Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. |
Milan Broz | 784aae7 | 2009-01-06 03:05:12 +0000 | [diff] [blame] | 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4 | * |
| 5 | * This file is released under the GPL. |
| 6 | */ |
| 7 | |
| 8 | #include "dm.h" |
Mike Anderson | 51e5b2b | 2007-10-19 22:48:00 +0100 | [diff] [blame] | 9 | #include "dm-uevent.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 10 | |
| 11 | #include <linux/init.h> |
| 12 | #include <linux/module.h> |
Arjan van de Ven | 48c9c27 | 2006-03-27 01:18:20 -0800 | [diff] [blame] | 13 | #include <linux/mutex.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 14 | #include <linux/moduleparam.h> |
| 15 | #include <linux/blkpg.h> |
| 16 | #include <linux/bio.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 17 | #include <linux/mempool.h> |
| 18 | #include <linux/slab.h> |
| 19 | #include <linux/idr.h> |
Darrick J. Wong | 3ac51e7 | 2006-03-27 01:17:54 -0800 | [diff] [blame] | 20 | #include <linux/hdreg.h> |
Kiyoshi Ueda | 3f77316d | 2010-08-12 04:13:56 +0100 | [diff] [blame] | 21 | #include <linux/delay.h> |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 22 | #include <linux/wait.h> |
Keith Busch | 2eb6e1e | 2014-10-17 17:46:36 -0600 | [diff] [blame] | 23 | #include <linux/kthread.h> |
Li Zefan | 5578213 | 2009-06-09 13:43:05 +0800 | [diff] [blame] | 24 | |
| 25 | #include <trace/events/block.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 26 | |
Alasdair G Kergon | 72d9486 | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 27 | #define DM_MSG_PREFIX "core" |
| 28 | |
Namhyung Kim | 71a1673 | 2011-10-31 20:18:54 +0000 | [diff] [blame] | 29 | #ifdef CONFIG_PRINTK |
| 30 | /* |
| 31 | * ratelimit state to be used in DMXXX_LIMIT(). |
| 32 | */ |
| 33 | DEFINE_RATELIMIT_STATE(dm_ratelimit_state, |
| 34 | DEFAULT_RATELIMIT_INTERVAL, |
| 35 | DEFAULT_RATELIMIT_BURST); |
| 36 | EXPORT_SYMBOL(dm_ratelimit_state); |
| 37 | #endif |
| 38 | |
Milan Broz | 60935eb | 2009-06-22 10:12:30 +0100 | [diff] [blame] | 39 | /* |
| 40 | * Cookies are numeric values sent with CHANGE and REMOVE |
| 41 | * uevents while resuming, removing or renaming the device. |
| 42 | */ |
| 43 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" |
| 44 | #define DM_COOKIE_LENGTH 24 |
| 45 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 46 | static const char *_name = DM_NAME; |
| 47 | |
| 48 | static unsigned int major = 0; |
| 49 | static unsigned int _major = 0; |
| 50 | |
Alasdair G Kergon | d15b774 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 51 | static DEFINE_IDR(_minor_idr); |
| 52 | |
Jeff Mahoney | f32c10b | 2006-06-26 00:27:22 -0700 | [diff] [blame] | 53 | static DEFINE_SPINLOCK(_minor_lock); |
Mikulas Patocka | 2c140a2 | 2013-11-01 18:27:41 -0400 | [diff] [blame] | 54 | |
| 55 | static void do_deferred_remove(struct work_struct *w); |
| 56 | |
| 57 | static DECLARE_WORK(deferred_remove_work, do_deferred_remove); |
| 58 | |
Mikulas Patocka | acfe0ad | 2014-06-14 13:44:31 -0400 | [diff] [blame] | 59 | static struct workqueue_struct *deferred_remove_workqueue; |
| 60 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 61 | /* |
Kiyoshi Ueda | 8fbf26a | 2009-01-06 03:05:06 +0000 | [diff] [blame] | 62 | * For bio-based dm. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 63 | * One of these is allocated per bio. |
| 64 | */ |
| 65 | struct dm_io { |
| 66 | struct mapped_device *md; |
| 67 | int error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 68 | atomic_t io_count; |
Richard Kennedy | 6ae2fa6 | 2008-07-21 12:00:28 +0100 | [diff] [blame] | 69 | struct bio *bio; |
Jun'ichi "Nick" Nomura | 3eaf840 | 2006-02-01 03:04:53 -0800 | [diff] [blame] | 70 | unsigned long start_time; |
Kiyoshi Ueda | f88fb98 | 2009-10-16 23:18:15 +0100 | [diff] [blame] | 71 | spinlock_t endio_lock; |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 72 | struct dm_stats_aux stats_aux; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 73 | }; |
| 74 | |
| 75 | /* |
Kiyoshi Ueda | 8fbf26a | 2009-01-06 03:05:06 +0000 | [diff] [blame] | 76 | * For request-based dm. |
| 77 | * One of these is allocated per request. |
| 78 | */ |
| 79 | struct dm_rq_target_io { |
| 80 | struct mapped_device *md; |
| 81 | struct dm_target *ti; |
Mike Snitzer | 1ae49ea | 2014-12-05 17:11:05 -0500 | [diff] [blame] | 82 | struct request *orig, *clone; |
Keith Busch | 2eb6e1e | 2014-10-17 17:46:36 -0600 | [diff] [blame] | 83 | struct kthread_work work; |
Kiyoshi Ueda | 8fbf26a | 2009-01-06 03:05:06 +0000 | [diff] [blame] | 84 | int error; |
| 85 | union map_info info; |
| 86 | }; |
| 87 | |
| 88 | /* |
Kent Overstreet | 9481874 | 2012-09-07 13:44:01 -0700 | [diff] [blame] | 89 | * For request-based dm - the bio clones we allocate are embedded in these |
| 90 | * structs. |
| 91 | * |
| 92 | * We allocate these with bio_alloc_bioset, using the front_pad parameter when |
| 93 | * the bioset is created - this means the bio has to come at the end of the |
| 94 | * struct. |
Kiyoshi Ueda | 8fbf26a | 2009-01-06 03:05:06 +0000 | [diff] [blame] | 95 | */ |
| 96 | struct dm_rq_clone_bio_info { |
| 97 | struct bio *orig; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 98 | struct dm_rq_target_io *tio; |
Kent Overstreet | 9481874 | 2012-09-07 13:44:01 -0700 | [diff] [blame] | 99 | struct bio clone; |
Kiyoshi Ueda | 8fbf26a | 2009-01-06 03:05:06 +0000 | [diff] [blame] | 100 | }; |
| 101 | |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 102 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
| 103 | { |
| 104 | if (rq && rq->end_io_data) |
| 105 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; |
| 106 | return NULL; |
| 107 | } |
| 108 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); |
| 109 | |
Jeff Mahoney | ba61fdd | 2006-06-26 00:27:21 -0700 | [diff] [blame] | 110 | #define MINOR_ALLOCED ((void *)-1) |
| 111 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 112 | /* |
| 113 | * Bits for the md->flags field. |
| 114 | */ |
Alasdair G Kergon | 1eb787e | 2009-04-09 00:27:14 +0100 | [diff] [blame] | 115 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 116 | #define DMF_SUSPENDED 1 |
Alasdair G Kergon | aa8d7c2 | 2006-01-06 00:20:06 -0800 | [diff] [blame] | 117 | #define DMF_FROZEN 2 |
Jeff Mahoney | fba9f90 | 2006-06-26 00:27:23 -0700 | [diff] [blame] | 118 | #define DMF_FREEING 3 |
Alasdair G Kergon | 5c6bd75 | 2006-06-26 00:27:34 -0700 | [diff] [blame] | 119 | #define DMF_DELETING 4 |
Kiyoshi Ueda | 2e93ccc | 2006-12-08 02:41:09 -0800 | [diff] [blame] | 120 | #define DMF_NOFLUSH_SUSPENDING 5 |
Mikulas Patocka | d5b9dd0 | 2011-08-02 12:32:04 +0100 | [diff] [blame] | 121 | #define DMF_MERGE_IS_OPTIONAL 6 |
Mikulas Patocka | 2c140a2 | 2013-11-01 18:27:41 -0400 | [diff] [blame] | 122 | #define DMF_DEFERRED_REMOVE 7 |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 123 | #define DMF_SUSPENDED_INTERNALLY 8 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 124 | |
Milan Broz | 304f3f6 | 2008-02-08 02:11:17 +0000 | [diff] [blame] | 125 | /* |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 126 | * A dummy definition to make RCU happy. |
| 127 | * struct dm_table should never be dereferenced in this file. |
| 128 | */ |
| 129 | struct dm_table { |
| 130 | int undefined__; |
| 131 | }; |
| 132 | |
| 133 | /* |
Milan Broz | 304f3f6 | 2008-02-08 02:11:17 +0000 | [diff] [blame] | 134 | * Work processed by per-device workqueue. |
| 135 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 136 | struct mapped_device { |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 137 | struct srcu_struct io_barrier; |
Daniel Walker | e61290a | 2008-02-08 02:10:08 +0000 | [diff] [blame] | 138 | struct mutex suspend_lock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 139 | atomic_t holders; |
Alasdair G Kergon | 5c6bd75 | 2006-06-26 00:27:34 -0700 | [diff] [blame] | 140 | atomic_t open_count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 141 | |
Mikulas Patocka | 2a7faeb | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 142 | /* |
| 143 | * The current mapping. |
| 144 | * Use dm_get_live_table{_fast} or take suspend_lock for |
| 145 | * dereference. |
| 146 | */ |
Pranith Kumar | 6fa9952 | 2014-10-28 15:09:57 -0700 | [diff] [blame] | 147 | struct dm_table __rcu *map; |
Mikulas Patocka | 2a7faeb | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 148 | |
Benjamin Marzinski | 86f1152 | 2014-08-13 13:53:43 -0500 | [diff] [blame] | 149 | struct list_head table_devices; |
| 150 | struct mutex table_devices_lock; |
| 151 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 152 | unsigned long flags; |
| 153 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 154 | struct request_queue *queue; |
Mike Snitzer | a5664da | 2010-08-12 04:14:01 +0100 | [diff] [blame] | 155 | unsigned type; |
Mike Snitzer | 4a0b4dd | 2010-08-12 04:14:02 +0100 | [diff] [blame] | 156 | /* Protect queue and type against concurrent access. */ |
Mike Snitzer | a5664da | 2010-08-12 04:14:01 +0100 | [diff] [blame] | 157 | struct mutex type_lock; |
| 158 | |
Alasdair G Kergon | 36a0456 | 2011-10-31 20:19:04 +0000 | [diff] [blame] | 159 | struct target_type *immutable_target_type; |
| 160 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 161 | struct gendisk *disk; |
Mike Anderson | 7e51f25 | 2006-03-27 01:17:52 -0800 | [diff] [blame] | 162 | char name[16]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 163 | |
| 164 | void *interface_ptr; |
| 165 | |
| 166 | /* |
| 167 | * A list of ios that arrived while we were suspended. |
| 168 | */ |
Nikanth Karthikesan | 316d315 | 2009-10-06 20:16:55 +0200 | [diff] [blame] | 169 | atomic_t pending[2]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 170 | wait_queue_head_t wait; |
Mikulas Patocka | 53d5914 | 2009-04-02 19:55:37 +0100 | [diff] [blame] | 171 | struct work_struct work; |
Kiyoshi Ueda | 7485936 | 2006-12-08 02:41:02 -0800 | [diff] [blame] | 172 | struct bio_list deferred; |
Mikulas Patocka | 022c261 | 2009-04-02 19:55:39 +0100 | [diff] [blame] | 173 | spinlock_t deferred_lock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 174 | |
| 175 | /* |
Tejun Heo | 29e4013 | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 176 | * Processing queue (flush) |
Milan Broz | 304f3f6 | 2008-02-08 02:11:17 +0000 | [diff] [blame] | 177 | */ |
| 178 | struct workqueue_struct *wq; |
| 179 | |
| 180 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 181 | * io objects are allocated from here. |
| 182 | */ |
| 183 | mempool_t *io_pool; |
Mike Snitzer | 1ae49ea | 2014-12-05 17:11:05 -0500 | [diff] [blame] | 184 | mempool_t *rq_pool; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 185 | |
Stefan Bader | 9faf400 | 2006-10-03 01:15:41 -0700 | [diff] [blame] | 186 | struct bio_set *bs; |
| 187 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 188 | /* |
| 189 | * Event handling. |
| 190 | */ |
| 191 | atomic_t event_nr; |
| 192 | wait_queue_head_t eventq; |
Mike Anderson | 7a8c3d3 | 2007-10-19 22:48:01 +0100 | [diff] [blame] | 193 | atomic_t uevent_seq; |
| 194 | struct list_head uevent_list; |
| 195 | spinlock_t uevent_lock; /* Protect access to uevent_list */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 196 | |
| 197 | /* |
| 198 | * freeze/thaw support require holding onto a super block |
| 199 | */ |
| 200 | struct super_block *frozen_sb; |
Mikulas Patocka | db8fef4 | 2009-06-22 10:12:15 +0100 | [diff] [blame] | 201 | struct block_device *bdev; |
Darrick J. Wong | 3ac51e7 | 2006-03-27 01:17:54 -0800 | [diff] [blame] | 202 | |
| 203 | /* forced geometry settings */ |
| 204 | struct hd_geometry geometry; |
Milan Broz | 784aae7 | 2009-01-06 03:05:12 +0000 | [diff] [blame] | 205 | |
Mikulas Patocka | 2995fa7 | 2014-01-13 19:37:54 -0500 | [diff] [blame] | 206 | /* kobject and completion */ |
| 207 | struct dm_kobject_holder kobj_holder; |
Mikulas Patocka | be35f48 | 2014-01-06 23:01:22 -0500 | [diff] [blame] | 208 | |
Tejun Heo | d87f4c1 | 2010-09-03 11:56:19 +0200 | [diff] [blame] | 209 | /* zero-length flush that will be cloned and submitted to targets */ |
| 210 | struct bio flush_bio; |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 211 | |
Mikulas Patocka | 96b26c8 | 2015-01-08 18:52:26 -0500 | [diff] [blame] | 212 | /* the number of internal suspends */ |
| 213 | unsigned internal_suspend_count; |
| 214 | |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 215 | struct dm_stats stats; |
Keith Busch | 2eb6e1e | 2014-10-17 17:46:36 -0600 | [diff] [blame] | 216 | |
| 217 | struct kthread_worker kworker; |
| 218 | struct task_struct *kworker_task; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 219 | }; |
| 220 | |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 221 | /* |
| 222 | * For mempools pre-allocation at the table loading time. |
| 223 | */ |
| 224 | struct dm_md_mempools { |
| 225 | mempool_t *io_pool; |
Mike Snitzer | 1ae49ea | 2014-12-05 17:11:05 -0500 | [diff] [blame] | 226 | mempool_t *rq_pool; |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 227 | struct bio_set *bs; |
| 228 | }; |
| 229 | |
Benjamin Marzinski | 86f1152 | 2014-08-13 13:53:43 -0500 | [diff] [blame] | 230 | struct table_device { |
| 231 | struct list_head list; |
| 232 | atomic_t count; |
| 233 | struct dm_dev dm_dev; |
| 234 | }; |
| 235 | |
Mike Snitzer | 6cfa585 | 2013-09-12 18:06:11 -0400 | [diff] [blame] | 236 | #define RESERVED_BIO_BASED_IOS 16 |
| 237 | #define RESERVED_REQUEST_BASED_IOS 256 |
Mike Snitzer | f479082 | 2013-09-12 18:06:12 -0400 | [diff] [blame] | 238 | #define RESERVED_MAX_IOS 1024 |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 239 | static struct kmem_cache *_io_cache; |
Kiyoshi Ueda | 8fbf26a | 2009-01-06 03:05:06 +0000 | [diff] [blame] | 240 | static struct kmem_cache *_rq_tio_cache; |
Mike Snitzer | 1ae49ea | 2014-12-05 17:11:05 -0500 | [diff] [blame] | 241 | static struct kmem_cache *_rq_cache; |
Kent Overstreet | 9481874 | 2012-09-07 13:44:01 -0700 | [diff] [blame] | 242 | |
Mike Snitzer | f479082 | 2013-09-12 18:06:12 -0400 | [diff] [blame] | 243 | /* |
Mike Snitzer | e860313 | 2013-09-12 18:06:12 -0400 | [diff] [blame] | 244 | * Bio-based DM's mempools' reserved IOs set by the user. |
| 245 | */ |
| 246 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; |
| 247 | |
| 248 | /* |
Mike Snitzer | f479082 | 2013-09-12 18:06:12 -0400 | [diff] [blame] | 249 | * Request-based DM's mempools' reserved IOs set by the user. |
| 250 | */ |
| 251 | static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS; |
| 252 | |
| 253 | static unsigned __dm_get_reserved_ios(unsigned *reserved_ios, |
| 254 | unsigned def, unsigned max) |
| 255 | { |
| 256 | unsigned ios = ACCESS_ONCE(*reserved_ios); |
| 257 | unsigned modified_ios = 0; |
| 258 | |
| 259 | if (!ios) |
| 260 | modified_ios = def; |
| 261 | else if (ios > max) |
| 262 | modified_ios = max; |
| 263 | |
| 264 | if (modified_ios) { |
| 265 | (void)cmpxchg(reserved_ios, ios, modified_ios); |
| 266 | ios = modified_ios; |
| 267 | } |
| 268 | |
| 269 | return ios; |
| 270 | } |
| 271 | |
Mike Snitzer | e860313 | 2013-09-12 18:06:12 -0400 | [diff] [blame] | 272 | unsigned dm_get_reserved_bio_based_ios(void) |
| 273 | { |
| 274 | return __dm_get_reserved_ios(&reserved_bio_based_ios, |
| 275 | RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS); |
| 276 | } |
| 277 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); |
| 278 | |
Mike Snitzer | f479082 | 2013-09-12 18:06:12 -0400 | [diff] [blame] | 279 | unsigned dm_get_reserved_rq_based_ios(void) |
| 280 | { |
| 281 | return __dm_get_reserved_ios(&reserved_rq_based_ios, |
| 282 | RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS); |
| 283 | } |
| 284 | EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios); |
| 285 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 286 | static int __init local_init(void) |
| 287 | { |
Kiyoshi Ueda | 51157b4 | 2008-10-21 17:45:08 +0100 | [diff] [blame] | 288 | int r = -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 289 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 290 | /* allocate a slab for the dm_ios */ |
Alasdair G Kergon | 028867a | 2007-07-12 17:26:32 +0100 | [diff] [blame] | 291 | _io_cache = KMEM_CACHE(dm_io, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 292 | if (!_io_cache) |
Kiyoshi Ueda | 51157b4 | 2008-10-21 17:45:08 +0100 | [diff] [blame] | 293 | return r; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 294 | |
Kiyoshi Ueda | 8fbf26a | 2009-01-06 03:05:06 +0000 | [diff] [blame] | 295 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
| 296 | if (!_rq_tio_cache) |
Mikulas Patocka | dba1416 | 2012-10-12 21:02:15 +0100 | [diff] [blame] | 297 | goto out_free_io_cache; |
Kiyoshi Ueda | 8fbf26a | 2009-01-06 03:05:06 +0000 | [diff] [blame] | 298 | |
Mike Snitzer | 1ae49ea | 2014-12-05 17:11:05 -0500 | [diff] [blame] | 299 | _rq_cache = kmem_cache_create("dm_clone_request", sizeof(struct request), |
| 300 | __alignof__(struct request), 0, NULL); |
| 301 | if (!_rq_cache) |
| 302 | goto out_free_rq_tio_cache; |
| 303 | |
Mike Anderson | 51e5b2b | 2007-10-19 22:48:00 +0100 | [diff] [blame] | 304 | r = dm_uevent_init(); |
Kiyoshi Ueda | 51157b4 | 2008-10-21 17:45:08 +0100 | [diff] [blame] | 305 | if (r) |
Mike Snitzer | 1ae49ea | 2014-12-05 17:11:05 -0500 | [diff] [blame] | 306 | goto out_free_rq_cache; |
Mike Anderson | 51e5b2b | 2007-10-19 22:48:00 +0100 | [diff] [blame] | 307 | |
Mikulas Patocka | acfe0ad | 2014-06-14 13:44:31 -0400 | [diff] [blame] | 308 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
| 309 | if (!deferred_remove_workqueue) { |
| 310 | r = -ENOMEM; |
| 311 | goto out_uevent_exit; |
| 312 | } |
| 313 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 314 | _major = major; |
| 315 | r = register_blkdev(_major, _name); |
Kiyoshi Ueda | 51157b4 | 2008-10-21 17:45:08 +0100 | [diff] [blame] | 316 | if (r < 0) |
Mikulas Patocka | acfe0ad | 2014-06-14 13:44:31 -0400 | [diff] [blame] | 317 | goto out_free_workqueue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 318 | |
| 319 | if (!_major) |
| 320 | _major = r; |
| 321 | |
| 322 | return 0; |
Kiyoshi Ueda | 51157b4 | 2008-10-21 17:45:08 +0100 | [diff] [blame] | 323 | |
Mikulas Patocka | acfe0ad | 2014-06-14 13:44:31 -0400 | [diff] [blame] | 324 | out_free_workqueue: |
| 325 | destroy_workqueue(deferred_remove_workqueue); |
Kiyoshi Ueda | 51157b4 | 2008-10-21 17:45:08 +0100 | [diff] [blame] | 326 | out_uevent_exit: |
| 327 | dm_uevent_exit(); |
Mike Snitzer | 1ae49ea | 2014-12-05 17:11:05 -0500 | [diff] [blame] | 328 | out_free_rq_cache: |
| 329 | kmem_cache_destroy(_rq_cache); |
Kiyoshi Ueda | 8fbf26a | 2009-01-06 03:05:06 +0000 | [diff] [blame] | 330 | out_free_rq_tio_cache: |
| 331 | kmem_cache_destroy(_rq_tio_cache); |
Kiyoshi Ueda | 51157b4 | 2008-10-21 17:45:08 +0100 | [diff] [blame] | 332 | out_free_io_cache: |
| 333 | kmem_cache_destroy(_io_cache); |
| 334 | |
| 335 | return r; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 336 | } |
| 337 | |
| 338 | static void local_exit(void) |
| 339 | { |
Mikulas Patocka | 2c140a2 | 2013-11-01 18:27:41 -0400 | [diff] [blame] | 340 | flush_scheduled_work(); |
Mikulas Patocka | acfe0ad | 2014-06-14 13:44:31 -0400 | [diff] [blame] | 341 | destroy_workqueue(deferred_remove_workqueue); |
Mikulas Patocka | 2c140a2 | 2013-11-01 18:27:41 -0400 | [diff] [blame] | 342 | |
Mike Snitzer | 1ae49ea | 2014-12-05 17:11:05 -0500 | [diff] [blame] | 343 | kmem_cache_destroy(_rq_cache); |
Kiyoshi Ueda | 8fbf26a | 2009-01-06 03:05:06 +0000 | [diff] [blame] | 344 | kmem_cache_destroy(_rq_tio_cache); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 345 | kmem_cache_destroy(_io_cache); |
Akinobu Mita | 00d5940 | 2007-07-17 04:03:46 -0700 | [diff] [blame] | 346 | unregister_blkdev(_major, _name); |
Mike Anderson | 51e5b2b | 2007-10-19 22:48:00 +0100 | [diff] [blame] | 347 | dm_uevent_exit(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 348 | |
| 349 | _major = 0; |
| 350 | |
| 351 | DMINFO("cleaned up"); |
| 352 | } |
| 353 | |
Alasdair G Kergon | b9249e5 | 2008-02-08 02:09:51 +0000 | [diff] [blame] | 354 | static int (*_inits[])(void) __initdata = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 355 | local_init, |
| 356 | dm_target_init, |
| 357 | dm_linear_init, |
| 358 | dm_stripe_init, |
Mikulas Patocka | 952b355 | 2009-12-10 23:51:57 +0000 | [diff] [blame] | 359 | dm_io_init, |
Mikulas Patocka | 945fa4d | 2008-04-24 21:43:49 +0100 | [diff] [blame] | 360 | dm_kcopyd_init, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 361 | dm_interface_init, |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 362 | dm_statistics_init, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 363 | }; |
| 364 | |
Alasdair G Kergon | b9249e5 | 2008-02-08 02:09:51 +0000 | [diff] [blame] | 365 | static void (*_exits[])(void) = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 366 | local_exit, |
| 367 | dm_target_exit, |
| 368 | dm_linear_exit, |
| 369 | dm_stripe_exit, |
Mikulas Patocka | 952b355 | 2009-12-10 23:51:57 +0000 | [diff] [blame] | 370 | dm_io_exit, |
Mikulas Patocka | 945fa4d | 2008-04-24 21:43:49 +0100 | [diff] [blame] | 371 | dm_kcopyd_exit, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 372 | dm_interface_exit, |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 373 | dm_statistics_exit, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 374 | }; |
| 375 | |
| 376 | static int __init dm_init(void) |
| 377 | { |
| 378 | const int count = ARRAY_SIZE(_inits); |
| 379 | |
| 380 | int r, i; |
| 381 | |
| 382 | for (i = 0; i < count; i++) { |
| 383 | r = _inits[i](); |
| 384 | if (r) |
| 385 | goto bad; |
| 386 | } |
| 387 | |
| 388 | return 0; |
| 389 | |
| 390 | bad: |
| 391 | while (i--) |
| 392 | _exits[i](); |
| 393 | |
| 394 | return r; |
| 395 | } |
| 396 | |
| 397 | static void __exit dm_exit(void) |
| 398 | { |
| 399 | int i = ARRAY_SIZE(_exits); |
| 400 | |
| 401 | while (i--) |
| 402 | _exits[i](); |
Alasdair G Kergon | d15b774 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 403 | |
| 404 | /* |
| 405 | * Should be empty by this point. |
| 406 | */ |
Alasdair G Kergon | d15b774 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 407 | idr_destroy(&_minor_idr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 408 | } |
| 409 | |
| 410 | /* |
| 411 | * Block device functions |
| 412 | */ |
Mike Anderson | 432a212 | 2009-12-10 23:52:20 +0000 | [diff] [blame] | 413 | int dm_deleting_md(struct mapped_device *md) |
| 414 | { |
| 415 | return test_bit(DMF_DELETING, &md->flags); |
| 416 | } |
| 417 | |
Al Viro | fe5f9f2 | 2008-03-02 10:29:31 -0500 | [diff] [blame] | 418 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 419 | { |
| 420 | struct mapped_device *md; |
| 421 | |
Jeff Mahoney | fba9f90 | 2006-06-26 00:27:23 -0700 | [diff] [blame] | 422 | spin_lock(&_minor_lock); |
| 423 | |
Al Viro | fe5f9f2 | 2008-03-02 10:29:31 -0500 | [diff] [blame] | 424 | md = bdev->bd_disk->private_data; |
Jeff Mahoney | fba9f90 | 2006-06-26 00:27:23 -0700 | [diff] [blame] | 425 | if (!md) |
| 426 | goto out; |
| 427 | |
Alasdair G Kergon | 5c6bd75 | 2006-06-26 00:27:34 -0700 | [diff] [blame] | 428 | if (test_bit(DMF_FREEING, &md->flags) || |
Mike Anderson | 432a212 | 2009-12-10 23:52:20 +0000 | [diff] [blame] | 429 | dm_deleting_md(md)) { |
Jeff Mahoney | fba9f90 | 2006-06-26 00:27:23 -0700 | [diff] [blame] | 430 | md = NULL; |
| 431 | goto out; |
