 0)
874 /* The target wants to complete the I/O */
875 dm_end_request(clone, r);
876 else if (r == DM_ENDIO_INCOMPLETE)
877 /* The target will handle the I/O */
878 return;
879 else if (r == DM_ENDIO_REQUEUE)
880 /* The target wants to requeue the I/O */
881 dm_requeue_unmapped_request(clone);
882 else {
883 DMWARN("unimplemented target endio return value: %d", r);
884 BUG();
885 }
886}
887
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100888/*
889 * Request completion handler for request-based dm
890 */
891static void dm_softirq_done(struct request *rq)
892{
Kiyoshi Ueda11a68242009-12-10 23:52:17 +0000893 bool mapped = true;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100894 struct request *clone = rq->completion_data;
895 struct dm_rq_target_io *tio = clone->end_io_data;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100896
Kiyoshi Ueda11a68242009-12-10 23:52:17 +0000897 if (rq->cmd_flags & REQ_FAILED)
898 mapped = false;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100899
Kiyoshi Ueda11a68242009-12-10 23:52:17 +0000900 dm_done(clone, tio->error, mapped);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +0100901}
902
903/*
904 * Complete the clone and the original request with the error status
905 * through softirq context.
906 */
907static void dm_complete_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 tio->error = error;
913 rq->completion_data = clone;
914 blk_complete_request(rq);
915}
916
917/*
918 * Complete the not-mapped clone and the original request with the error status
919 * through softirq context.
920 * Target's rq_end_io() function isn't called.
921 * This may be used when the target's map_rq() function fails.
922 */
923void dm_kill_unmapped_request(struct request *clone, int error)
924{
925 struct dm_rq_target_io *tio = clone->end_io_data;
926 struct request *rq = tio->orig;
927
928 rq->cmd_flags |= REQ_FAILED;
929 dm_complete_request(clone, error);
930}
931EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
932
933/*
934 * Called with the queue lock held
935 */
936static void end_clone_request(struct request *clone, int error)
937{
938 /*
939 * For just cleaning up the information of the queue in which
940 * the clone was dispatched.
941 * The clone is *NOT* freed actually here because it is alloced from
942 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
943 */
944 __blk_put_request(clone->q, clone);
945
946 /*
947 * Actual request completion is done in a softirq context which doesn't
948 * hold the queue lock. Otherwise, deadlock could occur because:
949 * - another request may be submitted by the upper level driver
950 * of the stacking during the completion
951 * - the submission which requires queue lock may be done
952 * against this queue
953 */
954 dm_complete_request(clone, error);
955}
956
Mike Snitzer56a67df2010-08-12 04:14:10 +0100957/*
958 * Return maximum size of I/O possible at the supplied sector up to the current
959 * target boundary.
960 */
961static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962{
Mike Snitzer56a67df2010-08-12 04:14:10 +0100963 sector_t target_offset = dm_target_offset(ti, sector);
964
965 return ti->len - target_offset;
966}
967
968static sector_t max_io_len(sector_t sector, struct dm_target *ti)
969{
970 sector_t len = max_io_len_target_boundary(sector, ti);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
972 /*
973 * Does the target need to split even further ?
974 */
975 if (ti->split_io) {
976 sector_t boundary;
Mike Snitzer56a67df2010-08-12 04:14:10 +0100977 sector_t offset = dm_target_offset(ti, sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
979 - offset;
980 if (len > boundary)
981 len = boundary;
982 }
983
984 return len;
985}
986
987static void __map_bio(struct dm_target *ti, struct bio *clone,
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100988 struct dm_target_io *tio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989{
990 int r;
Jens Axboe2056a782006-03-23 20:00:26 +0100991 sector_t sector;
Stefan Bader9faf4002006-10-03 01:15:41 -0700992 struct mapped_device *md;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 clone->bi_end_io = clone_endio;
995 clone->bi_private = tio;
996
997 /*
998 * Map the clone. If r == 0 we don't need to do
999 * anything, the target has assumed ownership of
1000 * this io.
1001 */
1002 atomic_inc(&tio->io->io_count);
Jens Axboe2056a782006-03-23 20:00:26 +01001003 sector = clone->bi_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 r = ti->type->map(ti, clone, &tio->info);
Kiyoshi Ueda45cbcd72006-12-08 02:41:05 -08001005 if (r == DM_MAPIO_REMAPPED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 /* the bio has been remapped so dispatch it */
Jens Axboe2056a782006-03-23 20:00:26 +01001007
Mike Snitzerd07335e2010-11-16 12:52:38 +01001008 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1009 tio->io->bio->bi_bdev->bd_dev, sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001010
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 generic_make_request(clone);
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -08001012 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1013 /* error the io and bail out, or requeue it if needed */
Stefan Bader9faf4002006-10-03 01:15:41 -07001014 md = tio->io->md;
1015 dec_pending(tio->io, r);
1016 /*
1017 * Store bio_set for cleanup.
1018 */
Hannes Reinecke4d7b38b2012-03-28 18:41:25 +01001019 clone->bi_end_io = NULL;
Stefan Bader9faf4002006-10-03 01:15:41 -07001020 clone->bi_private = md->bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 bio_put(clone);
Stefan Bader9faf4002006-10-03 01:15:41 -07001022 free_tio(md, tio);
Kiyoshi Ueda45cbcd72006-12-08 02:41:05 -08001023 } else if (r) {
1024 DMWARN("unimplemented target map return value: %d", r);
1025 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 }
1027}
1028
1029struct clone_info {
1030 struct mapped_device *md;
1031 struct dm_table *map;
1032 struct bio *bio;
1033 struct dm_io *io;
1034 sector_t sector;
1035 sector_t sector_count;
1036 unsigned short idx;
1037};
1038
Peter Osterlund36763472005-09-06 15:16:42 -07001039static void dm_bio_destructor(struct bio *bio)
1040{
Stefan Bader9faf4002006-10-03 01:15:41 -07001041 struct bio_set *bs = bio->bi_private;
1042
1043 bio_free(bio, bs);
Peter Osterlund36763472005-09-06 15:16:42 -07001044}
1045
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046/*
Tejun Heod87f4c12010-09-03 11:56:19 +02001047 * Creates a little bio that just does part of a bvec.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 */
1049static struct bio *split_bvec(struct bio *bio, sector_t sector,
1050 unsigned short idx, unsigned int offset,
Stefan Bader9faf4002006-10-03 01:15:41 -07001051 unsigned int len, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052{
1053 struct bio *clone;
1054 struct bio_vec *bv = bio->bi_io_vec + idx;
1055
Stefan Bader9faf4002006-10-03 01:15:41 -07001056 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
Peter Osterlund36763472005-09-06 15:16:42 -07001057 clone->bi_destructor = dm_bio_destructor;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 *clone->bi_io_vec = *bv;
1059
1060 clone->bi_sector = sector;
1061 clone->bi_bdev = bio->bi_bdev;
Tejun Heod87f4c12010-09-03 11:56:19 +02001062 clone->bi_rw = bio->bi_rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 clone->bi_vcnt = 1;
1064 clone->bi_size = to_bytes(len);
1065 clone->bi_io_vec->bv_offset = offset;
1066 clone->bi_io_vec->bv_len = clone->bi_size;
Martin K. Petersenf3e1d262008-10-21 17:45:04 +01001067 clone->bi_flags |= 1 << BIO_CLONED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068
Martin K. Petersen9c470082009-04-09 00:27:12 +01001069 if (bio_integrity(bio)) {
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001070 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
Martin K. Petersen9c470082009-04-09 00:27:12 +01001071 bio_integrity_trim(clone,
1072 bio_sector_offset(bio, idx, offset), len);
1073 }
1074
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 return clone;
1076}
1077
1078/*
1079 * Creates a bio that consists of range of complete bvecs.
1080 */
1081static struct bio *clone_bio(struct bio *bio, sector_t sector,
1082 unsigned short idx, unsigned short bv_count,
Stefan Bader9faf4002006-10-03 01:15:41 -07001083 unsigned int len, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084{
1085 struct bio *clone;
1086
Stefan Bader9faf4002006-10-03 01:15:41 -07001087 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
1088 __bio_clone(clone, bio);
1089 clone->bi_destructor = dm_bio_destructor;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 clone->bi_sector = sector;
1091 clone->bi_idx = idx;
1092 clone->bi_vcnt = idx + bv_count;
1093 clone->bi_size = to_bytes(len);
1094 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1095
Martin K. Petersen9c470082009-04-09 00:27:12 +01001096 if (bio_integrity(bio)) {
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001097 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
Martin K. Petersen9c470082009-04-09 00:27:12 +01001098
1099 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1100 bio_integrity_trim(clone,
1101 bio_sector_offset(bio, idx, 0), len);
1102 }
1103
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 return clone;
1105}
1106
Alasdair G Kergon9015df22009-06-22 10:12:21 +01001107static struct dm_target_io *alloc_tio(struct clone_info *ci,
1108 struct dm_target *ti)
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001109{
Alasdair G Kergon9015df22009-06-22 10:12:21 +01001110 struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO);
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001111
1112 tio->io = ci->io;
1113 tio->ti = ti;
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001114 memset(&tio->info, 0, sizeof(tio->info));
Alasdair G Kergon9015df22009-06-22 10:12:21 +01001115
1116 return tio;
1117}
1118
Mike Snitzer06a426c2010-08-12 04:14:09 +01001119static void __issue_target_request(struct clone_info *ci, struct dm_target *ti,
Mike Snitzera79245b2010-08-12 04:14:24 +01001120 unsigned request_nr, sector_t len)
Alasdair G Kergon9015df22009-06-22 10:12:21 +01001121{
1122 struct dm_target_io *tio = alloc_tio(ci, ti);
1123 struct bio *clone;
1124
Mike Snitzer57cba5d2010-08-12 04:14:04 +01001125 tio->info.target_request_nr = request_nr;
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001126
Mike Snitzer06a426c2010-08-12 04:14:09 +01001127 /*
1128 * Discard requests require the bio's inline iovecs be initialized.
