blob: 7bac564f3faa6e1b49899b72e1d5286efcff840b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
Milan Broz784aae72009-01-06 03:05:12 +00003 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
5 * This file is released under the GPL.
6 */
7
Mike Snitzer4cc96132016-05-12 16:28:10 -04008#include "dm-core.h"
9#include "dm-rq.h"
Mike Anderson51e5b2b2007-10-19 22:48:00 +010010#include "dm-uevent.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070011
12#include <linux/init.h>
13#include <linux/module.h>
Arjan van de Ven48c9c272006-03-27 01:18:20 -080014#include <linux/mutex.h>
Mikulas Patocka6958c1c2020-07-08 12:25:20 -040015#include <linux/sched/mm.h>
Ingo Molnar174cd4b2017-02-02 19:15:33 +010016#include <linux/sched/signal.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/blkpg.h>
18#include <linux/bio.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/mempool.h>
Dan Williamsf26c5712017-04-12 12:35:44 -070020#include <linux/dax.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/slab.h>
22#include <linux/idr.h>
Dan Williams7e026c82017-05-29 12:57:56 -070023#include <linux/uio.h>
Darrick J. Wong3ac51e72006-03-27 01:17:54 -080024#include <linux/hdreg.h>
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +010025#include <linux/delay.h>
Mike Snitzerffcc3932014-10-28 18:34:52 -040026#include <linux/wait.h>
Christoph Hellwig71cdb692015-10-15 14:10:51 +020027#include <linux/pr.h>
Elena Reshetovab0b4d7c2017-10-20 10:37:39 +030028#include <linux/refcount.h>
Christoph Hellwigc6a564ff2020-03-25 16:48:42 +010029#include <linux/part_stat.h>
Satya Tangiralaa892c8d2020-05-14 00:37:18 +000030#include <linux/blk-crypto.h>
Li Zefan55782132009-06-09 13:43:05 +080031
Alasdair G Kergon72d94862006-06-26 00:27:35 -070032#define DM_MSG_PREFIX "core"
33
Milan Broz60935eb2009-06-22 10:12:30 +010034/*
35 * Cookies are numeric values sent with CHANGE and REMOVE
36 * uevents while resuming, removing or renaming the device.
37 */
38#define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
39#define DM_COOKIE_LENGTH 24
40
Linus Torvalds1da177e2005-04-16 15:20:36 -070041static const char *_name = DM_NAME;
42
43static unsigned int major = 0;
44static unsigned int _major = 0;
45
Alasdair G Kergond15b7742011-08-02 12:32:01 +010046static DEFINE_IDR(_minor_idr);
47
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -070048static DEFINE_SPINLOCK(_minor_lock);
Mikulas Patocka2c140a22013-11-01 18:27:41 -040049
50static void do_deferred_remove(struct work_struct *w);
51
52static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
53
Mikulas Patockaacfe0ad2014-06-14 13:44:31 -040054static struct workqueue_struct *deferred_remove_workqueue;
55
Mikulas Patocka93e64422017-01-16 16:05:59 -050056atomic_t dm_global_event_nr = ATOMIC_INIT(0);
57DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
58
Mikulas Patocka62e08242017-09-20 07:29:49 -040059void dm_issue_global_event(void)
60{
61 atomic_inc(&dm_global_event_nr);
62 wake_up(&dm_global_eventq);
63}
64
Linus Torvalds1da177e2005-04-16 15:20:36 -070065/*
Mike Snitzer64f52b02017-12-11 23:17:47 -050066 * One of these is allocated (on-stack) per original bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 */
Mike Snitzer64f52b02017-12-11 23:17:47 -050068struct clone_info {
Mike Snitzer64f52b02017-12-11 23:17:47 -050069 struct dm_table *map;
70 struct bio *bio;
71 struct dm_io *io;
72 sector_t sector;
73 unsigned sector_count;
74};
75
76/*
77 * One of these is allocated per clone bio.
78 */
79#define DM_TIO_MAGIC 7282014
80struct dm_target_io {
81 unsigned magic;
82 struct dm_io *io;
83 struct dm_target *ti;
84 unsigned target_bio_nr;
85 unsigned *len_ptr;
86 bool inside_dm_io;
87 struct bio clone;
88};
89
90/*
91 * One of these is allocated per original bio.
92 * It contains the first clone used for that original.
93 */
94#define DM_IO_MAGIC 5191977
Linus Torvalds1da177e2005-04-16 15:20:36 -070095struct dm_io {
Mike Snitzer64f52b02017-12-11 23:17:47 -050096 unsigned magic;
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 struct mapped_device *md;
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +020098 blk_status_t status;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 atomic_t io_count;
Mike Snitzer745dc572017-12-11 20:51:50 -0500100 struct bio *orig_bio;
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -0800101 unsigned long start_time;
Kiyoshi Uedaf88fb982009-10-16 23:18:15 +0100102 spinlock_t endio_lock;
Mikulas Patockafd2ed4d2013-08-16 10:54:23 -0400103 struct dm_stats_aux stats_aux;
Mike Snitzer64f52b02017-12-11 23:17:47 -0500104 /* last member of dm_target_io is 'struct bio' */
105 struct dm_target_io tio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106};
107
Mike Snitzer64f52b02017-12-11 23:17:47 -0500108void *dm_per_bio_data(struct bio *bio, size_t data_size)
109{
110 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
111 if (!tio->inside_dm_io)
112 return (char *)bio - offsetof(struct dm_target_io, clone) - data_size;
113 return (char *)bio - offsetof(struct dm_target_io, clone) - offsetof(struct dm_io, tio) - data_size;
114}
115EXPORT_SYMBOL_GPL(dm_per_bio_data);
116
117struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
118{
119 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
120 if (io->magic == DM_IO_MAGIC)
121 return (struct bio *)((char *)io + offsetof(struct dm_io, tio) + offsetof(struct dm_target_io, clone));
122 BUG_ON(io->magic != DM_TIO_MAGIC);
123 return (struct bio *)((char *)io + offsetof(struct dm_target_io, clone));
124}
125EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
126
127unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
128{
129 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
130}
131EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
132
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -0700133#define MINOR_ALLOCED ((void *)-1)
134
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135/*
136 * Bits for the md->flags field.
137 */
Alasdair G Kergon1eb787e2009-04-09 00:27:14 +0100138#define DMF_BLOCK_IO_FOR_SUSPEND 0
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139#define DMF_SUSPENDED 1
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -0800140#define DMF_FROZEN 2
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700141#define DMF_FREEING 3
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700142#define DMF_DELETING 4
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800143#define DMF_NOFLUSH_SUSPENDING 5
Kent Overstreet8ae12662015-04-27 23:48:34 -0700144#define DMF_DEFERRED_REMOVE 6
145#define DMF_SUSPENDED_INTERNALLY 7
Mikulas Patocka5df96f22020-07-23 10:42:09 -0400146#define DMF_POST_SUSPENDING 8
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
Mike Snitzer115485e2016-02-22 12:16:21 -0500148#define DM_NUMA_NODE NUMA_NO_NODE
Mike Snitzer115485e2016-02-22 12:16:21 -0500149static int dm_numa_node = DM_NUMA_NODE;
Mike Snitzerfaad87d2016-01-28 16:52:56 -0500150
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100151/*
152 * For mempools pre-allocation at the table loading time.
153 */
154struct dm_md_mempools {
Kent Overstreet6f1c8192018-05-20 18:25:53 -0400155 struct bio_set bs;
156 struct bio_set io_bs;
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100157};
158
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500159struct table_device {
160 struct list_head list;
Elena Reshetovab0b4d7c2017-10-20 10:37:39 +0300161 refcount_t count;
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500162 struct dm_dev dm_dev;
163};
164
Mike Snitzerf4790822013-09-12 18:06:12 -0400165/*
Mike Snitzere8603132013-09-12 18:06:12 -0400166 * Bio-based DM's mempools' reserved IOs set by the user.
167 */
Mike Snitzer4cc96132016-05-12 16:28:10 -0400168#define RESERVED_BIO_BASED_IOS 16
Mike Snitzere8603132013-09-12 18:06:12 -0400169static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
170
Mike Snitzer115485e2016-02-22 12:16:21 -0500171static int __dm_get_module_param_int(int *module_param, int min, int max)
172{
Mark Rutland6aa7de02017-10-23 14:07:29 -0700173 int param = READ_ONCE(*module_param);
Mike Snitzer115485e2016-02-22 12:16:21 -0500174 int modified_param = 0;
175 bool modified = true;
176
177 if (param < min)
178 modified_param = min;
179 else if (param > max)
180 modified_param = max;
181 else
182 modified = false;
183
184 if (modified) {
185 (void)cmpxchg(module_param, param, modified_param);
186 param = modified_param;
187 }
188
189 return param;
190}
191
Mike Snitzer4cc96132016-05-12 16:28:10 -0400192unsigned __dm_get_module_param(unsigned *module_param,
193 unsigned def, unsigned max)
Mike Snitzerf4790822013-09-12 18:06:12 -0400194{
Mark Rutland6aa7de02017-10-23 14:07:29 -0700195 unsigned param = READ_ONCE(*module_param);
Mike Snitzer09c2d532015-02-27 22:25:26 -0500196 unsigned modified_param = 0;
Mike Snitzerf4790822013-09-12 18:06:12 -0400197
Mike Snitzer09c2d532015-02-27 22:25:26 -0500198 if (!param)
199 modified_param = def;
200 else if (param > max)
201 modified_param = max;
Mike Snitzerf4790822013-09-12 18:06:12 -0400202
Mike Snitzer09c2d532015-02-27 22:25:26 -0500203 if (modified_param) {
204 (void)cmpxchg(module_param, param, modified_param);
205 param = modified_param;
Mike Snitzerf4790822013-09-12 18:06:12 -0400206 }
207
Mike Snitzer09c2d532015-02-27 22:25:26 -0500208 return param;
Mike Snitzerf4790822013-09-12 18:06:12 -0400209}
210
Mike Snitzere8603132013-09-12 18:06:12 -0400211unsigned dm_get_reserved_bio_based_ios(void)
212{
Mike Snitzer09c2d532015-02-27 22:25:26 -0500213 return __dm_get_module_param(&reserved_bio_based_ios,
Mike Snitzer4cc96132016-05-12 16:28:10 -0400214 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
Mike Snitzere8603132013-09-12 18:06:12 -0400215}
216EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
217
Mike Snitzer115485e2016-02-22 12:16:21 -0500218static unsigned dm_get_numa_node(void)
219{
220 return __dm_get_module_param_int(&dm_numa_node,
221 DM_NUMA_NODE, num_online_nodes() - 1);
222}
223
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224static int __init local_init(void)
225{
Mike Snitzere689fba2019-02-20 15:37:44 -0500226 int r;
Mike Snitzer1ae49ea2014-12-05 17:11:05 -0500227
Mike Anderson51e5b2b2007-10-19 22:48:00 +0100228 r = dm_uevent_init();
Kiyoshi Ueda51157b42008-10-21 17:45:08 +0100229 if (r)
Mike Snitzere689fba2019-02-20 15:37:44 -0500230 return r;
Mike Anderson51e5b2b2007-10-19 22:48:00 +0100231
Mikulas Patockaacfe0ad2014-06-14 13:44:31 -0400232 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
233 if (!deferred_remove_workqueue) {
234 r = -ENOMEM;
235 goto out_uevent_exit;
236 }
237
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 _major = major;
239 r = register_blkdev(_major, _name);
Kiyoshi Ueda51157b42008-10-21 17:45:08 +0100240 if (r < 0)
Mikulas Patockaacfe0ad2014-06-14 13:44:31 -0400241 goto out_free_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
243 if (!_major)
244 _major = r;
245
246 return 0;
Kiyoshi Ueda51157b42008-10-21 17:45:08 +0100247
Mikulas Patockaacfe0ad2014-06-14 13:44:31 -0400248out_free_workqueue:
249 destroy_workqueue(deferred_remove_workqueue);
Kiyoshi Ueda51157b42008-10-21 17:45:08 +0100250out_uevent_exit:
251 dm_uevent_exit();
Kiyoshi Ueda51157b42008-10-21 17:45:08 +0100252
253 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254}
255
256static void local_exit(void)
257{
Mikulas Patocka2c140a22013-11-01 18:27:41 -0400258 flush_scheduled_work();
Mikulas Patockaacfe0ad2014-06-14 13:44:31 -0400259 destroy_workqueue(deferred_remove_workqueue);
Mikulas Patocka2c140a22013-11-01 18:27:41 -0400260
Akinobu Mita00d59402007-07-17 04:03:46 -0700261 unregister_blkdev(_major, _name);
Mike Anderson51e5b2b2007-10-19 22:48:00 +0100262 dm_uevent_exit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263
264 _major = 0;
265
266 DMINFO("cleaned up");
267}
268
Alasdair G Kergonb9249e52008-02-08 02:09:51 +0000269static int (*_inits[])(void) __initdata = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 local_init,
271 dm_target_init,
272 dm_linear_init,
273 dm_stripe_init,
Mikulas Patocka952b3552009-12-10 23:51:57 +0000274 dm_io_init,
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100275 dm_kcopyd_init,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 dm_interface_init,
Mikulas Patockafd2ed4d2013-08-16 10:54:23 -0400277 dm_statistics_init,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278};
279
Alasdair G Kergonb9249e52008-02-08 02:09:51 +0000280static void (*_exits[])(void) = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 local_exit,
282 dm_target_exit,
283 dm_linear_exit,
284 dm_stripe_exit,
Mikulas Patocka952b3552009-12-10 23:51:57 +0000285 dm_io_exit,
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100286 dm_kcopyd_exit,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 dm_interface_exit,
Mikulas Patockafd2ed4d2013-08-16 10:54:23 -0400288 dm_statistics_exit,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289};
290
291static int __init dm_init(void)
292{
293 const int count = ARRAY_SIZE(_inits);
294
295 int r, i;
296
297 for (i = 0; i < count; i++) {
298 r = _inits[i]();
299 if (r)
300 goto bad;
301 }
302
303 return 0;
304
305 bad:
306 while (i--)
307 _exits[i]();
308
309 return r;
310}
311
312static void __exit dm_exit(void)
313{
314 int i = ARRAY_SIZE(_exits);
315
316 while (i--)
317 _exits[i]();
Alasdair G Kergond15b7742011-08-02 12:32:01 +0100318
319 /*
320 * Should be empty by this point.
321 */
Alasdair G Kergond15b7742011-08-02 12:32:01 +0100322 idr_destroy(&_minor_idr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323}
324
325/*
326 * Block device functions
327 */
Mike Anderson432a2122009-12-10 23:52:20 +0000328int dm_deleting_md(struct mapped_device *md)
329{
330 return test_bit(DMF_DELETING, &md->flags);
331}
332
Al Virofe5f9f22008-03-02 10:29:31 -0500333static int dm_blk_open(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334{
335 struct mapped_device *md;
336
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700337 spin_lock(&_minor_lock);
338
Al Virofe5f9f22008-03-02 10:29:31 -0500339 md = bdev->bd_disk->private_data;
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700340 if (!md)
341 goto out;
342
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700343 if (test_bit(DMF_FREEING, &md->flags) ||
Mike Anderson432a2122009-12-10 23:52:20 +0000344 dm_deleting_md(md)) {
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700345 md = NULL;
346 goto out;
347 }
348
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 dm_get(md);
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700350 atomic_inc(&md->open_count);
Jeff Mahoneyfba9f902006-06-26 00:27:23 -0700351out:
352 spin_unlock(&_minor_lock);
353
354 return md ? 0 : -ENXIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355}
356
Al Virodb2a1442013-05-05 21:52:57 -0400357static void dm_blk_close(struct gendisk *disk, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358{
Mike Snitzer63a4f062015-03-23 17:01:43 -0400359 struct mapped_device *md;
Arnd Bergmann6e9624b2010-08-07 18:25:34 +0200360
Milan Broz4a1aeb92011-01-13 19:59:48 +0000361 spin_lock(&_minor_lock);
362
Mike Snitzer63a4f062015-03-23 17:01:43 -0400363 md = disk->private_data;
364 if (WARN_ON(!md))
365 goto out;
366
Mikulas Patocka2c140a22013-11-01 18:27:41 -0400367 if (atomic_dec_and_test(&md->open_count) &&
368 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
Mikulas Patockaacfe0ad2014-06-14 13:44:31 -0400369 queue_work(deferred_remove_workqueue, &deferred_remove_work);
Mikulas Patocka2c140a22013-11-01 18:27:41 -0400370
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 dm_put(md);
Mike Snitzer63a4f062015-03-23 17:01:43 -0400372out:
Milan Broz4a1aeb92011-01-13 19:59:48 +0000373 spin_unlock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374}
375
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700376int dm_open_count(struct mapped_device *md)
377{
378 return atomic_read(&md->open_count);
379}
380
381/*
382 * Guarantees nothing is using the device before it's deleted.
383 */
Mikulas Patocka2c140a22013-11-01 18:27:41 -0400384int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700385{
386 int r = 0;
387
388 spin_lock(&_minor_lock);
389
Mikulas Patocka2c140a22013-11-01 18:27:41 -0400390 if (dm_open_count(md)) {
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700391 r = -EBUSY;
Mikulas Patocka2c140a22013-11-01 18:27:41 -0400392 if (mark_deferred)
393 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
394 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
395 r = -EEXIST;
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -0700396 else
397 set_bit(DMF_DELETING, &md->flags);
398
399 spin_unlock(&_minor_lock);
400
401 return r;
402}
403
Mikulas Patocka2c140a22013-11-01 18:27:41 -0400404int dm_cancel_deferred_remove(struct mapped_device *md)
405{
406 int r = 0;
407
408 spin_lock(&_minor_lock);
409
410 if (test_bit(DMF_DELETING, &md->flags))
411 r = -EBUSY;
412 else
413 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
414
415 spin_unlock(&_minor_lock);
416
417 return r;
418}
419
420static void do_deferred_remove(struct work_struct *w)
421{
422 dm_deferred_remove();
423}
424
Darrick J. Wong3ac51e72006-03-27 01:17:54 -0800425static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
426{
427 struct mapped_device *md = bdev->bd_disk->private_data;
428
429 return dm_get_geometry(md, geo);
430}
431
Christoph Hellwige76239a2018-10-12 19:08:49 +0900432#ifdef CONFIG_BLK_DEV_ZONED
Christoph Hellwigd4100352019-11-11 11:39:30 +0900433int dm_report_zones_cb(struct blk_zone *zone, unsigned int idx, void *data)
434{
435 struct dm_report_zones_args *args = data;
436 sector_t sector_diff = args->tgt->begin - args->start;
437
438 /*
439 * Ignore zones beyond the target range.