| 432 | } |
| 433 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 434 | dm_get(md); |
Alasdair G Kergon | 5c6bd75 | 2006-06-26 00:27:34 -0700 | [diff] [blame] | 435 | atomic_inc(&md->open_count); |
Jeff Mahoney | fba9f90 | 2006-06-26 00:27:23 -0700 | [diff] [blame] | 436 | |
| 437 | out: |
| 438 | spin_unlock(&_minor_lock); |
| 439 | |
| 440 | return md ? 0 : -ENXIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 441 | } |
| 442 | |
Al Viro | db2a144 | 2013-05-05 21:52:57 -0400 | [diff] [blame] | 443 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 444 | { |
Al Viro | fe5f9f2 | 2008-03-02 10:29:31 -0500 | [diff] [blame] | 445 | struct mapped_device *md = disk->private_data; |
Arnd Bergmann | 6e9624b | 2010-08-07 18:25:34 +0200 | [diff] [blame] | 446 | |
Milan Broz | 4a1aeb9 | 2011-01-13 19:59:48 +0000 | [diff] [blame] | 447 | spin_lock(&_minor_lock); |
| 448 | |
Mikulas Patocka | 2c140a2 | 2013-11-01 18:27:41 -0400 | [diff] [blame] | 449 | if (atomic_dec_and_test(&md->open_count) && |
| 450 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) |
Mikulas Patocka | acfe0ad | 2014-06-14 13:44:31 -0400 | [diff] [blame] | 451 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
Mikulas Patocka | 2c140a2 | 2013-11-01 18:27:41 -0400 | [diff] [blame] | 452 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 453 | dm_put(md); |
Milan Broz | 4a1aeb9 | 2011-01-13 19:59:48 +0000 | [diff] [blame] | 454 | |
| 455 | spin_unlock(&_minor_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 456 | } |
| 457 | |
Alasdair G Kergon | 5c6bd75 | 2006-06-26 00:27:34 -0700 | [diff] [blame] | 458 | int dm_open_count(struct mapped_device *md) |
| 459 | { |
| 460 | return atomic_read(&md->open_count); |
| 461 | } |
| 462 | |
| 463 | /* |
| 464 | * Guarantees nothing is using the device before it's deleted. |
| 465 | */ |
Mikulas Patocka | 2c140a2 | 2013-11-01 18:27:41 -0400 | [diff] [blame] | 466 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
Alasdair G Kergon | 5c6bd75 | 2006-06-26 00:27:34 -0700 | [diff] [blame] | 467 | { |
| 468 | int r = 0; |
| 469 | |
| 470 | spin_lock(&_minor_lock); |
| 471 | |
Mikulas Patocka | 2c140a2 | 2013-11-01 18:27:41 -0400 | [diff] [blame] | 472 | if (dm_open_count(md)) { |
Alasdair G Kergon | 5c6bd75 | 2006-06-26 00:27:34 -0700 | [diff] [blame] | 473 | r = -EBUSY; |
Mikulas Patocka | 2c140a2 | 2013-11-01 18:27:41 -0400 | [diff] [blame] | 474 | if (mark_deferred) |
| 475 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); |
| 476 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) |
| 477 | r = -EEXIST; |
Alasdair G Kergon | 5c6bd75 | 2006-06-26 00:27:34 -0700 | [diff] [blame] | 478 | else |
| 479 | set_bit(DMF_DELETING, &md->flags); |
| 480 | |
| 481 | spin_unlock(&_minor_lock); |
| 482 | |
| 483 | return r; |
| 484 | } |
| 485 | |
Mikulas Patocka | 2c140a2 | 2013-11-01 18:27:41 -0400 | [diff] [blame] | 486 | int dm_cancel_deferred_remove(struct mapped_device *md) |
| 487 | { |
| 488 | int r = 0; |
| 489 | |
| 490 | spin_lock(&_minor_lock); |
| 491 | |
| 492 | if (test_bit(DMF_DELETING, &md->flags)) |
| 493 | r = -EBUSY; |
| 494 | else |
| 495 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); |
| 496 | |
| 497 | spin_unlock(&_minor_lock); |
| 498 | |
| 499 | return r; |
| 500 | } |
| 501 | |
| 502 | static void do_deferred_remove(struct work_struct *w) |
| 503 | { |
| 504 | dm_deferred_remove(); |
| 505 | } |
| 506 | |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 507 | sector_t dm_get_size(struct mapped_device *md) |
| 508 | { |
| 509 | return get_capacity(md->disk); |
| 510 | } |
| 511 | |
Mike Snitzer | 9974fa2 | 2014-02-28 15:33:43 +0100 | [diff] [blame] | 512 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
| 513 | { |
| 514 | return md->queue; |
| 515 | } |
| 516 | |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 517 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
| 518 | { |
| 519 | return &md->stats; |
| 520 | } |
| 521 | |
Darrick J. Wong | 3ac51e7 | 2006-03-27 01:17:54 -0800 | [diff] [blame] | 522 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
| 523 | { |
| 524 | struct mapped_device *md = bdev->bd_disk->private_data; |
| 525 | |
| 526 | return dm_get_geometry(md, geo); |
| 527 | } |
| 528 | |
Al Viro | fe5f9f2 | 2008-03-02 10:29:31 -0500 | [diff] [blame] | 529 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
Milan Broz | aa129a2 | 2006-10-03 01:15:15 -0700 | [diff] [blame] | 530 | unsigned int cmd, unsigned long arg) |
| 531 | { |
Al Viro | fe5f9f2 | 2008-03-02 10:29:31 -0500 | [diff] [blame] | 532 | struct mapped_device *md = bdev->bd_disk->private_data; |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 533 | int srcu_idx; |
Hannes Reinecke | 6c182cd | 2013-07-10 23:41:15 +0100 | [diff] [blame] | 534 | struct dm_table *map; |
Milan Broz | aa129a2 | 2006-10-03 01:15:15 -0700 | [diff] [blame] | 535 | struct dm_target *tgt; |
| 536 | int r = -ENOTTY; |
| 537 | |
Hannes Reinecke | 6c182cd | 2013-07-10 23:41:15 +0100 | [diff] [blame] | 538 | retry: |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 539 | map = dm_get_live_table(md, &srcu_idx); |
| 540 | |
Milan Broz | aa129a2 | 2006-10-03 01:15:15 -0700 | [diff] [blame] | 541 | if (!map || !dm_table_get_size(map)) |
| 542 | goto out; |
| 543 | |
| 544 | /* We only support devices that have a single target */ |
| 545 | if (dm_table_get_num_targets(map) != 1) |
| 546 | goto out; |
| 547 | |
| 548 | tgt = dm_table_get_target(map, 0); |
Mike Snitzer | 4d341d8 | 2014-11-16 14:21:47 -0500 | [diff] [blame] | 549 | if (!tgt->type->ioctl) |
| 550 | goto out; |
Milan Broz | aa129a2 | 2006-10-03 01:15:15 -0700 | [diff] [blame] | 551 | |
Kiyoshi Ueda | 4f186f8 | 2009-12-10 23:52:26 +0000 | [diff] [blame] | 552 | if (dm_suspended_md(md)) { |
Milan Broz | aa129a2 | 2006-10-03 01:15:15 -0700 | [diff] [blame] | 553 | r = -EAGAIN; |
| 554 | goto out; |
| 555 | } |
| 556 | |
Mike Snitzer | 4d341d8 | 2014-11-16 14:21:47 -0500 | [diff] [blame] | 557 | r = tgt->type->ioctl(tgt, cmd, arg); |
Milan Broz | aa129a2 | 2006-10-03 01:15:15 -0700 | [diff] [blame] | 558 | |
| 559 | out: |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 560 | dm_put_live_table(md, srcu_idx); |
Milan Broz | aa129a2 | 2006-10-03 01:15:15 -0700 | [diff] [blame] | 561 | |
Hannes Reinecke | 6c182cd | 2013-07-10 23:41:15 +0100 | [diff] [blame] | 562 | if (r == -ENOTCONN) { |
| 563 | msleep(10); |
| 564 | goto retry; |
| 565 | } |
| 566 | |
Milan Broz | aa129a2 | 2006-10-03 01:15:15 -0700 | [diff] [blame] | 567 | return r; |
| 568 | } |
| 569 | |
Alasdair G Kergon | 028867a | 2007-07-12 17:26:32 +0100 | [diff] [blame] | 570 | static struct dm_io *alloc_io(struct mapped_device *md) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 571 | { |
| 572 | return mempool_alloc(md->io_pool, GFP_NOIO); |
| 573 | } |
| 574 | |
Alasdair G Kergon | 028867a | 2007-07-12 17:26:32 +0100 | [diff] [blame] | 575 | static void free_io(struct mapped_device *md, struct dm_io *io) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 576 | { |
| 577 | mempool_free(io, md->io_pool); |
| 578 | } |
| 579 | |
Alasdair G Kergon | 028867a | 2007-07-12 17:26:32 +0100 | [diff] [blame] | 580 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 581 | { |
Mikulas Patocka | dba1416 | 2012-10-12 21:02:15 +0100 | [diff] [blame] | 582 | bio_put(&tio->clone); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 583 | } |
| 584 | |
Kiyoshi Ueda | 0888564 | 2009-12-10 23:52:15 +0000 | [diff] [blame] | 585 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
| 586 | gfp_t gfp_mask) |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 587 | { |
Jun'ichi Nomura | 5f01520 | 2013-03-01 22:45:48 +0000 | [diff] [blame] | 588 | return mempool_alloc(md->io_pool, gfp_mask); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 589 | } |
| 590 | |
| 591 | static void free_rq_tio(struct dm_rq_target_io *tio) |
| 592 | { |
Jun'ichi Nomura | 5f01520 | 2013-03-01 22:45:48 +0000 | [diff] [blame] | 593 | mempool_free(tio, tio->md->io_pool); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 594 | } |
| 595 | |
Mike Snitzer | 1ae49ea | 2014-12-05 17:11:05 -0500 | [diff] [blame] | 596 | static struct request *alloc_clone_request(struct mapped_device *md, |
| 597 | gfp_t gfp_mask) |
| 598 | { |
| 599 | return mempool_alloc(md->rq_pool, gfp_mask); |
| 600 | } |
| 601 | |
| 602 | static void free_clone_request(struct mapped_device *md, struct request *rq) |
| 603 | { |
| 604 | mempool_free(rq, md->rq_pool); |
| 605 | } |
| 606 | |
Kiyoshi Ueda | 90abb8c | 2009-12-10 23:52:13 +0000 | [diff] [blame] | 607 | static int md_in_flight(struct mapped_device *md) |
| 608 | { |
| 609 | return atomic_read(&md->pending[READ]) + |
| 610 | atomic_read(&md->pending[WRITE]); |
| 611 | } |
| 612 | |
Jun'ichi "Nick" Nomura | 3eaf840 | 2006-02-01 03:04:53 -0800 | [diff] [blame] | 613 | static void start_io_acct(struct dm_io *io) |
| 614 | { |
| 615 | struct mapped_device *md = io->md; |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 616 | struct bio *bio = io->bio; |
Tejun Heo | c995905 | 2008-08-25 19:47:21 +0900 | [diff] [blame] | 617 | int cpu; |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 618 | int rw = bio_data_dir(bio); |
Jun'ichi "Nick" Nomura | 3eaf840 | 2006-02-01 03:04:53 -0800 | [diff] [blame] | 619 | |
| 620 | io->start_time = jiffies; |
| 621 | |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 622 | cpu = part_stat_lock(); |
| 623 | part_round_stats(cpu, &dm_disk(md)->part0); |
| 624 | part_stat_unlock(); |
Shaohua Li | 1e9bb88 | 2011-03-22 08:35:35 +0100 | [diff] [blame] | 625 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
| 626 | atomic_inc_return(&md->pending[rw])); |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 627 | |
| 628 | if (unlikely(dm_stats_used(&md->stats))) |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 629 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 630 | bio_sectors(bio), false, 0, &io->stats_aux); |
Jun'ichi "Nick" Nomura | 3eaf840 | 2006-02-01 03:04:53 -0800 | [diff] [blame] | 631 | } |
| 632 | |
Mikulas Patocka | d221d2e | 2008-11-13 23:39:10 +0000 | [diff] [blame] | 633 | static void end_io_acct(struct dm_io *io) |
Jun'ichi "Nick" Nomura | 3eaf840 | 2006-02-01 03:04:53 -0800 | [diff] [blame] | 634 | { |
| 635 | struct mapped_device *md = io->md; |
| 636 | struct bio *bio = io->bio; |
| 637 | unsigned long duration = jiffies - io->start_time; |
Gu Zheng | 18c0b22 | 2014-11-24 11:05:26 +0800 | [diff] [blame] | 638 | int pending; |
Jun'ichi "Nick" Nomura | 3eaf840 | 2006-02-01 03:04:53 -0800 | [diff] [blame] | 639 | int rw = bio_data_dir(bio); |
| 640 | |
Gu Zheng | 18c0b22 | 2014-11-24 11:05:26 +0800 | [diff] [blame] | 641 | generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time); |
Jun'ichi "Nick" Nomura | 3eaf840 | 2006-02-01 03:04:53 -0800 | [diff] [blame] | 642 | |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 643 | if (unlikely(dm_stats_used(&md->stats))) |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 644 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 645 | bio_sectors(bio), true, duration, &io->stats_aux); |
| 646 | |
Mikulas Patocka | af7e466 | 2009-04-09 00:27:16 +0100 | [diff] [blame] | 647 | /* |
| 648 | * After this is decremented the bio must not be touched if it is |
Tejun Heo | d87f4c1 | 2010-09-03 11:56:19 +0200 | [diff] [blame] | 649 | * a flush. |
Mikulas Patocka | af7e466 | 2009-04-09 00:27:16 +0100 | [diff] [blame] | 650 | */ |
Shaohua Li | 1e9bb88 | 2011-03-22 08:35:35 +0100 | [diff] [blame] | 651 | pending = atomic_dec_return(&md->pending[rw]); |
| 652 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); |
Nikanth Karthikesan | 316d315 | 2009-10-06 20:16:55 +0200 | [diff] [blame] | 653 | pending += atomic_read(&md->pending[rw^0x1]); |
Jun'ichi "Nick" Nomura | 3eaf840 | 2006-02-01 03:04:53 -0800 | [diff] [blame] | 654 | |
Mikulas Patocka | d221d2e | 2008-11-13 23:39:10 +0000 | [diff] [blame] | 655 | /* nudge anyone waiting on suspend queue */ |
| 656 | if (!pending) |
| 657 | wake_up(&md->wait); |
Jun'ichi "Nick" Nomura | 3eaf840 | 2006-02-01 03:04:53 -0800 | [diff] [blame] | 658 | } |
| 659 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 660 | /* |
| 661 | * Add the bio to the list of deferred io. |
| 662 | */ |
Mikulas Patocka | 92c6390 | 2009-04-09 00:27:15 +0100 | [diff] [blame] | 663 | static void queue_io(struct mapped_device *md, struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 664 | { |
Kiyoshi Ueda | 05447420 | 2010-09-08 18:07:01 +0200 | [diff] [blame] | 665 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 666 | |
Kiyoshi Ueda | 05447420 | 2010-09-08 18:07:01 +0200 | [diff] [blame] | 667 | spin_lock_irqsave(&md->deferred_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 668 | bio_list_add(&md->deferred, bio); |
Kiyoshi Ueda | 05447420 | 2010-09-08 18:07:01 +0200 | [diff] [blame] | 669 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
Tejun Heo | 6a8736d | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 670 | queue_work(md->wq, &md->work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 671 | } |
| 672 | |
| 673 | /* |
| 674 | * Everyone (including functions in this file), should use this |
| 675 | * function to access the md->map field, and make sure they call |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 676 | * dm_put_live_table() when finished. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 677 | */ |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 678 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 679 | { |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 680 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 681 | |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 682 | return srcu_dereference(md->map, &md->io_barrier); |
| 683 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 684 | |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 685 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
| 686 | { |
| 687 | srcu_read_unlock(&md->io_barrier, srcu_idx); |
| 688 | } |
| 689 | |
| 690 | void dm_sync_table(struct mapped_device *md) |
| 691 | { |
| 692 | synchronize_srcu(&md->io_barrier); |
| 693 | synchronize_rcu_expedited(); |
| 694 | } |
| 695 | |
| 696 | /* |
| 697 | * A fast alternative to dm_get_live_table/dm_put_live_table. |
| 698 | * The caller must not block between these two functions. |
| 699 | */ |
| 700 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) |
| 701 | { |
| 702 | rcu_read_lock(); |
| 703 | return rcu_dereference(md->map); |
| 704 | } |
| 705 | |
| 706 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
| 707 | { |
| 708 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 709 | } |
| 710 | |
Darrick J. Wong | 3ac51e7 | 2006-03-27 01:17:54 -0800 | [diff] [blame] | 711 | /* |
Benjamin Marzinski | 86f1152 | 2014-08-13 13:53:43 -0500 | [diff] [blame] | 712 | * Open a table device so we can use it as a map destination. |
| 713 | */ |
| 714 | static int open_table_device(struct table_device *td, dev_t dev, |
| 715 | struct mapped_device *md) |
| 716 | { |
| 717 | static char *_claim_ptr = "I belong to device-mapper"; |
| 718 | struct block_device *bdev; |
| 719 | |
| 720 | int r; |
| 721 | |
| 722 | BUG_ON(td->dm_dev.bdev); |
| 723 | |
| 724 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr); |
| 725 | if (IS_ERR(bdev)) |
| 726 | return PTR_ERR(bdev); |
| 727 | |
| 728 | r = bd_link_disk_holder(bdev, dm_disk(md)); |
| 729 | if (r) { |
| 730 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); |
| 731 | return r; |
| 732 | } |
| 733 | |
| 734 | td->dm_dev.bdev = bdev; |
| 735 | return 0; |
| 736 | } |
| 737 | |
| 738 | /* |
| 739 | * Close a table device that we've been using. |
| 740 | */ |
| 741 | static void close_table_device(struct table_device *td, struct mapped_device *md) |
| 742 | { |
| 743 | if (!td->dm_dev.bdev) |
| 744 | return; |
| 745 | |
| 746 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); |
| 747 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); |
| 748 | td->dm_dev.bdev = NULL; |
| 749 | } |
| 750 | |
| 751 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, |
| 752 | fmode_t mode) { |
| 753 | struct table_device *td; |
| 754 | |
| 755 | list_for_each_entry(td, l, list) |
| 756 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) |
| 757 | return td; |
| 758 | |
| 759 | return NULL; |
| 760 | } |
| 761 | |
| 762 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, |
| 763 | struct dm_dev **result) { |
| 764 | int r; |
| 765 | struct table_device *td; |
| 766 | |
| 767 | mutex_lock(&md->table_devices_lock); |
| 768 | td = find_table_device(&md->table_devices, dev, mode); |
| 769 | if (!td) { |
| 770 | td = kmalloc(sizeof(*td), GFP_KERNEL); |
| 771 | if (!td) { |
| 772 | mutex_unlock(&md->table_devices_lock); |
| 773 | return -ENOMEM; |
| 774 | } |
| 775 | |
| 776 | td->dm_dev.mode = mode; |
| 777 | td->dm_dev.bdev = NULL; |
| 778 | |
| 779 | if ((r = open_table_device(td, dev, md))) { |
| 780 | mutex_unlock(&md->table_devices_lock); |
| 781 | kfree(td); |
| 782 | return r; |
| 783 | } |
| 784 | |
| 785 | format_dev_t(td->dm_dev.name, dev); |
| 786 | |
| 787 | atomic_set(&td->count, 0); |
| 788 | list_add(&td->list, &md->table_devices); |
| 789 | } |
| 790 | atomic_inc(&td->count); |
| 791 | mutex_unlock(&md->table_devices_lock); |
| 792 | |
| 793 | *result = &td->dm_dev; |
| 794 | return 0; |
| 795 | } |
| 796 | EXPORT_SYMBOL_GPL(dm_get_table_device); |
| 797 | |
| 798 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) |
| 799 | { |
| 800 | struct table_device *td = container_of(d, struct table_device, dm_dev); |
| 801 | |
| 802 | mutex_lock(&md->table_devices_lock); |
| 803 | if (atomic_dec_and_test(&td->count)) { |
| 804 | close_table_device(td, md); |
| 805 | list_del(&td->list); |
| 806 | kfree(td); |
| 807 | } |
| 808 | mutex_unlock(&md->table_devices_lock); |
| 809 | } |
| 810 | EXPORT_SYMBOL(dm_put_table_device); |
| 811 | |
| 812 | static void free_table_devices(struct list_head *devices) |
| 813 | { |
| 814 | struct list_head *tmp, *next; |
| 815 | |
| 816 | list_for_each_safe(tmp, next, devices) { |
| 817 | struct table_device *td = list_entry(tmp, struct table_device, list); |
| 818 | |
| 819 | DMWARN("dm_destroy: %s still exists with %d references", |
| 820 | td->dm_dev.name, atomic_read(&td->count)); |
| 821 | kfree(td); |
| 822 | } |
| 823 | } |
| 824 | |
| 825 | /* |
Darrick J. Wong | 3ac51e7 | 2006-03-27 01:17:54 -0800 | [diff] [blame] | 826 | * Get the geometry associated with a dm device |
| 827 | */ |
| 828 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) |
| 829 | { |
| 830 | *geo = md->geometry; |
| 831 | |
| 832 | return 0; |
| 833 | } |
| 834 | |
| 835 | /* |
| 836 | * Set the geometry of a device. |
| 837 | */ |
| 838 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) |
| 839 | { |
| 840 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; |
| 841 | |
| 842 | if (geo->start > sz) { |
| 843 | DMWARN("Start sector is beyond the geometry limits."); |
| 844 | return -EINVAL; |
| 845 | } |
| 846 | |
| 847 | md->geometry = *geo; |
| 848 | |
| 849 | return 0; |
| 850 | } |
| 851 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 852 | /*----------------------------------------------------------------- |
| 853 | * CRUD START: |
| 854 | * A more elegant soln is in the works that uses the queue |
| 855 | * merge fn, unfortunately there are a couple of changes to |
| 856 | * the block layer that I want to make for this. So in the |
| 857 | * interests of getting something for people to use I give |
| 858 | * you this clearly demarcated crap. |
| 859 | *---------------------------------------------------------------*/ |
| 860 | |
Kiyoshi Ueda | 2e93ccc | 2006-12-08 02:41:09 -0800 | [diff] [blame] | 861 | static int __noflush_suspending(struct mapped_device *md) |
| 862 | { |
| 863 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
| 864 | } |
| 865 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 866 | /* |
| 867 | * Decrements the number of outstanding ios that a bio has been |
| 868 | * cloned into, completing the original io if necc. |
| 869 | */ |
Arjan van de Ven | 858119e | 2006-01-14 13:20:43 -0800 | [diff] [blame] | 870 | static void dec_pending(struct dm_io *io, int error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 871 | { |
Kiyoshi Ueda | 2e93ccc | 2006-12-08 02:41:09 -0800 | [diff] [blame] | 872 | unsigned long flags; |
Milan Broz | b35f8ca | 2009-03-16 17:44:36 +0000 | [diff] [blame] | 873 | int io_error; |
| 874 | struct bio *bio; |
| 875 | struct mapped_device *md = io->md; |
Kiyoshi Ueda | 2e93ccc | 2006-12-08 02:41:09 -0800 | [diff] [blame] | 876 | |
| 877 | /* Push-back supersedes any I/O errors */ |
Kiyoshi Ueda | f88fb98 | 2009-10-16 23:18:15 +0100 | [diff] [blame] | 878 | if (unlikely(error)) { |
| 879 | spin_lock_irqsave(&io->endio_lock, flags); |
| 880 | if (!(io->error > 0 && __noflush_suspending(md))) |
| 881 | io->error = error; |
| 882 | spin_unlock_irqrestore(&io->endio_lock, flags); |
| 883 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 884 | |
| 885 | if (atomic_dec_and_test(&io->io_count)) { |
Kiyoshi Ueda | 2e93ccc | 2006-12-08 02:41:09 -0800 | [diff] [blame] | 886 | if (io->error == DM_ENDIO_REQUEUE) { |
| 887 | /* |
| 888 | * Target requested pushing back the I/O. |
Kiyoshi Ueda | 2e93ccc | 2006-12-08 02:41:09 -0800 | [diff] [blame] | 889 | */ |
Mikulas Patocka | 022c261 | 2009-04-02 19:55:39 +0100 | [diff] [blame] | 890 | spin_lock_irqsave(&md->deferred_lock, flags); |
Tejun Heo | 6a8736d | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 891 | if (__noflush_suspending(md)) |
| 892 | bio_list_add_head(&md->deferred, io->bio); |
| 893 | else |
Kiyoshi Ueda | 2e93ccc | 2006-12-08 02:41:09 -0800 | [diff] [blame] | 894 | /* noflush suspend was interrupted. */ |
| 895 | io->error = -EIO; |
Mikulas Patocka | 022c261 | 2009-04-02 19:55:39 +0100 | [diff] [blame] | 896 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
Kiyoshi Ueda | 2e93ccc | 2006-12-08 02:41:09 -0800 | [diff] [blame] | 897 | } |
| 898 | |
Milan Broz | b35f8ca | 2009-03-16 17:44:36 +0000 | [diff] [blame] | 899 | io_error = io->error; |
| 900 | bio = io->bio; |
Tejun Heo | 6a8736d | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 901 | end_io_acct(io); |
| 902 | free_io(md, io); |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 903 | |
Tejun Heo | 6a8736d | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 904 | if (io_error == DM_ENDIO_REQUEUE) |
| 905 | return; |
| 906 | |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 907 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) { |
Mikulas Patocka | af7e466 | 2009-04-09 00:27:16 +0100 | [diff] [blame] | 908 | /* |
Tejun Heo | 6a8736d | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 909 | * Preflush done for flush with data, reissue |
| 910 | * without REQ_FLUSH. |
Mikulas Patocka | af7e466 | 2009-04-09 00:27:16 +0100 | [diff] [blame] | 911 | */ |
Tejun Heo | 6a8736d | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 912 | bio->bi_rw &= ~REQ_FLUSH; |
| 913 | queue_io(md, bio); |
Mikulas Patocka | af7e466 | 2009-04-09 00:27:16 +0100 | [diff] [blame] | 914 | } else { |
Mike Snitzer | b372d36 | 2010-09-08 18:07:01 +0200 | [diff] [blame] | 915 | /* done with normal IO or empty flush */ |
Linus Torvalds | 0a82a8d | 2013-04-18 09:00:26 -0700 | [diff] [blame] | 916 | trace_block_bio_complete(md->queue, bio, io_error); |