1129 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1130 * and discard, so no need for concern about wasted bvec allocations.
1131 */
1132 clone = bio_alloc_bioset(GFP_NOIO, ci->bio->bi_max_vecs, ci->md->bs);
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001133 __bio_clone(clone, ci->bio);
1134 clone->bi_destructor = dm_bio_destructor;
Mike Snitzera79245b2010-08-12 04:14:24 +01001135 if (len) {
1136 clone->bi_sector = ci->sector;
1137 clone->bi_size = to_bytes(len);
1138 }
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001139
1140 __map_bio(ti, clone, tio);
1141}
1142
Mike Snitzer06a426c2010-08-12 04:14:09 +01001143static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti,
Mike Snitzera79245b2010-08-12 04:14:24 +01001144 unsigned num_requests, sector_t len)
Mike Snitzer06a426c2010-08-12 04:14:09 +01001145{
1146 unsigned request_nr;
1147
1148 for (request_nr = 0; request_nr < num_requests; request_nr++)
Mike Snitzera79245b2010-08-12 04:14:24 +01001149 __issue_target_request(ci, ti, request_nr, len);
Mike Snitzer06a426c2010-08-12 04:14:09 +01001150}
1151
Mike Snitzerb372d362010-09-08 18:07:01 +02001152static int __clone_and_map_empty_flush(struct clone_info *ci)
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001153{
Mike Snitzer06a426c2010-08-12 04:14:09 +01001154 unsigned target_nr = 0;
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001155 struct dm_target *ti;
1156
Mike Snitzerb372d362010-09-08 18:07:01 +02001157 BUG_ON(bio_has_data(ci->bio));
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001158 while ((ti = dm_table_get_target(ci->map, target_nr++)))
Mike Snitzera79245b2010-08-12 04:14:24 +01001159 __issue_target_requests(ci, ti, ti->num_flush_requests, 0);
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001160
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001161 return 0;
1162}
1163
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001164/*
1165 * Perform all io with a single clone.
1166 */
1167static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti)
1168{
1169 struct bio *clone, *bio = ci->bio;
1170 struct dm_target_io *tio;
1171
1172 tio = alloc_tio(ci, ti);
1173 clone = clone_bio(bio, ci->sector, ci->idx,
1174 bio->bi_vcnt - ci->idx, ci->sector_count,
1175 ci->md->bs);
1176 __map_bio(ti, clone, tio);
1177 ci->sector_count = 0;
1178}
1179
1180static int __clone_and_map_discard(struct clone_info *ci)
1181{
1182 struct dm_target *ti;
Mike Snitzera79245b2010-08-12 04:14:24 +01001183 sector_t len;
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001184
Mike Snitzera79245b2010-08-12 04:14:24 +01001185 do {
1186 ti = dm_table_find_target(ci->map, ci->sector);
1187 if (!dm_target_is_valid(ti))
1188 return -EIO;
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001189
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001190 /*
Mike Snitzera79245b2010-08-12 04:14:24 +01001191 * Even though the device advertised discard support,
Mike Snitzer936688d2011-08-02 12:32:01 +01001192 * that does not mean every target supports it, and
1193 * reconfiguration might also have changed that since the
Mike Snitzera79245b2010-08-12 04:14:24 +01001194 * check was performed.
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001195 */
Mike Snitzera79245b2010-08-12 04:14:24 +01001196 if (!ti->num_discard_requests)
1197 return -EOPNOTSUPP;
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001198
Mike Snitzera79245b2010-08-12 04:14:24 +01001199 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
Mike Snitzer06a426c2010-08-12 04:14:09 +01001200
Mike Snitzera79245b2010-08-12 04:14:24 +01001201 __issue_target_requests(ci, ti, ti->num_discard_requests, len);
1202
1203 ci->sector += len;
1204 } while (ci->sector_count -= len);
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001205
1206 return 0;
1207}
1208
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001209static int __clone_and_map(struct clone_info *ci)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210{
1211 struct bio *clone, *bio = ci->bio;
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001212 struct dm_target *ti;
1213 sector_t len = 0, max;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001214 struct dm_target_io *tio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001216 if (unlikely(bio->bi_rw & REQ_DISCARD))
1217 return __clone_and_map_discard(ci);
1218
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001219 ti = dm_table_find_target(ci->map, ci->sector);
1220 if (!dm_target_is_valid(ti))
1221 return -EIO;
1222
Mike Snitzer56a67df2010-08-12 04:14:10 +01001223 max = max_io_len(ci->sector, ti);
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001224
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 if (ci->sector_count <= max) {
1226 /*
1227 * Optimise for the simple case where we can do all of
1228 * the remaining io with a single clone.
1229 */
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001230 __clone_and_map_simple(ci, ti);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231
1232 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1233 /*
1234 * There are some bvecs that don't span targets.
1235 * Do as many of these as possible.
1236 */
1237 int i;
1238 sector_t remaining = max;
1239 sector_t bv_len;
1240
1241 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1242 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1243
1244 if (bv_len > remaining)
1245 break;
1246
1247 remaining -= bv_len;
1248 len += bv_len;
1249 }
1250
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001251 tio = alloc_tio(ci, ti);
Stefan Bader9faf4002006-10-03 01:15:41 -07001252 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
1253 ci->md->bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 __map_bio(ti, clone, tio);
1255
1256 ci->sector += len;
1257 ci->sector_count -= len;
1258 ci->idx = i;
1259
1260 } else {
1261 /*
Alasdair G Kergond2044a92006-03-22 00:07:42 -08001262 * Handle a bvec that must be split between two or more targets.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 */
1264 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
Alasdair G Kergond2044a92006-03-22 00:07:42 -08001265 sector_t remaining = to_sector(bv->bv_len);
1266 unsigned int offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267
Alasdair G Kergond2044a92006-03-22 00:07:42 -08001268 do {
1269 if (offset) {
1270 ti = dm_table_find_target(ci->map, ci->sector);
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001271 if (!dm_target_is_valid(ti))
1272 return -EIO;
1273
Mike Snitzer56a67df2010-08-12 04:14:10 +01001274 max = max_io_len(ci->sector, ti);
Alasdair G Kergond2044a92006-03-22 00:07:42 -08001275 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276
Alasdair G Kergond2044a92006-03-22 00:07:42 -08001277 len = min(remaining, max);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001279 tio = alloc_tio(ci, ti);
Alasdair G Kergond2044a92006-03-22 00:07:42 -08001280 clone = split_bvec(bio, ci->sector, ci->idx,
Stefan Bader9faf4002006-10-03 01:15:41 -07001281 bv->bv_offset + offset, len,
1282 ci->md->bs);
Alasdair G Kergond2044a92006-03-22 00:07:42 -08001283
1284 __map_bio(ti, clone, tio);
1285
1286 ci->sector += len;
1287 ci->sector_count -= len;
1288 offset += to_bytes(len);
1289 } while (remaining -= len);
1290
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 ci->idx++;
1292 }
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001293
1294 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295}
1296
1297/*
Mikulas Patocka8a53c282009-04-02 19:55:37 +01001298 * Split the bio into several clones and submit it to targets.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 */
Mikulas Patockaf0b9a452009-04-02 19:55:38 +01001300static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301{
1302 struct clone_info ci;
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001303 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
Alasdair G Kergon7c666412009-12-10 23:52:19 +00001305 ci.map = dm_get_live_table(md);
Mikulas Patockaf0b9a452009-04-02 19:55:38 +01001306 if (unlikely(!ci.map)) {
Tejun Heo6a8736d2010-09-08 18:07:00 +02001307 bio_io_error(bio);
Mikulas Patockaf0b9a452009-04-02 19:55:38 +01001308 return;
1309 }
Mikulas Patocka692d0eb2009-04-09 00:27:13 +01001310
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 ci.md = md;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 ci.io = alloc_io(md);
1313 ci.io->error = 0;
1314 atomic_set(&ci.io->io_count, 1);
1315 ci.io->bio = bio;
1316 ci.io->md = md;
Kiyoshi Uedaf88fb982009-10-16 23:18:15 +01001317 spin_lock_init(&ci.io->endio_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 ci.sector = bio->bi_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 ci.idx = bio->bi_idx;
1320
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -08001321 start_io_acct(ci.io);
Mike Snitzerb372d362010-09-08 18:07:01 +02001322 if (bio->bi_rw & REQ_FLUSH) {
1323 ci.bio = &ci.md->flush_bio;
1324 ci.sector_count = 0;
1325 error = __clone_and_map_empty_flush(&ci);
1326 /* dec_pending submits any data associated with flush */
1327 } else {
Tejun Heo6a8736d2010-09-08 18:07:00 +02001328 ci.bio = bio;
Tejun Heod87f4c12010-09-03 11:56:19 +02001329 ci.sector_count = bio_sectors(bio);
Mike Snitzerb372d362010-09-08 18:07:01 +02001330 while (ci.sector_count && !error)
Tejun Heod87f4c12010-09-03 11:56:19 +02001331 error = __clone_and_map(&ci);
Tejun Heod87f4c12010-09-03 11:56:19 +02001332 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
1334 /* drop the extra reference count */
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001335 dec_pending(ci.io, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 dm_table_put(ci.map);
1337}
1338/*-----------------------------------------------------------------
1339 * CRUD END
1340 *---------------------------------------------------------------*/
1341
Milan Brozf6fccb12008-07-21 12:00:37 +01001342static int dm_merge_bvec(struct request_queue *q,
1343 struct bvec_merge_data *bvm,
1344 struct bio_vec *biovec)
1345{
1346 struct mapped_device *md = q->queuedata;
Alasdair G Kergon7c666412009-12-10 23:52:19 +00001347 struct dm_table *map = dm_get_live_table(md);
Milan Brozf6fccb12008-07-21 12:00:37 +01001348 struct dm_target *ti;
1349 sector_t max_sectors;
Mikulas Patocka50371082008-10-01 14:39:17 +01001350 int max_size = 0;
Milan Brozf6fccb12008-07-21 12:00:37 +01001351
1352 if (unlikely(!map))
Mikulas Patocka50371082008-10-01 14:39:17 +01001353 goto out;
Milan Brozf6fccb12008-07-21 12:00:37 +01001354
1355 ti = dm_table_find_target(map, bvm->bi_sector);
Mikulas Patockab01cd5a2008-10-01 14:39:24 +01001356 if (!dm_target_is_valid(ti))
1357 goto out_table;
Milan Brozf6fccb12008-07-21 12:00:37 +01001358
1359 /*
1360 * Find maximum amount of I/O that won't need splitting
1361 */
Mike Snitzer56a67df2010-08-12 04:14:10 +01001362 max_sectors = min(max_io_len(bvm->bi_sector, ti),
Milan Brozf6fccb12008-07-21 12:00:37 +01001363 (sector_t) BIO_MAX_SECTORS);
1364 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1365 if (max_size < 0)
1366 max_size = 0;
1367
1368 /*
1369 * merge_bvec_fn() returns number of bytes
1370 * it can accept at this offset
1371 * max is precomputed maximal io size
1372 */
1373 if (max_size && ti->type->merge)
1374 max_size = ti->type->merge(ti, bvm, biovec, max_size);
Mikulas Patocka8cbeb672009-06-22 10:12:14 +01001375 /*
1376 * If the target doesn't support merge method and some of the devices
1377 * provided their merge_bvec method (we know this by looking at
1378 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1379 * entries. So always set max_size to 0, and the code below allows
1380 * just one page.