440 */
441 if (zone->start >= args->start + args->tgt->len)
442 return 0;
443
444 /*
445 * Remap the start sector and write pointer position of the zone
446 * to match its position in the target range.
447 */
448 zone->start += sector_diff;
449 if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
450 if (zone->cond == BLK_ZONE_COND_FULL)
451 zone->wp = zone->start + zone->len;
452 else if (zone->cond == BLK_ZONE_COND_EMPTY)
453 zone->wp = zone->start;
454 else
455 zone->wp += sector_diff;
456 }
457
458 args->next_sector = zone->start + zone->len;
459 return args->orig_cb(zone, args->zone_idx++, args->orig_data);
460}
461EXPORT_SYMBOL_GPL(dm_report_zones_cb);
462
463static int dm_blk_report_zones(struct gendisk *disk, sector_t sector,
464 unsigned int nr_zones, report_zones_cb cb, void *data)
465{
Christoph Hellwige76239a2018-10-12 19:08:49 +0900466 struct mapped_device *md = disk->private_data;
Christoph Hellwige76239a2018-10-12 19:08:49 +0900467 struct dm_table *map;
468 int srcu_idx, ret;
Christoph Hellwigd4100352019-11-11 11:39:30 +0900469 struct dm_report_zones_args args = {
470 .next_sector = sector,
471 .orig_data = data,
472 .orig_cb = cb,
473 };
Christoph Hellwige76239a2018-10-12 19:08:49 +0900474
475 if (dm_suspended_md(md))
476 return -EAGAIN;
477
478 map = dm_get_live_table(md, &srcu_idx);
Sergei Shtepa89478332020-11-11 15:55:46 +0300479 if (!map) {
480 ret = -EIO;
481 goto out;
482 }
Christoph Hellwige76239a2018-10-12 19:08:49 +0900483
Christoph Hellwigd4100352019-11-11 11:39:30 +0900484 do {
485 struct dm_target *tgt;
Christoph Hellwige76239a2018-10-12 19:08:49 +0900486
Christoph Hellwigd4100352019-11-11 11:39:30 +0900487 tgt = dm_table_find_target(map, args.next_sector);
488 if (WARN_ON_ONCE(!tgt->type->report_zones)) {
489 ret = -EIO;
490 goto out;
491 }
Christoph Hellwige76239a2018-10-12 19:08:49 +0900492
Christoph Hellwigd4100352019-11-11 11:39:30 +0900493 args.tgt = tgt;
Johannes Thumshirna9cb9f42020-08-04 18:25:01 +0900494 ret = tgt->type->report_zones(tgt, &args,
495 nr_zones - args.zone_idx);
Christoph Hellwigd4100352019-11-11 11:39:30 +0900496 if (ret < 0)
497 goto out;
498 } while (args.zone_idx < nr_zones &&
499 args.next_sector < get_capacity(disk));
Christoph Hellwige76239a2018-10-12 19:08:49 +0900500
Christoph Hellwigd4100352019-11-11 11:39:30 +0900501 ret = args.zone_idx;
Christoph Hellwige76239a2018-10-12 19:08:49 +0900502out:
503 dm_put_live_table(md, srcu_idx);
504 return ret;
Christoph Hellwige76239a2018-10-12 19:08:49 +0900505}
Christoph Hellwigd4100352019-11-11 11:39:30 +0900506#else
507#define dm_blk_report_zones NULL
508#endif /* CONFIG_BLK_DEV_ZONED */
Christoph Hellwige76239a2018-10-12 19:08:49 +0900509
Mike Snitzer971888c2018-04-03 15:05:12 -0400510static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
Mike Snitzer5bd5e8d2018-04-03 16:54:10 -0400511 struct block_device **bdev)
Milan Brozaa129a22006-10-03 01:15:15 -0700512{
Mike Snitzer66482022016-02-18 15:44:39 -0500513 struct dm_target *tgt;
Hannes Reinecke6c182cd2013-07-10 23:41:15 +0100514 struct dm_table *map;
Mike Snitzer971888c2018-04-03 15:05:12 -0400515 int r;
Milan Brozaa129a22006-10-03 01:15:15 -0700516
Hannes Reinecke6c182cd2013-07-10 23:41:15 +0100517retry:
Christoph Hellwige56f81e2015-10-15 14:10:50 +0200518 r = -ENOTTY;
Mike Snitzer971888c2018-04-03 15:05:12 -0400519 map = dm_get_live_table(md, srcu_idx);
Milan Brozaa129a22006-10-03 01:15:15 -0700520 if (!map || !dm_table_get_size(map))
Mike Snitzer971888c2018-04-03 15:05:12 -0400521 return r;
Milan Brozaa129a22006-10-03 01:15:15 -0700522
523 /* We only support devices that have a single target */
524 if (dm_table_get_num_targets(map) != 1)
Mike Snitzer971888c2018-04-03 15:05:12 -0400525 return r;
Milan Brozaa129a22006-10-03 01:15:15 -0700526
Mike Snitzer66482022016-02-18 15:44:39 -0500527 tgt = dm_table_get_target(map, 0);
528 if (!tgt->type->prepare_ioctl)
Mike Snitzer971888c2018-04-03 15:05:12 -0400529 return r;
Milan Brozaa129a22006-10-03 01:15:15 -0700530
Mike Snitzer971888c2018-04-03 15:05:12 -0400531 if (dm_suspended_md(md))
532 return -EAGAIN;
Milan Brozaa129a22006-10-03 01:15:15 -0700533
Mike Snitzer5bd5e8d2018-04-03 16:54:10 -0400534 r = tgt->type->prepare_ioctl(tgt, bdev);
Junichi Nomura5bbbfdf2015-11-17 09:39:26 +0000535 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
Mike Snitzer971888c2018-04-03 15:05:12 -0400536 dm_put_live_table(md, *srcu_idx);
Hannes Reinecke6c182cd2013-07-10 23:41:15 +0100537 msleep(10);
538 goto retry;
539 }
Mike Snitzer971888c2018-04-03 15:05:12 -0400540
Christoph Hellwige56f81e2015-10-15 14:10:50 +0200541 return r;
542}
Hannes Reinecke6c182cd2013-07-10 23:41:15 +0100543
Mike Snitzer971888c2018-04-03 15:05:12 -0400544static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
Mike Snitzer971888c2018-04-03 15:05:12 -0400545{
546 dm_put_live_table(md, srcu_idx);
547}
548
Christoph Hellwige56f81e2015-10-15 14:10:50 +0200549static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
550 unsigned int cmd, unsigned long arg)
551{
552 struct mapped_device *md = bdev->bd_disk->private_data;
Mike Snitzer971888c2018-04-03 15:05:12 -0400553 int r, srcu_idx;
Christoph Hellwige56f81e2015-10-15 14:10:50 +0200554
Mike Snitzer5bd5e8d2018-04-03 16:54:10 -0400555 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
Christoph Hellwige56f81e2015-10-15 14:10:50 +0200556 if (r < 0)
Mike Snitzer971888c2018-04-03 15:05:12 -0400557 goto out;
Christoph Hellwige56f81e2015-10-15 14:10:50 +0200558
559 if (r > 0) {
560 /*
Christoph Hellwige980f622017-02-04 10:45:03 +0100561 * Target determined this ioctl is being issued against a
562 * subset of the parent bdev; require extra privileges.
Christoph Hellwige56f81e2015-10-15 14:10:50 +0200563 */
Christoph Hellwige980f622017-02-04 10:45:03 +0100564 if (!capable(CAP_SYS_RAWIO)) {
Mike Snitzer0378c622021-01-06 18:19:05 -0500565 DMDEBUG_LIMIT(
Christoph Hellwige980f622017-02-04 10:45:03 +0100566 "%s: sending ioctl %x to DM device without required privilege.",
567 current->comm, cmd);
568 r = -ENOIOCTLCMD;
Christoph Hellwige56f81e2015-10-15 14:10:50 +0200569 goto out;
Christoph Hellwige980f622017-02-04 10:45:03 +0100570 }
Christoph Hellwige56f81e2015-10-15 14:10:50 +0200571 }
572
Christoph Hellwiga7cb3d2f2020-11-03 11:00:18 +0100573 if (!bdev->bd_disk->fops->ioctl)
574 r = -ENOTTY;
575 else
576 r = bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
Christoph Hellwige56f81e2015-10-15 14:10:50 +0200577out:
Mike Snitzer971888c2018-04-03 15:05:12 -0400578 dm_unprepare_ioctl(md, srcu_idx);
Milan Brozaa129a22006-10-03 01:15:15 -0700579 return r;
580}
581
Mike Snitzer7465d7a2020-09-17 12:59:36 -0400582u64 dm_start_time_ns_from_clone(struct bio *bio)
583{
584 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
585 struct dm_io *io = tio->io;
586
587 return jiffies_to_nsecs(io->start_time);
588}
589EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
590
591static void start_io_acct(struct dm_io *io)
592{
593 struct mapped_device *md = io->md;
594 struct bio *bio = io->orig_bio;
595
596 io->start_time = bio_start_io_acct(bio);
597 if (unlikely(dm_stats_used(&md->stats)))
598 dm_stats_account_io(&md->stats, bio_data_dir(bio),
599 bio->bi_iter.bi_sector, bio_sectors(bio),
600 false, 0, &io->stats_aux);
601}
602
603static void end_io_acct(struct dm_io *io)
604{
605 struct mapped_device *md = io->md;
606 struct bio *bio = io->orig_bio;
607 unsigned long duration = jiffies - io->start_time;
608
609 bio_end_io_acct(bio, io->start_time);
610
611 if (unlikely(dm_stats_used(&md->stats)))
612 dm_stats_account_io(&md->stats, bio_data_dir(bio),
613 bio->bi_iter.bi_sector, bio_sectors(bio),
614 true, duration, &io->stats_aux);
615
616 /* nudge anyone waiting on suspend queue */
617 if (unlikely(wq_has_sleeper(&md->wait)))
618 wake_up(&md->wait);
619}
Mike Snitzer978e51b2017-12-09 15:16:42 -0500620
621static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622{
Mike Snitzer64f52b02017-12-11 23:17:47 -0500623 struct dm_io *io;
624 struct dm_target_io *tio;
625 struct bio *clone;
626
Kent Overstreet6f1c8192018-05-20 18:25:53 -0400627 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
Mike Snitzer64f52b02017-12-11 23:17:47 -0500628 if (!clone)
629 return NULL;
630
631 tio = container_of(clone, struct dm_target_io, clone);
632 tio->inside_dm_io = true;
633 tio->io = NULL;
634
635 io = container_of(tio, struct dm_io, tio);
636 io->magic = DM_IO_MAGIC;
Mike Snitzer978e51b2017-12-09 15:16:42 -0500637 io->status = 0;
638 atomic_set(&io->io_count, 1);
639 io->orig_bio = bio;
640 io->md = md;
641 spin_lock_init(&io->endio_lock);
642
643 start_io_acct(io);
Mike Snitzer64f52b02017-12-11 23:17:47 -0500644
645 return io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646}
647
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100648static void free_io(struct mapped_device *md, struct dm_io *io)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649{
Mike Snitzer64f52b02017-12-11 23:17:47 -0500650 bio_put(&io->tio.clone);
651}
652
653static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
654 unsigned target_bio_nr, gfp_t gfp_mask)
655{
656 struct dm_target_io *tio;
657
658 if (!ci->io->tio.io) {
659 /* the dm_target_io embedded in ci->io is available */
660 tio = &ci->io->tio;
661 } else {
Kent Overstreet6f1c8192018-05-20 18:25:53 -0400662 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
Mike Snitzer64f52b02017-12-11 23:17:47 -0500663 if (!clone)
664 return NULL;
665
666 tio = container_of(clone, struct dm_target_io, clone);
667 tio->inside_dm_io = false;
668 }
669
670 tio->magic = DM_TIO_MAGIC;
671 tio->io = ci->io;
672 tio->ti = ti;
673 tio->target_bio_nr = target_bio_nr;
674
675 return tio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676}
677
Mike Snitzercfae7522016-04-11 12:05:38 -0400678static void free_tio(struct dm_target_io *tio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
Mike Snitzer64f52b02017-12-11 23:17:47 -0500680 if (tio->inside_dm_io)
681 return;
Mikulas Patockadba14162012-10-12 21:02:15 +0100682 bio_put(&tio->clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683}
684
685/*
686 * Add the bio to the list of deferred io.
687 */
Mikulas Patocka92c63902009-04-09 00:27:15 +0100688static void queue_io(struct mapped_device *md, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689{
Kiyoshi Ueda054474202010-09-08 18:07:01 +0200690 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691
Kiyoshi Ueda054474202010-09-08 18:07:01 +0200692 spin_lock_irqsave(&md->deferred_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 bio_list_add(&md->deferred, bio);
Kiyoshi Ueda054474202010-09-08 18:07:01 +0200694 spin_unlock_irqrestore(&md->deferred_lock, flags);
Tejun Heo6a8736d2010-09-08 18:07:00 +0200695 queue_work(md->wq, &md->work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696}
697
698/*
699 * Everyone (including functions in this file), should use this
700 * function to access the md->map field, and make sure they call
Mikulas Patocka83d5e5b2013-07-10 23:41:18 +0100701 * dm_put_live_table() when finished.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 */
Mikulas Patocka83d5e5b2013-07-10 23:41:18 +0100703struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704{
Mikulas Patocka83d5e5b2013-07-10 23:41:18 +0100705 *srcu_idx = srcu_read_lock(&md->io_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706
Mikulas Patocka83d5e5b2013-07-10 23:41:18 +0100707 return srcu_dereference(md->map, &md->io_barrier);
708}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709
Mikulas Patocka83d5e5b2013-07-10 23:41:18 +0100710void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
711{
712 srcu_read_unlock(&md->io_barrier, srcu_idx);
713}
714
715void dm_sync_table(struct mapped_device *md)
716{
717 synchronize_srcu(&md->io_barrier);
718 synchronize_rcu_expedited();
719}
720
721/*
722 * A fast alternative to dm_get_live_table/dm_put_live_table.
723 * The caller must not block between these two functions.
724 */
725static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
726{
727 rcu_read_lock();
728 return rcu_dereference(md->map);
729}
730
731static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
732{
733 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734}
735
Mike Snitzer971888c2018-04-03 15:05:12 -0400736static char *_dm_claim_ptr = "I belong to device-mapper";
737
Darrick J. Wong3ac51e72006-03-27 01:17:54 -0800738/*
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500739 * Open a table device so we can use it as a map destination.
740 */
741static int open_table_device(struct table_device *td, dev_t dev,
742 struct mapped_device *md)
743{
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500744 struct block_device *bdev;
745
746 int r;
747
748 BUG_ON(td->dm_dev.bdev);
749
Mike Snitzer519049a2018-02-22 13:31:20 -0500750 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500751 if (IS_ERR(bdev))
752 return PTR_ERR(bdev);
753
754 r = bd_link_disk_holder(bdev, dm_disk(md));
755 if (r) {
756 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
757 return r;
758 }
759
760 td->dm_dev.bdev = bdev;
Dan Williams817bf402017-04-12 13:37:44 -0700761 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500762 return 0;
763}
764
765/*
766 * Close a table device that we've been using.
767 */
768static void close_table_device(struct table_device *td, struct mapped_device *md)
769{
770 if (!td->dm_dev.bdev)
771 return;
772
773 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
774 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
Dan Williams817bf402017-04-12 13:37:44 -0700775 put_dax(td->dm_dev.dax_dev);
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500776 td->dm_dev.bdev = NULL;
Dan Williams817bf402017-04-12 13:37:44 -0700777 td->dm_dev.dax_dev = NULL;
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500778}
779
780static struct table_device *find_table_device(struct list_head *l, dev_t dev,
Sheetal Singala8454fca2019-05-10 23:18:37 +0530781 fmode_t mode)
782{
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500783 struct table_device *td;
784
785 list_for_each_entry(td, l, list)
786 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
787 return td;
788
789 return NULL;
790}
791
792int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
Sheetal Singala8454fca2019-05-10 23:18:37 +0530793 struct dm_dev **result)
794{
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500795 int r;
796 struct table_device *td;
797
798 mutex_lock(&md->table_devices_lock);
799 td = find_table_device(&md->table_devices, dev, mode);
800 if (!td) {
Mike Snitzer115485e2016-02-22 12:16:21 -0500801 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500802 if (!td) {
803 mutex_unlock(&md->table_devices_lock);
804 return -ENOMEM;
805 }
806
807 td->dm_dev.mode = mode;
808 td->dm_dev.bdev = NULL;
809
810 if ((r = open_table_device(td, dev, md))) {
811 mutex_unlock(&md->table_devices_lock);
812 kfree(td);
813 return r;
814 }
815
816 format_dev_t(td->dm_dev.name, dev);
817
Elena Reshetovab0b4d7c2017-10-20 10:37:39 +0300818 refcount_set(&td->count, 1);
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500819 list_add(&td->list, &md->table_devices);
Elena Reshetovab0b4d7c2017-10-20 10:37:39 +0300820 } else {
821 refcount_inc(&td->count);
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500822 }
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500823 mutex_unlock(&md->table_devices_lock);
824
825 *result = &td->dm_dev;
826 return 0;
827}
828EXPORT_SYMBOL_GPL(dm_get_table_device);
829
830void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
831{
832 struct table_device *td = container_of(d, struct table_device, dm_dev);
833
834 mutex_lock(&md->table_devices_lock);
Elena Reshetovab0b4d7c2017-10-20 10:37:39 +0300835 if (refcount_dec_and_test(&td->count)) {
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500836 close_table_device(td, md);
837 list_del(&td->list);
838 kfree(td);
839 }
840 mutex_unlock(&md->table_devices_lock);
841}
842EXPORT_SYMBOL(dm_put_table_device);
843
844static void free_table_devices(struct list_head *devices)
845{
846 struct list_head *tmp, *next;
847
848 list_for_each_safe(tmp, next, devices) {
849 struct table_device *td = list_entry(tmp, struct table_device, list);
850
851 DMWARN("dm_destroy: %s still exists with %d references",
Elena Reshetovab0b4d7c2017-10-20 10:37:39 +0300852 td->dm_dev.name, refcount_read(&td->count));
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500853 kfree(td);
854 }
855}
856
857/*
Darrick J. Wong3ac51e72006-03-27 01:17:54 -0800858 * Get the geometry associated with a dm device
859 */
860int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
861{
862 *geo = md->geometry;
863
864 return 0;
865}
866
867/*
868 * Set the geometry of a device.