Mike Snitzer | b372d36 | 2010-09-08 18:07:01 +0200 | [diff] [blame] | 917 | bio_endio(bio, io_error); |
Kiyoshi Ueda | 2e93ccc | 2006-12-08 02:41:09 -0800 | [diff] [blame] | 918 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 919 | } |
| 920 | } |
| 921 | |
Mike Snitzer | 7eee4ae | 2014-06-02 15:50:06 -0400 | [diff] [blame] | 922 | static void disable_write_same(struct mapped_device *md) |
| 923 | { |
| 924 | struct queue_limits *limits = dm_get_queue_limits(md); |
| 925 | |
| 926 | /* device doesn't really support WRITE SAME, disable it */ |
| 927 | limits->max_write_same_sectors = 0; |
| 928 | } |
| 929 | |
NeilBrown | 6712ecf | 2007-09-27 12:47:43 +0200 | [diff] [blame] | 930 | static void clone_endio(struct bio *bio, int error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 931 | { |
zhendong chen | 5164bec | 2014-12-17 14:37:04 +0800 | [diff] [blame] | 932 | int r = error; |
Mikulas Patocka | bfc6d41 | 2014-03-04 18:24:49 -0500 | [diff] [blame] | 933 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
Milan Broz | b35f8ca | 2009-03-16 17:44:36 +0000 | [diff] [blame] | 934 | struct dm_io *io = tio->io; |
Stefan Bader | 9faf400 | 2006-10-03 01:15:41 -0700 | [diff] [blame] | 935 | struct mapped_device *md = tio->io->md; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 936 | dm_endio_fn endio = tio->ti->type->end_io; |
| 937 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 938 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
| 939 | error = -EIO; |
| 940 | |
| 941 | if (endio) { |
Mikulas Patocka | 7de3ee5 | 2012-12-21 20:23:41 +0000 | [diff] [blame] | 942 | r = endio(tio->ti, bio, error); |
Kiyoshi Ueda | 2e93ccc | 2006-12-08 02:41:09 -0800 | [diff] [blame] | 943 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
| 944 | /* |
| 945 | * error and requeue request are handled |
| 946 | * in dec_pending(). |
| 947 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 948 | error = r; |
Kiyoshi Ueda | 45cbcd7 | 2006-12-08 02:41:05 -0800 | [diff] [blame] | 949 | else if (r == DM_ENDIO_INCOMPLETE) |
| 950 | /* The target will handle the io */ |
NeilBrown | 6712ecf | 2007-09-27 12:47:43 +0200 | [diff] [blame] | 951 | return; |
Kiyoshi Ueda | 45cbcd7 | 2006-12-08 02:41:05 -0800 | [diff] [blame] | 952 | else if (r) { |
| 953 | DMWARN("unimplemented target endio return value: %d", r); |
| 954 | BUG(); |
| 955 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 956 | } |
| 957 | |
Mike Snitzer | 7eee4ae | 2014-06-02 15:50:06 -0400 | [diff] [blame] | 958 | if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) && |
| 959 | !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)) |
| 960 | disable_write_same(md); |
| 961 | |
Stefan Bader | 9faf400 | 2006-10-03 01:15:41 -0700 | [diff] [blame] | 962 | free_tio(md, tio); |
Milan Broz | b35f8ca | 2009-03-16 17:44:36 +0000 | [diff] [blame] | 963 | dec_pending(io, error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 964 | } |
| 965 | |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 966 | /* |
| 967 | * Partial completion handling for request-based dm |
| 968 | */ |
| 969 | static void end_clone_bio(struct bio *clone, int error) |
| 970 | { |
Mikulas Patocka | bfc6d41 | 2014-03-04 18:24:49 -0500 | [diff] [blame] | 971 | struct dm_rq_clone_bio_info *info = |
| 972 | container_of(clone, struct dm_rq_clone_bio_info, clone); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 973 | struct dm_rq_target_io *tio = info->tio; |
| 974 | struct bio *bio = info->orig; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 975 | unsigned int nr_bytes = info->orig->bi_iter.bi_size; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 976 | |
| 977 | bio_put(clone); |
| 978 | |
| 979 | if (tio->error) |
| 980 | /* |
| 981 | * An error has already been detected on the request. |
| 982 | * Once error occurred, just let clone->end_io() handle |
| 983 | * the remainder. |
| 984 | */ |
| 985 | return; |
| 986 | else if (error) { |
| 987 | /* |
| 988 | * Don't notice the error to the upper layer yet. |
| 989 | * The error handling decision is made by the target driver, |
| 990 | * when the request is completed. |
| 991 | */ |
| 992 | tio->error = error; |
| 993 | return; |
| 994 | } |
| 995 | |
| 996 | /* |
| 997 | * I/O for the bio successfully completed. |
| 998 | * Notice the data completion to the upper layer. |
| 999 | */ |
| 1000 | |
| 1001 | /* |
| 1002 | * bios are processed from the head of the list. |
| 1003 | * So the completing bio should always be rq->bio. |
| 1004 | * If it's not, something wrong is happening. |
| 1005 | */ |
| 1006 | if (tio->orig->bio != bio) |
| 1007 | DMERR("bio completion is going in the middle of the request"); |
| 1008 | |
| 1009 | /* |
| 1010 | * Update the original request. |
| 1011 | * Do not use blk_end_request() here, because it may complete |
| 1012 | * the original request before the clone, and break the ordering. |
| 1013 | */ |
| 1014 | blk_update_request(tio->orig, 0, nr_bytes); |
| 1015 | } |
| 1016 | |
| 1017 | /* |
| 1018 | * Don't touch any member of the md after calling this function because |
| 1019 | * the md may be freed in dm_put() at the end of this function. |
| 1020 | * Or do dm_get() before calling this function and dm_put() later. |
| 1021 | */ |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1022 | static void rq_completed(struct mapped_device *md, int rw, bool run_queue) |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1023 | { |
Kiyoshi Ueda | b4324fe | 2009-12-10 23:52:16 +0000 | [diff] [blame] | 1024 | atomic_dec(&md->pending[rw]); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1025 | |
| 1026 | /* nudge anyone waiting on suspend queue */ |
Kiyoshi Ueda | b4324fe | 2009-12-10 23:52:16 +0000 | [diff] [blame] | 1027 | if (!md_in_flight(md)) |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1028 | wake_up(&md->wait); |
| 1029 | |
Jens Axboe | a8c32a5 | 2012-11-06 12:24:26 +0100 | [diff] [blame] | 1030 | /* |
| 1031 | * Run this off this callpath, as drivers could invoke end_io while |
| 1032 | * inside their request_fn (and holding the queue lock). Calling |
| 1033 | * back into ->request_fn() could deadlock attempting to grab the |
| 1034 | * queue lock again. |
| 1035 | */ |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1036 | if (run_queue) |
Jens Axboe | a8c32a5 | 2012-11-06 12:24:26 +0100 | [diff] [blame] | 1037 | blk_run_queue_async(md->queue); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1038 | |
| 1039 | /* |
| 1040 | * dm_put() must be at the end of this function. See the comment above |
| 1041 | */ |
| 1042 | dm_put(md); |
| 1043 | } |
| 1044 | |
Kiyoshi Ueda | a77e28c | 2009-09-04 20:40:16 +0100 | [diff] [blame] | 1045 | static void free_rq_clone(struct request *clone) |
| 1046 | { |
| 1047 | struct dm_rq_target_io *tio = clone->end_io_data; |
| 1048 | |
| 1049 | blk_rq_unprep_clone(clone); |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 1050 | if (clone->q && clone->q->mq_ops) |
| 1051 | tio->ti->type->release_clone_rq(clone); |
| 1052 | else |
| 1053 | free_clone_request(tio->md, clone); |
Kiyoshi Ueda | a77e28c | 2009-09-04 20:40:16 +0100 | [diff] [blame] | 1054 | free_rq_tio(tio); |
| 1055 | } |
| 1056 | |
Kiyoshi Ueda | 980691e | 2009-12-10 23:52:17 +0000 | [diff] [blame] | 1057 | /* |
| 1058 | * Complete the clone and the original request. |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1059 | * Must be called without clone's queue lock held, |
| 1060 | * see end_clone_request() for more details. |
Kiyoshi Ueda | 980691e | 2009-12-10 23:52:17 +0000 | [diff] [blame] | 1061 | */ |
| 1062 | static void dm_end_request(struct request *clone, int error) |
| 1063 | { |
| 1064 | int rw = rq_data_dir(clone); |
| 1065 | struct dm_rq_target_io *tio = clone->end_io_data; |
| 1066 | struct mapped_device *md = tio->md; |
| 1067 | struct request *rq = tio->orig; |
| 1068 | |
Tejun Heo | 29e4013 | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 1069 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
Kiyoshi Ueda | 980691e | 2009-12-10 23:52:17 +0000 | [diff] [blame] | 1070 | rq->errors = clone->errors; |
| 1071 | rq->resid_len = clone->resid_len; |
| 1072 | |
| 1073 | if (rq->sense) |
| 1074 | /* |
| 1075 | * We are using the sense buffer of the original |
| 1076 | * request. |
| 1077 | * So setting the length of the sense data is enough. |
| 1078 | */ |
| 1079 | rq->sense_len = clone->sense_len; |
| 1080 | } |
| 1081 | |
| 1082 | free_rq_clone(clone); |
Tejun Heo | 29e4013 | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 1083 | blk_end_request_all(rq, error); |
| 1084 | rq_completed(md, rw, true); |
Kiyoshi Ueda | 980691e | 2009-12-10 23:52:17 +0000 | [diff] [blame] | 1085 | } |
| 1086 | |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1087 | static void dm_unprep_request(struct request *rq) |
| 1088 | { |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1089 | struct dm_rq_target_io *tio = rq->special; |
| 1090 | struct request *clone = tio->clone; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1091 | |
| 1092 | rq->special = NULL; |
| 1093 | rq->cmd_flags &= ~REQ_DONTPREP; |
| 1094 | |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 1095 | if (clone) |
| 1096 | free_rq_clone(clone); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1097 | } |
| 1098 | |
| 1099 | /* |
| 1100 | * Requeue the original request of a clone. |
| 1101 | */ |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1102 | static void dm_requeue_unmapped_original_request(struct mapped_device *md, |
| 1103 | struct request *rq) |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1104 | { |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1105 | int rw = rq_data_dir(rq); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1106 | struct request_queue *q = rq->q; |
| 1107 | unsigned long flags; |
| 1108 | |
| 1109 | dm_unprep_request(rq); |
| 1110 | |
| 1111 | spin_lock_irqsave(q->queue_lock, flags); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1112 | blk_requeue_request(q, rq); |
| 1113 | spin_unlock_irqrestore(q->queue_lock, flags); |
| 1114 | |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1115 | rq_completed(md, rw, false); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1116 | } |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1117 | |
| 1118 | static void dm_requeue_unmapped_request(struct request *clone) |
| 1119 | { |
| 1120 | struct dm_rq_target_io *tio = clone->end_io_data; |
| 1121 | |
| 1122 | dm_requeue_unmapped_original_request(tio->md, tio->orig); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1123 | } |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1124 | |
| 1125 | static void __stop_queue(struct request_queue *q) |
| 1126 | { |
| 1127 | blk_stop_queue(q); |
| 1128 | } |
| 1129 | |
| 1130 | static void stop_queue(struct request_queue *q) |
| 1131 | { |
| 1132 | unsigned long flags; |
| 1133 | |
| 1134 | spin_lock_irqsave(q->queue_lock, flags); |
| 1135 | __stop_queue(q); |
| 1136 | spin_unlock_irqrestore(q->queue_lock, flags); |
| 1137 | } |
| 1138 | |
| 1139 | static void __start_queue(struct request_queue *q) |
| 1140 | { |
| 1141 | if (blk_queue_stopped(q)) |
| 1142 | blk_start_queue(q); |
| 1143 | } |
| 1144 | |
| 1145 | static void start_queue(struct request_queue *q) |
| 1146 | { |
| 1147 | unsigned long flags; |
| 1148 | |
| 1149 | spin_lock_irqsave(q->queue_lock, flags); |
| 1150 | __start_queue(q); |
| 1151 | spin_unlock_irqrestore(q->queue_lock, flags); |
| 1152 | } |
| 1153 | |
Kiyoshi Ueda | 11a6824 | 2009-12-10 23:52:17 +0000 | [diff] [blame] | 1154 | static void dm_done(struct request *clone, int error, bool mapped) |
| 1155 | { |
| 1156 | int r = error; |
| 1157 | struct dm_rq_target_io *tio = clone->end_io_data; |
Mike Snitzer | ba1cbad | 2012-09-26 23:45:42 +0100 | [diff] [blame] | 1158 | dm_request_endio_fn rq_end_io = NULL; |
Kiyoshi Ueda | 11a6824 | 2009-12-10 23:52:17 +0000 | [diff] [blame] | 1159 | |
Mike Snitzer | ba1cbad | 2012-09-26 23:45:42 +0100 | [diff] [blame] | 1160 | if (tio->ti) { |
| 1161 | rq_end_io = tio->ti->type->rq_end_io; |
| 1162 | |
| 1163 | if (mapped && rq_end_io) |
| 1164 | r = rq_end_io(tio->ti, clone, error, &tio->info); |
| 1165 | } |
Kiyoshi Ueda | 11a6824 | 2009-12-10 23:52:17 +0000 | [diff] [blame] | 1166 | |
Mike Snitzer | 7eee4ae | 2014-06-02 15:50:06 -0400 | [diff] [blame] | 1167 | if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) && |
| 1168 | !clone->q->limits.max_write_same_sectors)) |
| 1169 | disable_write_same(tio->md); |
| 1170 | |
Kiyoshi Ueda | 11a6824 | 2009-12-10 23:52:17 +0000 | [diff] [blame] | 1171 | if (r <= 0) |
| 1172 | /* The target wants to complete the I/O */ |
| 1173 | dm_end_request(clone, r); |
| 1174 | else if (r == DM_ENDIO_INCOMPLETE) |
| 1175 | /* The target will handle the I/O */ |
| 1176 | return; |
| 1177 | else if (r == DM_ENDIO_REQUEUE) |
| 1178 | /* The target wants to requeue the I/O */ |
| 1179 | dm_requeue_unmapped_request(clone); |
| 1180 | else { |
| 1181 | DMWARN("unimplemented target endio return value: %d", r); |
| 1182 | BUG(); |
| 1183 | } |
| 1184 | } |
| 1185 | |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1186 | /* |
| 1187 | * Request completion handler for request-based dm |
| 1188 | */ |
| 1189 | static void dm_softirq_done(struct request *rq) |
| 1190 | { |
Kiyoshi Ueda | 11a6824 | 2009-12-10 23:52:17 +0000 | [diff] [blame] | 1191 | bool mapped = true; |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1192 | struct dm_rq_target_io *tio = rq->special; |
| 1193 | struct request *clone = tio->clone; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1194 | |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 1195 | if (!clone) { |
| 1196 | blk_end_request_all(rq, tio->error); |
| 1197 | rq_completed(tio->md, rq_data_dir(rq), false); |
| 1198 | free_rq_tio(tio); |
| 1199 | return; |
| 1200 | } |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1201 | |
Kiyoshi Ueda | 11a6824 | 2009-12-10 23:52:17 +0000 | [diff] [blame] | 1202 | if (rq->cmd_flags & REQ_FAILED) |
| 1203 | mapped = false; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1204 | |
Kiyoshi Ueda | 11a6824 | 2009-12-10 23:52:17 +0000 | [diff] [blame] | 1205 | dm_done(clone, tio->error, mapped); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1206 | } |
| 1207 | |
| 1208 | /* |
| 1209 | * Complete the clone and the original request with the error status |
| 1210 | * through softirq context. |
| 1211 | */ |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1212 | static void dm_complete_request(struct request *rq, int error) |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1213 | { |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1214 | struct dm_rq_target_io *tio = rq->special; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1215 | |
| 1216 | tio->error = error; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1217 | blk_complete_request(rq); |
| 1218 | } |
| 1219 | |
| 1220 | /* |
| 1221 | * Complete the not-mapped clone and the original request with the error status |
| 1222 | * through softirq context. |
| 1223 | * Target's rq_end_io() function isn't called. |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 1224 | * This may be used when the target's map_rq() or clone_and_map_rq() functions fail. |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1225 | */ |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1226 | static void dm_kill_unmapped_request(struct request *rq, int error) |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1227 | { |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1228 | rq->cmd_flags |= REQ_FAILED; |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1229 | dm_complete_request(rq, error); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1230 | } |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1231 | |
| 1232 | /* |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1233 | * Called with the clone's queue lock held |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1234 | */ |
| 1235 | static void end_clone_request(struct request *clone, int error) |
| 1236 | { |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1237 | struct dm_rq_target_io *tio = clone->end_io_data; |
| 1238 | |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 1239 | if (!clone->q->mq_ops) { |
| 1240 | /* |
| 1241 | * For just cleaning up the information of the queue in which |
| 1242 | * the clone was dispatched. |
| 1243 | * The clone is *NOT* freed actually here because it is alloced |
| 1244 | * from dm own mempool (REQ_ALLOCED isn't set). |
| 1245 | */ |
| 1246 | __blk_put_request(clone->q, clone); |
| 1247 | } |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1248 | |
| 1249 | /* |
| 1250 | * Actual request completion is done in a softirq context which doesn't |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1251 | * hold the clone's queue lock. Otherwise, deadlock could occur because: |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1252 | * - another request may be submitted by the upper level driver |
| 1253 | * of the stacking during the completion |
| 1254 | * - the submission which requires queue lock may be done |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1255 | * against this clone's queue |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1256 | */ |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1257 | dm_complete_request(tio->orig, error); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1258 | } |
| 1259 | |
Mike Snitzer | 56a67df | 2010-08-12 04:14:10 +0100 | [diff] [blame] | 1260 | /* |
| 1261 | * Return maximum size of I/O possible at the supplied sector up to the current |
| 1262 | * target boundary. |
| 1263 | */ |
| 1264 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1265 | { |
Mike Snitzer | 56a67df | 2010-08-12 04:14:10 +0100 | [diff] [blame] | 1266 | sector_t target_offset = dm_target_offset(ti, sector); |
| 1267 | |
| 1268 | return ti->len - target_offset; |
| 1269 | } |
| 1270 | |
| 1271 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) |
| 1272 | { |
| 1273 | sector_t len = max_io_len_target_boundary(sector, ti); |
Mike Snitzer | 542f903 | 2012-07-27 15:08:00 +0100 | [diff] [blame] | 1274 | sector_t offset, max_len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1275 | |
| 1276 | /* |
Mike Snitzer | 542f903 | 2012-07-27 15:08:00 +0100 | [diff] [blame] | 1277 | * Does the target need to split even further? |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1278 | */ |
Mike Snitzer | 542f903 | 2012-07-27 15:08:00 +0100 | [diff] [blame] | 1279 | if (ti->max_io_len) { |
| 1280 | offset = dm_target_offset(ti, sector); |
| 1281 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) |
| 1282 | max_len = sector_div(offset, ti->max_io_len); |
| 1283 | else |
| 1284 | max_len = offset & (ti->max_io_len - 1); |
| 1285 | max_len = ti->max_io_len - max_len; |
| 1286 | |
| 1287 | if (len > max_len) |
| 1288 | len = max_len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1289 | } |
| 1290 | |
| 1291 | return len; |
| 1292 | } |
| 1293 | |
Mike Snitzer | 542f903 | 2012-07-27 15:08:00 +0100 | [diff] [blame] | 1294 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
| 1295 | { |
| 1296 | if (len > UINT_MAX) { |
| 1297 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", |
| 1298 | (unsigned long long)len, UINT_MAX); |
| 1299 | ti->error = "Maximum size of target IO is too large"; |
| 1300 | return -EINVAL; |
| 1301 | } |
| 1302 | |
| 1303 | ti->max_io_len = (uint32_t) len; |
| 1304 | |
| 1305 | return 0; |
| 1306 | } |
| 1307 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); |
| 1308 | |
Mikulas Patocka | 1dd40c3 | 2014-03-14 18:41:24 -0400 | [diff] [blame] | 1309 | /* |
| 1310 | * A target may call dm_accept_partial_bio only from the map routine. It is |
| 1311 | * allowed for all bio types except REQ_FLUSH. |
| 1312 | * |
| 1313 | * dm_accept_partial_bio informs the dm that the target only wants to process |
| 1314 | * additional n_sectors sectors of the bio and the rest of the data should be |
| 1315 | * sent in a next bio. |
| 1316 | * |
| 1317 | * A diagram that explains the arithmetics: |
| 1318 | * +--------------------+---------------+-------+ |
| 1319 | * | 1 | 2 | 3 | |
| 1320 | * +--------------------+---------------+-------+ |
| 1321 | * |
| 1322 | * <-------------- *tio->len_ptr ---------------> |
| 1323 | * <------- bi_size -------> |
| 1324 | * <-- n_sectors --> |
| 1325 | * |
| 1326 | * Region 1 was already iterated over with bio_advance or similar function. |
| 1327 | * (it may be empty if the target doesn't use bio_advance) |
| 1328 | * Region 2 is the remaining bio size that the target wants to process. |
| 1329 | * (it may be empty if region 1 is non-empty, although there is no reason |
| 1330 | * to make it empty) |
| 1331 | * The target requires that region 3 is to be sent in the next bio. |
| 1332 | * |
| 1333 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), |
| 1334 | * the partially processed part (the sum of regions 1+2) must be the same for all |
| 1335 | * copies of the bio. |
| 1336 | */ |
| 1337 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) |
| 1338 | { |
| 1339 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
| 1340 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; |
| 1341 | BUG_ON(bio->bi_rw & REQ_FLUSH); |
| 1342 | BUG_ON(bi_size > *tio->len_ptr); |
| 1343 | BUG_ON(n_sectors > bi_size); |
| 1344 | *tio->len_ptr -= bi_size - n_sectors; |
| 1345 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; |
| 1346 | } |
| 1347 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); |
| 1348 | |
Alasdair G Kergon | bd2a49b | 2013-03-01 22:45:46 +0000 | [diff] [blame] | 1349 | static void __map_bio(struct dm_target_io *tio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1350 | { |
| 1351 | int r; |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 1352 | sector_t sector; |
Stefan Bader | 9faf400 | 2006-10-03 01:15:41 -0700 | [diff] [blame] | 1353 | struct mapped_device *md; |
Mikulas Patocka | dba1416 | 2012-10-12 21:02:15 +0100 | [diff] [blame] | 1354 | struct bio *clone = &tio->clone; |
Alasdair G Kergon | bd2a49b | 2013-03-01 22:45:46 +0000 | [diff] [blame] | 1355 | struct dm_target *ti = tio->ti; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1356 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1357 | clone->bi_end_io = clone_endio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1358 | |
| 1359 | /* |
| 1360 | * Map the clone. If r == 0 we don't need to do |
| 1361 | * anything, the target has assumed ownership of |
| 1362 | * this io. |
| 1363 | */ |
| 1364 | atomic_inc(&tio->io->io_count); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1365 | sector = clone->bi_iter.bi_sector; |
Mikulas Patocka | 7de3ee5 | 2012-12-21 20:23:41 +0000 | [diff] [blame] | 1366 | r = ti->type->map(ti, clone); |
Kiyoshi Ueda | 45cbcd7 | 2006-12-08 02:41:05 -0800 | [diff] [blame] | 1367 | if (r == DM_MAPIO_REMAPPED) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1368 | /* the bio has been remapped so dispatch it */ |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 1369 | |
Mike Snitzer | d07335e | 2010-11-16 12:52:38 +0100 | [diff] [blame] | 1370 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