1381 */
1382 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1383
1384 max_size = 0;
Milan Brozf6fccb12008-07-21 12:00:37 +01001385
Mikulas Patockab01cd5a2008-10-01 14:39:24 +01001386out_table:
Mikulas Patocka50371082008-10-01 14:39:17 +01001387 dm_table_put(map);
1388
1389out:
Milan Brozf6fccb12008-07-21 12:00:37 +01001390 /*
1391 * Always allow an entire first page
1392 */
1393 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1394 max_size = biovec->bv_len;
1395
Milan Brozf6fccb12008-07-21 12:00:37 +01001396 return max_size;
1397}
1398
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399/*
1400 * The request function that just remaps the bio built up by
1401 * dm_merge_bvec.
1402 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001403static void _dm_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404{
Kevin Corry12f03a42006-02-01 03:04:52 -08001405 int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 struct mapped_device *md = q->queuedata;
Tejun Heoc9959052008-08-25 19:47:21 +09001407 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001409 down_read(&md->io_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410
Tejun Heo074a7ac2008-08-25 19:56:14 +09001411 cpu = part_stat_lock();
1412 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1413 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1414 part_stat_unlock();
Kevin Corry12f03a42006-02-01 03:04:52 -08001415
Tejun Heo6a8736d2010-09-08 18:07:00 +02001416 /* if we're suspended, we have to queue this io for later */
1417 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001418 up_read(&md->io_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
Tejun Heo6a8736d2010-09-08 18:07:00 +02001420 if (bio_rw(bio) != READA)
1421 queue_io(md, bio);
1422 else
Alasdair G Kergon54d9a1b2009-04-09 00:27:14 +01001423 bio_io_error(bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001424 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 }
1426
Mikulas Patockaf0b9a452009-04-02 19:55:38 +01001427 __split_and_process_bio(md, bio);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001428 up_read(&md->io_lock);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001429 return;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001430}
1431
1432static int dm_request_based(struct mapped_device *md)
1433{
1434 return blk_queue_stackable(md->queue);
1435}
1436
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001437static void dm_request(struct request_queue *q, struct bio *bio)
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001438{
1439 struct mapped_device *md = q->queuedata;
1440
1441 if (dm_request_based(md))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001442 blk_queue_bio(q, bio);
1443 else
1444 _dm_request(q, bio);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001445}
1446
1447void dm_dispatch_request(struct request *rq)
1448{
1449 int r;
1450
1451 if (blk_queue_io_stat(rq->q))
1452 rq->cmd_flags |= REQ_IO_STAT;
1453
1454 rq->start_time = jiffies;
1455 r = blk_insert_cloned_request(rq->q, rq);
1456 if (r)
1457 dm_complete_request(rq, r);
1458}
1459EXPORT_SYMBOL_GPL(dm_dispatch_request);
1460
1461static void dm_rq_bio_destructor(struct bio *bio)
1462{
1463 struct dm_rq_clone_bio_info *info = bio->bi_private;
1464 struct mapped_device *md = info->tio->md;
1465
1466 free_bio_info(info);
1467 bio_free(bio, md->bs);
1468}
1469
1470static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1471 void *data)
1472{
1473 struct dm_rq_target_io *tio = data;
1474 struct mapped_device *md = tio->md;
1475 struct dm_rq_clone_bio_info *info = alloc_bio_info(md);
1476
1477 if (!info)
1478 return -ENOMEM;
1479
1480 info->orig = bio_orig;
1481 info->tio = tio;
1482 bio->bi_end_io = end_clone_bio;
1483 bio->bi_private = info;
1484 bio->bi_destructor = dm_rq_bio_destructor;
1485
1486 return 0;
1487}
1488
1489static int setup_clone(struct request *clone, struct request *rq,
1490 struct dm_rq_target_io *tio)
1491{
Kiyoshi Uedad0bcb872009-12-10 23:52:18 +00001492 int r;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001493
Tejun Heo29e40132010-09-08 18:07:00 +02001494 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1495 dm_rq_bio_constructor, tio);
1496 if (r)
1497 return r;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001498
Tejun Heo29e40132010-09-08 18:07:00 +02001499 clone->cmd = rq->cmd;
1500 clone->cmd_len = rq->cmd_len;
1501 clone->sense = rq->sense;
1502 clone->buffer = rq->buffer;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001503 clone->end_io = end_clone_request;
1504 clone->end_io_data = tio;
1505
1506 return 0;
1507}
1508
Kiyoshi Ueda6facdaf2009-12-10 23:52:15 +00001509static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1510 gfp_t gfp_mask)
1511{
1512 struct request *clone;
1513 struct dm_rq_target_io *tio;
1514
1515 tio = alloc_rq_tio(md, gfp_mask);
1516 if (!tio)
1517 return NULL;
1518
1519 tio->md = md;
1520 tio->ti = NULL;
1521 tio->orig = rq;
1522 tio->error = 0;
1523 memset(&tio->info, 0, sizeof(tio->info));
1524
1525 clone = &tio->clone;
1526 if (setup_clone(clone, rq, tio)) {
1527 /* -ENOMEM */
1528 free_rq_tio(tio);
1529 return NULL;
1530 }
1531
1532 return clone;
1533}
1534
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001535/*
1536 * Called with the queue lock held.
1537 */
1538static int dm_prep_fn(struct request_queue *q, struct request *rq)
1539{
1540 struct mapped_device *md = q->queuedata;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001541 struct request *clone;
1542
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001543 if (unlikely(rq->special)) {
1544 DMWARN("Already has something in rq->special.");
1545 return BLKPREP_KILL;
1546 }
1547
Kiyoshi Ueda6facdaf2009-12-10 23:52:15 +00001548 clone = clone_rq(rq, md, GFP_ATOMIC);
1549 if (!clone)
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001550 return BLKPREP_DEFER;
1551
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001552 rq->special = clone;
1553 rq->cmd_flags |= REQ_DONTPREP;
1554
1555 return BLKPREP_OK;
1556}
1557
Kiyoshi Ueda9eef87d2010-02-16 18:43:01 +00001558/*
1559 * Returns:
1560 * 0 : the request has been processed (not requeued)
1561 * !0 : the request has been requeued
1562 */
1563static int map_request(struct dm_target *ti, struct request *clone,
1564 struct mapped_device *md)
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001565{
Kiyoshi Ueda9eef87d2010-02-16 18:43:01 +00001566 int r, requeued = 0;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001567 struct dm_rq_target_io *tio = clone->end_io_data;
1568
1569 /*
1570 * Hold the md reference here for the in-flight I/O.
1571 * We can't rely on the reference count by device opener,
1572 * because the device may be closed during the request completion
1573 * when all bios are completed.
1574 * See the comment in rq_completed() too.
1575 */
1576 dm_get(md);
1577
1578 tio->ti = ti;
1579 r = ti->type->map_rq(ti, clone, &tio->info);
1580 switch (r) {
1581 case DM_MAPIO_SUBMITTED:
1582 /* The target has taken the I/O to submit by itself later */
1583 break;
1584 case DM_MAPIO_REMAPPED:
1585 /* The target has remapped the I/O so dispatch it */
Jun'ichi Nomura6db4ccd2009-12-10 23:52:25 +00001586 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1587 blk_rq_pos(tio->orig));
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001588 dm_dispatch_request(clone);
1589 break;
1590 case DM_MAPIO_REQUEUE:
1591 /* The target wants to requeue the I/O */
1592 dm_requeue_unmapped_request(clone);
Kiyoshi Ueda9eef87d2010-02-16 18:43:01 +00001593 requeued = 1;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001594 break;
1595 default:
1596 if (r > 0) {
1597 DMWARN("unimplemented target map return value: %d", r);
1598 BUG();
1599 }
1600
1601 /* The target wants to complete the I/O */
1602 dm_kill_unmapped_request(clone, r);
1603 break;
1604 }
Kiyoshi Ueda9eef87d2010-02-16 18:43:01 +00001605
1606 return requeued;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001607}
1608
1609/*
1610 * q->request_fn for request-based dm.