869 */
870int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
871{
872 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
873
874 if (geo->start > sz) {
875 DMWARN("Start sector is beyond the geometry limits.");
876 return -EINVAL;
877 }
878
879 md->geometry = *geo;
880
881 return 0;
882}
883
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800884static int __noflush_suspending(struct mapped_device *md)
885{
886 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
887}
888
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889/*
890 * Decrements the number of outstanding ios that a bio has been
891 * cloned into, completing the original io if necc.
892 */
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200893static void dec_pending(struct dm_io *io, blk_status_t error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894{
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800895 unsigned long flags;
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200896 blk_status_t io_error;
Milan Brozb35f8ca2009-03-16 17:44:36 +0000897 struct bio *bio;
898 struct mapped_device *md = io->md;
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800899
900 /* Push-back supersedes any I/O errors */
Kiyoshi Uedaf88fb982009-10-16 23:18:15 +0100901 if (unlikely(error)) {
902 spin_lock_irqsave(&io->endio_lock, flags);
Mike Snitzer745dc572017-12-11 20:51:50 -0500903 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200904 io->status = error;
Kiyoshi Uedaf88fb982009-10-16 23:18:15 +0100905 spin_unlock_irqrestore(&io->endio_lock, flags);
906 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
908 if (atomic_dec_and_test(&io->io_count)) {
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200909 if (io->status == BLK_STS_DM_REQUEUE) {
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800910 /*
911 * Target requested pushing back the I/O.
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800912 */
Mikulas Patocka022c2612009-04-02 19:55:39 +0100913 spin_lock_irqsave(&md->deferred_lock, flags);
Tejun Heo6a8736d2010-09-08 18:07:00 +0200914 if (__noflush_suspending(md))
Mike Snitzer745dc572017-12-11 20:51:50 -0500915 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
916 bio_list_add_head(&md->deferred, io->orig_bio);
Tejun Heo6a8736d2010-09-08 18:07:00 +0200917 else
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800918 /* noflush suspend was interrupted. */
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200919 io->status = BLK_STS_IOERR;
Mikulas Patocka022c2612009-04-02 19:55:39 +0100920 spin_unlock_irqrestore(&md->deferred_lock, flags);
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800921 }
922
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200923 io_error = io->status;
Mike Snitzer745dc572017-12-11 20:51:50 -0500924 bio = io->orig_bio;
Tejun Heo6a8736d2010-09-08 18:07:00 +0200925 end_io_acct(io);
926 free_io(md, io);
Jens Axboe2056a782006-03-23 20:00:26 +0100927
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200928 if (io_error == BLK_STS_DM_REQUEUE)
Tejun Heo6a8736d2010-09-08 18:07:00 +0200929 return;
930
Jens Axboe1eff9d32016-08-05 15:35:16 -0600931 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
Mikulas Patockaaf7e4662009-04-09 00:27:16 +0100932 /*
Tejun Heo6a8736d2010-09-08 18:07:00 +0200933 * Preflush done for flush with data, reissue
Mike Christie28a8f0d2016-06-05 14:32:25 -0500934 * without REQ_PREFLUSH.
Mikulas Patockaaf7e4662009-04-09 00:27:16 +0100935 */
Jens Axboe1eff9d32016-08-05 15:35:16 -0600936 bio->bi_opf &= ~REQ_PREFLUSH;
Tejun Heo6a8736d2010-09-08 18:07:00 +0200937 queue_io(md, bio);
Mikulas Patockaaf7e4662009-04-09 00:27:16 +0100938 } else {
Mike Snitzerb372d362010-09-08 18:07:01 +0200939 /* done with normal IO or empty flush */
NeilBrown8dd601f2018-02-15 20:00:15 +1100940 if (io_error)
941 bio->bi_status = io_error;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200942 bio_endio(bio);
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -0800943 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 }
945}
946
Mike Snitzerbcb44432019-04-03 12:23:11 -0400947void disable_discard(struct mapped_device *md)
948{
949 struct queue_limits *limits = dm_get_queue_limits(md);
950
951 /* device doesn't really support DISCARD, disable it */
952 limits->max_discard_sectors = 0;
953 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
954}
955
Mike Snitzer4cc96132016-05-12 16:28:10 -0400956void disable_write_same(struct mapped_device *md)
Mike Snitzer7eee4ae2014-06-02 15:50:06 -0400957{
958 struct queue_limits *limits = dm_get_queue_limits(md);
959
960 /* device doesn't really support WRITE SAME, disable it */
961 limits->max_write_same_sectors = 0;
962}
963
Christoph Hellwigac62d622017-04-05 19:21:05 +0200964void disable_write_zeroes(struct mapped_device *md)
965{
966 struct queue_limits *limits = dm_get_queue_limits(md);
967
968 /* device doesn't really support WRITE ZEROES, disable it */
969 limits->max_write_zeroes_sectors = 0;
970}
971
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200972static void clone_endio(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973{
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200974 blk_status_t error = bio->bi_status;
Mikulas Patockabfc6d412014-03-04 18:24:49 -0500975 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
Milan Brozb35f8ca2009-03-16 17:44:36 +0000976 struct dm_io *io = tio->io;
Stefan Bader9faf4002006-10-03 01:15:41 -0700977 struct mapped_device *md = tio->io->md;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 dm_endio_fn endio = tio->ti->type->end_io;
Johannes Thumshirn415c79e2020-06-19 15:59:04 +0900979 struct bio *orig_bio = io->orig_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
Mike Snitzer9c37de22020-10-07 15:15:08 -0400981 if (unlikely(error == BLK_STS_TARGET)) {
Mike Snitzerbcb44432019-04-03 12:23:11 -0400982 if (bio_op(bio) == REQ_OP_DISCARD &&
983 !bio->bi_disk->queue->limits.max_discard_sectors)
984 disable_discard(md);
985 else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
986 !bio->bi_disk->queue->limits.max_write_same_sectors)
Christoph Hellwigac62d622017-04-05 19:21:05 +0200987 disable_write_same(md);
Mike Snitzerbcb44432019-04-03 12:23:11 -0400988 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
989 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
Christoph Hellwigac62d622017-04-05 19:21:05 +0200990 disable_write_zeroes(md);
991 }
Mike Snitzer7eee4ae2014-06-02 15:50:06 -0400992
Johannes Thumshirn415c79e2020-06-19 15:59:04 +0900993 /*
994 * For zone-append bios get offset in zone of the written
995 * sector and add that to the original bio sector pos.
996 */
997 if (bio_op(orig_bio) == REQ_OP_ZONE_APPEND) {
998 sector_t written_sector = bio->bi_iter.bi_sector;
999 struct request_queue *q = orig_bio->bi_disk->queue;
1000 u64 mask = (u64)blk_queue_zone_sectors(q) - 1;
1001
1002 orig_bio->bi_iter.bi_sector += written_sector & mask;
1003 }
1004
Christoph Hellwig1be56902017-06-03 09:38:03 +02001005 if (endio) {
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02001006 int r = endio(tio->ti, bio, &error);
Christoph Hellwig1be56902017-06-03 09:38:03 +02001007 switch (r) {
1008 case DM_ENDIO_REQUEUE:
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02001009 error = BLK_STS_DM_REQUEUE;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05001010 fallthrough;
Christoph Hellwig1be56902017-06-03 09:38:03 +02001011 case DM_ENDIO_DONE:
1012 break;
1013 case DM_ENDIO_INCOMPLETE:
1014 /* The target will handle the io */
1015 return;
1016 default:
1017 DMWARN("unimplemented target endio return value: %d", r);
1018 BUG();
1019 }
1020 }
1021
Mike Snitzercfae7522016-04-11 12:05:38 -04001022 free_tio(tio);
Milan Brozb35f8ca2009-03-16 17:44:36 +00001023 dec_pending(io, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024}
1025
Mike Snitzer78d8e582015-06-26 10:01:13 -04001026/*
Mike Snitzer56a67df2010-08-12 04:14:10 +01001027 * Return maximum size of I/O possible at the supplied sector up to the current
1028 * target boundary.
1029 */
Mike Snitzer37202812020-09-19 13:12:48 -04001030static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
1031 sector_t target_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032{
Mike Snitzer56a67df2010-08-12 04:14:10 +01001033 return ti->len - target_offset;
1034}
1035
Mike Snitzer37202812020-09-19 13:12:48 -04001036static sector_t max_io_len(struct dm_target *ti, sector_t sector)
Mike Snitzer56a67df2010-08-12 04:14:10 +01001037{
Mike Snitzer37202812020-09-19 13:12:48 -04001038 sector_t target_offset = dm_target_offset(ti, sector);
1039 sector_t len = max_io_len_target_boundary(ti, target_offset);
Mike Snitzer5091cde2020-09-18 20:22:30 -04001040 sector_t max_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
1042 /*
Mike Snitzer3ee16db2020-11-30 10:57:43 -05001043 * Does the target need to split IO even further?
1044 * - varied (per target) IO splitting is a tenet of DM; this
1045 * explains why stacked chunk_sectors based splitting via
1046 * blk_max_size_offset() isn't possible here. So pass in
1047 * ti->max_io_len to override stacked chunk_sectors.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 */
Mike Snitzer3ee16db2020-11-30 10:57:43 -05001049 if (ti->max_io_len) {
1050 max_len = blk_max_size_offset(ti->table->md->queue,
1051 target_offset, ti->max_io_len);
1052 if (len > max_len)
1053 len = max_len;
1054 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
1056 return len;
1057}
1058
Mike Snitzer542f9032012-07-27 15:08:00 +01001059int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1060{
1061 if (len > UINT_MAX) {
1062 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1063 (unsigned long long)len, UINT_MAX);
1064 ti->error = "Maximum size of target IO is too large";
1065 return -EINVAL;
1066 }
1067
Mikulas Patocka75ae1932019-03-21 16:46:12 -04001068 ti->max_io_len = (uint32_t) len;
Mike Snitzer542f9032012-07-27 15:08:00 +01001069
1070 return 0;
1071}
1072EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1073
Dan Williamsf26c5712017-04-12 12:35:44 -07001074static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
Mike Snitzer3d97c822018-04-30 16:06:28 -04001075 sector_t sector, int *srcu_idx)
1076 __acquires(md->io_barrier)
Toshi Kani545ed202016-06-22 17:54:53 -06001077{
Toshi Kani545ed202016-06-22 17:54:53 -06001078 struct dm_table *map;
1079 struct dm_target *ti;
Toshi Kani545ed202016-06-22 17:54:53 -06001080
Dan Williamsf26c5712017-04-12 12:35:44 -07001081 map = dm_get_live_table(md, srcu_idx);
Toshi Kani545ed202016-06-22 17:54:53 -06001082 if (!map)
Dan Williamsf26c5712017-04-12 12:35:44 -07001083 return NULL;
Toshi Kani545ed202016-06-22 17:54:53 -06001084
1085 ti = dm_table_find_target(map, sector);
Mikulas Patocka123d87d2019-08-23 09:55:26 -04001086 if (!ti)
Dan Williamsf26c5712017-04-12 12:35:44 -07001087 return NULL;
1088
1089 return ti;
1090}
1091
1092static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
Mike Snitzer3d97c822018-04-30 16:06:28 -04001093 long nr_pages, void **kaddr, pfn_t *pfn)
Dan Williamsf26c5712017-04-12 12:35:44 -07001094{
1095 struct mapped_device *md = dax_get_private(dax_dev);
1096 sector_t sector = pgoff * PAGE_SECTORS;
1097 struct dm_target *ti;
1098 long len, ret = -EIO;
1099 int srcu_idx;
1100
1101 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1102
1103 if (!ti)
Toshi Kani545ed202016-06-22 17:54:53 -06001104 goto out;
Dan Williamsf26c5712017-04-12 12:35:44 -07001105 if (!ti->type->direct_access)
1106 goto out;
Mike Snitzer37202812020-09-19 13:12:48 -04001107 len = max_io_len(ti, sector) / PAGE_SECTORS;
Dan Williamsf26c5712017-04-12 12:35:44 -07001108 if (len < 1)
1109 goto out;
1110 nr_pages = min(len, nr_pages);
Ross Zwislerdbc62652018-06-26 16:30:41 -06001111 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
Dan Williams817bf402017-04-12 13:37:44 -07001112
Dan Williamsf26c5712017-04-12 12:35:44 -07001113 out:
Toshi Kani545ed202016-06-22 17:54:53 -06001114 dm_put_live_table(md, srcu_idx);
Dan Williamsf26c5712017-04-12 12:35:44 -07001115
1116 return ret;
Toshi Kani545ed202016-06-22 17:54:53 -06001117}
1118
Dan Williams7bf7eac2019-05-16 13:26:29 -07001119static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
1120 int blocksize, sector_t start, sector_t len)
1121{
1122 struct mapped_device *md = dax_get_private(dax_dev);
1123 struct dm_table *map;
Dan Williams02186d82020-09-18 12:51:15 -07001124 bool ret = false;
Dan Williams7bf7eac2019-05-16 13:26:29 -07001125 int srcu_idx;
Dan Williams7bf7eac2019-05-16 13:26:29 -07001126
1127 map = dm_get_live_table(md, &srcu_idx);
1128 if (!map)
Dan Williams02186d82020-09-18 12:51:15 -07001129 goto out;
Dan Williams7bf7eac2019-05-16 13:26:29 -07001130
Pankaj Gupta2e9ee092019-07-05 19:33:25 +05301131 ret = dm_table_supports_dax(map, device_supports_dax, &blocksize);
Dan Williams7bf7eac2019-05-16 13:26:29 -07001132
Dan Williams02186d82020-09-18 12:51:15 -07001133out:
Dan Williams7bf7eac2019-05-16 13:26:29 -07001134 dm_put_live_table(md, srcu_idx);
1135
1136 return ret;
1137}
1138
Dan Williams7e026c82017-05-29 12:57:56 -07001139static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
Mike Snitzer3d97c822018-04-30 16:06:28 -04001140 void *addr, size_t bytes, struct iov_iter *i)
Dan Williams7e026c82017-05-29 12:57:56 -07001141{
1142 struct mapped_device *md = dax_get_private(dax_dev);
1143 sector_t sector = pgoff * PAGE_SECTORS;
1144 struct dm_target *ti;
1145 long ret = 0;
1146 int srcu_idx;
1147
1148 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1149
1150 if (!ti)
1151 goto out;
1152 if (!ti->type->dax_copy_from_iter) {
1153 ret = copy_from_iter(addr, bytes, i);
1154 goto out;
1155 }
1156 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1157 out:
1158 dm_put_live_table(md, srcu_idx);
1159
1160 return ret;
1161}
1162
Dan Williamsb3a9a0c2018-05-02 06:46:33 -07001163static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1164 void *addr, size_t bytes, struct iov_iter *i)
1165{
1166 struct mapped_device *md = dax_get_private(dax_dev);
1167 sector_t sector = pgoff * PAGE_SECTORS;
1168 struct dm_target *ti;
1169 long ret = 0;
1170 int srcu_idx;
1171
1172 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1173
1174 if (!ti)
1175 goto out;
1176 if (!ti->type->dax_copy_to_iter) {
1177 ret = copy_to_iter(addr, bytes, i);
1178 goto out;
1179 }
1180 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1181 out:
1182 dm_put_live_table(md, srcu_idx);
1183
1184 return ret;
1185}
1186
Vivek Goyalcdf6cdc2020-02-28 11:34:54 -05001187static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1188 size_t nr_pages)
1189{
1190 struct mapped_device *md = dax_get_private(dax_dev);
1191 sector_t sector = pgoff * PAGE_SECTORS;
1192 struct dm_target *ti;
1193 int ret = -EIO;
1194 int srcu_idx;
1195
1196 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1197
1198 if (!ti)
1199 goto out;
1200 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1201 /*
1202 * ->zero_page_range() is mandatory dax operation. If we are
1203 * here, something is wrong.
1204 */
Vivek Goyalcdf6cdc2020-02-28 11:34:54 -05001205 goto out;
1206 }
1207 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
Vivek Goyalcdf6cdc2020-02-28 11:34:54 -05001208 out:
1209 dm_put_live_table(md, srcu_idx);
1210
1211 return ret;
1212}
1213
Mikulas Patocka1dd40c32014-03-14 18:41:24 -04001214/*
1215 * A target may call dm_accept_partial_bio only from the map routine. It is
Ajay Joshi2e2d6f72019-10-27 23:05:48 +09001216 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_RESET,
1217 * REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE and REQ_OP_ZONE_FINISH.
Mikulas Patocka1dd40c32014-03-14 18:41:24 -04001218 *
1219 * dm_accept_partial_bio informs the dm that the target only wants to process
1220 * additional n_sectors sectors of the bio and the rest of the data should be
1221 * sent in a next bio.
1222 *
1223 * A diagram that explains the arithmetics:
1224 * +--------------------+---------------+-------+
1225 * | 1 | 2 | 3 |
1226 * +--------------------+---------------+-------+
1227 *
1228 * <-------------- *tio->len_ptr --------------->
1229 * <------- bi_size ------->
1230 * <-- n_sectors -->
1231 *
1232 * Region 1 was already iterated over with bio_advance or similar function.
1233 * (it may be empty if the target doesn't use bio_advance)
1234 * Region 2 is the remaining bio size that the target wants to process.
1235 * (it may be empty if region 1 is non-empty, although there is no reason
1236 * to make it empty)
1237 * The target requires that region 3 is to be sent in the next bio.
1238 *
1239 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1240 * the partially processed part (the sum of regions 1+2) must be the same for all
1241 * copies of the bio.