| 1371 | tio->io->bio->bi_bdev->bd_dev, sector); |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 1372 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1373 | generic_make_request(clone); |
Kiyoshi Ueda | 2e93ccc | 2006-12-08 02:41:09 -0800 | [diff] [blame] | 1374 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
| 1375 | /* error the io and bail out, or requeue it if needed */ |
Stefan Bader | 9faf400 | 2006-10-03 01:15:41 -0700 | [diff] [blame] | 1376 | md = tio->io->md; |
| 1377 | dec_pending(tio->io, r); |
Stefan Bader | 9faf400 | 2006-10-03 01:15:41 -0700 | [diff] [blame] | 1378 | free_tio(md, tio); |
Kiyoshi Ueda | 45cbcd7 | 2006-12-08 02:41:05 -0800 | [diff] [blame] | 1379 | } else if (r) { |
| 1380 | DMWARN("unimplemented target map return value: %d", r); |
| 1381 | BUG(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1382 | } |
| 1383 | } |
| 1384 | |
| 1385 | struct clone_info { |
| 1386 | struct mapped_device *md; |
| 1387 | struct dm_table *map; |
| 1388 | struct bio *bio; |
| 1389 | struct dm_io *io; |
| 1390 | sector_t sector; |
Mikulas Patocka | e0d6609 | 2014-03-14 18:40:39 -0400 | [diff] [blame] | 1391 | unsigned sector_count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1392 | }; |
| 1393 | |
Mikulas Patocka | e0d6609 | 2014-03-14 18:40:39 -0400 | [diff] [blame] | 1394 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
Alasdair G Kergon | bd2a49b | 2013-03-01 22:45:46 +0000 | [diff] [blame] | 1395 | { |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1396 | bio->bi_iter.bi_sector = sector; |
| 1397 | bio->bi_iter.bi_size = to_bytes(len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1398 | } |
| 1399 | |
| 1400 | /* |
| 1401 | * Creates a bio that consists of range of complete bvecs. |
| 1402 | */ |
Mikulas Patocka | dba1416 | 2012-10-12 21:02:15 +0100 | [diff] [blame] | 1403 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
Kent Overstreet | 1c3b13e | 2013-10-29 17:17:49 -0700 | [diff] [blame] | 1404 | sector_t sector, unsigned len) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1405 | { |
Mikulas Patocka | dba1416 | 2012-10-12 21:02:15 +0100 | [diff] [blame] | 1406 | struct bio *clone = &tio->clone; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1407 | |
Kent Overstreet | 1c3b13e | 2013-10-29 17:17:49 -0700 | [diff] [blame] | 1408 | __bio_clone_fast(clone, bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1409 | |
Kent Overstreet | 1c3b13e | 2013-10-29 17:17:49 -0700 | [diff] [blame] | 1410 | if (bio_integrity(bio)) |
| 1411 | bio_integrity_clone(clone, bio, GFP_NOIO); |
| 1412 | |
| 1413 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
| 1414 | clone->bi_iter.bi_size = to_bytes(len); |
| 1415 | |
| 1416 | if (bio_integrity(bio)) |
| 1417 | bio_integrity_trim(clone, 0, len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1418 | } |
| 1419 | |
Alasdair G Kergon | 9015df2 | 2009-06-22 10:12:21 +0100 | [diff] [blame] | 1420 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
Junichi Nomura | 9977827 | 2014-10-03 11:55:16 +0000 | [diff] [blame] | 1421 | struct dm_target *ti, |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1422 | unsigned target_bio_nr) |
Mikulas Patocka | f9ab94c | 2009-06-22 10:12:20 +0100 | [diff] [blame] | 1423 | { |
Mikulas Patocka | dba1416 | 2012-10-12 21:02:15 +0100 | [diff] [blame] | 1424 | struct dm_target_io *tio; |
| 1425 | struct bio *clone; |
| 1426 | |
Junichi Nomura | 9977827 | 2014-10-03 11:55:16 +0000 | [diff] [blame] | 1427 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); |
Mikulas Patocka | dba1416 | 2012-10-12 21:02:15 +0100 | [diff] [blame] | 1428 | tio = container_of(clone, struct dm_target_io, clone); |
Mikulas Patocka | f9ab94c | 2009-06-22 10:12:20 +0100 | [diff] [blame] | 1429 | |
| 1430 | tio->io = ci->io; |
| 1431 | tio->ti = ti; |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1432 | tio->target_bio_nr = target_bio_nr; |
Alasdair G Kergon | 9015df2 | 2009-06-22 10:12:21 +0100 | [diff] [blame] | 1433 | |
| 1434 | return tio; |
| 1435 | } |
| 1436 | |
Alasdair G Kergon | 14fe594 | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1437 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
| 1438 | struct dm_target *ti, |
Mikulas Patocka | 1dd40c3 | 2014-03-14 18:41:24 -0400 | [diff] [blame] | 1439 | unsigned target_bio_nr, unsigned *len) |
Alasdair G Kergon | 9015df2 | 2009-06-22 10:12:21 +0100 | [diff] [blame] | 1440 | { |
Junichi Nomura | 9977827 | 2014-10-03 11:55:16 +0000 | [diff] [blame] | 1441 | struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); |
Mikulas Patocka | dba1416 | 2012-10-12 21:02:15 +0100 | [diff] [blame] | 1442 | struct bio *clone = &tio->clone; |
Alasdair G Kergon | 9015df2 | 2009-06-22 10:12:21 +0100 | [diff] [blame] | 1443 | |
Mikulas Patocka | 1dd40c3 | 2014-03-14 18:41:24 -0400 | [diff] [blame] | 1444 | tio->len_ptr = len; |
| 1445 | |
Junichi Nomura | 9977827 | 2014-10-03 11:55:16 +0000 | [diff] [blame] | 1446 | __bio_clone_fast(clone, ci->bio); |
Alasdair G Kergon | bd2a49b | 2013-03-01 22:45:46 +0000 | [diff] [blame] | 1447 | if (len) |
Mikulas Patocka | 1dd40c3 | 2014-03-14 18:41:24 -0400 | [diff] [blame] | 1448 | bio_setup_sector(clone, ci->sector, *len); |
Mikulas Patocka | f9ab94c | 2009-06-22 10:12:20 +0100 | [diff] [blame] | 1449 | |
Alasdair G Kergon | bd2a49b | 2013-03-01 22:45:46 +0000 | [diff] [blame] | 1450 | __map_bio(tio); |
Mikulas Patocka | f9ab94c | 2009-06-22 10:12:20 +0100 | [diff] [blame] | 1451 | } |
| 1452 | |
Alasdair G Kergon | 14fe594 | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1453 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
Mikulas Patocka | 1dd40c3 | 2014-03-14 18:41:24 -0400 | [diff] [blame] | 1454 | unsigned num_bios, unsigned *len) |
Mike Snitzer | 06a426c | 2010-08-12 04:14:09 +0100 | [diff] [blame] | 1455 | { |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1456 | unsigned target_bio_nr; |
Mike Snitzer | 06a426c | 2010-08-12 04:14:09 +0100 | [diff] [blame] | 1457 | |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1458 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
Alasdair G Kergon | 14fe594 | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1459 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
Mike Snitzer | 06a426c | 2010-08-12 04:14:09 +0100 | [diff] [blame] | 1460 | } |
| 1461 | |
Alasdair G Kergon | 14fe594 | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1462 | static int __send_empty_flush(struct clone_info *ci) |
Mikulas Patocka | f9ab94c | 2009-06-22 10:12:20 +0100 | [diff] [blame] | 1463 | { |
Mike Snitzer | 06a426c | 2010-08-12 04:14:09 +0100 | [diff] [blame] | 1464 | unsigned target_nr = 0; |
Mikulas Patocka | f9ab94c | 2009-06-22 10:12:20 +0100 | [diff] [blame] | 1465 | struct dm_target *ti; |
| 1466 | |
Mike Snitzer | b372d36 | 2010-09-08 18:07:01 +0200 | [diff] [blame] | 1467 | BUG_ON(bio_has_data(ci->bio)); |
Mikulas Patocka | f9ab94c | 2009-06-22 10:12:20 +0100 | [diff] [blame] | 1468 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
Mikulas Patocka | 1dd40c3 | 2014-03-14 18:41:24 -0400 | [diff] [blame] | 1469 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
Mikulas Patocka | f9ab94c | 2009-06-22 10:12:20 +0100 | [diff] [blame] | 1470 | |
Mikulas Patocka | f9ab94c | 2009-06-22 10:12:20 +0100 | [diff] [blame] | 1471 | return 0; |
| 1472 | } |
| 1473 | |
Alasdair G Kergon | e4c9381 | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1474 | static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
Mikulas Patocka | 1dd40c3 | 2014-03-14 18:41:24 -0400 | [diff] [blame] | 1475 | sector_t sector, unsigned *len) |
Mike Snitzer | 5ae89a8 | 2010-08-12 04:14:08 +0100 | [diff] [blame] | 1476 | { |
Mikulas Patocka | dba1416 | 2012-10-12 21:02:15 +0100 | [diff] [blame] | 1477 | struct bio *bio = ci->bio; |
Mike Snitzer | 5ae89a8 | 2010-08-12 04:14:08 +0100 | [diff] [blame] | 1478 | struct dm_target_io *tio; |
Alasdair G Kergon | b0d8ed4 | 2013-03-01 22:45:49 +0000 | [diff] [blame] | 1479 | unsigned target_bio_nr; |
| 1480 | unsigned num_target_bios = 1; |
Mike Snitzer | 5ae89a8 | 2010-08-12 04:14:08 +0100 | [diff] [blame] | 1481 | |
Alasdair G Kergon | b0d8ed4 | 2013-03-01 22:45:49 +0000 | [diff] [blame] | 1482 | /* |
| 1483 | * Does the target want to receive duplicate copies of the bio? |
| 1484 | */ |
| 1485 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) |
| 1486 | num_target_bios = ti->num_write_bios(ti, bio); |
Alasdair G Kergon | e4c9381 | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1487 | |
Alasdair G Kergon | b0d8ed4 | 2013-03-01 22:45:49 +0000 | [diff] [blame] | 1488 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
Junichi Nomura | 9977827 | 2014-10-03 11:55:16 +0000 | [diff] [blame] | 1489 | tio = alloc_tio(ci, ti, target_bio_nr); |
Mikulas Patocka | 1dd40c3 | 2014-03-14 18:41:24 -0400 | [diff] [blame] | 1490 | tio->len_ptr = len; |
| 1491 | clone_bio(tio, bio, sector, *len); |
Alasdair G Kergon | b0d8ed4 | 2013-03-01 22:45:49 +0000 | [diff] [blame] | 1492 | __map_bio(tio); |
| 1493 | } |
Mike Snitzer | 5ae89a8 | 2010-08-12 04:14:08 +0100 | [diff] [blame] | 1494 | } |
| 1495 | |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1496 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
Mike Snitzer | 23508a9 | 2012-12-21 20:23:37 +0000 | [diff] [blame] | 1497 | |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1498 | static unsigned get_num_discard_bios(struct dm_target *ti) |
Mike Snitzer | 23508a9 | 2012-12-21 20:23:37 +0000 | [diff] [blame] | 1499 | { |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1500 | return ti->num_discard_bios; |
Mike Snitzer | 23508a9 | 2012-12-21 20:23:37 +0000 | [diff] [blame] | 1501 | } |
| 1502 | |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1503 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
Mike Snitzer | 23508a9 | 2012-12-21 20:23:37 +0000 | [diff] [blame] | 1504 | { |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1505 | return ti->num_write_same_bios; |
Mike Snitzer | 23508a9 | 2012-12-21 20:23:37 +0000 | [diff] [blame] | 1506 | } |
| 1507 | |
| 1508 | typedef bool (*is_split_required_fn)(struct dm_target *ti); |
| 1509 | |
| 1510 | static bool is_split_required_for_discard(struct dm_target *ti) |
| 1511 | { |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1512 | return ti->split_discard_bios; |
Mike Snitzer | 23508a9 | 2012-12-21 20:23:37 +0000 | [diff] [blame] | 1513 | } |
| 1514 | |
Alasdair G Kergon | 14fe594 | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1515 | static int __send_changing_extent_only(struct clone_info *ci, |
| 1516 | get_num_bios_fn get_num_bios, |
| 1517 | is_split_required_fn is_split_required) |
Mike Snitzer | 5ae89a8 | 2010-08-12 04:14:08 +0100 | [diff] [blame] | 1518 | { |
| 1519 | struct dm_target *ti; |
Mikulas Patocka | e0d6609 | 2014-03-14 18:40:39 -0400 | [diff] [blame] | 1520 | unsigned len; |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1521 | unsigned num_bios; |
Mike Snitzer | 5ae89a8 | 2010-08-12 04:14:08 +0100 | [diff] [blame] | 1522 | |
Mike Snitzer | a79245b | 2010-08-12 04:14:24 +0100 | [diff] [blame] | 1523 | do { |
| 1524 | ti = dm_table_find_target(ci->map, ci->sector); |
| 1525 | if (!dm_target_is_valid(ti)) |
| 1526 | return -EIO; |
Mike Snitzer | 5ae89a8 | 2010-08-12 04:14:08 +0100 | [diff] [blame] | 1527 | |
Mike Snitzer | 5ae89a8 | 2010-08-12 04:14:08 +0100 | [diff] [blame] | 1528 | /* |
Mike Snitzer | 23508a9 | 2012-12-21 20:23:37 +0000 | [diff] [blame] | 1529 | * Even though the device advertised support for this type of |
| 1530 | * request, that does not mean every target supports it, and |
Mike Snitzer | 936688d | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 1531 | * reconfiguration might also have changed that since the |
Mike Snitzer | a79245b | 2010-08-12 04:14:24 +0100 | [diff] [blame] | 1532 | * check was performed. |
Mike Snitzer | 5ae89a8 | 2010-08-12 04:14:08 +0100 | [diff] [blame] | 1533 | */ |
Alasdair G Kergon | 55a62ee | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1534 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
| 1535 | if (!num_bios) |
Mike Snitzer | a79245b | 2010-08-12 04:14:24 +0100 | [diff] [blame] | 1536 | return -EOPNOTSUPP; |
Mike Snitzer | 5ae89a8 | 2010-08-12 04:14:08 +0100 | [diff] [blame] | 1537 | |
Mike Snitzer | 23508a9 | 2012-12-21 20:23:37 +0000 | [diff] [blame] | 1538 | if (is_split_required && !is_split_required(ti)) |
Mikulas Patocka | e0d6609 | 2014-03-14 18:40:39 -0400 | [diff] [blame] | 1539 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
Mikulas Patocka | 7acf027 | 2012-07-27 15:08:03 +0100 | [diff] [blame] | 1540 | else |
Mikulas Patocka | e0d6609 | 2014-03-14 18:40:39 -0400 | [diff] [blame] | 1541 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
Mike Snitzer | 06a426c | 2010-08-12 04:14:09 +0100 | [diff] [blame] | 1542 | |
Mikulas Patocka | 1dd40c3 | 2014-03-14 18:41:24 -0400 | [diff] [blame] | 1543 | __send_duplicate_bios(ci, ti, num_bios, &len); |
Mike Snitzer | a79245b | 2010-08-12 04:14:24 +0100 | [diff] [blame] | 1544 | |
| 1545 | ci->sector += len; |
| 1546 | } while (ci->sector_count -= len); |
Mike Snitzer | 5ae89a8 | 2010-08-12 04:14:08 +0100 | [diff] [blame] | 1547 | |
| 1548 | return 0; |
| 1549 | } |
| 1550 | |
Alasdair G Kergon | 14fe594 | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1551 | static int __send_discard(struct clone_info *ci) |
Mike Snitzer | 23508a9 | 2012-12-21 20:23:37 +0000 | [diff] [blame] | 1552 | { |
Alasdair G Kergon | 14fe594 | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1553 | return __send_changing_extent_only(ci, get_num_discard_bios, |
| 1554 | is_split_required_for_discard); |
Mike Snitzer | 23508a9 | 2012-12-21 20:23:37 +0000 | [diff] [blame] | 1555 | } |
| 1556 | |
Alasdair G Kergon | 14fe594 | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1557 | static int __send_write_same(struct clone_info *ci) |
Mike Snitzer | 23508a9 | 2012-12-21 20:23:37 +0000 | [diff] [blame] | 1558 | { |
Alasdair G Kergon | 14fe594 | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1559 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
Mike Snitzer | 23508a9 | 2012-12-21 20:23:37 +0000 | [diff] [blame] | 1560 | } |
| 1561 | |
Alasdair G Kergon | e4c9381 | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1562 | /* |
Alasdair G Kergon | e4c9381 | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1563 | * Select the correct strategy for processing a non-flush bio. |
| 1564 | */ |
Alasdair G Kergon | 14fe594 | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1565 | static int __split_and_process_non_flush(struct clone_info *ci) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1566 | { |
Mikulas Patocka | dba1416 | 2012-10-12 21:02:15 +0100 | [diff] [blame] | 1567 | struct bio *bio = ci->bio; |
Jun'ichi Nomura | 512875b | 2007-12-13 14:15:25 +0000 | [diff] [blame] | 1568 | struct dm_target *ti; |
Kent Overstreet | 1c3b13e | 2013-10-29 17:17:49 -0700 | [diff] [blame] | 1569 | unsigned len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1570 | |
Mike Snitzer | 5ae89a8 | 2010-08-12 04:14:08 +0100 | [diff] [blame] | 1571 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
Alasdair G Kergon | 14fe594 | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1572 | return __send_discard(ci); |
Mike Snitzer | 23508a9 | 2012-12-21 20:23:37 +0000 | [diff] [blame] | 1573 | else if (unlikely(bio->bi_rw & REQ_WRITE_SAME)) |
Alasdair G Kergon | 14fe594 | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1574 | return __send_write_same(ci); |
Mike Snitzer | 5ae89a8 | 2010-08-12 04:14:08 +0100 | [diff] [blame] | 1575 | |
Jun'ichi Nomura | 512875b | 2007-12-13 14:15:25 +0000 | [diff] [blame] | 1576 | ti = dm_table_find_target(ci->map, ci->sector); |
| 1577 | if (!dm_target_is_valid(ti)) |
| 1578 | return -EIO; |
| 1579 | |
Kent Overstreet | 1c3b13e | 2013-10-29 17:17:49 -0700 | [diff] [blame] | 1580 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
Jun'ichi Nomura | 512875b | 2007-12-13 14:15:25 +0000 | [diff] [blame] | 1581 | |
Mikulas Patocka | 1dd40c3 | 2014-03-14 18:41:24 -0400 | [diff] [blame] | 1582 | __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1583 | |
Kent Overstreet | 1c3b13e | 2013-10-29 17:17:49 -0700 | [diff] [blame] | 1584 | ci->sector += len; |
| 1585 | ci->sector_count -= len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1586 | |
Kent Overstreet | 1c3b13e | 2013-10-29 17:17:49 -0700 | [diff] [blame] | 1587 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1588 | } |
| 1589 | |
| 1590 | /* |
Alasdair G Kergon | 14fe594 | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1591 | * Entry point to split a bio into clones and submit them to the targets. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1592 | */ |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 1593 | static void __split_and_process_bio(struct mapped_device *md, |
| 1594 | struct dm_table *map, struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1595 | { |
| 1596 | struct clone_info ci; |
Jun'ichi Nomura | 512875b | 2007-12-13 14:15:25 +0000 | [diff] [blame] | 1597 | int error = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1598 | |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 1599 | if (unlikely(!map)) { |
Tejun Heo | 6a8736d | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 1600 | bio_io_error(bio); |
Mikulas Patocka | f0b9a45 | 2009-04-02 19:55:38 +0100 | [diff] [blame] | 1601 | return; |
| 1602 | } |
Mikulas Patocka | 692d0eb | 2009-04-09 00:27:13 +0100 | [diff] [blame] | 1603 | |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 1604 | ci.map = map; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1605 | ci.md = md; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1606 | ci.io = alloc_io(md); |
| 1607 | ci.io->error = 0; |
| 1608 | atomic_set(&ci.io->io_count, 1); |
| 1609 | ci.io->bio = bio; |
| 1610 | ci.io->md = md; |
Kiyoshi Ueda | f88fb98 | 2009-10-16 23:18:15 +0100 | [diff] [blame] | 1611 | spin_lock_init(&ci.io->endio_lock); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1612 | ci.sector = bio->bi_iter.bi_sector; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1613 | |
Jun'ichi "Nick" Nomura | 3eaf840 | 2006-02-01 03:04:53 -0800 | [diff] [blame] | 1614 | start_io_acct(ci.io); |
Alasdair G Kergon | bd2a49b | 2013-03-01 22:45:46 +0000 | [diff] [blame] | 1615 | |
Mike Snitzer | b372d36 | 2010-09-08 18:07:01 +0200 | [diff] [blame] | 1616 | if (bio->bi_rw & REQ_FLUSH) { |
| 1617 | ci.bio = &ci.md->flush_bio; |
| 1618 | ci.sector_count = 0; |
Alasdair G Kergon | 14fe594 | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1619 | error = __send_empty_flush(&ci); |
Mike Snitzer | b372d36 | 2010-09-08 18:07:01 +0200 | [diff] [blame] | 1620 | /* dec_pending submits any data associated with flush */ |
| 1621 | } else { |
Tejun Heo | 6a8736d | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 1622 | ci.bio = bio; |
Tejun Heo | d87f4c1 | 2010-09-03 11:56:19 +0200 | [diff] [blame] | 1623 | ci.sector_count = bio_sectors(bio); |
Mike Snitzer | b372d36 | 2010-09-08 18:07:01 +0200 | [diff] [blame] | 1624 | while (ci.sector_count && !error) |
Alasdair G Kergon | 14fe594 | 2013-03-01 22:45:47 +0000 | [diff] [blame] | 1625 | error = __split_and_process_non_flush(&ci); |
Tejun Heo | d87f4c1 | 2010-09-03 11:56:19 +0200 | [diff] [blame] | 1626 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1627 | |
| 1628 | /* drop the extra reference count */ |
Jun'ichi Nomura | 512875b | 2007-12-13 14:15:25 +0000 | [diff] [blame] | 1629 | dec_pending(ci.io, error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1630 | } |
| 1631 | /*----------------------------------------------------------------- |
| 1632 | * CRUD END |
| 1633 | *---------------------------------------------------------------*/ |
| 1634 | |
Milan Broz | f6fccb1 | 2008-07-21 12:00:37 +0100 | [diff] [blame] | 1635 | static int dm_merge_bvec(struct request_queue *q, |
| 1636 | struct bvec_merge_data *bvm, |
| 1637 | struct bio_vec *biovec) |
| 1638 | { |
| 1639 | struct mapped_device *md = q->queuedata; |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 1640 | struct dm_table *map = dm_get_live_table_fast(md); |
Milan Broz | f6fccb1 | 2008-07-21 12:00:37 +0100 | [diff] [blame] | 1641 | struct dm_target *ti; |
| 1642 | sector_t max_sectors; |
Mikulas Patocka | 5037108 | 2008-10-01 14:39:17 +0100 | [diff] [blame] | 1643 | int max_size = 0; |
Milan Broz | f6fccb1 | 2008-07-21 12:00:37 +0100 | [diff] [blame] | 1644 | |
| 1645 | if (unlikely(!map)) |
Mikulas Patocka | 5037108 | 2008-10-01 14:39:17 +0100 | [diff] [blame] | 1646 | goto out; |
Milan Broz | f6fccb1 | 2008-07-21 12:00:37 +0100 | [diff] [blame] | 1647 | |
| 1648 | ti = dm_table_find_target(map, bvm->bi_sector); |
Mikulas Patocka | b01cd5a | 2008-10-01 14:39:24 +0100 | [diff] [blame] | 1649 | if (!dm_target_is_valid(ti)) |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 1650 | goto out; |
Milan Broz | f6fccb1 | 2008-07-21 12:00:37 +0100 | [diff] [blame] | 1651 | |
| 1652 | /* |
| 1653 | * Find maximum amount of I/O that won't need splitting |
| 1654 | */ |
Mike Snitzer | 56a67df | 2010-08-12 04:14:10 +0100 | [diff] [blame] | 1655 | max_sectors = min(max_io_len(bvm->bi_sector, ti), |
Mike Snitzer | 148e51b | 2014-10-09 19:32:22 -0400 | [diff] [blame] | 1656 | (sector_t) queue_max_sectors(q)); |
Milan Broz | f6fccb1 | 2008-07-21 12:00:37 +0100 | [diff] [blame] | 1657 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; |
Mike Snitzer | 148e51b | 2014-10-09 19:32:22 -0400 | [diff] [blame] | 1658 | if (unlikely(max_size < 0)) /* this shouldn't _ever_ happen */ |
Milan Broz | f6fccb1 | 2008-07-21 12:00:37 +0100 | [diff] [blame] | 1659 | max_size = 0; |
| 1660 | |
| 1661 | /* |
| 1662 | * merge_bvec_fn() returns number of bytes |
| 1663 | * it can accept at this offset |
| 1664 | * max is precomputed maximal io size |
| 1665 | */ |
| 1666 | if (max_size && ti->type->merge) |
| 1667 | max_size = ti->type->merge(ti, bvm, biovec, max_size); |
Mikulas Patocka | 8cbeb67 | 2009-06-22 10:12:14 +0100 | [diff] [blame] | 1668 | /* |
| 1669 | * If the target doesn't support merge method and some of the devices |
Mike Snitzer | 148e51b | 2014-10-09 19:32:22 -0400 | [diff] [blame] | 1670 | * provided their merge_bvec method (we know this by looking for the |
| 1671 | * max_hw_sectors that dm_set_device_limits may set), then we can't |
| 1672 | * allow bios with multiple vector entries. So always set max_size |
| 1673 | * to 0, and the code below allows just one page. |
Mikulas Patocka | 8cbeb67 | 2009-06-22 10:12:14 +0100 | [diff] [blame] | 1674 | */ |
| 1675 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) |
Mikulas Patocka | 8cbeb67 | 2009-06-22 10:12:14 +0100 | [diff] [blame] | 1676 | max_size = 0; |
Milan Broz | f6fccb1 | 2008-07-21 12:00:37 +0100 | [diff] [blame] | 1677 | |
Mikulas Patocka | 5037108 | 2008-10-01 14:39:17 +0100 | [diff] [blame] | 1678 | out: |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 1679 | dm_put_live_table_fast(md); |
Milan Broz | f6fccb1 | 2008-07-21 12:00:37 +0100 | [diff] [blame] | 1680 | /* |
| 1681 | * Always allow an entire first page |
| 1682 | */ |
| 1683 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) |
| 1684 | max_size = biovec->bv_len; |
| 1685 | |
Milan Broz | f6fccb1 | 2008-07-21 12:00:37 +0100 | [diff] [blame] | 1686 | return max_size; |
| 1687 | } |
| 1688 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1689 | /* |