1611 * Called with the queue lock held.
1612 */
1613static void dm_request_fn(struct request_queue *q)
1614{
1615 struct mapped_device *md = q->queuedata;
Alasdair G Kergon7c666412009-12-10 23:52:19 +00001616 struct dm_table *map = dm_get_live_table(md);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001617 struct dm_target *ti;
Kiyoshi Uedab4324fe2009-12-10 23:52:16 +00001618 struct request *rq, *clone;
Tejun Heo29e40132010-09-08 18:07:00 +02001619 sector_t pos;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001620
1621 /*
Kiyoshi Uedab4324fe2009-12-10 23:52:16 +00001622 * For suspend, check blk_queue_stopped() and increment
1623 * ->pending within a single queue_lock not to increment the
1624 * number of in-flight I/Os after the queue is stopped in
1625 * dm_suspend().
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001626 */
Jens Axboe7eaceac2011-03-10 08:52:07 +01001627 while (!blk_queue_stopped(q)) {
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001628 rq = blk_peek_request(q);
1629 if (!rq)
Jens Axboe7eaceac2011-03-10 08:52:07 +01001630 goto delay_and_out;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001631
Tejun Heo29e40132010-09-08 18:07:00 +02001632 /* always use block 0 to find the target for flushes for now */
1633 pos = 0;
1634 if (!(rq->cmd_flags & REQ_FLUSH))
1635 pos = blk_rq_pos(rq);
Kiyoshi Uedad0bcb872009-12-10 23:52:18 +00001636
Tejun Heo29e40132010-09-08 18:07:00 +02001637 ti = dm_table_find_target(map, pos);
1638 BUG_ON(!dm_target_is_valid(ti));
1639
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001640 if (ti->type->busy && ti->type->busy(ti))
Jens Axboe7eaceac2011-03-10 08:52:07 +01001641 goto delay_and_out;
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001642
1643 blk_start_request(rq);
Kiyoshi Uedab4324fe2009-12-10 23:52:16 +00001644 clone = rq->special;
1645 atomic_inc(&md->pending[rq_data_dir(clone)]);
1646
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001647 spin_unlock(q->queue_lock);
Kiyoshi Ueda9eef87d2010-02-16 18:43:01 +00001648 if (map_request(ti, clone, md))
1649 goto requeued;
1650
Kiyoshi Ueda052189a2011-01-13 20:00:00 +00001651 BUG_ON(!irqs_disabled());
1652 spin_lock(q->queue_lock);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001653 }
1654
1655 goto out;
1656
Kiyoshi Ueda9eef87d2010-02-16 18:43:01 +00001657requeued:
Kiyoshi Ueda052189a2011-01-13 20:00:00 +00001658 BUG_ON(!irqs_disabled());
1659 spin_lock(q->queue_lock);
Kiyoshi Ueda9eef87d2010-02-16 18:43:01 +00001660
Jens Axboe7eaceac2011-03-10 08:52:07 +01001661delay_and_out:
1662 blk_delay_queue(q, HZ / 10);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001663out:
1664 dm_table_put(map);
1665
1666 return;
1667}
1668
1669int dm_underlying_device_busy(struct request_queue *q)
1670{
1671 return blk_lld_busy(q);
1672}
1673EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1674
1675static int dm_lld_busy(struct request_queue *q)
1676{
1677 int r;
1678 struct mapped_device *md = q->queuedata;
Alasdair G Kergon7c666412009-12-10 23:52:19 +00001679 struct dm_table *map = dm_get_live_table(md);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001680
1681 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1682 r = 1;
1683 else
1684 r = dm_table_any_busy_target(map);
1685
1686 dm_table_put(map);
1687
1688 return r;
1689}
1690
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691static int dm_any_congested(void *congested_data, int bdi_bits)
1692{
Chandra Seetharaman8a57dfc2008-11-13 23:39:14 +00001693 int r = bdi_bits;
1694 struct mapped_device *md = congested_data;
1695 struct dm_table *map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696
Alasdair G Kergon1eb787e2009-04-09 00:27:14 +01001697 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
Alasdair G Kergon7c666412009-12-10 23:52:19 +00001698 map = dm_get_live_table(md);
Chandra Seetharaman8a57dfc2008-11-13 23:39:14 +00001699 if (map) {
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001700 /*
1701 * Request-based dm cares about only own queue for
1702 * the query about congestion status of request_queue
1703 */
1704 if (dm_request_based(md))
1705 r = md->queue->backing_dev_info.state &
1706 bdi_bits;
1707 else
1708 r = dm_table_any_congested(map, bdi_bits);
1709
Chandra Seetharaman8a57dfc2008-11-13 23:39:14 +00001710 dm_table_put(map);
1711 }
1712 }
1713
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 return r;
1715}
1716
1717/*-----------------------------------------------------------------
1718 * An IDR is used to keep track of allocated minor numbers.
1719 *---------------------------------------------------------------*/
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07001720static void free_minor(int minor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721{
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001722 spin_lock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 idr_remove(&_minor_idr, minor);
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001724 spin_unlock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725}
1726
1727/*
1728 * See if the device with a specific minor # is free.
1729 */
Frederik Deweerdtcf13ab82008-04-24 22:10:59 +01001730static int specific_minor(int minor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731{
1732 int r, m;
1733
1734 if (minor >= (1 << MINORBITS))
1735 return -EINVAL;
1736
Jeff Mahoney62f75c22006-06-26 00:27:21 -07001737 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1738 if (!r)
1739 return -ENOMEM;
1740
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001741 spin_lock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742
1743 if (idr_find(&_minor_idr, minor)) {
1744 r = -EBUSY;
1745 goto out;
1746 }
1747
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -07001748 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
Jeff Mahoney62f75c22006-06-26 00:27:21 -07001749 if (r)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751
1752 if (m != minor) {
1753 idr_remove(&_minor_idr, m);
1754 r = -EBUSY;
1755 goto out;
1756 }
1757
1758out:
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001759 spin_unlock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 return r;
1761}
1762
Frederik Deweerdtcf13ab82008-04-24 22:10:59 +01001763static int next_free_minor(int *minor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764{
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07001765 int r, m;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
Jeff Mahoney62f75c22006-06-26 00:27:21 -07001768 if (!r)
1769 return -ENOMEM;
1770
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001771 spin_lock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -07001773 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
Frederik Deweerdtcf13ab82008-04-24 22:10:59 +01001774 if (r)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776
1777 if (m >= (1 << MINORBITS)) {
1778 idr_remove(&_minor_idr, m);
1779 r = -ENOSPC;
1780 goto out;
1781 }
1782
1783 *minor = m;
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
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001790static const struct block_device_operations dm_blk_dops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791
Mikulas Patocka53d59142009-04-02 19:55:37 +01001792static void dm_wq_work(struct work_struct *work);
1793
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +01001794static void dm_init_md_queue(struct mapped_device *md)
1795{
1796 /*
1797 * Request-based dm devices cannot be stacked on top of bio-based dm
1798 * devices. The type of this dm device has not been decided yet.
1799 * The type is decided at the first table loading time.
1800 * To prevent problematic device stacking, clear the queue flag
1801 * for request stacking support until then.
1802 *
1803 * This queue is new, so no concurrency on the queue_flags.
1804 */
1805 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1806
1807 md->queue->queuedata = md;
1808 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1809 md->queue->backing_dev_info.congested_data = md;
1810 blk_queue_make_request(md->queue, dm_request);
1811 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +01001812 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1813}
1814
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815/*
1816 * Allocate and initialise a blank device with a given minor.