1242 */
1243void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1244{
1245 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1246 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
Jens Axboe1eff9d32016-08-05 15:35:16 -06001247 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
Mikulas Patocka1dd40c32014-03-14 18:41:24 -04001248 BUG_ON(bi_size > *tio->len_ptr);
1249 BUG_ON(n_sectors > bi_size);
1250 *tio->len_ptr -= bi_size - n_sectors;
1251 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1252}
1253EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1254
Mike Snitzer978e51b2017-12-09 15:16:42 -05001255static blk_qc_t __map_bio(struct dm_target_io *tio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256{
1257 int r;
Jens Axboe2056a782006-03-23 20:00:26 +01001258 sector_t sector;
Mikulas Patockadba14162012-10-12 21:02:15 +01001259 struct bio *clone = &tio->clone;
Mike Snitzer64f52b02017-12-11 23:17:47 -05001260 struct dm_io *io = tio->io;
Alasdair G Kergonbd2a49b2013-03-01 22:45:46 +00001261 struct dm_target *ti = tio->ti;
Mike Snitzer978e51b2017-12-09 15:16:42 -05001262 blk_qc_t ret = BLK_QC_T_NONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 clone->bi_end_io = clone_endio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265
1266 /*
1267 * Map the clone. If r == 0 we don't need to do
1268 * anything, the target has assumed ownership of
1269 * this io.
1270 */
Mike Snitzer64f52b02017-12-11 23:17:47 -05001271 atomic_inc(&io->io_count);
Kent Overstreet4f024f32013-10-11 15:44:27 -07001272 sector = clone->bi_iter.bi_sector;
Mikulas Patockad67a5f42017-02-15 11:26:10 -05001273
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00001274 r = ti->type->map(ti, clone);
Christoph Hellwig846785e2017-06-03 09:38:02 +02001275 switch (r) {
1276 case DM_MAPIO_SUBMITTED:
1277 break;
1278 case DM_MAPIO_REMAPPED:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 /* the bio has been remapped so dispatch it */
Christoph Hellwig1c02fca2020-12-03 17:21:38 +01001280 trace_block_bio_remap(clone, bio_dev(io->orig_bio), sector);
Christoph Hellwig5a6c35f2020-07-01 10:59:47 +02001281 ret = submit_bio_noacct(clone);
Christoph Hellwig846785e2017-06-03 09:38:02 +02001282 break;
1283 case DM_MAPIO_KILL:
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02001284 free_tio(tio);
Mike Snitzer64f52b02017-12-11 23:17:47 -05001285 dec_pending(io, BLK_STS_IOERR);
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02001286 break;
Christoph Hellwig846785e2017-06-03 09:38:02 +02001287 case DM_MAPIO_REQUEUE:
Mike Snitzercfae7522016-04-11 12:05:38 -04001288 free_tio(tio);
Mike Snitzer64f52b02017-12-11 23:17:47 -05001289 dec_pending(io, BLK_STS_DM_REQUEUE);
Christoph Hellwig846785e2017-06-03 09:38:02 +02001290 break;
1291 default:
Kiyoshi Ueda45cbcd72006-12-08 02:41:05 -08001292 DMWARN("unimplemented target map return value: %d", r);
1293 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
Mike Snitzer978e51b2017-12-09 15:16:42 -05001296 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298
Mikulas Patockae0d66092014-03-14 18:40:39 -04001299static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
Alasdair G Kergonbd2a49b2013-03-01 22:45:46 +00001300{
Kent Overstreet4f024f32013-10-11 15:44:27 -07001301 bio->bi_iter.bi_sector = sector;
1302 bio->bi_iter.bi_size = to_bytes(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303}
1304
1305/*
1306 * Creates a bio that consists of range of complete bvecs.
1307 */
Mike Snitzerc80914e2016-03-02 12:33:03 -05001308static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1309 sector_t sector, unsigned len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310{
Mikulas Patockadba14162012-10-12 21:02:15 +01001311 struct bio *clone = &tio->clone;
Eric Biggers07560152020-09-15 20:53:13 -07001312 int r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313
Kent Overstreet1c3b13e2013-10-29 17:17:49 -07001314 __bio_clone_fast(clone, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315
Eric Biggers07560152020-09-15 20:53:13 -07001316 r = bio_crypt_clone(clone, bio, GFP_NOIO);
1317 if (r < 0)
1318 return r;
Satya Tangiralaa892c8d2020-05-14 00:37:18 +00001319
Mike Snitzer57c36512019-01-16 18:53:26 -05001320 if (bio_integrity(bio)) {
Mikulas Patockae2460f22017-04-18 16:51:48 -04001321 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1322 !dm_target_passes_integrity(tio->ti->type))) {
1323 DMWARN("%s: the target %s doesn't support integrity data.",
1324 dm_device_name(tio->io->md),
1325 tio->ti->type->name);
1326 return -EIO;
1327 }
1328
1329 r = bio_integrity_clone(clone, bio, GFP_NOIO);
Mike Snitzerc80914e2016-03-02 12:33:03 -05001330 if (r < 0)
1331 return r;
1332 }
Kent Overstreet1c3b13e2013-10-29 17:17:49 -07001333
Mike Snitzerfa8db492019-02-05 17:07:58 -05001334 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1335 clone->bi_iter.bi_size = to_bytes(len);
1336
1337 if (bio_integrity(bio))
1338 bio_integrity_trim(clone);
Mike Snitzerc80914e2016-03-02 12:33:03 -05001339
1340 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341}
1342
Mike Snitzer318716d2017-11-22 14:56:12 -05001343static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1344 struct dm_target *ti, unsigned num_bios)
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001345{
Mikulas Patockadba14162012-10-12 21:02:15 +01001346 struct dm_target_io *tio;
Mike Snitzer318716d2017-11-22 14:56:12 -05001347 int try;
Mikulas Patockadba14162012-10-12 21:02:15 +01001348
Mike Snitzer318716d2017-11-22 14:56:12 -05001349 if (!num_bios)
1350 return;
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001351
Mike Snitzer318716d2017-11-22 14:56:12 -05001352 if (num_bios == 1) {
1353 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1354 bio_list_add(blist, &tio->clone);
1355 return;
1356 }
Alasdair G Kergon9015df22009-06-22 10:12:21 +01001357
Mike Snitzer318716d2017-11-22 14:56:12 -05001358 for (try = 0; try < 2; try++) {
1359 int bio_nr;
1360 struct bio *bio;
1361
1362 if (try)
Mike Snitzerbc02cdb2017-12-14 16:30:42 -05001363 mutex_lock(&ci->io->md->table_devices_lock);
Mike Snitzer318716d2017-11-22 14:56:12 -05001364 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1365 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1366 if (!tio)
1367 break;
1368
1369 bio_list_add(blist, &tio->clone);
1370 }
1371 if (try)
Mike Snitzerbc02cdb2017-12-14 16:30:42 -05001372 mutex_unlock(&ci->io->md->table_devices_lock);
Mike Snitzer318716d2017-11-22 14:56:12 -05001373 if (bio_nr == num_bios)
1374 return;
1375
1376 while ((bio = bio_list_pop(blist))) {
1377 tio = container_of(bio, struct dm_target_io, clone);
1378 free_tio(tio);
1379 }
1380 }
Alasdair G Kergon9015df22009-06-22 10:12:21 +01001381}
1382
Mike Snitzer978e51b2017-12-09 15:16:42 -05001383static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1384 struct dm_target_io *tio, unsigned *len)
Alasdair G Kergon9015df22009-06-22 10:12:21 +01001385{
Mikulas Patockadba14162012-10-12 21:02:15 +01001386 struct bio *clone = &tio->clone;
Alasdair G Kergon9015df22009-06-22 10:12:21 +01001387
Mikulas Patocka1dd40c32014-03-14 18:41:24 -04001388 tio->len_ptr = len;
1389
Junichi Nomura99778272014-10-03 11:55:16 +00001390 __bio_clone_fast(clone, ci->bio);
Alasdair G Kergonbd2a49b2013-03-01 22:45:46 +00001391 if (len)
Mikulas Patocka1dd40c32014-03-14 18:41:24 -04001392 bio_setup_sector(clone, ci->sector, *len);
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001393
Mike Snitzer978e51b2017-12-09 15:16:42 -05001394 return __map_bio(tio);
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001395}
1396
Alasdair G Kergon14fe5942013-03-01 22:45:47 +00001397static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
Mikulas Patocka1dd40c32014-03-14 18:41:24 -04001398 unsigned num_bios, unsigned *len)
Mike Snitzer06a426c2010-08-12 04:14:09 +01001399{
Mike Snitzer318716d2017-11-22 14:56:12 -05001400 struct bio_list blist = BIO_EMPTY_LIST;
1401 struct bio *bio;
1402 struct dm_target_io *tio;
Mike Snitzer06a426c2010-08-12 04:14:09 +01001403
Mike Snitzer318716d2017-11-22 14:56:12 -05001404 alloc_multiple_bios(&blist, ci, ti, num_bios);
1405
1406 while ((bio = bio_list_pop(&blist))) {
1407 tio = container_of(bio, struct dm_target_io, clone);
Mike Snitzer978e51b2017-12-09 15:16:42 -05001408 (void) __clone_and_map_simple_bio(ci, tio, len);
Mike Snitzer318716d2017-11-22 14:56:12 -05001409 }
Mike Snitzer06a426c2010-08-12 04:14:09 +01001410}
1411
Alasdair G Kergon14fe5942013-03-01 22:45:47 +00001412static int __send_empty_flush(struct clone_info *ci)
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001413{
Mike Snitzer06a426c2010-08-12 04:14:09 +01001414 unsigned target_nr = 0;
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001415 struct dm_target *ti;
Mike Snitzer828678b2020-09-14 13:59:53 -04001416 struct bio flush_bio;
1417
1418 /*
1419 * Use an on-stack bio for this, it's safe since we don't
1420 * need to reference it after submit. It's just used as
1421 * the basis for the clone(s).
1422 */
1423 bio_init(&flush_bio, NULL, 0);
1424 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
Christoph Hellwig47d95102020-11-07 09:30:05 +01001425 flush_bio.bi_disk = ci->io->md->disk;
1426 bio_associate_blkg(&flush_bio);
1427
Mike Snitzer828678b2020-09-14 13:59:53 -04001428 ci->bio = &flush_bio;
1429 ci->sector_count = 0;
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001430
Mike Snitzerb372d362010-09-08 18:07:01 +02001431 BUG_ON(bio_has_data(ci->bio));
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001432 while ((ti = dm_table_get_target(ci->map, target_nr++)))
Mikulas Patocka1dd40c32014-03-14 18:41:24 -04001433 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
Mike Snitzer828678b2020-09-14 13:59:53 -04001434
1435 bio_uninit(ci->bio);
Mikulas Patockaf9ab94c2009-06-22 10:12:20 +01001436 return 0;
1437}
1438
Mike Snitzerc80914e2016-03-02 12:33:03 -05001439static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
NeilBrownf31c21e2017-11-22 14:25:18 +11001440 sector_t sector, unsigned *len)
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001441{
Mikulas Patockadba14162012-10-12 21:02:15 +01001442 struct bio *bio = ci->bio;
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001443 struct dm_target_io *tio;
NeilBrownf31c21e2017-11-22 14:25:18 +11001444 int r;
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001445
Mike Snitzer318716d2017-11-22 14:56:12 -05001446 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
NeilBrownf31c21e2017-11-22 14:25:18 +11001447 tio->len_ptr = len;
1448 r = clone_bio(tio, bio, sector, *len);
1449 if (r < 0) {
1450 free_tio(tio);
1451 return r;
Alasdair G Kergonb0d8ed42013-03-01 22:45:49 +00001452 }
Mike Snitzer978e51b2017-12-09 15:16:42 -05001453 (void) __map_bio(tio);
Mike Snitzerc80914e2016-03-02 12:33:03 -05001454
NeilBrownf31c21e2017-11-22 14:25:18 +11001455 return 0;
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001456}
1457
Mike Snitzer3d7f4562017-12-08 15:02:11 -05001458static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
Mike Snitzer61697a62019-01-18 14:19:26 -05001459 unsigned num_bios)
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001460{
Michael Lass51b86f92019-05-21 21:58:07 +02001461 unsigned len;
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001462
Mike Snitzer3d7f4562017-12-08 15:02:11 -05001463 /*
1464 * Even though the device advertised support for this type of
1465 * request, that does not mean every target supports it, and
1466 * reconfiguration might also have changed that since the
1467 * check was performed.
1468 */
Mike Snitzer3d7f4562017-12-08 15:02:11 -05001469 if (!num_bios)
1470 return -EOPNOTSUPP;
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001471
Mike Snitzer37202812020-09-19 13:12:48 -04001472 len = min_t(sector_t, ci->sector_count,
1473 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
Michael Lass51b86f92019-05-21 21:58:07 +02001474
Mike Snitzer3d7f4562017-12-08 15:02:11 -05001475 __send_duplicate_bios(ci, ti, num_bios, &len);
Mike Snitzer06a426c2010-08-12 04:14:09 +01001476
Mike Snitzer3d7f4562017-12-08 15:02:11 -05001477 ci->sector += len;
1478 ci->sector_count -= len;
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001479
1480 return 0;
1481}
1482
Mike Snitzer568c73a2019-01-18 14:10:37 -05001483static bool is_abnormal_io(struct bio *bio)
1484{
1485 bool r = false;
1486
1487 switch (bio_op(bio)) {
1488 case REQ_OP_DISCARD:
1489 case REQ_OP_SECURE_ERASE:
1490 case REQ_OP_WRITE_SAME:
1491 case REQ_OP_WRITE_ZEROES:
1492 r = true;
1493 break;
1494 }
1495
1496 return r;
1497}
1498
Mike Snitzer0519c712018-03-26 11:49:16 -04001499static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1500 int *result)
1501{
1502 struct bio *bio = ci->bio;
Mike Snitzer9679b5a2020-09-15 21:56:29 -04001503 unsigned num_bios = 0;
Mike Snitzer0519c712018-03-26 11:49:16 -04001504
Mike Snitzer9679b5a2020-09-15 21:56:29 -04001505 switch (bio_op(bio)) {
1506 case REQ_OP_DISCARD:
1507 num_bios = ti->num_discard_bios;
1508 break;
1509 case REQ_OP_SECURE_ERASE:
1510 num_bios = ti->num_secure_erase_bios;
1511 break;
1512 case REQ_OP_WRITE_SAME:
1513 num_bios = ti->num_write_same_bios;
1514 break;
1515 case REQ_OP_WRITE_ZEROES:
1516 num_bios = ti->num_write_zeroes_bios;
1517 break;
1518 default:
Mike Snitzer0519c712018-03-26 11:49:16 -04001519 return false;
Mike Snitzer9679b5a2020-09-15 21:56:29 -04001520 }
Mike Snitzer0519c712018-03-26 11:49:16 -04001521
Mike Snitzer9679b5a2020-09-15 21:56:29 -04001522 *result = __send_changing_extent_only(ci, ti, num_bios);
Mike Snitzer0519c712018-03-26 11:49:16 -04001523 return true;
1524}
1525
Alasdair G Kergone4c93812013-03-01 22:45:47 +00001526/*
Alasdair G Kergone4c93812013-03-01 22:45:47 +00001527 * Select the correct strategy for processing a non-flush bio.
1528 */
Alasdair G Kergon14fe5942013-03-01 22:45:47 +00001529static int __split_and_process_non_flush(struct clone_info *ci)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530{
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001531 struct dm_target *ti;
Kent Overstreet1c3b13e2013-10-29 17:17:49 -07001532 unsigned len;
Mike Snitzerc80914e2016-03-02 12:33:03 -05001533 int r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001535 ti = dm_table_find_target(ci->map, ci->sector);
Mikulas Patocka123d87d2019-08-23 09:55:26 -04001536 if (!ti)
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001537 return -EIO;
1538
Mike Snitzer568c73a2019-01-18 14:10:37 -05001539 if (__process_abnormal_io(ci, ti, &r))
Mike Snitzer0519c712018-03-26 11:49:16 -04001540 return r;
Mike Snitzer3d7f4562017-12-08 15:02:11 -05001541
Mike Snitzer37202812020-09-19 13:12:48 -04001542 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001543
Mike Snitzerc80914e2016-03-02 12:33:03 -05001544 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1545 if (r < 0)
1546 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
Kent Overstreet1c3b13e2013-10-29 17:17:49 -07001548 ci->sector += len;
1549 ci->sector_count -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550
Kent Overstreet1c3b13e2013-10-29 17:17:49 -07001551 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552}
1553
Mike Snitzer978e51b2017-12-09 15:16:42 -05001554static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1555 struct dm_table *map, struct bio *bio)
1556{
1557 ci->map = map;
1558 ci->io = alloc_io(md, bio);
1559 ci->sector = bio->bi_iter.bi_sector;
1560}
1561
Mike Snitzera1e1cb72019-01-17 10:48:01 -05001562#define __dm_part_stat_sub(part, field, subnd) \
1563 (part_stat_get(part, field) -= (subnd))
1564
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565/*
Alasdair G Kergon14fe5942013-03-01 22:45:47 +00001566 * Entry point to split a bio into clones and submit them to the targets.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 */
Mike Snitzer978e51b2017-12-09 15:16:42 -05001568static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1569 struct dm_table *map, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570{
1571 struct clone_info ci;
Mike Snitzer978e51b2017-12-09 15:16:42 -05001572 blk_qc_t ret = BLK_QC_T_NONE;
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001573 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574
Mike Snitzer978e51b2017-12-09 15:16:42 -05001575 init_clone_info(&ci, md, map, bio);
Alasdair G Kergonbd2a49b2013-03-01 22:45:46 +00001576
Jens Axboe1eff9d32016-08-05 15:35:16 -06001577 if (bio->bi_opf & REQ_PREFLUSH) {
Alasdair G Kergon14fe5942013-03-01 22:45:47 +00001578 error = __send_empty_flush(&ci);
Mike Snitzerb372d362010-09-08 18:07:01 +02001579 /* dec_pending submits any data associated with flush */
Ajay Joshi2e2d6f72019-10-27 23:05:48 +09001580 } else if (op_is_zone_mgmt(bio_op(bio))) {
Damien Le Moala4aa5e52017-05-08 16:40:46 -07001581 ci.bio = bio;
1582 ci.sector_count = 0;
1583 error = __split_and_process_non_flush(&ci);
Mike Snitzerb372d362010-09-08 18:07:01 +02001584 } else {
Tejun Heo6a8736d2010-09-08 18:07:00 +02001585 ci.bio = bio;
Tejun Heod87f4c12010-09-03 11:56:19 +02001586 ci.sector_count = bio_sectors(bio);
NeilBrown18a25da2017-09-06 09:43:28 +10001587 while (ci.sector_count && !error) {
Alasdair G Kergon14fe5942013-03-01 22:45:47 +00001588 error = __split_and_process_non_flush(&ci);
Jeffle Xu985eabd2020-11-04 11:57:56 +08001589 if (ci.sector_count && !error) {
NeilBrown18a25da2017-09-06 09:43:28 +10001590 /*
Christoph Hellwiged00aab2020-07-01 10:59:44 +02001591 * Remainder must be passed to submit_bio_noacct()
NeilBrown18a25da2017-09-06 09:43:28 +10001592 * so that it gets handled *after* bios already submitted
1593 * have been completely processed.