| 1690 | * The request function that just remaps the bio built up by |
| 1691 | * dm_merge_bvec. |
| 1692 | */ |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1693 | static void _dm_request(struct request_queue *q, struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1694 | { |
Kevin Corry | 12f03a4 | 2006-02-01 03:04:52 -0800 | [diff] [blame] | 1695 | int rw = bio_data_dir(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1696 | struct mapped_device *md = q->queuedata; |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 1697 | int srcu_idx; |
| 1698 | struct dm_table *map; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1699 | |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 1700 | map = dm_get_live_table(md, &srcu_idx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1701 | |
Gu Zheng | 18c0b22 | 2014-11-24 11:05:26 +0800 | [diff] [blame] | 1702 | generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0); |
Kevin Corry | 12f03a4 | 2006-02-01 03:04:52 -0800 | [diff] [blame] | 1703 | |
Tejun Heo | 6a8736d | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 1704 | /* if we're suspended, we have to queue this io for later */ |
| 1705 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 1706 | dm_put_live_table(md, srcu_idx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1707 | |
Tejun Heo | 6a8736d | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 1708 | if (bio_rw(bio) != READA) |
| 1709 | queue_io(md, bio); |
| 1710 | else |
Alasdair G Kergon | 54d9a1b | 2009-04-09 00:27:14 +0100 | [diff] [blame] | 1711 | bio_io_error(bio); |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1712 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1713 | } |
| 1714 | |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 1715 | __split_and_process_bio(md, map, bio); |
| 1716 | dm_put_live_table(md, srcu_idx); |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1717 | return; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1718 | } |
| 1719 | |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 1720 | int dm_request_based(struct mapped_device *md) |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1721 | { |
| 1722 | return blk_queue_stackable(md->queue); |
| 1723 | } |
| 1724 | |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1725 | static void dm_request(struct request_queue *q, struct bio *bio) |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1726 | { |
| 1727 | struct mapped_device *md = q->queuedata; |
| 1728 | |
| 1729 | if (dm_request_based(md)) |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1730 | blk_queue_bio(q, bio); |
| 1731 | else |
| 1732 | _dm_request(q, bio); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1733 | } |
| 1734 | |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1735 | static void dm_dispatch_clone_request(struct request *clone, struct request *rq) |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1736 | { |
| 1737 | int r; |
| 1738 | |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1739 | if (blk_queue_io_stat(clone->q)) |
| 1740 | clone->cmd_flags |= REQ_IO_STAT; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1741 | |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1742 | clone->start_time = jiffies; |
| 1743 | r = blk_insert_cloned_request(clone->q, clone); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1744 | if (r) |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1745 | /* must complete clone in terms of original request */ |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1746 | dm_complete_request(rq, r); |
| 1747 | } |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1748 | |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1749 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
| 1750 | void *data) |
| 1751 | { |
| 1752 | struct dm_rq_target_io *tio = data; |
Kent Overstreet | 9481874 | 2012-09-07 13:44:01 -0700 | [diff] [blame] | 1753 | struct dm_rq_clone_bio_info *info = |
| 1754 | container_of(bio, struct dm_rq_clone_bio_info, clone); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1755 | |
| 1756 | info->orig = bio_orig; |
| 1757 | info->tio = tio; |
| 1758 | bio->bi_end_io = end_clone_bio; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1759 | |
| 1760 | return 0; |
| 1761 | } |
| 1762 | |
| 1763 | static int setup_clone(struct request *clone, struct request *rq, |
Mike Snitzer | 1ae49ea | 2014-12-05 17:11:05 -0500 | [diff] [blame] | 1764 | struct dm_rq_target_io *tio, gfp_t gfp_mask) |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1765 | { |
Kiyoshi Ueda | d0bcb87 | 2009-12-10 23:52:18 +0000 | [diff] [blame] | 1766 | int r; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1767 | |
Mike Snitzer | 1ae49ea | 2014-12-05 17:11:05 -0500 | [diff] [blame] | 1768 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask, |
Tejun Heo | 29e4013 | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 1769 | dm_rq_bio_constructor, tio); |
| 1770 | if (r) |
| 1771 | return r; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1772 | |
Tejun Heo | 29e4013 | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 1773 | clone->cmd = rq->cmd; |
| 1774 | clone->cmd_len = rq->cmd_len; |
| 1775 | clone->sense = rq->sense; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1776 | clone->end_io = end_clone_request; |
| 1777 | clone->end_io_data = tio; |
| 1778 | |
Mike Snitzer | 1ae49ea | 2014-12-05 17:11:05 -0500 | [diff] [blame] | 1779 | tio->clone = clone; |
| 1780 | |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1781 | return 0; |
| 1782 | } |
| 1783 | |
Kiyoshi Ueda | 6facdaf | 2009-12-10 23:52:15 +0000 | [diff] [blame] | 1784 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1785 | struct dm_rq_target_io *tio, gfp_t gfp_mask) |
Kiyoshi Ueda | 6facdaf | 2009-12-10 23:52:15 +0000 | [diff] [blame] | 1786 | { |
Mike Snitzer | 1ae49ea | 2014-12-05 17:11:05 -0500 | [diff] [blame] | 1787 | struct request *clone = alloc_clone_request(md, gfp_mask); |
| 1788 | |
| 1789 | if (!clone) |
| 1790 | return NULL; |
| 1791 | |
| 1792 | blk_rq_init(NULL, clone); |
| 1793 | if (setup_clone(clone, rq, tio, gfp_mask)) { |
| 1794 | /* -ENOMEM */ |
| 1795 | free_clone_request(md, clone); |
| 1796 | return NULL; |
| 1797 | } |
| 1798 | |
| 1799 | return clone; |
| 1800 | } |
| 1801 | |
Keith Busch | 2eb6e1e | 2014-10-17 17:46:36 -0600 | [diff] [blame] | 1802 | static void map_tio_request(struct kthread_work *work); |
| 1803 | |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1804 | static struct dm_rq_target_io *prep_tio(struct request *rq, |
| 1805 | struct mapped_device *md, gfp_t gfp_mask) |
Kiyoshi Ueda | 6facdaf | 2009-12-10 23:52:15 +0000 | [diff] [blame] | 1806 | { |
Kiyoshi Ueda | 6facdaf | 2009-12-10 23:52:15 +0000 | [diff] [blame] | 1807 | struct dm_rq_target_io *tio; |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 1808 | int srcu_idx; |
| 1809 | struct dm_table *table; |
Kiyoshi Ueda | 6facdaf | 2009-12-10 23:52:15 +0000 | [diff] [blame] | 1810 | |
| 1811 | tio = alloc_rq_tio(md, gfp_mask); |
| 1812 | if (!tio) |
| 1813 | return NULL; |
| 1814 | |
| 1815 | tio->md = md; |
| 1816 | tio->ti = NULL; |
Mike Snitzer | 1ae49ea | 2014-12-05 17:11:05 -0500 | [diff] [blame] | 1817 | tio->clone = NULL; |
Kiyoshi Ueda | 6facdaf | 2009-12-10 23:52:15 +0000 | [diff] [blame] | 1818 | tio->orig = rq; |
| 1819 | tio->error = 0; |
| 1820 | memset(&tio->info, 0, sizeof(tio->info)); |
Keith Busch | 2eb6e1e | 2014-10-17 17:46:36 -0600 | [diff] [blame] | 1821 | init_kthread_work(&tio->work, map_tio_request); |
Kiyoshi Ueda | 6facdaf | 2009-12-10 23:52:15 +0000 | [diff] [blame] | 1822 | |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 1823 | table = dm_get_live_table(md, &srcu_idx); |
| 1824 | if (!dm_table_mq_request_based(table)) { |
| 1825 | if (!clone_rq(rq, md, tio, gfp_mask)) { |
| 1826 | dm_put_live_table(md, srcu_idx); |
| 1827 | free_rq_tio(tio); |
| 1828 | return NULL; |
| 1829 | } |
Kiyoshi Ueda | 6facdaf | 2009-12-10 23:52:15 +0000 | [diff] [blame] | 1830 | } |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 1831 | dm_put_live_table(md, srcu_idx); |
Kiyoshi Ueda | 6facdaf | 2009-12-10 23:52:15 +0000 | [diff] [blame] | 1832 | |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1833 | return tio; |
Kiyoshi Ueda | 6facdaf | 2009-12-10 23:52:15 +0000 | [diff] [blame] | 1834 | } |
| 1835 | |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1836 | /* |
| 1837 | * Called with the queue lock held. |
| 1838 | */ |
| 1839 | static int dm_prep_fn(struct request_queue *q, struct request *rq) |
| 1840 | { |
| 1841 | struct mapped_device *md = q->queuedata; |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1842 | struct dm_rq_target_io *tio; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1843 | |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1844 | if (unlikely(rq->special)) { |
| 1845 | DMWARN("Already has something in rq->special."); |
| 1846 | return BLKPREP_KILL; |
| 1847 | } |
| 1848 | |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1849 | tio = prep_tio(rq, md, GFP_ATOMIC); |
| 1850 | if (!tio) |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1851 | return BLKPREP_DEFER; |
| 1852 | |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1853 | rq->special = tio; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1854 | rq->cmd_flags |= REQ_DONTPREP; |
| 1855 | |
| 1856 | return BLKPREP_OK; |
| 1857 | } |
| 1858 | |
Kiyoshi Ueda | 9eef87d | 2010-02-16 18:43:01 +0000 | [diff] [blame] | 1859 | /* |
| 1860 | * Returns: |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 1861 | * 0 : the request has been processed |
| 1862 | * DM_MAPIO_REQUEUE : the original request needs to be requeued |
| 1863 | * < 0 : the request was completed due to failure |
Kiyoshi Ueda | 9eef87d | 2010-02-16 18:43:01 +0000 | [diff] [blame] | 1864 | */ |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1865 | static int map_request(struct dm_target *ti, struct request *rq, |
Kiyoshi Ueda | 9eef87d | 2010-02-16 18:43:01 +0000 | [diff] [blame] | 1866 | struct mapped_device *md) |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1867 | { |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 1868 | int r; |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1869 | struct dm_rq_target_io *tio = rq->special; |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 1870 | struct request *clone = NULL; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1871 | |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 1872 | if (tio->clone) { |
| 1873 | clone = tio->clone; |
| 1874 | r = ti->type->map_rq(ti, clone, &tio->info); |
| 1875 | } else { |
| 1876 | r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone); |
| 1877 | if (r < 0) { |
| 1878 | /* The target wants to complete the I/O */ |
| 1879 | dm_kill_unmapped_request(rq, r); |
| 1880 | return r; |
| 1881 | } |
| 1882 | if (IS_ERR(clone)) |
| 1883 | return DM_MAPIO_REQUEUE; |
| 1884 | if (setup_clone(clone, rq, tio, GFP_KERNEL)) { |
| 1885 | /* -ENOMEM */ |
| 1886 | ti->type->release_clone_rq(clone); |
| 1887 | return DM_MAPIO_REQUEUE; |
| 1888 | } |
| 1889 | } |
| 1890 | |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1891 | switch (r) { |
| 1892 | case DM_MAPIO_SUBMITTED: |
| 1893 | /* The target has taken the I/O to submit by itself later */ |
| 1894 | break; |
| 1895 | case DM_MAPIO_REMAPPED: |
| 1896 | /* The target has remapped the I/O so dispatch it */ |
Jun'ichi Nomura | 6db4ccd | 2009-12-10 23:52:25 +0000 | [diff] [blame] | 1897 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1898 | blk_rq_pos(rq)); |
| 1899 | dm_dispatch_clone_request(clone, rq); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1900 | break; |
| 1901 | case DM_MAPIO_REQUEUE: |
| 1902 | /* The target wants to requeue the I/O */ |
| 1903 | dm_requeue_unmapped_request(clone); |
| 1904 | break; |
| 1905 | default: |
| 1906 | if (r > 0) { |
| 1907 | DMWARN("unimplemented target map return value: %d", r); |
| 1908 | BUG(); |
| 1909 | } |
| 1910 | |
| 1911 | /* The target wants to complete the I/O */ |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1912 | dm_kill_unmapped_request(rq, r); |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 1913 | return r; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1914 | } |
Kiyoshi Ueda | 9eef87d | 2010-02-16 18:43:01 +0000 | [diff] [blame] | 1915 | |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 1916 | return 0; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1917 | } |
| 1918 | |
Keith Busch | 2eb6e1e | 2014-10-17 17:46:36 -0600 | [diff] [blame] | 1919 | static void map_tio_request(struct kthread_work *work) |
Mike Snitzer | ba1cbad | 2012-09-26 23:45:42 +0100 | [diff] [blame] | 1920 | { |
Keith Busch | 2eb6e1e | 2014-10-17 17:46:36 -0600 | [diff] [blame] | 1921 | struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work); |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 1922 | struct request *rq = tio->orig; |
| 1923 | struct mapped_device *md = tio->md; |
Mike Snitzer | ba1cbad | 2012-09-26 23:45:42 +0100 | [diff] [blame] | 1924 | |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 1925 | if (map_request(tio->ti, rq, md) == DM_MAPIO_REQUEUE) |
| 1926 | dm_requeue_unmapped_original_request(md, rq); |
Keith Busch | 2eb6e1e | 2014-10-17 17:46:36 -0600 | [diff] [blame] | 1927 | } |
| 1928 | |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1929 | static void dm_start_request(struct mapped_device *md, struct request *orig) |
Mike Snitzer | ba1cbad | 2012-09-26 23:45:42 +0100 | [diff] [blame] | 1930 | { |
Mike Snitzer | ba1cbad | 2012-09-26 23:45:42 +0100 | [diff] [blame] | 1931 | blk_start_request(orig); |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1932 | atomic_inc(&md->pending[rq_data_dir(orig)]); |
Mike Snitzer | ba1cbad | 2012-09-26 23:45:42 +0100 | [diff] [blame] | 1933 | |
| 1934 | /* |
| 1935 | * Hold the md reference here for the in-flight I/O. |
| 1936 | * We can't rely on the reference count by device opener, |
| 1937 | * because the device may be closed during the request completion |
| 1938 | * when all bios are completed. |
| 1939 | * See the comment in rq_completed() too. |
| 1940 | */ |
| 1941 | dm_get(md); |
Mike Snitzer | ba1cbad | 2012-09-26 23:45:42 +0100 | [diff] [blame] | 1942 | } |
| 1943 | |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1944 | /* |
| 1945 | * q->request_fn for request-based dm. |
| 1946 | * Called with the queue lock held. |
| 1947 | */ |
| 1948 | static void dm_request_fn(struct request_queue *q) |
| 1949 | { |
| 1950 | struct mapped_device *md = q->queuedata; |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 1951 | int srcu_idx; |
| 1952 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1953 | struct dm_target *ti; |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1954 | struct request *rq; |
Keith Busch | 2eb6e1e | 2014-10-17 17:46:36 -0600 | [diff] [blame] | 1955 | struct dm_rq_target_io *tio; |
Tejun Heo | 29e4013 | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 1956 | sector_t pos; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1957 | |
| 1958 | /* |
Kiyoshi Ueda | b4324fe | 2009-12-10 23:52:16 +0000 | [diff] [blame] | 1959 | * For suspend, check blk_queue_stopped() and increment |
| 1960 | * ->pending within a single queue_lock not to increment the |
| 1961 | * number of in-flight I/Os after the queue is stopped in |
| 1962 | * dm_suspend(). |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1963 | */ |
Jens Axboe | 7eaceac | 2011-03-10 08:52:07 +0100 | [diff] [blame] | 1964 | while (!blk_queue_stopped(q)) { |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1965 | rq = blk_peek_request(q); |
| 1966 | if (!rq) |
Jens Axboe | 7eaceac | 2011-03-10 08:52:07 +0100 | [diff] [blame] | 1967 | goto delay_and_out; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1968 | |
Tejun Heo | 29e4013 | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 1969 | /* always use block 0 to find the target for flushes for now */ |
| 1970 | pos = 0; |
| 1971 | if (!(rq->cmd_flags & REQ_FLUSH)) |
| 1972 | pos = blk_rq_pos(rq); |
Kiyoshi Ueda | d0bcb87 | 2009-12-10 23:52:18 +0000 | [diff] [blame] | 1973 | |
Tejun Heo | 29e4013 | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 1974 | ti = dm_table_find_target(map, pos); |
Mike Snitzer | ba1cbad | 2012-09-26 23:45:42 +0100 | [diff] [blame] | 1975 | if (!dm_target_is_valid(ti)) { |
| 1976 | /* |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1977 | * Must perform setup, that rq_completed() requires, |
Mike Snitzer | ba1cbad | 2012-09-26 23:45:42 +0100 | [diff] [blame] | 1978 | * before calling dm_kill_unmapped_request |
| 1979 | */ |
| 1980 | DMERR_LIMIT("request attempted access beyond the end of device"); |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1981 | dm_start_request(md, rq); |
| 1982 | dm_kill_unmapped_request(rq, -EIO); |
Mike Snitzer | ba1cbad | 2012-09-26 23:45:42 +0100 | [diff] [blame] | 1983 | continue; |
| 1984 | } |
Tejun Heo | 29e4013 | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 1985 | |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1986 | if (ti->type->busy && ti->type->busy(ti)) |
Jens Axboe | 7eaceac | 2011-03-10 08:52:07 +0100 | [diff] [blame] | 1987 | goto delay_and_out; |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1988 | |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 1989 | dm_start_request(md, rq); |
Kiyoshi Ueda | b4324fe | 2009-12-10 23:52:16 +0000 | [diff] [blame] | 1990 | |
Keith Busch | 2eb6e1e | 2014-10-17 17:46:36 -0600 | [diff] [blame] | 1991 | tio = rq->special; |
| 1992 | /* Establish tio->ti before queuing work (map_tio_request) */ |
| 1993 | tio->ti = ti; |
| 1994 | queue_kthread_work(&md->kworker, &tio->work); |
Kiyoshi Ueda | 052189a | 2011-01-13 20:00:00 +0000 | [diff] [blame] | 1995 | BUG_ON(!irqs_disabled()); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 1996 | } |
| 1997 | |
| 1998 | goto out; |
| 1999 | |
Jens Axboe | 7eaceac | 2011-03-10 08:52:07 +0100 | [diff] [blame] | 2000 | delay_and_out: |
| 2001 | blk_delay_queue(q, HZ / 10); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 2002 | out: |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 2003 | dm_put_live_table(md, srcu_idx); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 2004 | } |
| 2005 | |
| 2006 | int dm_underlying_device_busy(struct request_queue *q) |
| 2007 | { |
| 2008 | return blk_lld_busy(q); |
| 2009 | } |
| 2010 | EXPORT_SYMBOL_GPL(dm_underlying_device_busy); |
| 2011 | |
| 2012 | static int dm_lld_busy(struct request_queue *q) |
| 2013 | { |
| 2014 | int r; |
| 2015 | struct mapped_device *md = q->queuedata; |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 2016 | struct dm_table *map = dm_get_live_table_fast(md); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 2017 | |
| 2018 | if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) |
| 2019 | r = 1; |
| 2020 | else |
| 2021 | r = dm_table_any_busy_target(map); |
| 2022 | |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 2023 | dm_put_live_table_fast(md); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 2024 | |
| 2025 | return r; |
| 2026 | } |
| 2027 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2028 | static int dm_any_congested(void *congested_data, int bdi_bits) |
| 2029 | { |
Chandra Seetharaman | 8a57dfc | 2008-11-13 23:39:14 +0000 | [diff] [blame] | 2030 | int r = bdi_bits; |
| 2031 | struct mapped_device *md = congested_data; |
| 2032 | struct dm_table *map; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2033 | |
Alasdair G Kergon | 1eb787e | 2009-04-09 00:27:14 +0100 | [diff] [blame] | 2034 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 2035 | map = dm_get_live_table_fast(md); |
Chandra Seetharaman | 8a57dfc | 2008-11-13 23:39:14 +0000 | [diff] [blame] | 2036 | if (map) { |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 2037 | /* |
| 2038 | * Request-based dm cares about only own queue for |
| 2039 | * the query about congestion status of request_queue |
| 2040 | */ |
| 2041 | if (dm_request_based(md)) |
| 2042 | r = md->queue->backing_dev_info.state & |
| 2043 | bdi_bits; |
| 2044 | else |
| 2045 | r = dm_table_any_congested(map, bdi_bits); |
Chandra Seetharaman | 8a57dfc | 2008-11-13 23:39:14 +0000 | [diff] [blame] | 2046 | } |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 2047 | dm_put_live_table_fast(md); |
Chandra Seetharaman | 8a57dfc | 2008-11-13 23:39:14 +0000 | [diff] [blame] | 2048 | } |
| 2049 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2050 | return r; |
| 2051 | } |
| 2052 | |
| 2053 | /*----------------------------------------------------------------- |
| 2054 | * An IDR is used to keep track of allocated minor numbers. |
| 2055 | *---------------------------------------------------------------*/ |
Alasdair G Kergon | 2b06cff | 2006-06-26 00:27:32 -0700 | [diff] [blame] | 2056 | static void free_minor(int minor) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2057 | { |
Jeff Mahoney | f32c10b | 2006-06-26 00:27:22 -0700 | [diff] [blame] | 2058 | spin_lock(&_minor_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2059 | idr_remove(&_minor_idr, minor); |
Jeff Mahoney | f32c10b | 2006-06-26 00:27:22 -0700 | [diff] [blame] | 2060 | spin_unlock(&_minor_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2061 | } |
| 2062 | |
| 2063 | /* |
| 2064 | * See if the device with a specific minor # is free. |
| 2065 | */ |
Frederik Deweerdt | cf13ab8 | 2008-04-24 22:10:59 +0100 | [diff] [blame] | 2066 | static int specific_minor(int minor) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2067 | { |
Tejun Heo | c9d76be | 2013-02-27 17:04:26 -0800 | [diff] [blame] | 2068 | int r; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2069 | |
| 2070 | if (minor >= (1 << MINORBITS)) |
| 2071 | return -EINVAL; |
| 2072 | |
Tejun Heo | c9d76be | 2013-02-27 17:04:26 -0800 | [diff] [blame] | 2073 | idr_preload(GFP_KERNEL); |
Jeff Mahoney | f32c10b | 2006-06-26 00:27:22 -0700 | [diff] [blame] | 2074 | spin_lock(&_minor_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2075 | |
Tejun Heo | c9d76be | 2013-02-27 17:04:26 -0800 | [diff] [blame] | 2076 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2077 | |
Jeff Mahoney | f32c10b | 2006-06-26 00:27:22 -0700 | [diff] [blame] | 2078 | spin_unlock(&_minor_lock); |
Tejun Heo | c9d76be | 2013-02-27 17:04:26 -0800 | [diff] [blame] | 2079 | idr_preload_end(); |
| 2080 | if (r < 0) |
| 2081 | return r == -ENOSPC ? -EBUSY : r; |
| 2082 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2083 | } |
| 2084 | |
Frederik Deweerdt | cf13ab8 | 2008-04-24 22:10:59 +0100 | [diff] [blame] | 2085 | static int next_free_minor(int *minor) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2086 | { |
Tejun Heo | c9d76be | 2013-02-27 17:04:26 -0800 | [diff] [blame] | 2087 | int r; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2088 | |
Tejun Heo | c9d76be | 2013-02-27 17:04:26 -0800 | [diff] [blame] | 2089 | idr_preload(GFP_KERNEL); |
Jeff Mahoney | f32c10b | 2006-06-26 00:27:22 -0700 | [diff] [blame] | 2090 | spin_lock(&_minor_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2091 | |
Tejun Heo | c9d76be | 2013-02-27 17:04:26 -0800 | [diff] [blame] | 2092 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2093 | |
Jeff Mahoney | f32c10b | 2006-06-26 00:27:22 -0700 | [diff] [blame] | 2094 | spin_unlock(&_minor_lock); |