1817 */
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07001818static struct mapped_device *alloc_dev(int minor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819{
1820 int r;
Frederik Deweerdtcf13ab82008-04-24 22:10:59 +01001821 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -07001822 void *old_md;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823
1824 if (!md) {
1825 DMWARN("unable to allocate device, out of memory.");
1826 return NULL;
1827 }
1828
Jeff Mahoney10da4f72006-06-26 00:27:25 -07001829 if (!try_module_get(THIS_MODULE))
Milan Broz6ed7ade2008-02-08 02:10:19 +00001830 goto bad_module_get;
Jeff Mahoney10da4f72006-06-26 00:27:25 -07001831
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 /* get a minor number for the dev */
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07001833 if (minor == DM_ANY_MINOR)
Frederik Deweerdtcf13ab82008-04-24 22:10:59 +01001834 r = next_free_minor(&minor);
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07001835 else
Frederik Deweerdtcf13ab82008-04-24 22:10:59 +01001836 r = specific_minor(minor);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 if (r < 0)
Milan Broz6ed7ade2008-02-08 02:10:19 +00001838 goto bad_minor;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839
Mike Snitzera5664da2010-08-12 04:14:01 +01001840 md->type = DM_TYPE_NONE;
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001841 init_rwsem(&md->io_lock);
Daniel Walkere61290a2008-02-08 02:10:08 +00001842 mutex_init(&md->suspend_lock);
Mike Snitzera5664da2010-08-12 04:14:01 +01001843 mutex_init(&md->type_lock);
Mikulas Patocka022c2612009-04-02 19:55:39 +01001844 spin_lock_init(&md->deferred_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 rwlock_init(&md->map_lock);
1846 atomic_set(&md->holders, 1);
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -07001847 atomic_set(&md->open_count, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 atomic_set(&md->event_nr, 0);
Mike Anderson7a8c3d32007-10-19 22:48:01 +01001849 atomic_set(&md->uevent_seq, 0);
1850 INIT_LIST_HEAD(&md->uevent_list);
1851 spin_lock_init(&md->uevent_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +01001853 md->queue = blk_alloc_queue(GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 if (!md->queue)
Milan Broz6ed7ade2008-02-08 02:10:19 +00001855 goto bad_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +01001857 dm_init_md_queue(md);
Stefan Bader9faf4002006-10-03 01:15:41 -07001858
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 md->disk = alloc_disk(1);
1860 if (!md->disk)
Milan Broz6ed7ade2008-02-08 02:10:19 +00001861 goto bad_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001863 atomic_set(&md->pending[0], 0);
1864 atomic_set(&md->pending[1], 0);
Jeff Mahoneyf0b04112006-06-26 00:27:25 -07001865 init_waitqueue_head(&md->wait);
Mikulas Patocka53d59142009-04-02 19:55:37 +01001866 INIT_WORK(&md->work, dm_wq_work);
Jeff Mahoneyf0b04112006-06-26 00:27:25 -07001867 init_waitqueue_head(&md->eventq);
1868
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 md->disk->major = _major;
1870 md->disk->first_minor = minor;
1871 md->disk->fops = &dm_blk_dops;
1872 md->disk->queue = md->queue;
1873 md->disk->private_data = md;
1874 sprintf(md->disk->disk_name, "dm-%d", minor);
1875 add_disk(md->disk);
Mike Anderson7e51f252006-03-27 01:17:52 -08001876 format_dev_t(md->name, MKDEV(_major, minor));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877
Tejun Heo9c4376d2011-01-13 19:59:58 +00001878 md->wq = alloc_workqueue("kdmflush",
1879 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
Milan Broz304f3f62008-02-08 02:11:17 +00001880 if (!md->wq)
1881 goto bad_thread;
1882
Mikulas Patocka32a926d2009-06-22 10:12:17 +01001883 md->bdev = bdget_disk(md->disk, 0);
1884 if (!md->bdev)
1885 goto bad_bdev;
1886
Tejun Heo6a8736d2010-09-08 18:07:00 +02001887 bio_init(&md->flush_bio);
1888 md->flush_bio.bi_bdev = md->bdev;
1889 md->flush_bio.bi_rw = WRITE_FLUSH;
1890
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -07001891 /* Populate the mapping, nobody knows we exist yet */
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001892 spin_lock(&_minor_lock);
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -07001893 old_md = idr_replace(&_minor_idr, md, minor);
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001894 spin_unlock(&_minor_lock);
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -07001895
1896 BUG_ON(old_md != MINOR_ALLOCED);
1897
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 return md;
1899
Mikulas Patocka32a926d2009-06-22 10:12:17 +01001900bad_bdev:
1901 destroy_workqueue(md->wq);
Milan Broz304f3f62008-02-08 02:11:17 +00001902bad_thread:
Zdenek Kabelac03022c52009-10-16 23:18:15 +01001903 del_gendisk(md->disk);
Milan Broz304f3f62008-02-08 02:11:17 +00001904 put_disk(md->disk);
Milan Broz6ed7ade2008-02-08 02:10:19 +00001905bad_disk:
Al Viro1312f402006-03-12 11:02:03 -05001906 blk_cleanup_queue(md->queue);
Milan Broz6ed7ade2008-02-08 02:10:19 +00001907bad_queue:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 free_minor(minor);
Milan Broz6ed7ade2008-02-08 02:10:19 +00001909bad_minor:
Jeff Mahoney10da4f72006-06-26 00:27:25 -07001910 module_put(THIS_MODULE);
Milan Broz6ed7ade2008-02-08 02:10:19 +00001911bad_module_get:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 kfree(md);
1913 return NULL;
1914}
1915
Jun'ichi Nomuraae9da832007-10-19 22:38:43 +01001916static void unlock_fs(struct mapped_device *md);
1917
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918static void free_dev(struct mapped_device *md)
1919{
Tejun Heof331c022008-09-03 09:01:48 +02001920 int minor = MINOR(disk_devt(md->disk));
Jun'ichi Nomura63d94e42006-02-24 13:04:25 -08001921
Mikulas Patocka32a926d2009-06-22 10:12:17 +01001922 unlock_fs(md);
1923 bdput(md->bdev);
Milan Broz304f3f62008-02-08 02:11:17 +00001924 destroy_workqueue(md->wq);
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001925 if (md->tio_pool)
1926 mempool_destroy(md->tio_pool);
1927 if (md->io_pool)
1928 mempool_destroy(md->io_pool);
1929 if (md->bs)
1930 bioset_free(md->bs);
Martin K. Petersen9c470082009-04-09 00:27:12 +01001931 blk_integrity_unregister(md->disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 del_gendisk(md->disk);
Jun'ichi Nomura63d94e42006-02-24 13:04:25 -08001933 free_minor(minor);
Jeff Mahoneyfba9f902006-06-26 00:27:23 -07001934
1935 spin_lock(&_minor_lock);
1936 md->disk->private_data = NULL;
1937 spin_unlock(&_minor_lock);
1938
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 put_disk(md->disk);
Al Viro1312f402006-03-12 11:02:03 -05001940 blk_cleanup_queue(md->queue);
Jeff Mahoney10da4f72006-06-26 00:27:25 -07001941 module_put(THIS_MODULE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 kfree(md);
1943}
1944
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001945static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1946{
1947 struct dm_md_mempools *p;
1948
1949 if (md->io_pool && md->tio_pool && md->bs)
1950 /* the md already has necessary mempools */
1951 goto out;
1952
1953 p = dm_table_get_md_mempools(t);
1954 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
1955
1956 md->io_pool = p->io_pool;
1957 p->io_pool = NULL;
1958 md->tio_pool = p->tio_pool;
1959 p->tio_pool = NULL;
1960 md->bs = p->bs;
1961 p->bs = NULL;
1962
1963out:
1964 /* mempool bind completed, now no need any mempools in the table */
1965 dm_table_free_md_mempools(t);
1966}
1967
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968/*
1969 * Bind a table to the device.
1970 */
1971static void event_callback(void *context)
1972{
Mike Anderson7a8c3d32007-10-19 22:48:01 +01001973 unsigned long flags;
1974 LIST_HEAD(uevents);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 struct mapped_device *md = (struct mapped_device *) context;
1976
Mike Anderson7a8c3d32007-10-19 22:48:01 +01001977 spin_lock_irqsave(&md->uevent_lock, flags);
1978 list_splice_init(&md->uevent_list, &uevents);
1979 spin_unlock_irqrestore(&md->uevent_lock, flags);
1980
Tejun Heoed9e1982008-08-25 19:56:05 +09001981 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
Mike Anderson7a8c3d32007-10-19 22:48:01 +01001982
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 atomic_inc(&md->event_nr);
1984 wake_up(&md->eventq);
1985}
1986
Mike Snitzerc2176492011-01-13 19:53:46 +00001987/*
1988 * Protected by md->suspend_lock obtained by dm_swap_table().
1989 */
Alasdair G Kergon4e90188be2005-07-28 21:15:59 -07001990static void __set_size(struct mapped_device *md, sector_t size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991{
Alasdair G Kergon4e90188be2005-07-28 21:15:59 -07001992 set_capacity(md->disk, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993
Mikulas Patockadb8fef42009-06-22 10:12:15 +01001994 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995}
1996
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00001997/*
Mikulas Patockad5b9dd02011-08-02 12:32:04 +01001998 * Return 1 if the queue has a compulsory merge_bvec_fn function.
1999 *
2000 * If this function returns 0, then the device is either a non-dm
2001 * device without a merge_bvec_fn, or it is a dm device that is
2002 * able to split any bios it receives that are too big.
2003 */
2004int dm_queue_merge_is_compulsory(struct request_queue *q)
2005{
2006 struct mapped_device *dev_md;
2007
2008 if (!q->merge_bvec_fn)
2009 return 0;
2010
2011 if (q->make_request_fn == dm_request) {
2012 dev_md = q->queuedata;
2013 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2014 return 0;
2015 }
2016
2017 return 1;
2018}
2019
2020static int dm_device_merge_is_compulsory(struct dm_target *ti,
2021 struct dm_dev *dev, sector_t start,
2022 sector_t len, void *data)
2023{
2024 struct block_device *bdev = dev->bdev;
2025 struct request_queue *q = bdev_get_queue(bdev);
2026
2027 return dm_queue_merge_is_compulsory(q);
2028}
2029
2030/*
2031 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2032 * on the properties of the underlying devices.
2033 */
2034static int dm_table_merge_is_optional(struct dm_table *table)
2035{
2036 unsigned i = 0;
2037 struct dm_target *ti;
2038
2039 while (i < dm_table_get_num_targets(table)) {
2040 ti = dm_table_get_target(table, i++);
2041
2042 if (ti->type->iterate_devices &&
2043 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2044 return 0;
2045 }
2046
2047 return 1;
2048}
2049
2050/*
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002051 * Returns old map, which caller must destroy.
2052 */
2053static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2054 struct queue_limits *limits)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055{
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002056 struct dm_table *old_map;
Jens Axboe165125e2007-07-24 09:28:11 +02002057 struct request_queue *q = md->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 sector_t size;
Kiyoshi Ueda523d9292009-06-22 10:12:37 +01002059 unsigned long flags;
Mikulas Patockad5b9dd02011-08-02 12:32:04 +01002060 int merge_is_optional;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061
2062 size = dm_table_get_size(t);
Darrick J. Wong3ac51e72006-03-27 01:17:54 -08002063
2064 /*
2065 * Wipe any geometry if the size of the table changed.
2066 */
2067 if (size != get_capacity(md->disk))
2068 memset(&md->geometry, 0, sizeof(md->geometry));
2069
Mikulas Patocka32a926d2009-06-22 10:12:17 +01002070 __set_size(md, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002072 dm_table_event_callback(t, event_callback, md);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002073
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002074 /*
2075 * The queue hasn't been stopped yet, if the old table type wasn't
2076 * for request-based during suspension. So stop it to prevent
2077 * I/O mapping before resume.
2078 * This must be done before setting the queue restrictions,
2079 * because request-based dm may be run just after the setting.