1594 * We take a clone of the original to store in
Mike Snitzer745dc572017-12-11 20:51:50 -05001595 * ci.io->orig_bio to be used by end_io_acct() and
NeilBrown18a25da2017-09-06 09:43:28 +10001596 * for dec_pending to use for completion handling.
NeilBrown18a25da2017-09-06 09:43:28 +10001597 */
Mike Snitzerf21c6012018-06-15 09:35:33 -04001598 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1599 GFP_NOIO, &md->queue->bio_split);
Mike Snitzer745dc572017-12-11 20:51:50 -05001600 ci.io->orig_bio = b;
Mike Snitzera1e1cb72019-01-17 10:48:01 -05001601
1602 /*
1603 * Adjust IO stats for each split, otherwise upon queue
1604 * reentry there will be redundant IO accounting.
1605 * NOTE: this is a stop-gap fix, a proper fix involves
1606 * significant refactoring of DM core's bio splitting
1607 * (by eliminating DM's splitting and just using bio_split)
1608 */
1609 part_stat_lock();
Christoph Hellwig8446fe92020-11-24 09:36:54 +01001610 __dm_part_stat_sub(dm_disk(md)->part0,
Mike Snitzera1e1cb72019-01-17 10:48:01 -05001611 sectors[op_stat_group(bio_op(bio))], ci.sector_count);
1612 part_stat_unlock();
1613
NeilBrown18a25da2017-09-06 09:43:28 +10001614 bio_chain(b, bio);
Christoph Hellwigeb6f7f72020-12-03 17:21:37 +01001615 trace_block_split(b, bio->bi_iter.bi_sector);
Christoph Hellwiged00aab2020-07-01 10:59:44 +02001616 ret = submit_bio_noacct(bio);
NeilBrown18a25da2017-09-06 09:43:28 +10001617 break;
1618 }
1619 }
Tejun Heod87f4c12010-09-03 11:56:19 +02001620 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621
1622 /* drop the extra reference count */
Bart Van Assche54385bf2017-08-09 11:32:10 -07001623 dec_pending(ci.io, errno_to_blk_status(error));
Mike Snitzer978e51b2017-12-09 15:16:42 -05001624 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626
Christoph Hellwigc62b37d2020-07-01 10:59:43 +02001627static blk_qc_t dm_submit_bio(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628{
Christoph Hellwigc4a59c42020-07-01 10:59:37 +02001629 struct mapped_device *md = bio->bi_disk->private_data;
Mike Snitzer978e51b2017-12-09 15:16:42 -05001630 blk_qc_t ret = BLK_QC_T_NONE;
Mikulas Patocka83d5e5b2013-07-10 23:41:18 +01001631 int srcu_idx;
1632 struct dm_table *map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633
Mikulas Patocka83d5e5b2013-07-10 23:41:18 +01001634 map = dm_get_live_table(md, &srcu_idx);
Mike Snitzerb2abdb12020-09-30 13:45:20 -04001635 if (unlikely(!map)) {
1636 DMERR_LIMIT("%s: mapping table unavailable, erroring io",
1637 dm_device_name(md));
1638 bio_io_error(bio);
1639 goto out;
Tejun Heo6a8736d2010-09-08 18:07:00 +02001640 }
1641
Mike Snitzerb2abdb12020-09-30 13:45:20 -04001642 /* If suspended, queue this IO for later */
Mikulas Patockaaf7e4662009-04-09 00:27:16 +01001643 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
Konstantin Khlebnikov6abc4942020-09-23 16:06:52 -04001644 if (bio->bi_opf & REQ_NOWAIT)
1645 bio_wouldblock_error(bio);
Mike Snitzerb2abdb12020-09-30 13:45:20 -04001646 else if (bio->bi_opf & REQ_RAHEAD)
Alasdair G Kergon54d9a1b2009-04-09 00:27:14 +01001647 bio_io_error(bio);
Mike Snitzerb2abdb12020-09-30 13:45:20 -04001648 else
1649 queue_io(md, bio);
1650 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 }
1652
Mike Snitzerb2abdb12020-09-30 13:45:20 -04001653 /*
1654 * Use blk_queue_split() for abnormal IO (e.g. discard, writesame, etc)
1655 * otherwise associated queue_limits won't be imposed.
1656 */
1657 if (is_abnormal_io(bio))
1658 blk_queue_split(&bio);
Mike Snitzer978e51b2017-12-09 15:16:42 -05001659
Mike Snitzer9c37de22020-10-07 15:15:08 -04001660 ret = __split_and_process_bio(md, map, bio);
Mike Snitzerb2abdb12020-09-30 13:45:20 -04001661out:
Mikulas Patocka83d5e5b2013-07-10 23:41:18 +01001662 dm_put_live_table(md, srcu_idx);
Mike Snitzer978e51b2017-12-09 15:16:42 -05001663 return ret;
1664}
1665
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666/*-----------------------------------------------------------------
1667 * An IDR is used to keep track of allocated minor numbers.
1668 *---------------------------------------------------------------*/
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07001669static void free_minor(int minor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670{
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001671 spin_lock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 idr_remove(&_minor_idr, minor);
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001673 spin_unlock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674}
1675
1676/*
1677 * See if the device with a specific minor # is free.
1678 */
Frederik Deweerdtcf13ab82008-04-24 22:10:59 +01001679static int specific_minor(int minor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680{
Tejun Heoc9d76be2013-02-27 17:04:26 -08001681 int r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682
1683 if (minor >= (1 << MINORBITS))
1684 return -EINVAL;
1685
Tejun Heoc9d76be2013-02-27 17:04:26 -08001686 idr_preload(GFP_KERNEL);
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001687 spin_lock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688
Tejun Heoc9d76be2013-02-27 17:04:26 -08001689 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001691 spin_unlock(&_minor_lock);
Tejun Heoc9d76be2013-02-27 17:04:26 -08001692 idr_preload_end();
1693 if (r < 0)
1694 return r == -ENOSPC ? -EBUSY : r;
1695 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696}
1697
Frederik Deweerdtcf13ab82008-04-24 22:10:59 +01001698static int next_free_minor(int *minor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699{
Tejun Heoc9d76be2013-02-27 17:04:26 -08001700 int r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701
Tejun Heoc9d76be2013-02-27 17:04:26 -08001702 idr_preload(GFP_KERNEL);
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001703 spin_lock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704
Tejun Heoc9d76be2013-02-27 17:04:26 -08001705 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001707 spin_unlock(&_minor_lock);
Tejun Heoc9d76be2013-02-27 17:04:26 -08001708 idr_preload_end();
1709 if (r < 0)
1710 return r;
1711 *minor = r;
1712 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713}
1714
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001715static const struct block_device_operations dm_blk_dops;
Mike Snitzer681cc5e2020-10-07 16:41:01 -04001716static const struct block_device_operations dm_rq_blk_dops;
Dan Williamsf26c5712017-04-12 12:35:44 -07001717static const struct dax_operations dm_dax_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718
Mikulas Patocka53d59142009-04-02 19:55:37 +01001719static void dm_wq_work(struct work_struct *work);
1720
Mike Snitzer0f209722015-04-28 11:50:29 -04001721static void cleanup_mapped_device(struct mapped_device *md)
1722{
Mike Snitzer0f209722015-04-28 11:50:29 -04001723 if (md->wq)
1724 destroy_workqueue(md->wq);
Kent Overstreet6f1c8192018-05-20 18:25:53 -04001725 bioset_exit(&md->bs);
1726 bioset_exit(&md->io_bs);
Mike Snitzer0f209722015-04-28 11:50:29 -04001727
Dan Williamsf26c5712017-04-12 12:35:44 -07001728 if (md->dax_dev) {
1729 kill_dax(md->dax_dev);
1730 put_dax(md->dax_dev);
1731 md->dax_dev = NULL;
1732 }
1733
Mike Snitzer0f209722015-04-28 11:50:29 -04001734 if (md->disk) {
1735 spin_lock(&_minor_lock);
1736 md->disk->private_data = NULL;
1737 spin_unlock(&_minor_lock);
Mike Snitzer0f209722015-04-28 11:50:29 -04001738 del_gendisk(md->disk);
1739 put_disk(md->disk);
1740 }
1741
1742 if (md->queue)
1743 blk_cleanup_queue(md->queue);
1744
Tahsin Erdogand09960b2016-10-10 05:35:19 -07001745 cleanup_srcu_struct(&md->io_barrier);
1746
Mike Snitzerd5ffebd2018-01-05 21:17:20 -05001747 mutex_destroy(&md->suspend_lock);
1748 mutex_destroy(&md->type_lock);
1749 mutex_destroy(&md->table_devices_lock);
1750
Mike Snitzer4cc96132016-05-12 16:28:10 -04001751 dm_mq_cleanup_mapped_device(md);
Mike Snitzer0f209722015-04-28 11:50:29 -04001752}
1753
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754/*
1755 * Allocate and initialise a blank device with a given minor.
1756 */
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07001757static struct mapped_device *alloc_dev(int minor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758{
Mike Snitzer115485e2016-02-22 12:16:21 -05001759 int r, numa_node_id = dm_get_numa_node();
1760 struct mapped_device *md;
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -07001761 void *old_md;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762
Mikulas Patocka856eb092017-10-31 19:33:02 -04001763 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 if (!md) {
1765 DMWARN("unable to allocate device, out of memory.");
1766 return NULL;
1767 }
1768
Jeff Mahoney10da4f72006-06-26 00:27:25 -07001769 if (!try_module_get(THIS_MODULE))
Milan Broz6ed7ade2008-02-08 02:10:19 +00001770 goto bad_module_get;
Jeff Mahoney10da4f72006-06-26 00:27:25 -07001771
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 /* get a minor number for the dev */
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07001773 if (minor == DM_ANY_MINOR)
Frederik Deweerdtcf13ab82008-04-24 22:10:59 +01001774 r = next_free_minor(&minor);
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07001775 else
Frederik Deweerdtcf13ab82008-04-24 22:10:59 +01001776 r = specific_minor(minor);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 if (r < 0)
Milan Broz6ed7ade2008-02-08 02:10:19 +00001778 goto bad_minor;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779
Mikulas Patocka83d5e5b2013-07-10 23:41:18 +01001780 r = init_srcu_struct(&md->io_barrier);
1781 if (r < 0)
1782 goto bad_io_barrier;
1783
Mike Snitzer115485e2016-02-22 12:16:21 -05001784 md->numa_node_id = numa_node_id;
Mike Snitzer591ddcfc2016-01-31 12:05:42 -05001785 md->init_tio_pdu = false;
Mike Snitzera5664da2010-08-12 04:14:01 +01001786 md->type = DM_TYPE_NONE;
Daniel Walkere61290a2008-02-08 02:10:08 +00001787 mutex_init(&md->suspend_lock);
Mike Snitzera5664da2010-08-12 04:14:01 +01001788 mutex_init(&md->type_lock);
Benjamin Marzinski86f11522014-08-13 13:53:43 -05001789 mutex_init(&md->table_devices_lock);
Mikulas Patocka022c2612009-04-02 19:55:39 +01001790 spin_lock_init(&md->deferred_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 atomic_set(&md->holders, 1);
Alasdair G Kergon5c6bd752006-06-26 00:27:34 -07001792 atomic_set(&md->open_count, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 atomic_set(&md->event_nr, 0);
Mike Anderson7a8c3d32007-10-19 22:48:01 +01001794 atomic_set(&md->uevent_seq, 0);
1795 INIT_LIST_HEAD(&md->uevent_list);
Benjamin Marzinski86f11522014-08-13 13:53:43 -05001796 INIT_LIST_HEAD(&md->table_devices);
Mike Anderson7a8c3d32007-10-19 22:48:01 +01001797 spin_lock_init(&md->uevent_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798
Mike Snitzer47ace7e2020-01-27 14:07:23 -05001799 /*
Christoph Hellwigc62b37d2020-07-01 10:59:43 +02001800 * default to bio-based until DM table is loaded and md->type
1801 * established. If request-based table is loaded: blk-mq will
1802 * override accordingly.
Mike Snitzer47ace7e2020-01-27 14:07:23 -05001803 */
Christoph Hellwigc62b37d2020-07-01 10:59:43 +02001804 md->queue = blk_alloc_queue(numa_node_id);
Christoph Hellwig3d745ea2020-03-27 09:30:11 +01001805 if (!md->queue)
1806 goto bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807
Mike Snitzerc12c9a32018-01-12 09:32:21 -05001808 md->disk = alloc_disk_node(1, md->numa_node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 if (!md->disk)
Mike Snitzer0f209722015-04-28 11:50:29 -04001810 goto bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811
Jeff Mahoneyf0b04112006-06-26 00:27:25 -07001812 init_waitqueue_head(&md->wait);
Mikulas Patocka53d59142009-04-02 19:55:37 +01001813 INIT_WORK(&md->work, dm_wq_work);
Jeff Mahoneyf0b04112006-06-26 00:27:25 -07001814 init_waitqueue_head(&md->eventq);
Mikulas Patocka2995fa72014-01-13 19:37:54 -05001815 init_completion(&md->kobj_holder.completion);
Jeff Mahoneyf0b04112006-06-26 00:27:25 -07001816
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 md->disk->major = _major;
1818 md->disk->first_minor = minor;
1819 md->disk->fops = &dm_blk_dops;
1820 md->disk->queue = md->queue;
1821 md->disk->private_data = md;
1822 sprintf(md->disk->disk_name, "dm-%d", minor);
Dan Williamsf26c5712017-04-12 12:35:44 -07001823
Dan Williams976431b2018-03-29 17:22:13 -07001824 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
Pankaj Guptafefc1d92019-07-05 19:33:24 +05301825 md->dax_dev = alloc_dax(md, md->disk->disk_name,
1826 &dm_dax_ops, 0);
Vivek Goyal4e4ced92020-04-01 12:11:25 -04001827 if (IS_ERR(md->dax_dev))
Dan Williams976431b2018-03-29 17:22:13 -07001828 goto bad;
1829 }
Dan Williamsf26c5712017-04-12 12:35:44 -07001830
Mike Snitzerc100ec42018-01-08 20:03:04 -05001831 add_disk_no_queue_reg(md->disk);
Mike Anderson7e51f252006-03-27 01:17:52 -08001832 format_dev_t(md->name, MKDEV(_major, minor));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833
Tejun Heo670368a2013-07-30 08:40:21 -04001834 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
Milan Broz304f3f62008-02-08 02:11:17 +00001835 if (!md->wq)
Mike Snitzer0f209722015-04-28 11:50:29 -04001836 goto bad;
Milan Broz304f3f62008-02-08 02:11:17 +00001837
Mikulas Patockafd2ed4d2013-08-16 10:54:23 -04001838 dm_stats_init(&md->stats);
1839
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -07001840 /* Populate the mapping, nobody knows we exist yet */
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001841 spin_lock(&_minor_lock);
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -07001842 old_md = idr_replace(&_minor_idr, md, minor);
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07001843 spin_unlock(&_minor_lock);
Jeff Mahoneyba61fdd2006-06-26 00:27:21 -07001844
1845 BUG_ON(old_md != MINOR_ALLOCED);
1846
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 return md;
1848
Mike Snitzer0f209722015-04-28 11:50:29 -04001849bad:
1850 cleanup_mapped_device(md);
Mikulas Patocka83d5e5b2013-07-10 23:41:18 +01001851bad_io_barrier:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 free_minor(minor);
Milan Broz6ed7ade2008-02-08 02:10:19 +00001853bad_minor:
Jeff Mahoney10da4f72006-06-26 00:27:25 -07001854 module_put(THIS_MODULE);
Milan Broz6ed7ade2008-02-08 02:10:19 +00001855bad_module_get:
Mikulas Patocka856eb092017-10-31 19:33:02 -04001856 kvfree(md);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 return NULL;
1858}
1859
Jun'ichi Nomuraae9da832007-10-19 22:38:43 +01001860static void unlock_fs(struct mapped_device *md);
1861
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862static void free_dev(struct mapped_device *md)
1863{
Tejun Heof331c022008-09-03 09:01:48 +02001864 int minor = MINOR(disk_devt(md->disk));
Jun'ichi Nomura63d94e42006-02-24 13:04:25 -08001865
Mikulas Patocka32a926d2009-06-22 10:12:17 +01001866 unlock_fs(md);
Keith Busch2eb6e1e2014-10-17 17:46:36 -06001867
Mike Snitzer0f209722015-04-28 11:50:29 -04001868 cleanup_mapped_device(md);
Mike Snitzer0f209722015-04-28 11:50:29 -04001869
1870 free_table_devices(&md->table_devices);
1871 dm_stats_cleanup(&md->stats);
Mike Snitzer63a4f062015-03-23 17:01:43 -04001872 free_minor(minor);
1873
Jeff Mahoney10da4f72006-06-26 00:27:25 -07001874 module_put(THIS_MODULE);
Mikulas Patocka856eb092017-10-31 19:33:02 -04001875 kvfree(md);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876}
1877
Jens Axboe2a2a4c52018-06-07 14:42:06 -06001878static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001879{
Mikulas Patockac0820cf2012-12-21 20:23:38 +00001880 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
Jens Axboe2a2a4c52018-06-07 14:42:06 -06001881 int ret = 0;
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001882
Mike Snitzer0776aa02017-12-08 14:40:52 -05001883 if (dm_table_bio_based(t)) {
Mike Snitzer64f52b02017-12-11 23:17:47 -05001884 /*
1885 * The md may already have mempools that need changing.
1886 * If so, reload bioset because front_pad may have changed
1887 * because a different table was loaded.