Tejun Heo | c9d76be | 2013-02-27 17:04:26 -0800 | [diff] [blame] | 2095 | idr_preload_end(); |
| 2096 | if (r < 0) |
| 2097 | return r; |
| 2098 | *minor = r; |
| 2099 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2100 | } |
| 2101 | |
Alexey Dobriyan | 83d5cde | 2009-09-21 17:01:13 -0700 | [diff] [blame] | 2102 | static const struct block_device_operations dm_blk_dops; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2103 | |
Mikulas Patocka | 53d5914 | 2009-04-02 19:55:37 +0100 | [diff] [blame] | 2104 | static void dm_wq_work(struct work_struct *work); |
| 2105 | |
Mike Snitzer | 4a0b4dd | 2010-08-12 04:14:02 +0100 | [diff] [blame] | 2106 | static void dm_init_md_queue(struct mapped_device *md) |
| 2107 | { |
| 2108 | /* |
| 2109 | * Request-based dm devices cannot be stacked on top of bio-based dm |
| 2110 | * devices. The type of this dm device has not been decided yet. |
| 2111 | * The type is decided at the first table loading time. |
| 2112 | * To prevent problematic device stacking, clear the queue flag |
| 2113 | * for request stacking support until then. |
| 2114 | * |
| 2115 | * This queue is new, so no concurrency on the queue_flags. |
| 2116 | */ |
| 2117 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); |
| 2118 | |
| 2119 | md->queue->queuedata = md; |
| 2120 | md->queue->backing_dev_info.congested_fn = dm_any_congested; |
| 2121 | md->queue->backing_dev_info.congested_data = md; |
| 2122 | blk_queue_make_request(md->queue, dm_request); |
| 2123 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
Mike Snitzer | 4a0b4dd | 2010-08-12 04:14:02 +0100 | [diff] [blame] | 2124 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); |
| 2125 | } |
| 2126 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2127 | /* |
| 2128 | * Allocate and initialise a blank device with a given minor. |
| 2129 | */ |
Alasdair G Kergon | 2b06cff | 2006-06-26 00:27:32 -0700 | [diff] [blame] | 2130 | static struct mapped_device *alloc_dev(int minor) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2131 | { |
| 2132 | int r; |
Frederik Deweerdt | cf13ab8 | 2008-04-24 22:10:59 +0100 | [diff] [blame] | 2133 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
Jeff Mahoney | ba61fdd | 2006-06-26 00:27:21 -0700 | [diff] [blame] | 2134 | void *old_md; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2135 | |
| 2136 | if (!md) { |
| 2137 | DMWARN("unable to allocate device, out of memory."); |
| 2138 | return NULL; |
| 2139 | } |
| 2140 | |
Jeff Mahoney | 10da4f7 | 2006-06-26 00:27:25 -0700 | [diff] [blame] | 2141 | if (!try_module_get(THIS_MODULE)) |
Milan Broz | 6ed7ade | 2008-02-08 02:10:19 +0000 | [diff] [blame] | 2142 | goto bad_module_get; |
Jeff Mahoney | 10da4f7 | 2006-06-26 00:27:25 -0700 | [diff] [blame] | 2143 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2144 | /* get a minor number for the dev */ |
Alasdair G Kergon | 2b06cff | 2006-06-26 00:27:32 -0700 | [diff] [blame] | 2145 | if (minor == DM_ANY_MINOR) |
Frederik Deweerdt | cf13ab8 | 2008-04-24 22:10:59 +0100 | [diff] [blame] | 2146 | r = next_free_minor(&minor); |
Alasdair G Kergon | 2b06cff | 2006-06-26 00:27:32 -0700 | [diff] [blame] | 2147 | else |
Frederik Deweerdt | cf13ab8 | 2008-04-24 22:10:59 +0100 | [diff] [blame] | 2148 | r = specific_minor(minor); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2149 | if (r < 0) |
Milan Broz | 6ed7ade | 2008-02-08 02:10:19 +0000 | [diff] [blame] | 2150 | goto bad_minor; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2151 | |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 2152 | r = init_srcu_struct(&md->io_barrier); |
| 2153 | if (r < 0) |
| 2154 | goto bad_io_barrier; |
| 2155 | |
Mike Snitzer | a5664da | 2010-08-12 04:14:01 +0100 | [diff] [blame] | 2156 | md->type = DM_TYPE_NONE; |
Daniel Walker | e61290a | 2008-02-08 02:10:08 +0000 | [diff] [blame] | 2157 | mutex_init(&md->suspend_lock); |
Mike Snitzer | a5664da | 2010-08-12 04:14:01 +0100 | [diff] [blame] | 2158 | mutex_init(&md->type_lock); |
Benjamin Marzinski | 86f1152 | 2014-08-13 13:53:43 -0500 | [diff] [blame] | 2159 | mutex_init(&md->table_devices_lock); |
Mikulas Patocka | 022c261 | 2009-04-02 19:55:39 +0100 | [diff] [blame] | 2160 | spin_lock_init(&md->deferred_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2161 | atomic_set(&md->holders, 1); |
Alasdair G Kergon | 5c6bd75 | 2006-06-26 00:27:34 -0700 | [diff] [blame] | 2162 | atomic_set(&md->open_count, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2163 | atomic_set(&md->event_nr, 0); |
Mike Anderson | 7a8c3d3 | 2007-10-19 22:48:01 +0100 | [diff] [blame] | 2164 | atomic_set(&md->uevent_seq, 0); |
| 2165 | INIT_LIST_HEAD(&md->uevent_list); |
Benjamin Marzinski | 86f1152 | 2014-08-13 13:53:43 -0500 | [diff] [blame] | 2166 | INIT_LIST_HEAD(&md->table_devices); |
Mike Anderson | 7a8c3d3 | 2007-10-19 22:48:01 +0100 | [diff] [blame] | 2167 | spin_lock_init(&md->uevent_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2168 | |
Mike Snitzer | 4a0b4dd | 2010-08-12 04:14:02 +0100 | [diff] [blame] | 2169 | md->queue = blk_alloc_queue(GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2170 | if (!md->queue) |
Milan Broz | 6ed7ade | 2008-02-08 02:10:19 +0000 | [diff] [blame] | 2171 | goto bad_queue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2172 | |
Mike Snitzer | 4a0b4dd | 2010-08-12 04:14:02 +0100 | [diff] [blame] | 2173 | dm_init_md_queue(md); |
Stefan Bader | 9faf400 | 2006-10-03 01:15:41 -0700 | [diff] [blame] | 2174 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2175 | md->disk = alloc_disk(1); |
| 2176 | if (!md->disk) |
Milan Broz | 6ed7ade | 2008-02-08 02:10:19 +0000 | [diff] [blame] | 2177 | goto bad_disk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2178 | |
Nikanth Karthikesan | 316d315 | 2009-10-06 20:16:55 +0200 | [diff] [blame] | 2179 | atomic_set(&md->pending[0], 0); |
| 2180 | atomic_set(&md->pending[1], 0); |
Jeff Mahoney | f0b0411 | 2006-06-26 00:27:25 -0700 | [diff] [blame] | 2181 | init_waitqueue_head(&md->wait); |
Mikulas Patocka | 53d5914 | 2009-04-02 19:55:37 +0100 | [diff] [blame] | 2182 | INIT_WORK(&md->work, dm_wq_work); |
Jeff Mahoney | f0b0411 | 2006-06-26 00:27:25 -0700 | [diff] [blame] | 2183 | init_waitqueue_head(&md->eventq); |
Mikulas Patocka | 2995fa7 | 2014-01-13 19:37:54 -0500 | [diff] [blame] | 2184 | init_completion(&md->kobj_holder.completion); |
Keith Busch | 2eb6e1e | 2014-10-17 17:46:36 -0600 | [diff] [blame] | 2185 | md->kworker_task = NULL; |
Jeff Mahoney | f0b0411 | 2006-06-26 00:27:25 -0700 | [diff] [blame] | 2186 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2187 | md->disk->major = _major; |
| 2188 | md->disk->first_minor = minor; |
| 2189 | md->disk->fops = &dm_blk_dops; |
| 2190 | md->disk->queue = md->queue; |
| 2191 | md->disk->private_data = md; |
| 2192 | sprintf(md->disk->disk_name, "dm-%d", minor); |
| 2193 | add_disk(md->disk); |
Mike Anderson | 7e51f25 | 2006-03-27 01:17:52 -0800 | [diff] [blame] | 2194 | format_dev_t(md->name, MKDEV(_major, minor)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2195 | |
Tejun Heo | 670368a | 2013-07-30 08:40:21 -0400 | [diff] [blame] | 2196 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
Milan Broz | 304f3f6 | 2008-02-08 02:11:17 +0000 | [diff] [blame] | 2197 | if (!md->wq) |
| 2198 | goto bad_thread; |
| 2199 | |
Mikulas Patocka | 32a926d | 2009-06-22 10:12:17 +0100 | [diff] [blame] | 2200 | md->bdev = bdget_disk(md->disk, 0); |
| 2201 | if (!md->bdev) |
| 2202 | goto bad_bdev; |
| 2203 | |
Tejun Heo | 6a8736d | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 2204 | bio_init(&md->flush_bio); |
| 2205 | md->flush_bio.bi_bdev = md->bdev; |
| 2206 | md->flush_bio.bi_rw = WRITE_FLUSH; |
| 2207 | |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 2208 | dm_stats_init(&md->stats); |
| 2209 | |
Jeff Mahoney | ba61fdd | 2006-06-26 00:27:21 -0700 | [diff] [blame] | 2210 | /* Populate the mapping, nobody knows we exist yet */ |
Jeff Mahoney | f32c10b | 2006-06-26 00:27:22 -0700 | [diff] [blame] | 2211 | spin_lock(&_minor_lock); |
Jeff Mahoney | ba61fdd | 2006-06-26 00:27:21 -0700 | [diff] [blame] | 2212 | old_md = idr_replace(&_minor_idr, md, minor); |
Jeff Mahoney | f32c10b | 2006-06-26 00:27:22 -0700 | [diff] [blame] | 2213 | spin_unlock(&_minor_lock); |
Jeff Mahoney | ba61fdd | 2006-06-26 00:27:21 -0700 | [diff] [blame] | 2214 | |
| 2215 | BUG_ON(old_md != MINOR_ALLOCED); |
| 2216 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2217 | return md; |
| 2218 | |
Mikulas Patocka | 32a926d | 2009-06-22 10:12:17 +0100 | [diff] [blame] | 2219 | bad_bdev: |
| 2220 | destroy_workqueue(md->wq); |
Milan Broz | 304f3f6 | 2008-02-08 02:11:17 +0000 | [diff] [blame] | 2221 | bad_thread: |
Zdenek Kabelac | 03022c5 | 2009-10-16 23:18:15 +0100 | [diff] [blame] | 2222 | del_gendisk(md->disk); |
Milan Broz | 304f3f6 | 2008-02-08 02:11:17 +0000 | [diff] [blame] | 2223 | put_disk(md->disk); |
Milan Broz | 6ed7ade | 2008-02-08 02:10:19 +0000 | [diff] [blame] | 2224 | bad_disk: |
Al Viro | 1312f40 | 2006-03-12 11:02:03 -0500 | [diff] [blame] | 2225 | blk_cleanup_queue(md->queue); |
Milan Broz | 6ed7ade | 2008-02-08 02:10:19 +0000 | [diff] [blame] | 2226 | bad_queue: |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 2227 | cleanup_srcu_struct(&md->io_barrier); |
| 2228 | bad_io_barrier: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2229 | free_minor(minor); |
Milan Broz | 6ed7ade | 2008-02-08 02:10:19 +0000 | [diff] [blame] | 2230 | bad_minor: |
Jeff Mahoney | 10da4f7 | 2006-06-26 00:27:25 -0700 | [diff] [blame] | 2231 | module_put(THIS_MODULE); |
Milan Broz | 6ed7ade | 2008-02-08 02:10:19 +0000 | [diff] [blame] | 2232 | bad_module_get: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2233 | kfree(md); |
| 2234 | return NULL; |
| 2235 | } |
| 2236 | |
Jun'ichi Nomura | ae9da83 | 2007-10-19 22:38:43 +0100 | [diff] [blame] | 2237 | static void unlock_fs(struct mapped_device *md); |
| 2238 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2239 | static void free_dev(struct mapped_device *md) |
| 2240 | { |
Tejun Heo | f331c02 | 2008-09-03 09:01:48 +0200 | [diff] [blame] | 2241 | int minor = MINOR(disk_devt(md->disk)); |
Jun'ichi Nomura | 63d94e4 | 2006-02-24 13:04:25 -0800 | [diff] [blame] | 2242 | |
Mikulas Patocka | 32a926d | 2009-06-22 10:12:17 +0100 | [diff] [blame] | 2243 | unlock_fs(md); |
| 2244 | bdput(md->bdev); |
Milan Broz | 304f3f6 | 2008-02-08 02:11:17 +0000 | [diff] [blame] | 2245 | destroy_workqueue(md->wq); |
Keith Busch | 2eb6e1e | 2014-10-17 17:46:36 -0600 | [diff] [blame] | 2246 | |
| 2247 | if (md->kworker_task) |
| 2248 | kthread_stop(md->kworker_task); |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 2249 | if (md->io_pool) |
| 2250 | mempool_destroy(md->io_pool); |
Mike Snitzer | 1ae49ea | 2014-12-05 17:11:05 -0500 | [diff] [blame] | 2251 | if (md->rq_pool) |
| 2252 | mempool_destroy(md->rq_pool); |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 2253 | if (md->bs) |
| 2254 | bioset_free(md->bs); |
Martin K. Petersen | 9c47008 | 2009-04-09 00:27:12 +0100 | [diff] [blame] | 2255 | blk_integrity_unregister(md->disk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2256 | del_gendisk(md->disk); |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 2257 | cleanup_srcu_struct(&md->io_barrier); |
Benjamin Marzinski | 86f1152 | 2014-08-13 13:53:43 -0500 | [diff] [blame] | 2258 | free_table_devices(&md->table_devices); |
Jun'ichi Nomura | 63d94e4 | 2006-02-24 13:04:25 -0800 | [diff] [blame] | 2259 | free_minor(minor); |
Jeff Mahoney | fba9f90 | 2006-06-26 00:27:23 -0700 | [diff] [blame] | 2260 | |
| 2261 | spin_lock(&_minor_lock); |
| 2262 | md->disk->private_data = NULL; |
| 2263 | spin_unlock(&_minor_lock); |
| 2264 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2265 | put_disk(md->disk); |
Al Viro | 1312f40 | 2006-03-12 11:02:03 -0500 | [diff] [blame] | 2266 | blk_cleanup_queue(md->queue); |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 2267 | dm_stats_cleanup(&md->stats); |
Jeff Mahoney | 10da4f7 | 2006-06-26 00:27:25 -0700 | [diff] [blame] | 2268 | module_put(THIS_MODULE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2269 | kfree(md); |
| 2270 | } |
| 2271 | |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 2272 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
| 2273 | { |
Mikulas Patocka | c0820cf | 2012-12-21 20:23:38 +0000 | [diff] [blame] | 2274 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 2275 | |
Jun'ichi Nomura | 5f01520 | 2013-03-01 22:45:48 +0000 | [diff] [blame] | 2276 | if (md->io_pool && md->bs) { |
Jun'ichi Nomura | 16245bd | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2277 | /* The md already has necessary mempools. */ |
| 2278 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { |
| 2279 | /* |
| 2280 | * Reload bioset because front_pad may have changed |
| 2281 | * because a different table was loaded. |
| 2282 | */ |
| 2283 | bioset_free(md->bs); |
| 2284 | md->bs = p->bs; |
| 2285 | p->bs = NULL; |
Jun'ichi Nomura | 16245bd | 2013-03-01 22:45:44 +0000 | [diff] [blame] | 2286 | } |
Keith Busch | 466d89a | 2014-10-17 17:46:37 -0600 | [diff] [blame] | 2287 | /* |
| 2288 | * There's no need to reload with request-based dm |
| 2289 | * because the size of front_pad doesn't change. |
| 2290 | * Note for future: If you are to reload bioset, |
| 2291 | * prep-ed requests in the queue may refer |
| 2292 | * to bio from the old bioset, so you must walk |
| 2293 | * through the queue to unprep. |
| 2294 | */ |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 2295 | goto out; |
Mikulas Patocka | c0820cf | 2012-12-21 20:23:38 +0000 | [diff] [blame] | 2296 | } |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 2297 | |
Mike Snitzer | 1ae49ea | 2014-12-05 17:11:05 -0500 | [diff] [blame] | 2298 | BUG_ON(!p || md->io_pool || md->rq_pool || md->bs); |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 2299 | |
| 2300 | md->io_pool = p->io_pool; |
| 2301 | p->io_pool = NULL; |
Mike Snitzer | 1ae49ea | 2014-12-05 17:11:05 -0500 | [diff] [blame] | 2302 | md->rq_pool = p->rq_pool; |
| 2303 | p->rq_pool = NULL; |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 2304 | md->bs = p->bs; |
| 2305 | p->bs = NULL; |
| 2306 | |
| 2307 | out: |
| 2308 | /* mempool bind completed, now no need any mempools in the table */ |
| 2309 | dm_table_free_md_mempools(t); |
| 2310 | } |
| 2311 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2312 | /* |
| 2313 | * Bind a table to the device. |
| 2314 | */ |
| 2315 | static void event_callback(void *context) |
| 2316 | { |
Mike Anderson | 7a8c3d3 | 2007-10-19 22:48:01 +0100 | [diff] [blame] | 2317 | unsigned long flags; |
| 2318 | LIST_HEAD(uevents); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2319 | struct mapped_device *md = (struct mapped_device *) context; |
| 2320 | |
Mike Anderson | 7a8c3d3 | 2007-10-19 22:48:01 +0100 | [diff] [blame] | 2321 | spin_lock_irqsave(&md->uevent_lock, flags); |
| 2322 | list_splice_init(&md->uevent_list, &uevents); |
| 2323 | spin_unlock_irqrestore(&md->uevent_lock, flags); |
| 2324 | |
Tejun Heo | ed9e198 | 2008-08-25 19:56:05 +0900 | [diff] [blame] | 2325 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
Mike Anderson | 7a8c3d3 | 2007-10-19 22:48:01 +0100 | [diff] [blame] | 2326 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2327 | atomic_inc(&md->event_nr); |
| 2328 | wake_up(&md->eventq); |
| 2329 | } |
| 2330 | |
Mike Snitzer | c217649 | 2011-01-13 19:53:46 +0000 | [diff] [blame] | 2331 | /* |
| 2332 | * Protected by md->suspend_lock obtained by dm_swap_table(). |
| 2333 | */ |
Alasdair G Kergon | 4e90188be | 2005-07-28 21:15:59 -0700 | [diff] [blame] | 2334 | static void __set_size(struct mapped_device *md, sector_t size) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2335 | { |
Alasdair G Kergon | 4e90188be | 2005-07-28 21:15:59 -0700 | [diff] [blame] | 2336 | set_capacity(md->disk, size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2337 | |
Mikulas Patocka | db8fef4 | 2009-06-22 10:12:15 +0100 | [diff] [blame] | 2338 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2339 | } |
| 2340 | |
Alasdair G Kergon | 042d2a9 | 2009-12-10 23:52:24 +0000 | [diff] [blame] | 2341 | /* |
Mikulas Patocka | d5b9dd0 | 2011-08-02 12:32:04 +0100 | [diff] [blame] | 2342 | * Return 1 if the queue has a compulsory merge_bvec_fn function. |
| 2343 | * |
| 2344 | * If this function returns 0, then the device is either a non-dm |
| 2345 | * device without a merge_bvec_fn, or it is a dm device that is |
| 2346 | * able to split any bios it receives that are too big. |
| 2347 | */ |
| 2348 | int dm_queue_merge_is_compulsory(struct request_queue *q) |
| 2349 | { |
| 2350 | struct mapped_device *dev_md; |
| 2351 | |
| 2352 | if (!q->merge_bvec_fn) |
| 2353 | return 0; |
| 2354 | |
| 2355 | if (q->make_request_fn == dm_request) { |
| 2356 | dev_md = q->queuedata; |
| 2357 | if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags)) |
| 2358 | return 0; |
| 2359 | } |
| 2360 | |
| 2361 | return 1; |
| 2362 | } |
| 2363 | |
| 2364 | static int dm_device_merge_is_compulsory(struct dm_target *ti, |
| 2365 | struct dm_dev *dev, sector_t start, |
| 2366 | sector_t len, void *data) |
| 2367 | { |
| 2368 | struct block_device *bdev = dev->bdev; |
| 2369 | struct request_queue *q = bdev_get_queue(bdev); |
| 2370 | |
| 2371 | return dm_queue_merge_is_compulsory(q); |
| 2372 | } |
| 2373 | |
| 2374 | /* |
| 2375 | * Return 1 if it is acceptable to ignore merge_bvec_fn based |
| 2376 | * on the properties of the underlying devices. |
| 2377 | */ |
| 2378 | static int dm_table_merge_is_optional(struct dm_table *table) |
| 2379 | { |
| 2380 | unsigned i = 0; |
| 2381 | struct dm_target *ti; |
| 2382 | |
| 2383 | while (i < dm_table_get_num_targets(table)) { |
| 2384 | ti = dm_table_get_target(table, i++); |
| 2385 | |
| 2386 | if (ti->type->iterate_devices && |
| 2387 | ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL)) |
| 2388 | return 0; |
| 2389 | } |
| 2390 | |
| 2391 | return 1; |
| 2392 | } |
| 2393 | |
| 2394 | /* |
Alasdair G Kergon | 042d2a9 | 2009-12-10 23:52:24 +0000 | [diff] [blame] | 2395 | * Returns old map, which caller must destroy. |
| 2396 | */ |
| 2397 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, |
| 2398 | struct queue_limits *limits) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2399 | { |
Alasdair G Kergon | 042d2a9 | 2009-12-10 23:52:24 +0000 | [diff] [blame] | 2400 | struct dm_table *old_map; |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 2401 | struct request_queue *q = md->queue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2402 | sector_t size; |
Mikulas Patocka | d5b9dd0 | 2011-08-02 12:32:04 +0100 | [diff] [blame] | 2403 | int merge_is_optional; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2404 | |
| 2405 | size = dm_table_get_size(t); |
Darrick J. Wong | 3ac51e7 | 2006-03-27 01:17:54 -0800 | [diff] [blame] | 2406 | |
| 2407 | /* |
| 2408 | * Wipe any geometry if the size of the table changed. |
| 2409 | */ |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 2410 | if (size != dm_get_size(md)) |
Darrick J. Wong | 3ac51e7 | 2006-03-27 01:17:54 -0800 | [diff] [blame] | 2411 | memset(&md->geometry, 0, sizeof(md->geometry)); |
| 2412 | |
Mikulas Patocka | 32a926d | 2009-06-22 10:12:17 +0100 | [diff] [blame] | 2413 | __set_size(md, size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2414 | |
Alasdair G Kergon | cf222b3 | 2005-07-28 21:15:57 -0700 | [diff] [blame] | 2415 | dm_table_event_callback(t, event_callback, md); |
Alasdair G Kergon | 2ca3310 | 2005-07-28 21:16:00 -0700 | [diff] [blame] | 2416 | |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 2417 | /* |
| 2418 | * The queue hasn't been stopped yet, if the old table type wasn't |
| 2419 | * for request-based during suspension. So stop it to prevent |
| 2420 | * I/O mapping before resume. |
| 2421 | * This must be done before setting the queue restrictions, |
| 2422 | * because request-based dm may be run just after the setting. |
| 2423 | */ |
| 2424 | if (dm_table_request_based(t) && !blk_queue_stopped(q)) |
| 2425 | stop_queue(q); |
| 2426 | |
| 2427 | __bind_mempools(md, t); |
| 2428 | |
Mikulas Patocka | d5b9dd0 | 2011-08-02 12:32:04 +0100 | [diff] [blame] | 2429 | merge_is_optional = dm_table_merge_is_optional(t); |
| 2430 | |
Eric Dumazet | a12f5d4 | 2014-11-23 09:34:29 -0800 | [diff] [blame] | 2431 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 2432 | rcu_assign_pointer(md->map, t); |
Alasdair G Kergon | 36a0456 | 2011-10-31 20:19:04 +0000 | [diff] [blame] | 2433 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
| 2434 | |
Mike Snitzer | 754c5fc | 2009-06-22 10:12:34 +0100 | [diff] [blame] | 2435 | dm_table_set_restrictions(t, q, limits); |
Mikulas Patocka | d5b9dd0 | 2011-08-02 12:32:04 +0100 | [diff] [blame] | 2436 | if (merge_is_optional) |
| 2437 | set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); |
| 2438 | else |
| 2439 | clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); |
Hannes Reinecke | 41abc4e | 2014-11-05 14:35:50 +0100 | [diff] [blame] | 2440 | if (old_map) |
| 2441 | dm_sync_table(md); |
Alasdair G Kergon | 2ca3310 | 2005-07-28 21:16:00 -0700 | [diff] [blame] | 2442 | |
Alasdair G Kergon | 042d2a9 | 2009-12-10 23:52:24 +0000 | [diff] [blame] | 2443 | return old_map; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2444 | } |
| 2445 | |
Alasdair G Kergon | a794015 | 2009-12-10 23:52:23 +0000 | [diff] [blame] | 2446 | /* |
| 2447 | * Returns unbound table for the caller to free. |
| 2448 | */ |
| 2449 | static struct dm_table *__unbind(struct mapped_device *md) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2450 | { |
Eric Dumazet | a12f5d4 | 2014-11-23 09:34:29 -0800 | [diff] [blame] | 2451 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2452 | |
| 2453 | if (!map) |
Alasdair G Kergon | a794015 | 2009-12-10 23:52:23 +0000 | [diff] [blame] | 2454 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2455 | |
| 2456 | dm_table_event_callback(map, NULL, NULL); |
Monam Agarwal | 9cdb852 | 2014-03-23 23:58:27 +0530 | [diff] [blame] | 2457 | RCU_INIT_POINTER(md->map, NULL); |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 2458 | dm_sync_table(md); |
Alasdair G Kergon | a794015 | 2009-12-10 23:52:23 +0000 | [diff] [blame] | 2459 | |
| 2460 | return map; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2461 | } |
| 2462 | |
| 2463 | /* |
| 2464 | * Constructor for a new device. |
| 2465 | */ |
Alasdair G Kergon | 2b06cff | 2006-06-26 00:27:32 -0700 | [diff] [blame] | 2466 | int dm_create(int minor, struct mapped_device **result) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2467 | { |
| 2468 | struct mapped_device *md; |
| 2469 | |
Alasdair G Kergon | 2b06cff | 2006-06-26 00:27:32 -0700 | [diff] [blame] | 2470 | md = alloc_dev(minor); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2471 | if (!md) |
| 2472 | return -ENXIO; |
| 2473 | |
Milan Broz | 784aae7 | 2009-01-06 03:05:12 +0000 | [diff] [blame] | 2474 | dm_sysfs_init(md); |
| 2475 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2476 | *result = md; |
| 2477 | return 0; |
| 2478 | } |
| 2479 | |
Mike Snitzer | a5664da | 2010-08-12 04:14:01 +0100 | [diff] [blame] | 2480 | /* |
| 2481 | * Functions to manage md->type. |
| 2482 | * All are required to hold md->type_lock. |
| 2483 | */ |