2080 */
2081 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2082 stop_queue(q);
2083
2084 __bind_mempools(md, t);
2085
Mikulas Patockad5b9dd02011-08-02 12:32:04 +01002086 merge_is_optional = dm_table_merge_is_optional(t);
2087
Kiyoshi Ueda523d9292009-06-22 10:12:37 +01002088 write_lock_irqsave(&md->map_lock, flags);
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002089 old_map = md->map;
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002090 md->map = t;
Alasdair G Kergon36a04562011-10-31 20:19:04 +00002091 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2092
Mike Snitzer754c5fc2009-06-22 10:12:34 +01002093 dm_table_set_restrictions(t, q, limits);
Mikulas Patockad5b9dd02011-08-02 12:32:04 +01002094 if (merge_is_optional)
2095 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2096 else
2097 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
Kiyoshi Ueda523d9292009-06-22 10:12:37 +01002098 write_unlock_irqrestore(&md->map_lock, flags);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002099
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002100 return old_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101}
2102
Alasdair G Kergona7940152009-12-10 23:52:23 +00002103/*
2104 * Returns unbound table for the caller to free.
2105 */
2106static struct dm_table *__unbind(struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107{
2108 struct dm_table *map = md->map;
Kiyoshi Ueda523d9292009-06-22 10:12:37 +01002109 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110
2111 if (!map)
Alasdair G Kergona7940152009-12-10 23:52:23 +00002112 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113
2114 dm_table_event_callback(map, NULL, NULL);
Kiyoshi Ueda523d9292009-06-22 10:12:37 +01002115 write_lock_irqsave(&md->map_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 md->map = NULL;
Kiyoshi Ueda523d9292009-06-22 10:12:37 +01002117 write_unlock_irqrestore(&md->map_lock, flags);
Alasdair G Kergona7940152009-12-10 23:52:23 +00002118
2119 return map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120}
2121
2122/*
2123 * Constructor for a new device.
2124 */
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07002125int dm_create(int minor, struct mapped_device **result)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126{
2127 struct mapped_device *md;
2128
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07002129 md = alloc_dev(minor);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 if (!md)
2131 return -ENXIO;
2132
Milan Broz784aae72009-01-06 03:05:12 +00002133 dm_sysfs_init(md);
2134
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 *result = md;
2136 return 0;
2137}
2138
Mike Snitzera5664da2010-08-12 04:14:01 +01002139/*
2140 * Functions to manage md->type.
2141 * All are required to hold md->type_lock.
2142 */
2143void dm_lock_md_type(struct mapped_device *md)
2144{
2145 mutex_lock(&md->type_lock);
2146}
2147
2148void dm_unlock_md_type(struct mapped_device *md)
2149{
2150 mutex_unlock(&md->type_lock);
2151}
2152
2153void dm_set_md_type(struct mapped_device *md, unsigned type)
2154{
2155 md->type = type;
2156}
2157
2158unsigned dm_get_md_type(struct mapped_device *md)
2159{
2160 return md->type;
2161}
2162
Alasdair G Kergon36a04562011-10-31 20:19:04 +00002163struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2164{
2165 return md->immutable_target_type;
2166}
2167
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +01002168/*
2169 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2170 */
2171static int dm_init_request_based_queue(struct mapped_device *md)
2172{
2173 struct request_queue *q = NULL;
2174
2175 if (md->queue->elevator)
2176 return 1;
2177
2178 /* Fully initialize the queue */
2179 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2180 if (!q)
2181 return 0;
2182
2183 md->queue = q;
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +01002184 dm_init_md_queue(md);
2185 blk_queue_softirq_done(md->queue, dm_softirq_done);
2186 blk_queue_prep_rq(md->queue, dm_prep_fn);
2187 blk_queue_lld_busy(md->queue, dm_lld_busy);
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +01002188
2189 elv_register_queue(md->queue);
2190
2191 return 1;
2192}
2193
2194/*
2195 * Setup the DM device's queue based on md's type
2196 */
2197int dm_setup_md_queue(struct mapped_device *md)
2198{
2199 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2200 !dm_init_request_based_queue(md)) {
2201 DMWARN("Cannot initialize queue for request-based mapped device");
2202 return -EINVAL;
2203 }
2204
2205 return 0;
2206}
2207
David Teigland637842c2006-01-06 00:20:00 -08002208static struct mapped_device *dm_find_md(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209{
2210 struct mapped_device *md;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 unsigned minor = MINOR(dev);
2212
2213 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2214 return NULL;
2215
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07002216 spin_lock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217
2218 md = idr_find(&_minor_idr, minor);
Jeff Mahoneyfba9f902006-06-26 00:27:23 -07002219 if (md && (md == MINOR_ALLOCED ||
Tejun Heof331c022008-09-03 09:01:48 +02002220 (MINOR(disk_devt(dm_disk(md))) != minor) ||
Kiyoshi Uedaabdc5682010-08-12 04:13:54 +01002221 dm_deleting_md(md) ||
Alasdair G Kergon17b2f662006-06-26 00:27:33 -07002222 test_bit(DMF_FREEING, &md->flags))) {
David Teigland637842c2006-01-06 00:20:00 -08002223 md = NULL;
Jeff Mahoneyfba9f902006-06-26 00:27:23 -07002224 goto out;
2225 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226
Jeff Mahoneyfba9f902006-06-26 00:27:23 -07002227out:
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07002228 spin_unlock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229
David Teigland637842c2006-01-06 00:20:00 -08002230 return md;
2231}
2232
David Teiglandd229a952006-01-06 00:20:01 -08002233struct mapped_device *dm_get_md(dev_t dev)
2234{
2235 struct mapped_device *md = dm_find_md(dev);
2236
2237 if (md)
2238 dm_get(md);
2239
2240 return md;
2241}
Alasdair G Kergon3cf2e4b2011-10-31 20:19:06 +00002242EXPORT_SYMBOL_GPL(dm_get_md);
David Teiglandd229a952006-01-06 00:20:01 -08002243
Alasdair G Kergon9ade92a2006-03-27 01:17:53 -08002244void *dm_get_mdptr(struct mapped_device *md)
David Teigland637842c2006-01-06 00:20:00 -08002245{
Alasdair G Kergon9ade92a2006-03-27 01:17:53 -08002246 return md->interface_ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247}
2248
2249void dm_set_mdptr(struct mapped_device *md, void *ptr)
2250{
2251 md->interface_ptr = ptr;
2252}
2253
2254void dm_get(struct mapped_device *md)
2255{
2256 atomic_inc(&md->holders);
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +01002257 BUG_ON(test_bit(DMF_FREEING, &md->flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258}
2259
Alasdair G Kergon72d94862006-06-26 00:27:35 -07002260const char *dm_device_name(struct mapped_device *md)
2261{
2262 return md->name;
2263}
2264EXPORT_SYMBOL_GPL(dm_device_name);
2265
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +01002266static void __dm_destroy(struct mapped_device *md, bool wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267{
Mike Anderson1134e5a2006-03-27 01:17:54 -08002268 struct dm_table *map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +01002270 might_sleep();
Jeff Mahoneyfba9f902006-06-26 00:27:23 -07002271
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +01002272 spin_lock(&_minor_lock);
2273 map = dm_get_live_table(md);
2274 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2275 set_bit(DMF_FREEING, &md->flags);
2276 spin_unlock(&_minor_lock);
2277
2278 if (!dm_suspended_md(md)) {
2279 dm_table_presuspend_targets(map);
2280 dm_table_postsuspend_targets(map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 }
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +01002282
2283 /*
2284 * Rare, but there may be I/O requests still going to complete,
2285 * for example. Wait for all references to disappear.
2286 * No one should increment the reference count of the mapped_device,
2287 * after the mapped_device state becomes DMF_FREEING.