1888 */
Kent Overstreet6f1c8192018-05-20 18:25:53 -04001889 bioset_exit(&md->bs);
1890 bioset_exit(&md->io_bs);
Mike Snitzer0776aa02017-12-08 14:40:52 -05001891
Kent Overstreet6f1c8192018-05-20 18:25:53 -04001892 } else if (bioset_initialized(&md->bs)) {
Mike Snitzer4e6e36c2015-06-26 09:42:57 -04001893 /*
1894 * There's no need to reload with request-based dm
1895 * because the size of front_pad doesn't change.
1896 * Note for future: If you are to reload bioset,
1897 * prep-ed requests in the queue may refer
1898 * to bio from the old bioset, so you must walk
1899 * through the queue to unprep.
1900 */
1901 goto out;
Mikulas Patockac0820cf2012-12-21 20:23:38 +00001902 }
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001903
Kent Overstreet6f1c8192018-05-20 18:25:53 -04001904 BUG_ON(!p ||
1905 bioset_initialized(&md->bs) ||
1906 bioset_initialized(&md->io_bs));
Mike Snitzercbc4e3c2015-04-27 16:37:50 -04001907
Jens Axboe2a2a4c52018-06-07 14:42:06 -06001908 ret = bioset_init_from_src(&md->bs, &p->bs);
1909 if (ret)
1910 goto out;
1911 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
1912 if (ret)
1913 bioset_exit(&md->bs);
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001914out:
Mike Snitzer02233342015-03-10 23:49:26 -04001915 /* mempool bind completed, no longer need any mempools in the table */
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001916 dm_table_free_md_mempools(t);
Jens Axboe2a2a4c52018-06-07 14:42:06 -06001917 return ret;
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001918}
1919
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920/*
1921 * Bind a table to the device.
1922 */
1923static void event_callback(void *context)
1924{
Mike Anderson7a8c3d32007-10-19 22:48:01 +01001925 unsigned long flags;
1926 LIST_HEAD(uevents);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 struct mapped_device *md = (struct mapped_device *) context;
1928
Mike Anderson7a8c3d32007-10-19 22:48:01 +01001929 spin_lock_irqsave(&md->uevent_lock, flags);
1930 list_splice_init(&md->uevent_list, &uevents);
1931 spin_unlock_irqrestore(&md->uevent_lock, flags);
1932
Tejun Heoed9e1982008-08-25 19:56:05 +09001933 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
Mike Anderson7a8c3d32007-10-19 22:48:01 +01001934
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 atomic_inc(&md->event_nr);
1936 wake_up(&md->eventq);
Mikulas Patocka62e08242017-09-20 07:29:49 -04001937 dm_issue_global_event();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938}
1939
Mike Snitzerc2176492011-01-13 19:53:46 +00001940/*
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00001941 * Returns old map, which caller must destroy.
1942 */
1943static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1944 struct queue_limits *limits)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945{
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00001946 struct dm_table *old_map;
Jens Axboe165125e2007-07-24 09:28:11 +02001947 struct request_queue *q = md->queue;
Mike Snitzer978e51b2017-12-09 15:16:42 -05001948 bool request_based = dm_table_request_based(t);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 sector_t size;
Jens Axboe2a2a4c52018-06-07 14:42:06 -06001950 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951
Bart Van Assche5a8f1f82016-08-31 15:17:04 -07001952 lockdep_assert_held(&md->suspend_lock);
1953
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 size = dm_table_get_size(t);
Darrick J. Wong3ac51e72006-03-27 01:17:54 -08001955
1956 /*
1957 * Wipe any geometry if the size of the table changed.
1958 */
Mikulas Patockafd2ed4d2013-08-16 10:54:23 -04001959 if (size != dm_get_size(md))
Darrick J. Wong3ac51e72006-03-27 01:17:54 -08001960 memset(&md->geometry, 0, sizeof(md->geometry));
1961
Christoph Hellwigf64d9b22020-11-16 15:57:03 +01001962 set_capacity_and_notify(md->disk, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001964 dm_table_event_callback(t, event_callback, md);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001965
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001966 /*
1967 * The queue hasn't been stopped yet, if the old table type wasn't
1968 * for request-based during suspension. So stop it to prevent
1969 * I/O mapping before resume.
1970 * This must be done before setting the queue restrictions,
1971 * because request-based dm may be run just after the setting.
1972 */
Mike Snitzer978e51b2017-12-09 15:16:42 -05001973 if (request_based)
Mike Snitzereca7ee62016-02-20 13:45:38 -05001974 dm_stop_queue(q);
Mike Snitzer978e51b2017-12-09 15:16:42 -05001975
Mike Snitzer9c37de22020-10-07 15:15:08 -04001976 if (request_based) {
Mike Snitzer16f12262016-01-31 17:22:27 -05001977 /*
Mike Snitzer9c37de22020-10-07 15:15:08 -04001978 * Leverage the fact that request-based DM targets are
1979 * immutable singletons - used to optimize dm_mq_queue_rq.
Mike Snitzer16f12262016-01-31 17:22:27 -05001980 */
1981 md->immutable_target = dm_table_get_immutable_target(t);
1982 }
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001983
Jens Axboe2a2a4c52018-06-07 14:42:06 -06001984 ret = __bind_mempools(md, t);
1985 if (ret) {
1986 old_map = ERR_PTR(ret);
1987 goto out;
1988 }
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001989
Eric Dumazeta12f5d42014-11-23 09:34:29 -08001990 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
Mike Snitzer1d3aa6f2016-02-22 14:14:24 -05001991 rcu_assign_pointer(md->map, (void *)t);
Alasdair G Kergon36a04562011-10-31 20:19:04 +00001992 md->immutable_target_type = dm_table_get_immutable_target_type(t);
1993
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001994 dm_table_set_restrictions(t, q, limits);
Hannes Reinecke41abc4e2014-11-05 14:35:50 +01001995 if (old_map)
1996 dm_sync_table(md);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07001997
Jens Axboe2a2a4c52018-06-07 14:42:06 -06001998out:
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00001999 return old_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000}
2001
Alasdair G Kergona7940152009-12-10 23:52:23 +00002002/*
2003 * Returns unbound table for the caller to free.
2004 */
2005static struct dm_table *__unbind(struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006{
Eric Dumazeta12f5d42014-11-23 09:34:29 -08002007 struct dm_table *map = rcu_dereference_protected(md->map, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008
2009 if (!map)
Alasdair G Kergona7940152009-12-10 23:52:23 +00002010 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011
2012 dm_table_event_callback(map, NULL, NULL);
Monam Agarwal9cdb8522014-03-23 23:58:27 +05302013 RCU_INIT_POINTER(md->map, NULL);
Mikulas Patocka83d5e5b2013-07-10 23:41:18 +01002014 dm_sync_table(md);
Alasdair G Kergona7940152009-12-10 23:52:23 +00002015
2016 return map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017}
2018
2019/*
2020 * Constructor for a new device.
2021 */
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07002022int dm_create(int minor, struct mapped_device **result)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023{
Mike Snitzerc12c9a32018-01-12 09:32:21 -05002024 int r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 struct mapped_device *md;
2026
Alasdair G Kergon2b06cff2006-06-26 00:27:32 -07002027 md = alloc_dev(minor);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 if (!md)
2029 return -ENXIO;
2030
Mike Snitzerc12c9a32018-01-12 09:32:21 -05002031 r = dm_sysfs_init(md);
2032 if (r) {
2033 free_dev(md);
2034 return r;
2035 }
Milan Broz784aae72009-01-06 03:05:12 +00002036
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 *result = md;
2038 return 0;
2039}
2040
Mike Snitzera5664da2010-08-12 04:14:01 +01002041/*
2042 * Functions to manage md->type.
2043 * All are required to hold md->type_lock.
2044 */
2045void dm_lock_md_type(struct mapped_device *md)
2046{
2047 mutex_lock(&md->type_lock);
2048}
2049
2050void dm_unlock_md_type(struct mapped_device *md)
2051{
2052 mutex_unlock(&md->type_lock);
2053}
2054
Bart Van Assche7e0d5742017-04-27 10:11:23 -07002055void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
Mike Snitzera5664da2010-08-12 04:14:01 +01002056{
Mike Snitzer00c4fc32013-08-27 18:57:03 -04002057 BUG_ON(!mutex_is_locked(&md->type_lock));
Mike Snitzera5664da2010-08-12 04:14:01 +01002058 md->type = type;
2059}
2060
Bart Van Assche7e0d5742017-04-27 10:11:23 -07002061enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
Mike Snitzera5664da2010-08-12 04:14:01 +01002062{
2063 return md->type;
2064}
2065
Alasdair G Kergon36a04562011-10-31 20:19:04 +00002066struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2067{
2068 return md->immutable_target_type;
2069}
2070
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +01002071/*
Mike Snitzerf84cb8a2013-09-19 12:13:58 -04002072 * The queue_limits are only valid as long as you have a reference
2073 * count on 'md'.
2074 */
2075struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2076{
2077 BUG_ON(!atomic_read(&md->holders));
2078 return &md->queue->limits;
2079}
2080EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2081
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +01002082/*
2083 * Setup the DM device's queue based on md's type
2084 */
Mike Snitzer591ddcfc2016-01-31 12:05:42 -05002085int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +01002086{
Mike Snitzerbfebd1c2015-03-08 00:51:47 -05002087 int r;
Mike Snitzerc100ec42018-01-08 20:03:04 -05002088 struct queue_limits limits;
Bart Van Assche7e0d5742017-04-27 10:11:23 -07002089 enum dm_queue_mode type = dm_get_md_type(md);
Mike Snitzerbfebd1c2015-03-08 00:51:47 -05002090
Toshi Kani545ed202016-06-22 17:54:53 -06002091 switch (type) {
Mike Snitzerbfebd1c2015-03-08 00:51:47 -05002092 case DM_TYPE_REQUEST_BASED:
Mike Snitzer681cc5e2020-10-07 16:41:01 -04002093 md->disk->fops = &dm_rq_blk_dops;
Mike Snitzere83068a2016-05-24 21:16:51 -04002094 r = dm_mq_init_request_queue(md, t);
Mike Snitzerbfebd1c2015-03-08 00:51:47 -05002095 if (r) {
Mike Snitzer681cc5e2020-10-07 16:41:01 -04002096 DMERR("Cannot initialize queue for request-based dm mapped device");
Mike Snitzerbfebd1c2015-03-08 00:51:47 -05002097 return r;
2098 }
2099 break;
2100 case DM_TYPE_BIO_BASED:
Toshi Kani545ed202016-06-22 17:54:53 -06002101 case DM_TYPE_DAX_BIO_BASED:
Mike Snitzerbfebd1c2015-03-08 00:51:47 -05002102 break;
Bart Van Assche7e0d5742017-04-27 10:11:23 -07002103 case DM_TYPE_NONE:
2104 WARN_ON_ONCE(true);
2105 break;
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +01002106 }
2107
Mike Snitzerc100ec42018-01-08 20:03:04 -05002108 r = dm_calculate_queue_limits(t, &limits);
2109 if (r) {
2110 DMERR("Cannot calculate initial queue limits");
2111 return r;
2112 }
2113 dm_table_set_restrictions(t, md->queue, &limits);
2114 blk_register_queue(md->disk);
2115
Mike Snitzer4a0b4dd2010-08-12 04:14:02 +01002116 return 0;
2117}
2118
Mikulas Patocka2bec1f42015-02-17 14:30:53 -05002119struct mapped_device *dm_get_md(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120{
2121 struct mapped_device *md;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 unsigned minor = MINOR(dev);
2123
2124 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2125 return NULL;
2126
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07002127 spin_lock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128
2129 md = idr_find(&_minor_idr, minor);
Mike Snitzer49de5762017-11-06 16:40:10 -05002130 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2131 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2132 md = NULL;
2133 goto out;
Jeff Mahoneyfba9f902006-06-26 00:27:23 -07002134 }
Mike Snitzer49de5762017-11-06 16:40:10 -05002135 dm_get(md);
Jeff Mahoneyfba9f902006-06-26 00:27:23 -07002136out:
Jeff Mahoneyf32c10b2006-06-26 00:27:22 -07002137 spin_unlock(&_minor_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138
David Teigland637842c2006-01-06 00:20:00 -08002139 return md;
2140}
Alasdair G Kergon3cf2e4b2011-10-31 20:19:06 +00002141EXPORT_SYMBOL_GPL(dm_get_md);
David Teiglandd229a952006-01-06 00:20:01 -08002142
Alasdair G Kergon9ade92a2006-03-27 01:17:53 -08002143void *dm_get_mdptr(struct mapped_device *md)
David Teigland637842c2006-01-06 00:20:00 -08002144{
Alasdair G Kergon9ade92a2006-03-27 01:17:53 -08002145 return md->interface_ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146}
2147
2148void dm_set_mdptr(struct mapped_device *md, void *ptr)
2149{
2150 md->interface_ptr = ptr;
2151}
2152
2153void dm_get(struct mapped_device *md)
2154{
2155 atomic_inc(&md->holders);
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +01002156 BUG_ON(test_bit(DMF_FREEING, &md->flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157}
2158
Mikulas Patocka09ee96b2015-02-26 11:41:28 -05002159int dm_hold(struct mapped_device *md)
2160{
2161 spin_lock(&_minor_lock);
2162 if (test_bit(DMF_FREEING, &md->flags)) {
2163 spin_unlock(&_minor_lock);
2164 return -EBUSY;
2165 }
2166 dm_get(md);
2167 spin_unlock(&_minor_lock);
2168 return 0;
2169}
2170EXPORT_SYMBOL_GPL(dm_hold);
2171
Alasdair G Kergon72d94862006-06-26 00:27:35 -07002172const char *dm_device_name(struct mapped_device *md)
2173{
2174 return md->name;
2175}
2176EXPORT_SYMBOL_GPL(dm_device_name);
2177
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +01002178static void __dm_destroy(struct mapped_device *md, bool wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179{
Mike Anderson1134e5ae2006-03-27 01:17:54 -08002180 struct dm_table *map;
Mikulas Patocka83d5e5b2013-07-10 23:41:18 +01002181 int srcu_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +01002183 might_sleep();
Jeff Mahoneyfba9f902006-06-26 00:27:23 -07002184
Mike Snitzer63a4f062015-03-23 17:01:43 -04002185 spin_lock(&_minor_lock);
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +01002186 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2187 set_bit(DMF_FREEING, &md->flags);
2188 spin_unlock(&_minor_lock);
2189
Mike Snitzerc12c9a32018-01-12 09:32:21 -05002190 blk_set_queue_dying(md->queue);
Bart Van Assche3b785fb2016-08-31 15:17:49 -07002191
Mikulas Patockaab7c7bb2015-02-27 14:04:27 -05002192 /*
2193 * Take suspend_lock so that presuspend and postsuspend methods
2194 * do not race with internal suspend.
2195 */
2196 mutex_lock(&md->suspend_lock);
Junichi Nomura2a708cf2015-10-01 08:31:51 +00002197 map = dm_get_live_table(md, &srcu_idx);
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +01002198 if (!dm_suspended_md(md)) {
2199 dm_table_presuspend_targets(map);
Mikulas Patockaadc0daa2020-02-24 10:20:28 +01002200 set_bit(DMF_SUSPENDED, &md->flags);
Mikulas Patocka5df96f22020-07-23 10:42:09 -04002201 set_bit(DMF_POST_SUSPENDING, &md->flags);
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +01002202 dm_table_postsuspend_targets(map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 }
Mikulas Patocka83d5e5b2013-07-10 23:41:18 +01002204 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2205 dm_put_live_table(md, srcu_idx);
Junichi Nomura2a708cf2015-10-01 08:31:51 +00002206 mutex_unlock(&md->suspend_lock);
Mikulas Patocka83d5e5b2013-07-10 23:41:18 +01002207
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +01002208 /*
2209 * Rare, but there may be I/O requests still going to complete,
2210 * for example. Wait for all references to disappear.
2211 * No one should increment the reference count of the mapped_device,
2212 * after the mapped_device state becomes DMF_FREEING.