| 2484 | void dm_lock_md_type(struct mapped_device *md) |
| 2485 | { |
| 2486 | mutex_lock(&md->type_lock); |
| 2487 | } |
| 2488 | |
| 2489 | void dm_unlock_md_type(struct mapped_device *md) |
| 2490 | { |
| 2491 | mutex_unlock(&md->type_lock); |
| 2492 | } |
| 2493 | |
| 2494 | void dm_set_md_type(struct mapped_device *md, unsigned type) |
| 2495 | { |
Mike Snitzer | 00c4fc3 | 2013-08-27 18:57:03 -0400 | [diff] [blame] | 2496 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
Mike Snitzer | a5664da | 2010-08-12 04:14:01 +0100 | [diff] [blame] | 2497 | md->type = type; |
| 2498 | } |
| 2499 | |
| 2500 | unsigned dm_get_md_type(struct mapped_device *md) |
| 2501 | { |
Mike Snitzer | 00c4fc3 | 2013-08-27 18:57:03 -0400 | [diff] [blame] | 2502 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
Mike Snitzer | a5664da | 2010-08-12 04:14:01 +0100 | [diff] [blame] | 2503 | return md->type; |
| 2504 | } |
| 2505 | |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 2506 | static bool dm_md_type_request_based(struct mapped_device *md) |
| 2507 | { |
| 2508 | unsigned table_type = dm_get_md_type(md); |
| 2509 | |
| 2510 | return (table_type == DM_TYPE_REQUEST_BASED || |
| 2511 | table_type == DM_TYPE_MQ_REQUEST_BASED); |
| 2512 | } |
| 2513 | |
Alasdair G Kergon | 36a0456 | 2011-10-31 20:19:04 +0000 | [diff] [blame] | 2514 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
| 2515 | { |
| 2516 | return md->immutable_target_type; |
| 2517 | } |
| 2518 | |
Mike Snitzer | 4a0b4dd | 2010-08-12 04:14:02 +0100 | [diff] [blame] | 2519 | /* |
Mike Snitzer | f84cb8a | 2013-09-19 12:13:58 -0400 | [diff] [blame] | 2520 | * The queue_limits are only valid as long as you have a reference |
| 2521 | * count on 'md'. |
| 2522 | */ |
| 2523 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) |
| 2524 | { |
| 2525 | BUG_ON(!atomic_read(&md->holders)); |
| 2526 | return &md->queue->limits; |
| 2527 | } |
| 2528 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); |
| 2529 | |
| 2530 | /* |
Mike Snitzer | 4a0b4dd | 2010-08-12 04:14:02 +0100 | [diff] [blame] | 2531 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). |
| 2532 | */ |
| 2533 | static int dm_init_request_based_queue(struct mapped_device *md) |
| 2534 | { |
| 2535 | struct request_queue *q = NULL; |
| 2536 | |
| 2537 | if (md->queue->elevator) |
| 2538 | return 1; |
| 2539 | |
| 2540 | /* Fully initialize the queue */ |
| 2541 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); |
| 2542 | if (!q) |
| 2543 | return 0; |
| 2544 | |
| 2545 | md->queue = q; |
Mike Snitzer | 4a0b4dd | 2010-08-12 04:14:02 +0100 | [diff] [blame] | 2546 | dm_init_md_queue(md); |
| 2547 | blk_queue_softirq_done(md->queue, dm_softirq_done); |
| 2548 | blk_queue_prep_rq(md->queue, dm_prep_fn); |
| 2549 | blk_queue_lld_busy(md->queue, dm_lld_busy); |
Mike Snitzer | 4a0b4dd | 2010-08-12 04:14:02 +0100 | [diff] [blame] | 2550 | |
Keith Busch | 2eb6e1e | 2014-10-17 17:46:36 -0600 | [diff] [blame] | 2551 | /* Also initialize the request-based DM worker thread */ |
| 2552 | init_kthread_worker(&md->kworker); |
| 2553 | md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker, |
| 2554 | "kdmwork-%s", dm_device_name(md)); |
| 2555 | |
Mike Snitzer | 4a0b4dd | 2010-08-12 04:14:02 +0100 | [diff] [blame] | 2556 | elv_register_queue(md->queue); |
| 2557 | |
| 2558 | return 1; |
| 2559 | } |
| 2560 | |
| 2561 | /* |
| 2562 | * Setup the DM device's queue based on md's type |
| 2563 | */ |
| 2564 | int dm_setup_md_queue(struct mapped_device *md) |
| 2565 | { |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 2566 | if (dm_md_type_request_based(md) && !dm_init_request_based_queue(md)) { |
Mike Snitzer | 4a0b4dd | 2010-08-12 04:14:02 +0100 | [diff] [blame] | 2567 | DMWARN("Cannot initialize queue for request-based mapped device"); |
| 2568 | return -EINVAL; |
| 2569 | } |
| 2570 | |
| 2571 | return 0; |
| 2572 | } |
| 2573 | |
Mikulas Patocka | 2bec1f4 | 2015-02-17 14:30:53 -0500 | [diff] [blame] | 2574 | struct mapped_device *dm_get_md(dev_t dev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2575 | { |
| 2576 | struct mapped_device *md; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2577 | unsigned minor = MINOR(dev); |
| 2578 | |
| 2579 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) |
| 2580 | return NULL; |
| 2581 | |
Jeff Mahoney | f32c10b | 2006-06-26 00:27:22 -0700 | [diff] [blame] | 2582 | spin_lock(&_minor_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2583 | |
| 2584 | md = idr_find(&_minor_idr, minor); |
Mikulas Patocka | 2bec1f4 | 2015-02-17 14:30:53 -0500 | [diff] [blame] | 2585 | if (md) { |
| 2586 | if ((md == MINOR_ALLOCED || |
| 2587 | (MINOR(disk_devt(dm_disk(md))) != minor) || |
| 2588 | dm_deleting_md(md) || |
| 2589 | test_bit(DMF_FREEING, &md->flags))) { |
| 2590 | md = NULL; |
| 2591 | goto out; |
| 2592 | } |
| 2593 | dm_get(md); |
Jeff Mahoney | fba9f90 | 2006-06-26 00:27:23 -0700 | [diff] [blame] | 2594 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2595 | |
Jeff Mahoney | fba9f90 | 2006-06-26 00:27:23 -0700 | [diff] [blame] | 2596 | out: |
Jeff Mahoney | f32c10b | 2006-06-26 00:27:22 -0700 | [diff] [blame] | 2597 | spin_unlock(&_minor_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2598 | |
David Teigland | 637842c | 2006-01-06 00:20:00 -0800 | [diff] [blame] | 2599 | return md; |
| 2600 | } |
Alasdair G Kergon | 3cf2e4b | 2011-10-31 20:19:06 +0000 | [diff] [blame] | 2601 | EXPORT_SYMBOL_GPL(dm_get_md); |
David Teigland | d229a95 | 2006-01-06 00:20:01 -0800 | [diff] [blame] | 2602 | |
Alasdair G Kergon | 9ade92a | 2006-03-27 01:17:53 -0800 | [diff] [blame] | 2603 | void *dm_get_mdptr(struct mapped_device *md) |
David Teigland | 637842c | 2006-01-06 00:20:00 -0800 | [diff] [blame] | 2604 | { |
Alasdair G Kergon | 9ade92a | 2006-03-27 01:17:53 -0800 | [diff] [blame] | 2605 | return md->interface_ptr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2606 | } |
| 2607 | |
| 2608 | void dm_set_mdptr(struct mapped_device *md, void *ptr) |
| 2609 | { |
| 2610 | md->interface_ptr = ptr; |
| 2611 | } |
| 2612 | |
| 2613 | void dm_get(struct mapped_device *md) |
| 2614 | { |
| 2615 | atomic_inc(&md->holders); |
Kiyoshi Ueda | 3f77316d | 2010-08-12 04:13:56 +0100 | [diff] [blame] | 2616 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2617 | } |
| 2618 | |
Alasdair G Kergon | 72d9486 | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 2619 | const char *dm_device_name(struct mapped_device *md) |
| 2620 | { |
| 2621 | return md->name; |
| 2622 | } |
| 2623 | EXPORT_SYMBOL_GPL(dm_device_name); |
| 2624 | |
Kiyoshi Ueda | 3f77316d | 2010-08-12 04:13:56 +0100 | [diff] [blame] | 2625 | static void __dm_destroy(struct mapped_device *md, bool wait) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2626 | { |
Mike Anderson | 1134e5a | 2006-03-27 01:17:54 -0800 | [diff] [blame] | 2627 | struct dm_table *map; |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 2628 | int srcu_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2629 | |
Kiyoshi Ueda | 3f77316d | 2010-08-12 04:13:56 +0100 | [diff] [blame] | 2630 | might_sleep(); |
Jeff Mahoney | fba9f90 | 2006-06-26 00:27:23 -0700 | [diff] [blame] | 2631 | |
Kiyoshi Ueda | 3f77316d | 2010-08-12 04:13:56 +0100 | [diff] [blame] | 2632 | spin_lock(&_minor_lock); |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 2633 | map = dm_get_live_table(md, &srcu_idx); |
Kiyoshi Ueda | 3f77316d | 2010-08-12 04:13:56 +0100 | [diff] [blame] | 2634 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
| 2635 | set_bit(DMF_FREEING, &md->flags); |
| 2636 | spin_unlock(&_minor_lock); |
| 2637 | |
Keith Busch | 2eb6e1e | 2014-10-17 17:46:36 -0600 | [diff] [blame] | 2638 | if (dm_request_based(md)) |
| 2639 | flush_kthread_worker(&md->kworker); |
| 2640 | |
Kiyoshi Ueda | 3f77316d | 2010-08-12 04:13:56 +0100 | [diff] [blame] | 2641 | if (!dm_suspended_md(md)) { |
| 2642 | dm_table_presuspend_targets(map); |
| 2643 | dm_table_postsuspend_targets(map); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2644 | } |
Kiyoshi Ueda | 3f77316d | 2010-08-12 04:13:56 +0100 | [diff] [blame] | 2645 | |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 2646 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
| 2647 | dm_put_live_table(md, srcu_idx); |
| 2648 | |
Kiyoshi Ueda | 3f77316d | 2010-08-12 04:13:56 +0100 | [diff] [blame] | 2649 | /* |
| 2650 | * Rare, but there may be I/O requests still going to complete, |
| 2651 | * for example. Wait for all references to disappear. |
| 2652 | * No one should increment the reference count of the mapped_device, |
| 2653 | * after the mapped_device state becomes DMF_FREEING. |
| 2654 | */ |
| 2655 | if (wait) |
| 2656 | while (atomic_read(&md->holders)) |
| 2657 | msleep(1); |
| 2658 | else if (atomic_read(&md->holders)) |
| 2659 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", |
| 2660 | dm_device_name(md), atomic_read(&md->holders)); |
| 2661 | |
| 2662 | dm_sysfs_exit(md); |
Kiyoshi Ueda | 3f77316d | 2010-08-12 04:13:56 +0100 | [diff] [blame] | 2663 | dm_table_destroy(__unbind(md)); |
| 2664 | free_dev(md); |
| 2665 | } |
| 2666 | |
| 2667 | void dm_destroy(struct mapped_device *md) |
| 2668 | { |
| 2669 | __dm_destroy(md, true); |
| 2670 | } |
| 2671 | |
| 2672 | void dm_destroy_immediate(struct mapped_device *md) |
| 2673 | { |
| 2674 | __dm_destroy(md, false); |
| 2675 | } |
| 2676 | |
| 2677 | void dm_put(struct mapped_device *md) |
| 2678 | { |
| 2679 | atomic_dec(&md->holders); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2680 | } |
Edward Goggin | 79eb885 | 2007-05-09 02:32:56 -0700 | [diff] [blame] | 2681 | EXPORT_SYMBOL_GPL(dm_put); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2682 | |
Mikulas Patocka | 401600d | 2009-04-02 19:55:38 +0100 | [diff] [blame] | 2683 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
Milan Broz | 46125c1 | 2008-02-08 02:10:30 +0000 | [diff] [blame] | 2684 | { |
| 2685 | int r = 0; |
Mikulas Patocka | b44ebeb | 2009-04-02 19:55:39 +0100 | [diff] [blame] | 2686 | DECLARE_WAITQUEUE(wait, current); |
| 2687 | |
Mikulas Patocka | b44ebeb | 2009-04-02 19:55:39 +0100 | [diff] [blame] | 2688 | add_wait_queue(&md->wait, &wait); |
Milan Broz | 46125c1 | 2008-02-08 02:10:30 +0000 | [diff] [blame] | 2689 | |
| 2690 | while (1) { |
Mikulas Patocka | 401600d | 2009-04-02 19:55:38 +0100 | [diff] [blame] | 2691 | set_current_state(interruptible); |
Milan Broz | 46125c1 | 2008-02-08 02:10:30 +0000 | [diff] [blame] | 2692 | |
Kiyoshi Ueda | b4324fe | 2009-12-10 23:52:16 +0000 | [diff] [blame] | 2693 | if (!md_in_flight(md)) |
Milan Broz | 46125c1 | 2008-02-08 02:10:30 +0000 | [diff] [blame] | 2694 | break; |
| 2695 | |
Mikulas Patocka | 401600d | 2009-04-02 19:55:38 +0100 | [diff] [blame] | 2696 | if (interruptible == TASK_INTERRUPTIBLE && |
| 2697 | signal_pending(current)) { |
Milan Broz | 46125c1 | 2008-02-08 02:10:30 +0000 | [diff] [blame] | 2698 | r = -EINTR; |
| 2699 | break; |
| 2700 | } |
| 2701 | |
| 2702 | io_schedule(); |
| 2703 | } |
| 2704 | set_current_state(TASK_RUNNING); |
| 2705 | |
Mikulas Patocka | b44ebeb | 2009-04-02 19:55:39 +0100 | [diff] [blame] | 2706 | remove_wait_queue(&md->wait, &wait); |
| 2707 | |
Milan Broz | 46125c1 | 2008-02-08 02:10:30 +0000 | [diff] [blame] | 2708 | return r; |
| 2709 | } |
| 2710 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2711 | /* |
| 2712 | * Process the deferred bios |
| 2713 | */ |
Mikulas Patocka | ef20858 | 2009-04-02 19:55:38 +0100 | [diff] [blame] | 2714 | static void dm_wq_work(struct work_struct *work) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2715 | { |
Mikulas Patocka | ef20858 | 2009-04-02 19:55:38 +0100 | [diff] [blame] | 2716 | struct mapped_device *md = container_of(work, struct mapped_device, |
| 2717 | work); |
Milan Broz | 6d6f10d | 2008-02-08 02:10:22 +0000 | [diff] [blame] | 2718 | struct bio *c; |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 2719 | int srcu_idx; |
| 2720 | struct dm_table *map; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2721 | |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 2722 | map = dm_get_live_table(md, &srcu_idx); |
Mikulas Patocka | ef20858 | 2009-04-02 19:55:38 +0100 | [diff] [blame] | 2723 | |
Mikulas Patocka | 3b00b20 | 2009-04-09 00:27:15 +0100 | [diff] [blame] | 2724 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
Alasdair G Kergon | df12ee9 | 2009-04-09 00:27:13 +0100 | [diff] [blame] | 2725 | spin_lock_irq(&md->deferred_lock); |
| 2726 | c = bio_list_pop(&md->deferred); |
| 2727 | spin_unlock_irq(&md->deferred_lock); |
Mikulas Patocka | 022c261 | 2009-04-02 19:55:39 +0100 | [diff] [blame] | 2728 | |
Tejun Heo | 6a8736d | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 2729 | if (!c) |
Alasdair G Kergon | df12ee9 | 2009-04-09 00:27:13 +0100 | [diff] [blame] | 2730 | break; |
Alasdair G Kergon | df12ee9 | 2009-04-09 00:27:13 +0100 | [diff] [blame] | 2731 | |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 2732 | if (dm_request_based(md)) |
| 2733 | generic_make_request(c); |
Tejun Heo | 6a8736d | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 2734 | else |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 2735 | __split_and_process_bio(md, map, c); |
Mikulas Patocka | 022c261 | 2009-04-02 19:55:39 +0100 | [diff] [blame] | 2736 | } |
Milan Broz | 73d410c | 2008-02-08 02:10:25 +0000 | [diff] [blame] | 2737 | |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 2738 | dm_put_live_table(md, srcu_idx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2739 | } |
| 2740 | |
Mikulas Patocka | 9a1fb46 | 2009-04-02 19:55:36 +0100 | [diff] [blame] | 2741 | static void dm_queue_flush(struct mapped_device *md) |
Milan Broz | 304f3f6 | 2008-02-08 02:11:17 +0000 | [diff] [blame] | 2742 | { |
Mikulas Patocka | 3b00b20 | 2009-04-09 00:27:15 +0100 | [diff] [blame] | 2743 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
Peter Zijlstra | 4e857c5 | 2014-03-17 18:06:10 +0100 | [diff] [blame] | 2744 | smp_mb__after_atomic(); |
Mikulas Patocka | 53d5914 | 2009-04-02 19:55:37 +0100 | [diff] [blame] | 2745 | queue_work(md->wq, &md->work); |
Milan Broz | 304f3f6 | 2008-02-08 02:11:17 +0000 | [diff] [blame] | 2746 | } |
| 2747 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2748 | /* |
Alasdair G Kergon | 042d2a9 | 2009-12-10 23:52:24 +0000 | [diff] [blame] | 2749 | * Swap in a new table, returning the old one for the caller to destroy. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2750 | */ |
Alasdair G Kergon | 042d2a9 | 2009-12-10 23:52:24 +0000 | [diff] [blame] | 2751 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2752 | { |
Mike Christie | 87eb5b2 | 2013-03-01 22:45:48 +0000 | [diff] [blame] | 2753 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
Mike Snitzer | 754c5fc | 2009-06-22 10:12:34 +0100 | [diff] [blame] | 2754 | struct queue_limits limits; |
Alasdair G Kergon | 042d2a9 | 2009-12-10 23:52:24 +0000 | [diff] [blame] | 2755 | int r; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2756 | |
Daniel Walker | e61290a | 2008-02-08 02:10:08 +0000 | [diff] [blame] | 2757 | mutex_lock(&md->suspend_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2758 | |
| 2759 | /* device must be suspended */ |
Kiyoshi Ueda | 4f186f8 | 2009-12-10 23:52:26 +0000 | [diff] [blame] | 2760 | if (!dm_suspended_md(md)) |
Alasdair G Kergon | 93c534a | 2005-07-12 15:53:05 -0700 | [diff] [blame] | 2761 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2762 | |
Mike Snitzer | 3ae7065 | 2012-09-26 23:45:45 +0100 | [diff] [blame] | 2763 | /* |
| 2764 | * If the new table has no data devices, retain the existing limits. |
| 2765 | * This helps multipath with queue_if_no_path if all paths disappear, |
| 2766 | * then new I/O is queued based on these limits, and then some paths |
| 2767 | * reappear. |
| 2768 | */ |
| 2769 | if (dm_table_has_no_data_devices(table)) { |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 2770 | live_map = dm_get_live_table_fast(md); |
Mike Snitzer | 3ae7065 | 2012-09-26 23:45:45 +0100 | [diff] [blame] | 2771 | if (live_map) |
| 2772 | limits = md->queue->limits; |
Mikulas Patocka | 83d5e5b | 2013-07-10 23:41:18 +0100 | [diff] [blame] | 2773 | dm_put_live_table_fast(md); |
Mike Snitzer | 3ae7065 | 2012-09-26 23:45:45 +0100 | [diff] [blame] | 2774 | } |
| 2775 | |
Mike Christie | 87eb5b2 | 2013-03-01 22:45:48 +0000 | [diff] [blame] | 2776 | if (!live_map) { |
| 2777 | r = dm_calculate_queue_limits(table, &limits); |
| 2778 | if (r) { |
| 2779 | map = ERR_PTR(r); |
| 2780 | goto out; |
| 2781 | } |
Alasdair G Kergon | 042d2a9 | 2009-12-10 23:52:24 +0000 | [diff] [blame] | 2782 | } |
Mike Snitzer | 754c5fc | 2009-06-22 10:12:34 +0100 | [diff] [blame] | 2783 | |
Alasdair G Kergon | 042d2a9 | 2009-12-10 23:52:24 +0000 | [diff] [blame] | 2784 | map = __bind(md, table, &limits); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2785 | |
Alasdair G Kergon | 93c534a | 2005-07-12 15:53:05 -0700 | [diff] [blame] | 2786 | out: |
Daniel Walker | e61290a | 2008-02-08 02:10:08 +0000 | [diff] [blame] | 2787 | mutex_unlock(&md->suspend_lock); |
Alasdair G Kergon | 042d2a9 | 2009-12-10 23:52:24 +0000 | [diff] [blame] | 2788 | return map; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2789 | } |
| 2790 | |
| 2791 | /* |
| 2792 | * Functions to lock and unlock any filesystem running on the |
| 2793 | * device. |
| 2794 | */ |
Alasdair G Kergon | 2ca3310 | 2005-07-28 21:16:00 -0700 | [diff] [blame] | 2795 | static int lock_fs(struct mapped_device *md) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2796 | { |
Alasdair G Kergon | e39e2e9 | 2006-01-06 00:20:05 -0800 | [diff] [blame] | 2797 | int r; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2798 | |
| 2799 | WARN_ON(md->frozen_sb); |
Alasdair G Kergon | dfbe03f | 2005-05-05 16:16:04 -0700 | [diff] [blame] | 2800 | |
Mikulas Patocka | db8fef4 | 2009-06-22 10:12:15 +0100 | [diff] [blame] | 2801 | md->frozen_sb = freeze_bdev(md->bdev); |
Alasdair G Kergon | dfbe03f | 2005-05-05 16:16:04 -0700 | [diff] [blame] | 2802 | if (IS_ERR(md->frozen_sb)) { |
Alasdair G Kergon | cf222b3 | 2005-07-28 21:15:57 -0700 | [diff] [blame] | 2803 | r = PTR_ERR(md->frozen_sb); |
Alasdair G Kergon | e39e2e9 | 2006-01-06 00:20:05 -0800 | [diff] [blame] | 2804 | md->frozen_sb = NULL; |
| 2805 | return r; |
Alasdair G Kergon | dfbe03f | 2005-05-05 16:16:04 -0700 | [diff] [blame] | 2806 | } |
| 2807 | |
Alasdair G Kergon | aa8d7c2 | 2006-01-06 00:20:06 -0800 | [diff] [blame] | 2808 | set_bit(DMF_FROZEN, &md->flags); |
| 2809 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2810 | return 0; |
| 2811 | } |
| 2812 | |
Alasdair G Kergon | 2ca3310 | 2005-07-28 21:16:00 -0700 | [diff] [blame] | 2813 | static void unlock_fs(struct mapped_device *md) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2814 | { |
Alasdair G Kergon | aa8d7c2 | 2006-01-06 00:20:06 -0800 | [diff] [blame] | 2815 | if (!test_bit(DMF_FROZEN, &md->flags)) |
| 2816 | return; |
| 2817 | |
Mikulas Patocka | db8fef4 | 2009-06-22 10:12:15 +0100 | [diff] [blame] | 2818 | thaw_bdev(md->bdev, md->frozen_sb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2819 | md->frozen_sb = NULL; |
Alasdair G Kergon | aa8d7c2 | 2006-01-06 00:20:06 -0800 | [diff] [blame] | 2820 | clear_bit(DMF_FROZEN, &md->flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2821 | } |
| 2822 | |
| 2823 | /* |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 2824 | * If __dm_suspend returns 0, the device is completely quiescent |
| 2825 | * now. There is no request-processing activity. All new requests |
| 2826 | * are being added to md->deferred list. |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 2827 | * |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 2828 | * Caller must hold md->suspend_lock |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 2829 | */ |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 2830 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
| 2831 | unsigned suspend_flags, int interruptible) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2832 | { |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 2833 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
| 2834 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; |
| 2835 | int r; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2836 | |
Kiyoshi Ueda | 2e93ccc | 2006-12-08 02:41:09 -0800 | [diff] [blame] | 2837 | /* |
| 2838 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. |
| 2839 | * This flag is cleared before dm_suspend returns. |
| 2840 | */ |
| 2841 | if (noflush) |
| 2842 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
| 2843 | |
Mike Snitzer | d67ee21 | 2014-10-28 20:13:31 -0400 | [diff] [blame] | 2844 | /* |
| 2845 | * This gets reverted if there's an error later and the targets |
| 2846 | * provide the .presuspend_undo hook. |
| 2847 | */ |
Alasdair G Kergon | cf222b3 | 2005-07-28 21:15:57 -0700 | [diff] [blame] | 2848 | dm_table_presuspend_targets(map); |
| 2849 | |
Mikulas Patocka | 32a926d | 2009-06-22 10:12:17 +0100 | [diff] [blame] | 2850 | /* |
Kiyoshi Ueda | 9f518b2 | 2009-12-10 23:52:16 +0000 | [diff] [blame] | 2851 | * Flush I/O to the device. |
| 2852 | * Any I/O submitted after lock_fs() may not be flushed. |
| 2853 | * noflush takes precedence over do_lockfs. |
| 2854 | * (lock_fs() flushes I/Os and waits for them to complete.) |
Mikulas Patocka | 32a926d | 2009-06-22 10:12:17 +0100 | [diff] [blame] | 2855 | */ |
| 2856 | if (!noflush && do_lockfs) { |
| 2857 | r = lock_fs(md); |
Mike Snitzer | d67ee21 | 2014-10-28 20:13:31 -0400 | [diff] [blame] | 2858 | if (r) { |
| 2859 | dm_table_presuspend_undo_targets(map); |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 2860 | return r; |
Mike Snitzer | d67ee21 | 2014-10-28 20:13:31 -0400 | [diff] [blame] | 2861 | } |
Alasdair G Kergon | aa8d7c2 | 2006-01-06 00:20:06 -0800 | [diff] [blame] | 2862 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2863 | |
| 2864 | /* |
Mikulas Patocka | 3b00b20 | 2009-04-09 00:27:15 +0100 | [diff] [blame] | 2865 | * Here we must make sure that no processes are submitting requests |
| 2866 | * to target drivers i.e. no one may be executing |
| 2867 | * __split_and_process_bio. This is called from dm_request and |
| 2868 | * dm_wq_work. |
| 2869 | * |
| 2870 | * To get all processes out of __split_and_process_bio in dm_request, |
| 2871 | * we take the write lock. To prevent any process from reentering |
Tejun Heo | 6a8736d | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 2872 | * __split_and_process_bio from dm_request and quiesce the thread |
| 2873 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call |
| 2874 | * flush_workqueue(md->wq). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2875 | */ |
Alasdair G Kergon | 1eb787e | 2009-04-09 00:27:14 +0100 | [diff] [blame] | 2876 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
Hannes Reinecke | 41abc4e | 2014-11-05 14:35:50 +0100 | [diff] [blame] | 2877 | if (map) |
| 2878 | synchronize_srcu(&md->io_barrier); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2879 | |
Kiyoshi Ueda | d0bcb87 | 2009-12-10 23:52:18 +0000 | [diff] [blame] | 2880 | /* |
Tejun Heo | 29e4013 | 2010-09-08 18:07:00 +0200 | [diff] [blame] | 2881 | * Stop md->queue before flushing md->wq in case request-based |
| 2882 | * dm defers requests to md->wq from md->queue. |
Kiyoshi Ueda | d0bcb87 | 2009-12-10 23:52:18 +0000 | [diff] [blame] | 2883 | */ |