2288 */
2289 if (wait)
2290 while (atomic_read(&md->holders))
2291 msleep(1);
2292 else if (atomic_read(&md->holders))
2293 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2294 dm_device_name(md), atomic_read(&md->holders));
2295
2296 dm_sysfs_exit(md);
2297 dm_table_put(map);
2298 dm_table_destroy(__unbind(md));
2299 free_dev(md);
2300}
2301
2302void dm_destroy(struct mapped_device *md)
2303{
2304 __dm_destroy(md, true);
2305}
2306
2307void dm_destroy_immediate(struct mapped_device *md)
2308{
2309 __dm_destroy(md, false);
2310}
2311
2312void dm_put(struct mapped_device *md)
2313{
2314 atomic_dec(&md->holders);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315}
Edward Goggin79eb8852007-05-09 02:32:56 -07002316EXPORT_SYMBOL_GPL(dm_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317
Mikulas Patocka401600d2009-04-02 19:55:38 +01002318static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
Milan Broz46125c12008-02-08 02:10:30 +00002319{
2320 int r = 0;
Mikulas Patockab44ebeb2009-04-02 19:55:39 +01002321 DECLARE_WAITQUEUE(wait, current);
2322
Mikulas Patockab44ebeb2009-04-02 19:55:39 +01002323 add_wait_queue(&md->wait, &wait);
Milan Broz46125c12008-02-08 02:10:30 +00002324
2325 while (1) {
Mikulas Patocka401600d2009-04-02 19:55:38 +01002326 set_current_state(interruptible);
Milan Broz46125c12008-02-08 02:10:30 +00002327
Kiyoshi Uedab4324fe2009-12-10 23:52:16 +00002328 if (!md_in_flight(md))
Milan Broz46125c12008-02-08 02:10:30 +00002329 break;
2330
Mikulas Patocka401600d2009-04-02 19:55:38 +01002331 if (interruptible == TASK_INTERRUPTIBLE &&
2332 signal_pending(current)) {
Milan Broz46125c12008-02-08 02:10:30 +00002333 r = -EINTR;
2334 break;
2335 }
2336
2337 io_schedule();
2338 }
2339 set_current_state(TASK_RUNNING);
2340
Mikulas Patockab44ebeb2009-04-02 19:55:39 +01002341 remove_wait_queue(&md->wait, &wait);
2342
Milan Broz46125c12008-02-08 02:10:30 +00002343 return r;
2344}
2345
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346/*
2347 * Process the deferred bios
2348 */
Mikulas Patockaef208582009-04-02 19:55:38 +01002349static void dm_wq_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350{
Mikulas Patockaef208582009-04-02 19:55:38 +01002351 struct mapped_device *md = container_of(work, struct mapped_device,
2352 work);
Milan Broz6d6f10d2008-02-08 02:10:22 +00002353 struct bio *c;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354
Tejun Heo6a8736d2010-09-08 18:07:00 +02002355 down_read(&md->io_lock);
Mikulas Patockaef208582009-04-02 19:55:38 +01002356
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002357 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
Alasdair G Kergondf12ee92009-04-09 00:27:13 +01002358 spin_lock_irq(&md->deferred_lock);
2359 c = bio_list_pop(&md->deferred);
2360 spin_unlock_irq(&md->deferred_lock);
Mikulas Patocka022c2612009-04-02 19:55:39 +01002361
Tejun Heo6a8736d2010-09-08 18:07:00 +02002362 if (!c)
Alasdair G Kergondf12ee92009-04-09 00:27:13 +01002363 break;
Alasdair G Kergondf12ee92009-04-09 00:27:13 +01002364
Tejun Heo6a8736d2010-09-08 18:07:00 +02002365 up_read(&md->io_lock);
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002366
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002367 if (dm_request_based(md))
2368 generic_make_request(c);
Tejun Heo6a8736d2010-09-08 18:07:00 +02002369 else
2370 __split_and_process_bio(md, c);
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002371
Tejun Heo6a8736d2010-09-08 18:07:00 +02002372 down_read(&md->io_lock);
Mikulas Patocka022c2612009-04-02 19:55:39 +01002373 }
Milan Broz73d410c2008-02-08 02:10:25 +00002374
Tejun Heo6a8736d2010-09-08 18:07:00 +02002375 up_read(&md->io_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376}
2377
Mikulas Patocka9a1fb462009-04-02 19:55:36 +01002378static void dm_queue_flush(struct mapped_device *md)
Milan Broz304f3f62008-02-08 02:11:17 +00002379{
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002380 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2381 smp_mb__after_clear_bit();
Mikulas Patocka53d59142009-04-02 19:55:37 +01002382 queue_work(md->wq, &md->work);
Milan Broz304f3f62008-02-08 02:11:17 +00002383}
2384
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385/*
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002386 * Swap in a new table, returning the old one for the caller to destroy.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 */
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002388struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389{
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002390 struct dm_table *map = ERR_PTR(-EINVAL);
Mike Snitzer754c5fc2009-06-22 10:12:34 +01002391 struct queue_limits limits;
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002392 int r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393
Daniel Walkere61290a2008-02-08 02:10:08 +00002394 mutex_lock(&md->suspend_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395
2396 /* device must be suspended */
Kiyoshi Ueda4f186f82009-12-10 23:52:26 +00002397 if (!dm_suspended_md(md))
Alasdair G Kergon93c534a2005-07-12 15:53:05 -07002398 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399
Mike Snitzer754c5fc2009-06-22 10:12:34 +01002400 r = dm_calculate_queue_limits(table, &limits);
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002401 if (r) {
2402 map = ERR_PTR(r);
Mike Snitzer754c5fc2009-06-22 10:12:34 +01002403 goto out;
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002404 }
Mike Snitzer754c5fc2009-06-22 10:12:34 +01002405
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002406 map = __bind(md, table, &limits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407
Alasdair G Kergon93c534a2005-07-12 15:53:05 -07002408out:
Daniel Walkere61290a2008-02-08 02:10:08 +00002409 mutex_unlock(&md->suspend_lock);
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002410 return map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411}
2412
2413/*
2414 * Functions to lock and unlock any filesystem running on the
2415 * device.
2416 */
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002417static int lock_fs(struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418{
Alasdair G Kergone39e2e92006-01-06 00:20:05 -08002419 int r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420
2421 WARN_ON(md->frozen_sb);
Alasdair G Kergondfbe03f2005-05-05 16:16:04 -07002422
Mikulas Patockadb8fef42009-06-22 10:12:15 +01002423 md->frozen_sb = freeze_bdev(md->bdev);
Alasdair G Kergondfbe03f2005-05-05 16:16:04 -07002424 if (IS_ERR(md->frozen_sb)) {
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002425 r = PTR_ERR(md->frozen_sb);
Alasdair G Kergone39e2e92006-01-06 00:20:05 -08002426 md->frozen_sb = NULL;
2427 return r;
Alasdair G Kergondfbe03f2005-05-05 16:16:04 -07002428 }
2429
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -08002430 set_bit(DMF_FROZEN, &md->flags);
2431
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432 return 0;
2433}
2434
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002435static void unlock_fs(struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436{
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -08002437 if (!test_bit(DMF_FROZEN, &md->flags))
2438 return;
2439
Mikulas Patockadb8fef42009-06-22 10:12:15 +01002440 thaw_bdev(md->bdev, md->frozen_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 md->frozen_sb = NULL;
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -08002442 clear_bit(DMF_FROZEN, &md->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443}
2444
2445/*
2446 * We need to be able to change a mapping table under a mounted
2447 * filesystem. For example we might want to move some data in
2448 * the background. Before the table can be swapped with
2449 * dm_bind_table, dm_suspend must be called to flush any in
2450 * flight bios and ensure that any further io gets deferred.
2451 */
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002452/*
2453 * Suspend mechanism in request-based dm.
2454 *
Kiyoshi Ueda9f518b22009-12-10 23:52:16 +00002455 * 1. Flush all I/Os by lock_fs() if needed.
2456 * 2. Stop dispatching any I/O by stopping the request_queue.
2457 * 3. Wait for all in-flight I/Os to be completed or requeued.
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002458 *
Kiyoshi Ueda9f518b22009-12-10 23:52:16 +00002459 * To abort suspend, start the request_queue.
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002460 */
Kiyoshi Uedaa3d77d32006-12-08 02:41:04 -08002461int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462{
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002463 struct dm_table *map = NULL;
Milan Broz46125c12008-02-08 02:10:30 +00002464 int r = 0;
Kiyoshi Uedaa3d77d32006-12-08 02:41:04 -08002465 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -08002466 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467
Daniel Walkere61290a2008-02-08 02:10:08 +00002468 mutex_lock(&md->suspend_lock);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002469
Kiyoshi Ueda4f186f82009-12-10 23:52:26 +00002470 if (dm_suspended_md(md)) {
Milan Broz73d410c2008-02-08 02:10:25 +00002471 r = -EINVAL;
Alasdair G Kergond2874832006-11-08 17:44:43 -08002472 goto out_unlock;
Milan Broz73d410c2008-02-08 02:10:25 +00002473 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474
Alasdair G Kergon7c666412009-12-10 23:52:19 +00002475 map = dm_get_live_table(md);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -08002477 /*
2478 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2479 * This flag is cleared before dm_suspend returns.
2480 */
2481 if (noflush)
2482 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2483
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002484 /* This does not get reverted if there's an error later. */
2485 dm_table_presuspend_targets(map);
2486
Mikulas Patocka32a926d2009-06-22 10:12:17 +01002487 /*
Kiyoshi Ueda9f518b22009-12-10 23:52:16 +00002488 * Flush I/O to the device.
2489 * Any I/O submitted after lock_fs() may not be flushed.
2490 * noflush takes precedence over do_lockfs.
2491 * (lock_fs() flushes I/Os and waits for them to complete.)
Mikulas Patocka32a926d2009-06-22 10:12:17 +01002492 */
2493 if (!noflush && do_lockfs) {
2494 r = lock_fs(md);
2495 if (r)
Kiyoshi Uedaf431d962008-10-21 17:45:07 +01002496 goto out;
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -08002497 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498
2499 /*
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002500 * Here we must make sure that no processes are submitting requests
2501 * to target drivers i.e. no one may be executing
2502 * __split_and_process_bio. This is called from dm_request and
2503 * dm_wq_work.
2504 *
2505 * To get all processes out of __split_and_process_bio in dm_request,
2506 * we take the write lock. To prevent any process from reentering
Tejun Heo6a8736d2010-09-08 18:07:00 +02002507 * __split_and_process_bio from dm_request and quiesce the thread
2508 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2509 * flush_workqueue(md->wq).
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 */
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002511 down_write(&md->io_lock);
Alasdair G Kergon1eb787e2009-04-09 00:27:14 +01002512 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002513 up_write(&md->io_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514
Kiyoshi Uedad0bcb872009-12-10 23:52:18 +00002515 /*
Tejun Heo29e40132010-09-08 18:07:00 +02002516 * Stop md->queue before flushing md->wq in case request-based
2517 * dm defers requests to md->wq from md->queue.
Kiyoshi Uedad0bcb872009-12-10 23:52:18 +00002518 */
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002519 if (dm_request_based(md))
Kiyoshi Ueda9f518b22009-12-10 23:52:16 +00002520 stop_queue(md->queue);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002521
Kiyoshi Uedad0bcb872009-12-10 23:52:18 +00002522 flush_workqueue(md->wq);
2523
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 /*
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002525 * At this point no more requests are entering target request routines.