2213 */
2214 if (wait)
2215 while (atomic_read(&md->holders))
2216 msleep(1);
2217 else if (atomic_read(&md->holders))
2218 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2219 dm_device_name(md), atomic_read(&md->holders));
2220
2221 dm_sysfs_exit(md);
Kiyoshi Ueda3f77316d2010-08-12 04:13:56 +01002222 dm_table_destroy(__unbind(md));
2223 free_dev(md);
2224}
2225
2226void dm_destroy(struct mapped_device *md)
2227{
2228 __dm_destroy(md, true);
2229}
2230
2231void dm_destroy_immediate(struct mapped_device *md)
2232{
2233 __dm_destroy(md, false);
2234}
2235
2236void dm_put(struct mapped_device *md)
2237{
2238 atomic_dec(&md->holders);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239}
Edward Goggin79eb8852007-05-09 02:32:56 -07002240EXPORT_SYMBOL_GPL(dm_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241
Ming Lei85067742020-06-24 16:00:58 -04002242static bool md_in_flight_bios(struct mapped_device *md)
2243{
2244 int cpu;
Christoph Hellwig8446fe92020-11-24 09:36:54 +01002245 struct block_device *part = dm_disk(md)->part0;
Ming Lei85067742020-06-24 16:00:58 -04002246 long sum = 0;
2247
2248 for_each_possible_cpu(cpu) {
2249 sum += part_stat_local_read_cpu(part, in_flight[0], cpu);
2250 sum += part_stat_local_read_cpu(part, in_flight[1], cpu);
2251 }
2252
2253 return sum != 0;
2254}
2255
2256static int dm_wait_for_bios_completion(struct mapped_device *md, long task_state)
Milan Broz46125c12008-02-08 02:10:30 +00002257{
2258 int r = 0;
Bart Van Assche9f4c3f82016-08-31 15:16:43 -07002259 DEFINE_WAIT(wait);
Milan Broz46125c12008-02-08 02:10:30 +00002260
Ming Lei85067742020-06-24 16:00:58 -04002261 while (true) {
Bart Van Assche9f4c3f82016-08-31 15:16:43 -07002262 prepare_to_wait(&md->wait, &wait, task_state);
Milan Broz46125c12008-02-08 02:10:30 +00002263
Ming Lei85067742020-06-24 16:00:58 -04002264 if (!md_in_flight_bios(md))
Milan Broz46125c12008-02-08 02:10:30 +00002265 break;
2266
Bart Van Asschee3fabdf2016-08-31 15:16:22 -07002267 if (signal_pending_state(task_state, current)) {
Milan Broz46125c12008-02-08 02:10:30 +00002268 r = -EINTR;
2269 break;
2270 }
2271
2272 io_schedule();
2273 }
Bart Van Assche9f4c3f82016-08-31 15:16:43 -07002274 finish_wait(&md->wait, &wait);
Mikulas Patockab44ebeb2009-04-02 19:55:39 +01002275
Milan Broz46125c12008-02-08 02:10:30 +00002276 return r;
2277}
2278
Ming Lei85067742020-06-24 16:00:58 -04002279static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2280{
2281 int r = 0;
2282
2283 if (!queue_is_mq(md->queue))
2284 return dm_wait_for_bios_completion(md, task_state);
2285
2286 while (true) {
2287 if (!blk_mq_queue_inflight(md->queue))
2288 break;
2289
2290 if (signal_pending_state(task_state, current)) {
2291 r = -EINTR;
2292 break;
2293 }
2294
2295 msleep(5);
2296 }
2297
2298 return r;
2299}
2300
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301/*
2302 * Process the deferred bios
2303 */
Mikulas Patockaef208582009-04-02 19:55:38 +01002304static void dm_wq_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305{
Mike Snitzer0c2915b2020-09-28 13:41:36 -04002306 struct mapped_device *md = container_of(work, struct mapped_device, work);
2307 struct bio *bio;
Mikulas Patockaef208582009-04-02 19:55:38 +01002308
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002309 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
Alasdair G Kergondf12ee92009-04-09 00:27:13 +01002310 spin_lock_irq(&md->deferred_lock);
Mike Snitzer0c2915b2020-09-28 13:41:36 -04002311 bio = bio_list_pop(&md->deferred);
Alasdair G Kergondf12ee92009-04-09 00:27:13 +01002312 spin_unlock_irq(&md->deferred_lock);
Mikulas Patocka022c2612009-04-02 19:55:39 +01002313
Mike Snitzer0c2915b2020-09-28 13:41:36 -04002314 if (!bio)
Alasdair G Kergondf12ee92009-04-09 00:27:13 +01002315 break;
Alasdair G Kergondf12ee92009-04-09 00:27:13 +01002316
Mike Snitzer0c2915b2020-09-28 13:41:36 -04002317 submit_bio_noacct(bio);
Mikulas Patocka022c2612009-04-02 19:55:39 +01002318 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319}
2320
Mikulas Patocka9a1fb462009-04-02 19:55:36 +01002321static void dm_queue_flush(struct mapped_device *md)
Milan Broz304f3f62008-02-08 02:11:17 +00002322{
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002323 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
Peter Zijlstra4e857c52014-03-17 18:06:10 +01002324 smp_mb__after_atomic();
Mikulas Patocka53d59142009-04-02 19:55:37 +01002325 queue_work(md->wq, &md->work);
Milan Broz304f3f62008-02-08 02:11:17 +00002326}
2327
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328/*
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002329 * Swap in a new table, returning the old one for the caller to destroy.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 */
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002331struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332{
Mike Christie87eb5b22013-03-01 22:45:48 +00002333 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
Mike Snitzer754c5fc2009-06-22 10:12:34 +01002334 struct queue_limits limits;
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002335 int r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336
Daniel Walkere61290a2008-02-08 02:10:08 +00002337 mutex_lock(&md->suspend_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338
2339 /* device must be suspended */
Kiyoshi Ueda4f186f82009-12-10 23:52:26 +00002340 if (!dm_suspended_md(md))
Alasdair G Kergon93c534a2005-07-12 15:53:05 -07002341 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342
Mike Snitzer3ae70652012-09-26 23:45:45 +01002343 /*
2344 * If the new table has no data devices, retain the existing limits.
2345 * This helps multipath with queue_if_no_path if all paths disappear,
2346 * then new I/O is queued based on these limits, and then some paths
2347 * reappear.
2348 */
2349 if (dm_table_has_no_data_devices(table)) {
Mikulas Patocka83d5e5b2013-07-10 23:41:18 +01002350 live_map = dm_get_live_table_fast(md);
Mike Snitzer3ae70652012-09-26 23:45:45 +01002351 if (live_map)
2352 limits = md->queue->limits;
Mikulas Patocka83d5e5b2013-07-10 23:41:18 +01002353 dm_put_live_table_fast(md);
Mike Snitzer3ae70652012-09-26 23:45:45 +01002354 }
2355
Mike Christie87eb5b22013-03-01 22:45:48 +00002356 if (!live_map) {
2357 r = dm_calculate_queue_limits(table, &limits);
2358 if (r) {
2359 map = ERR_PTR(r);
2360 goto out;
2361 }
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002362 }
Mike Snitzer754c5fc2009-06-22 10:12:34 +01002363
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002364 map = __bind(md, table, &limits);
Mikulas Patocka62e08242017-09-20 07:29:49 -04002365 dm_issue_global_event();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366
Alasdair G Kergon93c534a2005-07-12 15:53:05 -07002367out:
Daniel Walkere61290a2008-02-08 02:10:08 +00002368 mutex_unlock(&md->suspend_lock);
Alasdair G Kergon042d2a92009-12-10 23:52:24 +00002369 return map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370}
2371
2372/*
2373 * Functions to lock and unlock any filesystem running on the
2374 * device.
2375 */
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002376static int lock_fs(struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377{
Alasdair G Kergone39e2e92006-01-06 00:20:05 -08002378 int r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379
Christoph Hellwig040f04b2020-11-24 11:54:06 +01002380 WARN_ON(test_bit(DMF_FROZEN, &md->flags));
Alasdair G Kergondfbe03f2005-05-05 16:16:04 -07002381
Christoph Hellwig977115c2020-11-26 10:41:07 +01002382 r = freeze_bdev(md->disk->part0);
Christoph Hellwig040f04b2020-11-24 11:54:06 +01002383 if (!r)
2384 set_bit(DMF_FROZEN, &md->flags);
2385 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386}
2387
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002388static void unlock_fs(struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389{
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -08002390 if (!test_bit(DMF_FROZEN, &md->flags))
2391 return;
Christoph Hellwig977115c2020-11-26 10:41:07 +01002392 thaw_bdev(md->disk->part0);
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -08002393 clear_bit(DMF_FROZEN, &md->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394}
2395
2396/*
Bart Van Asscheb48633f2016-08-31 15:16:02 -07002397 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2398 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2399 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2400 *
Mike Snitzerffcc3932014-10-28 18:34:52 -04002401 * If __dm_suspend returns 0, the device is completely quiescent
2402 * now. There is no request-processing activity. All new requests
2403 * are being added to md->deferred list.
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002404 */
Mike Snitzerffcc3932014-10-28 18:34:52 -04002405static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
Bart Van Asscheb48633f2016-08-31 15:16:02 -07002406 unsigned suspend_flags, long task_state,
Mike Snitzereaf9a732016-08-02 13:07:20 -04002407 int dmf_suspended_flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408{
Mike Snitzerffcc3932014-10-28 18:34:52 -04002409 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2410 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2411 int r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412
Bart Van Assche5a8f1f82016-08-31 15:17:04 -07002413 lockdep_assert_held(&md->suspend_lock);
2414
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -08002415 /*
2416 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2417 * This flag is cleared before dm_suspend returns.
2418 */
2419 if (noflush)
2420 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
Bart Van Assche86331f32017-04-27 10:11:26 -07002421 else
Mike Snitzerac75b09f2020-05-14 12:55:39 -04002422 DMDEBUG("%s: suspending with flush", dm_device_name(md));
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -08002423
Mike Snitzerd67ee2132014-10-28 20:13:31 -04002424 /*
2425 * This gets reverted if there's an error later and the targets
2426 * provide the .presuspend_undo hook.
2427 */
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002428 dm_table_presuspend_targets(map);
2429
Mikulas Patocka32a926d2009-06-22 10:12:17 +01002430 /*
Kiyoshi Ueda9f518b22009-12-10 23:52:16 +00002431 * Flush I/O to the device.
2432 * Any I/O submitted after lock_fs() may not be flushed.
2433 * noflush takes precedence over do_lockfs.
2434 * (lock_fs() flushes I/Os and waits for them to complete.)
Mikulas Patocka32a926d2009-06-22 10:12:17 +01002435 */
2436 if (!noflush && do_lockfs) {
2437 r = lock_fs(md);
Mike Snitzerd67ee2132014-10-28 20:13:31 -04002438 if (r) {
2439 dm_table_presuspend_undo_targets(map);
Mike Snitzerffcc3932014-10-28 18:34:52 -04002440 return r;
Mike Snitzerd67ee2132014-10-28 20:13:31 -04002441 }
Alasdair G Kergonaa8d7c22006-01-06 00:20:06 -08002442 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443
2444 /*
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002445 * Here we must make sure that no processes are submitting requests
2446 * to target drivers i.e. no one may be executing
Mike Snitzer0cede372020-09-30 15:12:04 -04002447 * __split_and_process_bio from dm_submit_bio.
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002448 *
Mike Snitzer0cede372020-09-30 15:12:04 -04002449 * To get all processes out of __split_and_process_bio in dm_submit_bio,
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002450 * we take the write lock. To prevent any process from reentering
Mike Snitzer0cede372020-09-30 15:12:04 -04002451 * __split_and_process_bio from dm_submit_bio and quiesce the thread
2452 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
Tejun Heo6a8736d2010-09-08 18:07:00 +02002453 * flush_workqueue(md->wq).
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 */
Alasdair G Kergon1eb787e2009-04-09 00:27:14 +01002455 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
Hannes Reinecke41abc4e2014-11-05 14:35:50 +01002456 if (map)
2457 synchronize_srcu(&md->io_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458
Kiyoshi Uedad0bcb872009-12-10 23:52:18 +00002459 /*
Tejun Heo29e40132010-09-08 18:07:00 +02002460 * Stop md->queue before flushing md->wq in case request-based
2461 * dm defers requests to md->wq from md->queue.
Kiyoshi Uedad0bcb872009-12-10 23:52:18 +00002462 */
Jens Axboe6a23e052018-10-10 20:49:26 -06002463 if (dm_request_based(md))
Mike Snitzereca7ee62016-02-20 13:45:38 -05002464 dm_stop_queue(md->queue);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002465
Kiyoshi Uedad0bcb872009-12-10 23:52:18 +00002466 flush_workqueue(md->wq);
2467
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468 /*
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002469 * At this point no more requests are entering target request routines.
2470 * We call dm_wait_for_completion to wait for all existing requests
2471 * to finish.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 */
Bart Van Asscheb48633f2016-08-31 15:16:02 -07002473 r = dm_wait_for_completion(md, task_state);
Mike Snitzereaf9a732016-08-02 13:07:20 -04002474 if (!r)
2475 set_bit(dmf_suspended_flag, &md->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476
Milan Broz6d6f10d2008-02-08 02:10:22 +00002477 if (noflush)
Mikulas Patocka022c2612009-04-02 19:55:39 +01002478 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
Hannes Reinecke41abc4e2014-11-05 14:35:50 +01002479 if (map)
2480 synchronize_srcu(&md->io_barrier);
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -08002481
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 /* were we interrupted ? */
Milan Broz46125c12008-02-08 02:10:30 +00002483 if (r < 0) {
Mikulas Patocka9a1fb462009-04-02 19:55:36 +01002484 dm_queue_flush(md);
Milan Broz73d410c2008-02-08 02:10:25 +00002485
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002486 if (dm_request_based(md))
Mike Snitzereca7ee62016-02-20 13:45:38 -05002487 dm_start_queue(md->queue);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002488
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002489 unlock_fs(md);
Mike Snitzerd67ee2132014-10-28 20:13:31 -04002490 dm_table_presuspend_undo_targets(map);
Mike Snitzerffcc3932014-10-28 18:34:52 -04002491 /* pushback list is already flushed, so skip flush */
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002492 }
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002493
Mike Snitzerffcc3932014-10-28 18:34:52 -04002494 return r;
2495}
2496
2497/*
2498 * We need to be able to change a mapping table under a mounted
2499 * filesystem. For example we might want to move some data in
2500 * the background. Before the table can be swapped with
2501 * dm_bind_table, dm_suspend must be called to flush any in
2502 * flight bios and ensure that any further io gets deferred.
2503 */
2504/*
2505 * Suspend mechanism in request-based dm.
2506 *
2507 * 1. Flush all I/Os by lock_fs() if needed.
2508 * 2. Stop dispatching any I/O by stopping the request_queue.
2509 * 3. Wait for all in-flight I/Os to be completed or requeued.
2510 *
2511 * To abort suspend, start the request_queue.
2512 */
2513int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2514{
2515 struct dm_table *map = NULL;
2516 int r = 0;
2517
2518retry:
2519 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2520
2521 if (dm_suspended_md(md)) {
2522 r = -EINVAL;
2523 goto out_unlock;
2524 }
2525
2526 if (dm_suspended_internally_md(md)) {
2527 /* already internally suspended, wait for internal resume */
2528 mutex_unlock(&md->suspend_lock);
2529 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2530 if (r)
2531 return r;
2532 goto retry;
2533 }
2534
Eric Dumazeta12f5d42014-11-23 09:34:29 -08002535 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
Mike Snitzerffcc3932014-10-28 18:34:52 -04002536
Mike Snitzereaf9a732016-08-02 13:07:20 -04002537 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
Mike Snitzerffcc3932014-10-28 18:34:52 -04002538 if (r)
2539 goto out_unlock;
Mikulas Patocka3b00b202009-04-09 00:27:15 +01002540
Mikulas Patocka5df96f22020-07-23 10:42:09 -04002541 set_bit(DMF_POST_SUSPENDING, &md->flags);
Kiyoshi Ueda4d4471c2009-12-10 23:52:26 +00002542 dm_table_postsuspend_targets(map);
Mikulas Patocka5df96f22020-07-23 10:42:09 -04002543 clear_bit(DMF_POST_SUSPENDING, &md->flags);
Kiyoshi Ueda4d4471c2009-12-10 23:52:26 +00002544
Alasdair G Kergond2874832006-11-08 17:44:43 -08002545out_unlock:
Daniel Walkere61290a2008-02-08 02:10:08 +00002546 mutex_unlock(&md->suspend_lock);
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002547 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548}
2549
Mike Snitzerffcc3932014-10-28 18:34:52 -04002550static int __dm_resume(struct mapped_device *md, struct dm_table *map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551{
Mike Snitzerffcc3932014-10-28 18:34:52 -04002552 if (map) {
2553 int r = dm_table_resume_targets(map);
2554 if (r)
2555 return r;
2556 }
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002557
Mikulas Patocka9a1fb462009-04-02 19:55:36 +01002558 dm_queue_flush(md);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002559
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002560 /*
2561 * Flushing deferred I/Os must be done after targets are resumed
2562 * so that mapping of targets can work correctly.
2563 * Request-based dm is queueing the deferred I/Os in its request_queue.
2564 */
2565 if (dm_request_based(md))
Mike Snitzereca7ee62016-02-20 13:45:38 -05002566 dm_start_queue(md->queue);
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01002567
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002568 unlock_fs(md);
2569
Mike Snitzerffcc3932014-10-28 18:34:52 -04002570 return 0;
2571}
2572
2573int dm_resume(struct mapped_device *md)
2574{
Minfei Huang8dc23652016-09-06 16:00:29 +08002575 int r;
Mike Snitzerffcc3932014-10-28 18:34:52 -04002576 struct dm_table *map = NULL;
2577
2578retry:
Minfei Huang8dc23652016-09-06 16:00:29 +08002579 r = -EINVAL;
Mike Snitzerffcc3932014-10-28 18:34:52 -04002580 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2581
2582 if (!dm_suspended_md(md))
2583 goto out;
2584
2585 if (dm_suspended_internally_md(md)) {
2586 /* already internally suspended, wait for internal resume */
2587 mutex_unlock(&md->suspend_lock);
2588 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2589 if (r)
2590 return r;
2591 goto retry;
2592 }
2593
Eric Dumazeta12f5d42014-11-23 09:34:29 -08002594 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
Mike Snitzerffcc3932014-10-28 18:34:52 -04002595 if (!map || !dm_table_get_size(map))
2596 goto out;
2597
2598 r = __dm_resume(md, map);
2599 if (r)
2600 goto out;
2601
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002602 clear_bit(DMF_SUSPENDED, &md->flags);
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002603out:
Daniel Walkere61290a2008-02-08 02:10:08 +00002604 mutex_unlock(&md->suspend_lock);
Alasdair G Kergon2ca33102005-07-28 21:16:00 -07002605
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002606 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607}
2608
Mikulas Patockafd2ed4d2013-08-16 10:54:23 -04002609/*
2610 * Internal suspend/resume works like userspace-driven suspend. It waits
2611 * until all bios finish and prevents issuing new bios to the target drivers.
2612 * It may be used only from the kernel.
Mikulas Patockafd2ed4d2013-08-16 10:54:23 -04002613 */
2614
Mike Snitzerffcc3932014-10-28 18:34:52 -04002615static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2616{
2617 struct dm_table *map = NULL;
2618
Bart Van Assche1ea0654e2017-04-27 10:11:21 -07002619 lockdep_assert_held(&md->suspend_lock);
2620
Mikulas Patocka96b26c82015-01-08 18:52:26 -05002621 if (md->internal_suspend_count++)
Mike Snitzerffcc3932014-10-28 18:34:52 -04002622 return; /* nested internal suspend */
2623
2624 if (dm_suspended_md(md)) {
2625 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2626 return; /* nest suspend */
2627 }
2628
Eric Dumazeta12f5d42014-11-23 09:34:29 -08002629 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
Mike Snitzerffcc3932014-10-28 18:34:52 -04002630
2631 /*
2632 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2633 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2634 * would require changing .presuspend to return an error -- avoid this
2635 * until there is a need for more elaborate variants of internal suspend.