Keith Busch | 2eb6e1e | 2014-10-17 17:46:36 -0600 | [diff] [blame] | 2884 | if (dm_request_based(md)) { |
Kiyoshi Ueda | 9f518b2 | 2009-12-10 23:52:16 +0000 | [diff] [blame] | 2885 | stop_queue(md->queue); |
Keith Busch | 2eb6e1e | 2014-10-17 17:46:36 -0600 | [diff] [blame] | 2886 | flush_kthread_worker(&md->kworker); |
| 2887 | } |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 2888 | |
Kiyoshi Ueda | d0bcb87 | 2009-12-10 23:52:18 +0000 | [diff] [blame] | 2889 | flush_workqueue(md->wq); |
| 2890 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2891 | /* |
Mikulas Patocka | 3b00b20 | 2009-04-09 00:27:15 +0100 | [diff] [blame] | 2892 | * At this point no more requests are entering target request routines. |
| 2893 | * We call dm_wait_for_completion to wait for all existing requests |
| 2894 | * to finish. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2895 | */ |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 2896 | r = dm_wait_for_completion(md, interruptible); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2897 | |
Milan Broz | 6d6f10d | 2008-02-08 02:10:22 +0000 | [diff] [blame] | 2898 | if (noflush) |
Mikulas Patocka | 022c261 | 2009-04-02 19:55:39 +0100 | [diff] [blame] | 2899 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
Hannes Reinecke | 41abc4e | 2014-11-05 14:35:50 +0100 | [diff] [blame] | 2900 | if (map) |
| 2901 | synchronize_srcu(&md->io_barrier); |
Kiyoshi Ueda | 2e93ccc | 2006-12-08 02:41:09 -0800 | [diff] [blame] | 2902 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2903 | /* were we interrupted ? */ |
Milan Broz | 46125c1 | 2008-02-08 02:10:30 +0000 | [diff] [blame] | 2904 | if (r < 0) { |
Mikulas Patocka | 9a1fb46 | 2009-04-02 19:55:36 +0100 | [diff] [blame] | 2905 | dm_queue_flush(md); |
Milan Broz | 73d410c | 2008-02-08 02:10:25 +0000 | [diff] [blame] | 2906 | |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 2907 | if (dm_request_based(md)) |
Kiyoshi Ueda | 9f518b2 | 2009-12-10 23:52:16 +0000 | [diff] [blame] | 2908 | start_queue(md->queue); |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 2909 | |
Alasdair G Kergon | 2ca3310 | 2005-07-28 21:16:00 -0700 | [diff] [blame] | 2910 | unlock_fs(md); |
Mike Snitzer | d67ee21 | 2014-10-28 20:13:31 -0400 | [diff] [blame] | 2911 | dm_table_presuspend_undo_targets(map); |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 2912 | /* pushback list is already flushed, so skip flush */ |
Alasdair G Kergon | 2ca3310 | 2005-07-28 21:16:00 -0700 | [diff] [blame] | 2913 | } |
Alasdair G Kergon | 2ca3310 | 2005-07-28 21:16:00 -0700 | [diff] [blame] | 2914 | |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 2915 | return r; |
| 2916 | } |
| 2917 | |
| 2918 | /* |
| 2919 | * We need to be able to change a mapping table under a mounted |
| 2920 | * filesystem. For example we might want to move some data in |
| 2921 | * the background. Before the table can be swapped with |
| 2922 | * dm_bind_table, dm_suspend must be called to flush any in |
| 2923 | * flight bios and ensure that any further io gets deferred. |
| 2924 | */ |
| 2925 | /* |
| 2926 | * Suspend mechanism in request-based dm. |
| 2927 | * |
| 2928 | * 1. Flush all I/Os by lock_fs() if needed. |
| 2929 | * 2. Stop dispatching any I/O by stopping the request_queue. |
| 2930 | * 3. Wait for all in-flight I/Os to be completed or requeued. |
| 2931 | * |
| 2932 | * To abort suspend, start the request_queue. |
| 2933 | */ |
| 2934 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
| 2935 | { |
| 2936 | struct dm_table *map = NULL; |
| 2937 | int r = 0; |
| 2938 | |
| 2939 | retry: |
| 2940 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); |
| 2941 | |
| 2942 | if (dm_suspended_md(md)) { |
| 2943 | r = -EINVAL; |
| 2944 | goto out_unlock; |
| 2945 | } |
| 2946 | |
| 2947 | if (dm_suspended_internally_md(md)) { |
| 2948 | /* already internally suspended, wait for internal resume */ |
| 2949 | mutex_unlock(&md->suspend_lock); |
| 2950 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); |
| 2951 | if (r) |
| 2952 | return r; |
| 2953 | goto retry; |
| 2954 | } |
| 2955 | |
Eric Dumazet | a12f5d4 | 2014-11-23 09:34:29 -0800 | [diff] [blame] | 2956 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 2957 | |
| 2958 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE); |
| 2959 | if (r) |
| 2960 | goto out_unlock; |
Mikulas Patocka | 3b00b20 | 2009-04-09 00:27:15 +0100 | [diff] [blame] | 2961 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2962 | set_bit(DMF_SUSPENDED, &md->flags); |
| 2963 | |
Kiyoshi Ueda | 4d4471c | 2009-12-10 23:52:26 +0000 | [diff] [blame] | 2964 | dm_table_postsuspend_targets(map); |
| 2965 | |
Alasdair G Kergon | d287483 | 2006-11-08 17:44:43 -0800 | [diff] [blame] | 2966 | out_unlock: |
Daniel Walker | e61290a | 2008-02-08 02:10:08 +0000 | [diff] [blame] | 2967 | mutex_unlock(&md->suspend_lock); |
Alasdair G Kergon | cf222b3 | 2005-07-28 21:15:57 -0700 | [diff] [blame] | 2968 | return r; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2969 | } |
| 2970 | |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 2971 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2972 | { |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 2973 | if (map) { |
| 2974 | int r = dm_table_resume_targets(map); |
| 2975 | if (r) |
| 2976 | return r; |
| 2977 | } |
Alasdair G Kergon | 2ca3310 | 2005-07-28 21:16:00 -0700 | [diff] [blame] | 2978 | |
Mikulas Patocka | 9a1fb46 | 2009-04-02 19:55:36 +0100 | [diff] [blame] | 2979 | dm_queue_flush(md); |
Alasdair G Kergon | 2ca3310 | 2005-07-28 21:16:00 -0700 | [diff] [blame] | 2980 | |
Kiyoshi Ueda | cec47e3 | 2009-06-22 10:12:35 +0100 | [diff] [blame] | 2981 | /* |
| 2982 | * Flushing deferred I/Os must be done after targets are resumed |
| 2983 | * so that mapping of targets can work correctly. |
| 2984 | * Request-based dm is queueing the deferred I/Os in its request_queue. |
| 2985 | */ |
| 2986 | if (dm_request_based(md)) |
| 2987 | start_queue(md->queue); |
| 2988 | |
Alasdair G Kergon | 2ca3310 | 2005-07-28 21:16:00 -0700 | [diff] [blame] | 2989 | unlock_fs(md); |
| 2990 | |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 2991 | return 0; |
| 2992 | } |
| 2993 | |
| 2994 | int dm_resume(struct mapped_device *md) |
| 2995 | { |
| 2996 | int r = -EINVAL; |
| 2997 | struct dm_table *map = NULL; |
| 2998 | |
| 2999 | retry: |
| 3000 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); |
| 3001 | |
| 3002 | if (!dm_suspended_md(md)) |
| 3003 | goto out; |
| 3004 | |
| 3005 | if (dm_suspended_internally_md(md)) { |
| 3006 | /* already internally suspended, wait for internal resume */ |
| 3007 | mutex_unlock(&md->suspend_lock); |
| 3008 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); |
| 3009 | if (r) |
| 3010 | return r; |
| 3011 | goto retry; |
| 3012 | } |
| 3013 | |
Eric Dumazet | a12f5d4 | 2014-11-23 09:34:29 -0800 | [diff] [blame] | 3014 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 3015 | if (!map || !dm_table_get_size(map)) |
| 3016 | goto out; |
| 3017 | |
| 3018 | r = __dm_resume(md, map); |
| 3019 | if (r) |
| 3020 | goto out; |
| 3021 | |
Alasdair G Kergon | 2ca3310 | 2005-07-28 21:16:00 -0700 | [diff] [blame] | 3022 | clear_bit(DMF_SUSPENDED, &md->flags); |
| 3023 | |
Alasdair G Kergon | cf222b3 | 2005-07-28 21:15:57 -0700 | [diff] [blame] | 3024 | r = 0; |
| 3025 | out: |
Daniel Walker | e61290a | 2008-02-08 02:10:08 +0000 | [diff] [blame] | 3026 | mutex_unlock(&md->suspend_lock); |
Alasdair G Kergon | 2ca3310 | 2005-07-28 21:16:00 -0700 | [diff] [blame] | 3027 | |
Alasdair G Kergon | cf222b3 | 2005-07-28 21:15:57 -0700 | [diff] [blame] | 3028 | return r; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3029 | } |
| 3030 | |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 3031 | /* |
| 3032 | * Internal suspend/resume works like userspace-driven suspend. It waits |
| 3033 | * until all bios finish and prevents issuing new bios to the target drivers. |
| 3034 | * It may be used only from the kernel. |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 3035 | */ |
| 3036 | |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 3037 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
| 3038 | { |
| 3039 | struct dm_table *map = NULL; |
| 3040 | |
Mikulas Patocka | 96b26c8 | 2015-01-08 18:52:26 -0500 | [diff] [blame] | 3041 | if (md->internal_suspend_count++) |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 3042 | return; /* nested internal suspend */ |
| 3043 | |
| 3044 | if (dm_suspended_md(md)) { |
| 3045 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); |
| 3046 | return; /* nest suspend */ |
| 3047 | } |
| 3048 | |
Eric Dumazet | a12f5d4 | 2014-11-23 09:34:29 -0800 | [diff] [blame] | 3049 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 3050 | |
| 3051 | /* |
| 3052 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is |
| 3053 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend |
| 3054 | * would require changing .presuspend to return an error -- avoid this |
| 3055 | * until there is a need for more elaborate variants of internal suspend. |
| 3056 | */ |
| 3057 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE); |
| 3058 | |
| 3059 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); |
| 3060 | |
| 3061 | dm_table_postsuspend_targets(map); |
| 3062 | } |
| 3063 | |
| 3064 | static void __dm_internal_resume(struct mapped_device *md) |
| 3065 | { |
Mikulas Patocka | 96b26c8 | 2015-01-08 18:52:26 -0500 | [diff] [blame] | 3066 | BUG_ON(!md->internal_suspend_count); |
| 3067 | |
| 3068 | if (--md->internal_suspend_count) |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 3069 | return; /* resume from nested internal suspend */ |
| 3070 | |
| 3071 | if (dm_suspended_md(md)) |
| 3072 | goto done; /* resume from nested suspend */ |
| 3073 | |
| 3074 | /* |
| 3075 | * NOTE: existing callers don't need to call dm_table_resume_targets |
| 3076 | * (which may fail -- so best to avoid it for now by passing NULL map) |
| 3077 | */ |
| 3078 | (void) __dm_resume(md, NULL); |
| 3079 | |
| 3080 | done: |
| 3081 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); |
| 3082 | smp_mb__after_atomic(); |
| 3083 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); |
| 3084 | } |
| 3085 | |
| 3086 | void dm_internal_suspend_noflush(struct mapped_device *md) |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 3087 | { |
| 3088 | mutex_lock(&md->suspend_lock); |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 3089 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); |
| 3090 | mutex_unlock(&md->suspend_lock); |
| 3091 | } |
| 3092 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); |
| 3093 | |
| 3094 | void dm_internal_resume(struct mapped_device *md) |
| 3095 | { |
| 3096 | mutex_lock(&md->suspend_lock); |
| 3097 | __dm_internal_resume(md); |
| 3098 | mutex_unlock(&md->suspend_lock); |
| 3099 | } |
| 3100 | EXPORT_SYMBOL_GPL(dm_internal_resume); |
| 3101 | |
| 3102 | /* |
| 3103 | * Fast variants of internal suspend/resume hold md->suspend_lock, |
| 3104 | * which prevents interaction with userspace-driven suspend. |
| 3105 | */ |
| 3106 | |
| 3107 | void dm_internal_suspend_fast(struct mapped_device *md) |
| 3108 | { |
| 3109 | mutex_lock(&md->suspend_lock); |
| 3110 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 3111 | return; |
| 3112 | |
| 3113 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
| 3114 | synchronize_srcu(&md->io_barrier); |
| 3115 | flush_workqueue(md->wq); |
| 3116 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); |
| 3117 | } |
| 3118 | |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 3119 | void dm_internal_resume_fast(struct mapped_device *md) |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 3120 | { |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 3121 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
Mikulas Patocka | fd2ed4d | 2013-08-16 10:54:23 -0400 | [diff] [blame] | 3122 | goto done; |
| 3123 | |
| 3124 | dm_queue_flush(md); |
| 3125 | |
| 3126 | done: |
| 3127 | mutex_unlock(&md->suspend_lock); |
| 3128 | } |
| 3129 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3130 | /*----------------------------------------------------------------- |
| 3131 | * Event notification. |
| 3132 | *---------------------------------------------------------------*/ |
Peter Rajnoha | 3abf85b | 2010-03-06 02:32:31 +0000 | [diff] [blame] | 3133 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
Milan Broz | 60935eb | 2009-06-22 10:12:30 +0100 | [diff] [blame] | 3134 | unsigned cookie) |
Alasdair G Kergon | 69267a3 | 2007-12-13 14:15:57 +0000 | [diff] [blame] | 3135 | { |
Milan Broz | 60935eb | 2009-06-22 10:12:30 +0100 | [diff] [blame] | 3136 | char udev_cookie[DM_COOKIE_LENGTH]; |
| 3137 | char *envp[] = { udev_cookie, NULL }; |
| 3138 | |
| 3139 | if (!cookie) |
Peter Rajnoha | 3abf85b | 2010-03-06 02:32:31 +0000 | [diff] [blame] | 3140 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
Milan Broz | 60935eb | 2009-06-22 10:12:30 +0100 | [diff] [blame] | 3141 | else { |
| 3142 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", |
| 3143 | DM_COOKIE_ENV_VAR_NAME, cookie); |
Peter Rajnoha | 3abf85b | 2010-03-06 02:32:31 +0000 | [diff] [blame] | 3144 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
| 3145 | action, envp); |
Milan Broz | 60935eb | 2009-06-22 10:12:30 +0100 | [diff] [blame] | 3146 | } |
Alasdair G Kergon | 69267a3 | 2007-12-13 14:15:57 +0000 | [diff] [blame] | 3147 | } |
| 3148 | |
Mike Anderson | 7a8c3d3 | 2007-10-19 22:48:01 +0100 | [diff] [blame] | 3149 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
| 3150 | { |
| 3151 | return atomic_add_return(1, &md->uevent_seq); |
| 3152 | } |
| 3153 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3154 | uint32_t dm_get_event_nr(struct mapped_device *md) |
| 3155 | { |
| 3156 | return atomic_read(&md->event_nr); |
| 3157 | } |
| 3158 | |
| 3159 | int dm_wait_event(struct mapped_device *md, int event_nr) |
| 3160 | { |
| 3161 | return wait_event_interruptible(md->eventq, |
| 3162 | (event_nr != atomic_read(&md->event_nr))); |
| 3163 | } |
| 3164 | |
Mike Anderson | 7a8c3d3 | 2007-10-19 22:48:01 +0100 | [diff] [blame] | 3165 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
| 3166 | { |
| 3167 | unsigned long flags; |
| 3168 | |
| 3169 | spin_lock_irqsave(&md->uevent_lock, flags); |
| 3170 | list_add(elist, &md->uevent_list); |
| 3171 | spin_unlock_irqrestore(&md->uevent_lock, flags); |
| 3172 | } |
| 3173 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3174 | /* |
| 3175 | * The gendisk is only valid as long as you have a reference |
| 3176 | * count on 'md'. |
| 3177 | */ |
| 3178 | struct gendisk *dm_disk(struct mapped_device *md) |
| 3179 | { |
| 3180 | return md->disk; |
| 3181 | } |
| 3182 | |
Milan Broz | 784aae7 | 2009-01-06 03:05:12 +0000 | [diff] [blame] | 3183 | struct kobject *dm_kobject(struct mapped_device *md) |
| 3184 | { |
Mikulas Patocka | 2995fa7 | 2014-01-13 19:37:54 -0500 | [diff] [blame] | 3185 | return &md->kobj_holder.kobj; |
Milan Broz | 784aae7 | 2009-01-06 03:05:12 +0000 | [diff] [blame] | 3186 | } |
| 3187 | |
Milan Broz | 784aae7 | 2009-01-06 03:05:12 +0000 | [diff] [blame] | 3188 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
| 3189 | { |
| 3190 | struct mapped_device *md; |
| 3191 | |
Mikulas Patocka | 2995fa7 | 2014-01-13 19:37:54 -0500 | [diff] [blame] | 3192 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
Milan Broz | 784aae7 | 2009-01-06 03:05:12 +0000 | [diff] [blame] | 3193 | |
Milan Broz | 4d89b7b | 2009-06-22 10:12:11 +0100 | [diff] [blame] | 3194 | if (test_bit(DMF_FREEING, &md->flags) || |
Mike Anderson | 432a212 | 2009-12-10 23:52:20 +0000 | [diff] [blame] | 3195 | dm_deleting_md(md)) |
Milan Broz | 4d89b7b | 2009-06-22 10:12:11 +0100 | [diff] [blame] | 3196 | return NULL; |
| 3197 | |
Milan Broz | 784aae7 | 2009-01-06 03:05:12 +0000 | [diff] [blame] | 3198 | dm_get(md); |
| 3199 | return md; |
| 3200 | } |
| 3201 | |
Kiyoshi Ueda | 4f186f8 | 2009-12-10 23:52:26 +0000 | [diff] [blame] | 3202 | int dm_suspended_md(struct mapped_device *md) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3203 | { |
| 3204 | return test_bit(DMF_SUSPENDED, &md->flags); |
| 3205 | } |
| 3206 | |
Mike Snitzer | ffcc393 | 2014-10-28 18:34:52 -0400 | [diff] [blame] | 3207 | int dm_suspended_internally_md(struct mapped_device *md) |
| 3208 | { |
| 3209 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); |
| 3210 | } |
| 3211 | |
Mikulas Patocka | 2c140a2 | 2013-11-01 18:27:41 -0400 | [diff] [blame] | 3212 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
| 3213 | { |
| 3214 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); |
| 3215 | } |
| 3216 | |
Kiyoshi Ueda | 64dbce5 | 2009-12-10 23:52:27 +0000 | [diff] [blame] | 3217 | int dm_suspended(struct dm_target *ti) |
| 3218 | { |
Kiyoshi Ueda | ecdb2e2 | 2010-03-06 02:29:52 +0000 | [diff] [blame] | 3219 | return dm_suspended_md(dm_table_get_md(ti->table)); |
Kiyoshi Ueda | 64dbce5 | 2009-12-10 23:52:27 +0000 | [diff] [blame] | 3220 | } |
| 3221 | EXPORT_SYMBOL_GPL(dm_suspended); |
| 3222 | |
Kiyoshi Ueda | 2e93ccc | 2006-12-08 02:41:09 -0800 | [diff] [blame] | 3223 | int dm_noflush_suspending(struct dm_target *ti) |
| 3224 | { |
Kiyoshi Ueda | ecdb2e2 | 2010-03-06 02:29:52 +0000 | [diff] [blame] | 3225 | return __noflush_suspending(dm_table_get_md(ti->table)); |
Kiyoshi Ueda | 2e93ccc | 2006-12-08 02:41:09 -0800 | [diff] [blame] | 3226 | } |
| 3227 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); |
| 3228 | |
Mikulas Patocka | c0820cf | 2012-12-21 20:23:38 +0000 | [diff] [blame] | 3229 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size) |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 3230 | { |
Jun'ichi Nomura | 5f01520 | 2013-03-01 22:45:48 +0000 | [diff] [blame] | 3231 | struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL); |
| 3232 | struct kmem_cache *cachep; |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 3233 | unsigned int pool_size = 0; |
Jun'ichi Nomura | 5f01520 | 2013-03-01 22:45:48 +0000 | [diff] [blame] | 3234 | unsigned int front_pad; |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 3235 | |
| 3236 | if (!pools) |
| 3237 | return NULL; |
| 3238 | |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 3239 | switch (type) { |
| 3240 | case DM_TYPE_BIO_BASED: |
Jun'ichi Nomura | 5f01520 | 2013-03-01 22:45:48 +0000 | [diff] [blame] | 3241 | cachep = _io_cache; |
Mike Snitzer | e860313 | 2013-09-12 18:06:12 -0400 | [diff] [blame] | 3242 | pool_size = dm_get_reserved_bio_based_ios(); |
Jun'ichi Nomura | 5f01520 | 2013-03-01 22:45:48 +0000 | [diff] [blame] | 3243 | front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 3244 | break; |
| 3245 | case DM_TYPE_REQUEST_BASED: |
Mike Snitzer | f479082 | 2013-09-12 18:06:12 -0400 | [diff] [blame] | 3246 | pool_size = dm_get_reserved_rq_based_ios(); |
Mike Snitzer | 1ae49ea | 2014-12-05 17:11:05 -0500 | [diff] [blame] | 3247 | pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache); |
| 3248 | if (!pools->rq_pool) |
| 3249 | goto out; |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 3250 | /* fall through to setup remaining rq-based pools */ |
| 3251 | case DM_TYPE_MQ_REQUEST_BASED: |
| 3252 | cachep = _rq_tio_cache; |
| 3253 | if (!pool_size) |
| 3254 | pool_size = dm_get_reserved_rq_based_ios(); |
Jun'ichi Nomura | 5f01520 | 2013-03-01 22:45:48 +0000 | [diff] [blame] | 3255 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
| 3256 | /* per_bio_data_size is not used. See __bind_mempools(). */ |
| 3257 | WARN_ON(per_bio_data_size != 0); |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 3258 | break; |
| 3259 | default: |
Jun'ichi Nomura | 5f01520 | 2013-03-01 22:45:48 +0000 | [diff] [blame] | 3260 | goto out; |
Mike Snitzer | e5863d9 | 2014-12-17 21:08:12 -0500 | [diff] [blame] | 3261 | } |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 3262 | |
Mike Snitzer | 6cfa585 | 2013-09-12 18:06:11 -0400 | [diff] [blame] | 3263 | pools->io_pool = mempool_create_slab_pool(pool_size, cachep); |
Jun'ichi Nomura | 5f01520 | 2013-03-01 22:45:48 +0000 | [diff] [blame] | 3264 | if (!pools->io_pool) |
| 3265 | goto out; |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 3266 | |
Junichi Nomura | 3d8aab2 | 2014-10-03 11:55:26 +0000 | [diff] [blame] | 3267 | pools->bs = bioset_create_nobvec(pool_size, front_pad); |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 3268 | if (!pools->bs) |
Jun'ichi Nomura | 5f01520 | 2013-03-01 22:45:48 +0000 | [diff] [blame] | 3269 | goto out; |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 3270 | |
Martin K. Petersen | a91a278 | 2011-03-17 11:11:05 +0100 | [diff] [blame] | 3271 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
Jun'ichi Nomura | 5f01520 | 2013-03-01 22:45:48 +0000 | [diff] [blame] | 3272 | goto out; |
Martin K. Petersen | a91a278 | 2011-03-17 11:11:05 +0100 | [diff] [blame] | 3273 | |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 3274 | return pools; |
| 3275 | |
Jun'ichi Nomura | 5f01520 | 2013-03-01 22:45:48 +0000 | [diff] [blame] | 3276 | out: |
| 3277 | dm_free_md_mempools(pools); |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 3278 | |
| 3279 | return NULL; |
| 3280 | } |
| 3281 | |
| 3282 | void dm_free_md_mempools(struct dm_md_mempools *pools) |
| 3283 | { |
| 3284 | if (!pools) |
| 3285 | return; |
| 3286 | |
| 3287 | if (pools->io_pool) |
| 3288 | mempool_destroy(pools->io_pool); |
| 3289 | |
Mike Snitzer | 1ae49ea | 2014-12-05 17:11:05 -0500 | [diff] [blame] | 3290 | if (pools->rq_pool) |
| 3291 | mempool_destroy(pools->rq_pool); |
| 3292 | |
Kiyoshi Ueda | e6ee8c0 | 2009-06-22 10:12:36 +0100 | [diff] [blame] | 3293 | if (pools->bs) |
| 3294 | bioset_free(pools->bs); |
| 3295 | |
| 3296 | kfree(pools); |
| 3297 | } |
| 3298 | |
Alexey Dobriyan | 83d5cde | 2009-09-21 17:01:13 -0700 | [diff] [blame] | 3299 | static const struct block_device_operations dm_blk_dops = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3300 | .open = dm_blk_open, |
| 3301 | .release = dm_blk_close, |
Milan Broz | aa129a2 | 2006-10-03 01:15:15 -0700 | [diff] [blame] | 3302 | .ioctl = dm_blk_ioctl, |
Darrick J. Wong | 3ac51e7 | 2006-03-27 01:17:54 -0800 | [diff] [blame] | 3303 | .getgeo = dm_blk_getgeo, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3304 | .owner = THIS_MODULE |
| 3305 | }; |
| 3306 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3307 | /* |
| 3308 | * module hooks |
| 3309 | */ |
| 3310 | module_init(dm_init); |
| 3311 | module_exit(dm_exit); |
| 3312 | |
| 3313 | module_param(major, uint, 0); |
| 3314 | MODULE_PARM_DESC(major, "The major number of the device mapper"); |
Mike Snitzer | f479082 | 2013-09-12 18:06:12 -0400 | [diff] [blame] | 3315 | |
Mike Snitzer | e860313 | 2013-09-12 18:06:12 -0400 | [diff] [blame] | 3316 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
| 3317 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); |
| 3318 | |
Mike Snitzer | f479082 | 2013-09-12 18:06:12 -0400 | [diff] [blame] | 3319 | module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR); |
| 3320 | MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools"); |
| 3321 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3322 | MODULE_DESCRIPTION(DM_NAME " driver"); |
| 3323 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
| 3324 | MODULE_LICENSE("GPL"); |