2526 * We call dm_wait_for_completion to wait for all existing requests
2527 * to finish.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 */
Mikulas Patocka401600d2009-04-02 19:55:38 +01002529 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002531 down_write(&md->io_lock);
Milan Broz6d6f10d2008-02-08 02:10:22 +00002532 if (noflush)
Mikulas Patocka022c2612009-04-02 19:55:39 +01002533 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
Milan Broz94d63512008-02-08 02:10:27 +00002534 up_write(&md->io_lock);
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -08002535
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 /* were we interrupted ? */
Milan Broz46125c12008-02-08 02:10:30 +00002537 if (r < 0) {
Mikulas Patocka9a1fb462009-04-02 19:55:36 +01002538 dm_queue_flush(md);
Milan Broz73d410c2008-02-08 02:10:25 +00002539
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002540 if (dm_request_based(md))
Kiyoshi Ueda9f518b22009-12-10 23:52:16 +00002541 start_queue(md->queue);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002542
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002543 unlock_fs(md);
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -08002544 goto out; /* pushback list is already flushed, so skip flush */
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002545 }
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002546
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002547 /*
2548 * If dm_wait_for_completion returned 0, the device is completely
2549 * quiescent now. There is no request-processing activity. All new
2550 * requests are being added to md->deferred list.
2551 */
2552
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 set_bit(DMF_SUSPENDED, &md->flags);
2554
Kiyoshi Ueda4d4471c2009-12-10 23:52:26 +00002555 dm_table_postsuspend_targets(map);
2556
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002557out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 dm_table_put(map);
Alasdair G Kergond2874832006-11-08 17:44:43 -08002559
2560out_unlock:
Daniel Walkere61290a2008-02-08 02:10:08 +00002561 mutex_unlock(&md->suspend_lock);
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002562 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563}
2564
2565int dm_resume(struct mapped_device *md)
2566{
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002567 int r = -EINVAL;
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002568 struct dm_table *map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569
Daniel Walkere61290a2008-02-08 02:10:08 +00002570 mutex_lock(&md->suspend_lock);
Kiyoshi Ueda4f186f82009-12-10 23:52:26 +00002571 if (!dm_suspended_md(md))
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002572 goto out;
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002573
Alasdair G Kergon7c666412009-12-10 23:52:19 +00002574 map = dm_get_live_table(md);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002575 if (!map || !dm_table_get_size(map))
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002576 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577
Milan Broz8757b772006-10-03 01:15:36 -07002578 r = dm_table_resume_targets(map);
2579 if (r)
2580 goto out;
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002581
Mikulas Patocka9a1fb462009-04-02 19:55:36 +01002582 dm_queue_flush(md);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002583
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002584 /*
2585 * Flushing deferred I/Os must be done after targets are resumed
2586 * so that mapping of targets can work correctly.
2587 * Request-based dm is queueing the deferred I/Os in its request_queue.
2588 */
2589 if (dm_request_based(md))
2590 start_queue(md->queue);
2591
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002592 unlock_fs(md);
2593
2594 clear_bit(DMF_SUSPENDED, &md->flags);
2595
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002596 r = 0;
2597out:
2598 dm_table_put(map);
Daniel Walkere61290a2008-02-08 02:10:08 +00002599 mutex_unlock(&md->suspend_lock);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002600
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002601 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602}
2603
2604/*-----------------------------------------------------------------
2605 * Event notification.
2606 *---------------------------------------------------------------*/
Peter Rajnoha3abf85b2010-03-06 02:32:31 +00002607int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
Milan Broz60935eb2009-06-22 10:12:30 +01002608 unsigned cookie)
Alasdair G Kergon69267a32007-12-13 14:15:57 +00002609{
Milan Broz60935eb2009-06-22 10:12:30 +01002610 char udev_cookie[DM_COOKIE_LENGTH];
2611 char *envp[] = { udev_cookie, NULL };
2612
2613 if (!cookie)
Peter Rajnoha3abf85b2010-03-06 02:32:31 +00002614 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
Milan Broz60935eb2009-06-22 10:12:30 +01002615 else {
2616 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2617 DM_COOKIE_ENV_VAR_NAME, cookie);
Peter Rajnoha3abf85b2010-03-06 02:32:31 +00002618 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2619 action, envp);
Milan Broz60935eb2009-06-22 10:12:30 +01002620 }
Alasdair G Kergon69267a32007-12-13 14:15:57 +00002621}
2622
Mike Anderson7a8c3d32007-10-19 22:48:01 +01002623uint32_t dm_next_uevent_seq(struct mapped_device *md)
2624{
2625 return atomic_add_return(1, &md->uevent_seq);
2626}
2627
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628uint32_t dm_get_event_nr(struct mapped_device *md)
2629{
2630 return atomic_read(&md->event_nr);
2631}
2632
2633int dm_wait_event(struct mapped_device *md, int event_nr)
2634{
2635 return wait_event_interruptible(md->eventq,
2636 (event_nr != atomic_read(&md->event_nr)));
2637}
2638
Mike Anderson7a8c3d32007-10-19 22:48:01 +01002639void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2640{
2641 unsigned long flags;
2642
2643 spin_lock_irqsave(&md->uevent_lock, flags);
2644 list_add(elist, &md->uevent_list);
2645 spin_unlock_irqrestore(&md->uevent_lock, flags);
2646}
2647
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648/*
2649 * The gendisk is only valid as long as you have a reference
2650 * count on 'md'.
2651 */
2652struct gendisk *dm_disk(struct mapped_device *md)
2653{
2654 return md->disk;
2655}
2656
Milan Broz784aae72009-01-06 03:05:12 +00002657struct kobject *dm_kobject(struct mapped_device *md)
2658{
2659 return &md->kobj;
2660}
2661
2662/*
2663 * struct mapped_device should not be exported outside of dm.c
2664 * so use this check to verify that kobj is part of md structure
2665 */
2666struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2667{
2668 struct mapped_device *md;
2669
2670 md = container_of(kobj, struct mapped_device, kobj);
2671 if (&md->kobj != kobj)
2672 return NULL;
2673
Milan Broz4d89b7b2009-06-22 10:12:11 +01002674 if (test_bit(DMF_FREEING, &md->flags) ||
Mike Anderson432a2122009-12-10 23:52:20 +00002675 dm_deleting_md(md))
Milan Broz4d89b7b2009-06-22 10:12:11 +01002676 return NULL;
2677
Milan Broz784aae72009-01-06 03:05:12 +00002678 dm_get(md);
2679 return md;
2680}
2681
Kiyoshi Ueda4f186f82009-12-10 23:52:26 +00002682int dm_suspended_md(struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683{
2684 return test_bit(DMF_SUSPENDED, &md->flags);
2685}
2686
Kiyoshi Ueda64dbce52009-12-10 23:52:27 +00002687int dm_suspended(struct dm_target *ti)
2688{
Kiyoshi Uedaecdb2e22010-03-06 02:29:52 +00002689 return dm_suspended_md(dm_table_get_md(ti->table));
Kiyoshi Ueda64dbce52009-12-10 23:52:27 +00002690}
2691EXPORT_SYMBOL_GPL(dm_suspended);
2692
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -08002693int dm_noflush_suspending(struct dm_target *ti)
2694{
Kiyoshi Uedaecdb2e22010-03-06 02:29:52 +00002695 return __noflush_suspending(dm_table_get_md(ti->table));
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -08002696}
2697EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2698
Martin K. Petersena91a2782011-03-17 11:11:05 +01002699struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity)
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002700{
2701 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
Martin K. Petersena91a2782011-03-17 11:11:05 +01002702 unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS;
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002703
2704 if (!pools)
2705 return NULL;
2706
2707 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2708 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2709 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2710 if (!pools->io_pool)
2711 goto free_pools_and_out;
2712
2713 pools->tio_pool = (type == DM_TYPE_BIO_BASED) ?
2714 mempool_create_slab_pool(MIN_IOS, _tio_cache) :
2715 mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2716 if (!pools->tio_pool)
2717 goto free_io_pool_and_out;
2718
Martin K. Petersena91a2782011-03-17 11:11:05 +01002719 pools->bs = bioset_create(pool_size, 0);
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002720 if (!pools->bs)
2721 goto free_tio_pool_and_out;
2722
Martin K. Petersena91a2782011-03-17 11:11:05 +01002723 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2724 goto free_bioset_and_out;
2725
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002726 return pools;
2727
Martin K. Petersena91a2782011-03-17 11:11:05 +01002728free_bioset_and_out:
2729 bioset_free(pools->bs);
2730
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002731free_tio_pool_and_out:
2732 mempool_destroy(pools->tio_pool);
2733
2734free_io_pool_and_out:
2735 mempool_destroy(pools->io_pool);
2736
2737free_pools_and_out:
2738 kfree(pools);
2739
2740 return NULL;
2741}
2742
2743void dm_free_md_mempools(struct dm_md_mempools *pools)
2744{
2745 if (!pools)
2746 return;
2747
2748 if (pools->io_pool)
2749 mempool_destroy(pools->io_pool);
2750
2751 if (pools->tio_pool)
2752 mempool_destroy(pools->tio_pool);
2753
2754 if (pools->bs)
2755 bioset_free(pools->bs);
2756
2757 kfree(pools);
2758}
2759
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07002760static const struct block_device_operations dm_blk_dops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 .open = dm_blk_open,
2762 .release = dm_blk_close,
Milan Brozaa129a22006-10-03 01:15:15 -07002763 .ioctl = dm_blk_ioctl,
Darrick J. Wong3ac51e72006-03-27 01:17:54 -08002764 .getgeo = dm_blk_getgeo,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765 .owner = THIS_MODULE
2766};
2767
2768EXPORT_SYMBOL(dm_get_mapinfo);
2769
2770/*
2771 * module hooks
2772 */
2773module_init(dm_init);
2774module_exit(dm_exit);
2775
2776module_param(major, uint, 0);
2777MODULE_PARM_DESC(major, "The major number of the device mapper");
2778MODULE_DESCRIPTION(DM_NAME " driver");
2779MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2780MODULE_LICENSE("GPL");