2636 */
Mike Snitzereaf9a732016-08-02 13:07:20 -04002637 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2638 DMF_SUSPENDED_INTERNALLY);
Mike Snitzerffcc3932014-10-28 18:34:52 -04002639
Mikulas Patocka5df96f22020-07-23 10:42:09 -04002640 set_bit(DMF_POST_SUSPENDING, &md->flags);
Mike Snitzerffcc3932014-10-28 18:34:52 -04002641 dm_table_postsuspend_targets(map);
Mikulas Patocka5df96f22020-07-23 10:42:09 -04002642 clear_bit(DMF_POST_SUSPENDING, &md->flags);
Mike Snitzerffcc3932014-10-28 18:34:52 -04002643}
2644
2645static void __dm_internal_resume(struct mapped_device *md)
2646{
Mikulas Patocka96b26c82015-01-08 18:52:26 -05002647 BUG_ON(!md->internal_suspend_count);
2648
2649 if (--md->internal_suspend_count)
Mike Snitzerffcc3932014-10-28 18:34:52 -04002650 return; /* resume from nested internal suspend */
2651
2652 if (dm_suspended_md(md))
2653 goto done; /* resume from nested suspend */
2654
2655 /*
2656 * NOTE: existing callers don't need to call dm_table_resume_targets
2657 * (which may fail -- so best to avoid it for now by passing NULL map)
2658 */
2659 (void) __dm_resume(md, NULL);
2660
2661done:
2662 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2663 smp_mb__after_atomic();
2664 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2665}
2666
2667void dm_internal_suspend_noflush(struct mapped_device *md)
Mikulas Patockafd2ed4d2013-08-16 10:54:23 -04002668{
2669 mutex_lock(&md->suspend_lock);
Mike Snitzerffcc3932014-10-28 18:34:52 -04002670 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2671 mutex_unlock(&md->suspend_lock);
2672}
2673EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2674
2675void dm_internal_resume(struct mapped_device *md)
2676{
2677 mutex_lock(&md->suspend_lock);
2678 __dm_internal_resume(md);
2679 mutex_unlock(&md->suspend_lock);
2680}
2681EXPORT_SYMBOL_GPL(dm_internal_resume);
2682
2683/*
2684 * Fast variants of internal suspend/resume hold md->suspend_lock,
2685 * which prevents interaction with userspace-driven suspend.
2686 */
2687
2688void dm_internal_suspend_fast(struct mapped_device *md)
2689{
2690 mutex_lock(&md->suspend_lock);
2691 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
Mikulas Patockafd2ed4d2013-08-16 10:54:23 -04002692 return;
2693
2694 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2695 synchronize_srcu(&md->io_barrier);
2696 flush_workqueue(md->wq);
2697 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2698}
Mikulas Patockab735fed2015-02-26 11:40:35 -05002699EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
Mikulas Patockafd2ed4d2013-08-16 10:54:23 -04002700
Mike Snitzerffcc3932014-10-28 18:34:52 -04002701void dm_internal_resume_fast(struct mapped_device *md)
Mikulas Patockafd2ed4d2013-08-16 10:54:23 -04002702{
Mike Snitzerffcc3932014-10-28 18:34:52 -04002703 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
Mikulas Patockafd2ed4d2013-08-16 10:54:23 -04002704 goto done;
2705
2706 dm_queue_flush(md);
2707
2708done:
2709 mutex_unlock(&md->suspend_lock);
2710}
Mikulas Patockab735fed2015-02-26 11:40:35 -05002711EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
Mikulas Patockafd2ed4d2013-08-16 10:54:23 -04002712
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713/*-----------------------------------------------------------------
2714 * Event notification.
2715 *---------------------------------------------------------------*/
Peter Rajnoha3abf85b2010-03-06 02:32:31 +00002716int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
Milan Broz60935eb2009-06-22 10:12:30 +01002717 unsigned cookie)
Alasdair G Kergon69267a32007-12-13 14:15:57 +00002718{
Mikulas Patocka6958c1c2020-07-08 12:25:20 -04002719 int r;
2720 unsigned noio_flag;
Milan Broz60935eb2009-06-22 10:12:30 +01002721 char udev_cookie[DM_COOKIE_LENGTH];
2722 char *envp[] = { udev_cookie, NULL };
2723
Mikulas Patocka6958c1c2020-07-08 12:25:20 -04002724 noio_flag = memalloc_noio_save();
2725
Milan Broz60935eb2009-06-22 10:12:30 +01002726 if (!cookie)
Mikulas Patocka6958c1c2020-07-08 12:25:20 -04002727 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
Milan Broz60935eb2009-06-22 10:12:30 +01002728 else {
2729 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2730 DM_COOKIE_ENV_VAR_NAME, cookie);
Mikulas Patocka6958c1c2020-07-08 12:25:20 -04002731 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2732 action, envp);
Milan Broz60935eb2009-06-22 10:12:30 +01002733 }
Mikulas Patocka6958c1c2020-07-08 12:25:20 -04002734
2735 memalloc_noio_restore(noio_flag);
2736
2737 return r;
Alasdair G Kergon69267a32007-12-13 14:15:57 +00002738}
2739
Mike Anderson7a8c3d32007-10-19 22:48:01 +01002740uint32_t dm_next_uevent_seq(struct mapped_device *md)
2741{
2742 return atomic_add_return(1, &md->uevent_seq);
2743}
2744
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745uint32_t dm_get_event_nr(struct mapped_device *md)
2746{
2747 return atomic_read(&md->event_nr);
2748}
2749
2750int dm_wait_event(struct mapped_device *md, int event_nr)
2751{
2752 return wait_event_interruptible(md->eventq,
2753 (event_nr != atomic_read(&md->event_nr)));
2754}
2755
Mike Anderson7a8c3d32007-10-19 22:48:01 +01002756void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2757{
2758 unsigned long flags;
2759
2760 spin_lock_irqsave(&md->uevent_lock, flags);
2761 list_add(elist, &md->uevent_list);
2762 spin_unlock_irqrestore(&md->uevent_lock, flags);
2763}
2764
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765/*
2766 * The gendisk is only valid as long as you have a reference
2767 * count on 'md'.
2768 */
2769struct gendisk *dm_disk(struct mapped_device *md)
2770{
2771 return md->disk;
2772}
Sami Tolvanen65ff5b72015-03-18 15:52:14 +00002773EXPORT_SYMBOL_GPL(dm_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774
Milan Broz784aae72009-01-06 03:05:12 +00002775struct kobject *dm_kobject(struct mapped_device *md)
2776{
Mikulas Patocka2995fa72014-01-13 19:37:54 -05002777 return &md->kobj_holder.kobj;
Milan Broz784aae72009-01-06 03:05:12 +00002778}
2779
Milan Broz784aae72009-01-06 03:05:12 +00002780struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2781{
2782 struct mapped_device *md;
2783
Mikulas Patocka2995fa72014-01-13 19:37:54 -05002784 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
Milan Broz784aae72009-01-06 03:05:12 +00002785
Hou Taob9a41d22017-11-01 15:42:36 +08002786 spin_lock(&_minor_lock);
2787 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2788 md = NULL;
2789 goto out;
2790 }
Milan Broz784aae72009-01-06 03:05:12 +00002791 dm_get(md);
Hou Taob9a41d22017-11-01 15:42:36 +08002792out:
2793 spin_unlock(&_minor_lock);
2794
Milan Broz784aae72009-01-06 03:05:12 +00002795 return md;
2796}
2797
Kiyoshi Ueda4f186f82009-12-10 23:52:26 +00002798int dm_suspended_md(struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799{
2800 return test_bit(DMF_SUSPENDED, &md->flags);
2801}
2802
Mikulas Patocka5df96f22020-07-23 10:42:09 -04002803static int dm_post_suspending_md(struct mapped_device *md)
2804{
2805 return test_bit(DMF_POST_SUSPENDING, &md->flags);
2806}
2807
Mike Snitzerffcc3932014-10-28 18:34:52 -04002808int dm_suspended_internally_md(struct mapped_device *md)
2809{
2810 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2811}
2812
Mikulas Patocka2c140a22013-11-01 18:27:41 -04002813int dm_test_deferred_remove_flag(struct mapped_device *md)
2814{
2815 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2816}
2817
Kiyoshi Ueda64dbce52009-12-10 23:52:27 +00002818int dm_suspended(struct dm_target *ti)
2819{
Mike Snitzer33bd6f02020-09-19 13:09:11 -04002820 return dm_suspended_md(ti->table->md);
Kiyoshi Ueda64dbce52009-12-10 23:52:27 +00002821}
2822EXPORT_SYMBOL_GPL(dm_suspended);
2823
Mikulas Patocka5df96f22020-07-23 10:42:09 -04002824int dm_post_suspending(struct dm_target *ti)
2825{
Mike Snitzer33bd6f02020-09-19 13:09:11 -04002826 return dm_post_suspending_md(ti->table->md);
Mikulas Patocka5df96f22020-07-23 10:42:09 -04002827}
2828EXPORT_SYMBOL_GPL(dm_post_suspending);
2829
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -08002830int dm_noflush_suspending(struct dm_target *ti)
2831{
Mike Snitzer33bd6f02020-09-19 13:09:11 -04002832 return __noflush_suspending(ti->table->md);
Kiyoshi Ueda2e93ccc2006-12-08 02:41:09 -08002833}
2834EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2835
Bart Van Assche7e0d5742017-04-27 10:11:23 -07002836struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
Mike Snitzer0776aa02017-12-08 14:40:52 -05002837 unsigned integrity, unsigned per_io_data_size,
2838 unsigned min_pool_size)
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002839{
Mike Snitzer115485e2016-02-22 12:16:21 -05002840 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
Mike Snitzer78d8e582015-06-26 10:01:13 -04002841 unsigned int pool_size = 0;
Mike Snitzer64f52b02017-12-11 23:17:47 -05002842 unsigned int front_pad, io_front_pad;
Kent Overstreet6f1c8192018-05-20 18:25:53 -04002843 int ret;
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002844
2845 if (!pools)
Mike Snitzer4e6e36c2015-06-26 09:42:57 -04002846 return NULL;
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002847
Mike Snitzer78d8e582015-06-26 10:01:13 -04002848 switch (type) {
2849 case DM_TYPE_BIO_BASED:
Toshi Kani545ed202016-06-22 17:54:53 -06002850 case DM_TYPE_DAX_BIO_BASED:
Mike Snitzer0776aa02017-12-08 14:40:52 -05002851 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
Mike Snitzer30187e12016-01-31 13:28:26 -05002852 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
Mike Snitzer64f52b02017-12-11 23:17:47 -05002853 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
Kent Overstreet6f1c8192018-05-20 18:25:53 -04002854 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2855 if (ret)
Mike Snitzer64f52b02017-12-11 23:17:47 -05002856 goto out;
Kent Overstreet6f1c8192018-05-20 18:25:53 -04002857 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
Christoph Hellwigeb8db832017-01-22 18:32:46 +01002858 goto out;
Mike Snitzer78d8e582015-06-26 10:01:13 -04002859 break;
2860 case DM_TYPE_REQUEST_BASED:
Mike Snitzer0776aa02017-12-08 14:40:52 -05002861 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
Mike Snitzer78d8e582015-06-26 10:01:13 -04002862 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
Mike Snitzer591ddcfc2016-01-31 12:05:42 -05002863 /* per_io_data_size is used for blk-mq pdu at queue allocation */
Mike Snitzer78d8e582015-06-26 10:01:13 -04002864 break;
2865 default:
2866 BUG();
2867 }
2868
Kent Overstreet6f1c8192018-05-20 18:25:53 -04002869 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2870 if (ret)
Jun'ichi Nomura5f015202013-03-01 22:45:48 +00002871 goto out;
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002872
Kent Overstreet6f1c8192018-05-20 18:25:53 -04002873 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
Jun'ichi Nomura5f015202013-03-01 22:45:48 +00002874 goto out;
Martin K. Petersena91a2782011-03-17 11:11:05 +01002875
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002876 return pools;
Mike Snitzer78d8e582015-06-26 10:01:13 -04002877
Jun'ichi Nomura5f015202013-03-01 22:45:48 +00002878out:
2879 dm_free_md_mempools(pools);
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002880
Mike Snitzer4e6e36c2015-06-26 09:42:57 -04002881 return NULL;
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002882}
2883
2884void dm_free_md_mempools(struct dm_md_mempools *pools)
2885{
2886 if (!pools)
2887 return;
2888
Kent Overstreet6f1c8192018-05-20 18:25:53 -04002889 bioset_exit(&pools->bs);
2890 bioset_exit(&pools->io_bs);
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01002891
2892 kfree(pools);
2893}
2894
Christoph Hellwig9c72bad2016-07-08 21:23:51 +09002895struct dm_pr {
2896 u64 old_key;
2897 u64 new_key;
2898 u32 flags;
2899 bool fail_early;
2900};
2901
2902static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2903 void *data)
2904{
2905 struct mapped_device *md = bdev->bd_disk->private_data;
2906 struct dm_table *table;
2907 struct dm_target *ti;
2908 int ret = -ENOTTY, srcu_idx;
2909
2910 table = dm_get_live_table(md, &srcu_idx);
2911 if (!table || !dm_table_get_size(table))
2912 goto out;
2913
2914 /* We only support devices that have a single target */
2915 if (dm_table_get_num_targets(table) != 1)
2916 goto out;
2917 ti = dm_table_get_target(table, 0);
2918
2919 ret = -EINVAL;
2920 if (!ti->type->iterate_devices)
2921 goto out;
2922
2923 ret = ti->type->iterate_devices(ti, fn, data);
2924out:
2925 dm_put_live_table(md, srcu_idx);
2926 return ret;
2927}
2928
2929/*
2930 * For register / unregister we need to manually call out to every path.
2931 */
2932static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2933 sector_t start, sector_t len, void *data)
2934{
2935 struct dm_pr *pr = data;
2936 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2937
2938 if (!ops || !ops->pr_register)
2939 return -EOPNOTSUPP;
2940 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2941}
2942
Christoph Hellwig71cdb692015-10-15 14:10:51 +02002943static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
Mike Snitzer956a4022016-02-18 16:13:51 -05002944 u32 flags)
Christoph Hellwig71cdb692015-10-15 14:10:51 +02002945{
Christoph Hellwig9c72bad2016-07-08 21:23:51 +09002946 struct dm_pr pr = {
2947 .old_key = old_key,
2948 .new_key = new_key,
2949 .flags = flags,
2950 .fail_early = true,
2951 };
2952 int ret;
Christoph Hellwig71cdb692015-10-15 14:10:51 +02002953
Christoph Hellwig9c72bad2016-07-08 21:23:51 +09002954 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2955 if (ret && new_key) {
2956 /* unregister all paths if we failed to register any path */
2957 pr.old_key = new_key;
2958 pr.new_key = 0;
2959 pr.flags = 0;
2960 pr.fail_early = false;
2961 dm_call_pr(bdev, __dm_pr_register, &pr);
2962 }
Christoph Hellwig71cdb692015-10-15 14:10:51 +02002963
Christoph Hellwig9c72bad2016-07-08 21:23:51 +09002964 return ret;
Christoph Hellwig71cdb692015-10-15 14:10:51 +02002965}
2966
2967static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
Mike Snitzer956a4022016-02-18 16:13:51 -05002968 u32 flags)
Christoph Hellwig71cdb692015-10-15 14:10:51 +02002969{
2970 struct mapped_device *md = bdev->bd_disk->private_data;
2971 const struct pr_ops *ops;
Mike Snitzer971888c2018-04-03 15:05:12 -04002972 int r, srcu_idx;
Christoph Hellwig71cdb692015-10-15 14:10:51 +02002973
Mike Snitzer5bd5e8d2018-04-03 16:54:10 -04002974 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
Christoph Hellwig71cdb692015-10-15 14:10:51 +02002975 if (r < 0)
Mike Snitzer971888c2018-04-03 15:05:12 -04002976 goto out;
Christoph Hellwig71cdb692015-10-15 14:10:51 +02002977
2978 ops = bdev->bd_disk->fops->pr_ops;
2979 if (ops && ops->pr_reserve)
2980 r = ops->pr_reserve(bdev, key, type, flags);
2981 else
2982 r = -EOPNOTSUPP;
Mike Snitzer971888c2018-04-03 15:05:12 -04002983out:
2984 dm_unprepare_ioctl(md, srcu_idx);
Christoph Hellwig71cdb692015-10-15 14:10:51 +02002985 return r;
2986}
2987
2988static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2989{
2990 struct mapped_device *md = bdev->bd_disk->private_data;
2991 const struct pr_ops *ops;
Mike Snitzer971888c2018-04-03 15:05:12 -04002992 int r, srcu_idx;
Christoph Hellwig71cdb692015-10-15 14:10:51 +02002993
Mike Snitzer5bd5e8d2018-04-03 16:54:10 -04002994 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
Christoph Hellwig71cdb692015-10-15 14:10:51 +02002995 if (r < 0)
Mike Snitzer971888c2018-04-03 15:05:12 -04002996 goto out;
Christoph Hellwig71cdb692015-10-15 14:10:51 +02002997
2998 ops = bdev->bd_disk->fops->pr_ops;
2999 if (ops && ops->pr_release)
3000 r = ops->pr_release(bdev, key, type);
3001 else
3002 r = -EOPNOTSUPP;
Mike Snitzer971888c2018-04-03 15:05:12 -04003003out:
3004 dm_unprepare_ioctl(md, srcu_idx);
Christoph Hellwig71cdb692015-10-15 14:10:51 +02003005 return r;
3006}
3007
3008static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
Mike Snitzer956a4022016-02-18 16:13:51 -05003009 enum pr_type type, bool abort)
Christoph Hellwig71cdb692015-10-15 14:10:51 +02003010{
3011 struct mapped_device *md = bdev->bd_disk->private_data;
3012 const struct pr_ops *ops;
Mike Snitzer971888c2018-04-03 15:05:12 -04003013 int r, srcu_idx;
Christoph Hellwig71cdb692015-10-15 14:10:51 +02003014
Mike Snitzer5bd5e8d2018-04-03 16:54:10 -04003015 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
Christoph Hellwig71cdb692015-10-15 14:10:51 +02003016 if (r < 0)
Mike Snitzer971888c2018-04-03 15:05:12 -04003017 goto out;
Christoph Hellwig71cdb692015-10-15 14:10:51 +02003018
3019 ops = bdev->bd_disk->fops->pr_ops;
3020 if (ops && ops->pr_preempt)
3021 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3022 else
3023 r = -EOPNOTSUPP;
Mike Snitzer971888c2018-04-03 15:05:12 -04003024out:
3025 dm_unprepare_ioctl(md, srcu_idx);
Christoph Hellwig71cdb692015-10-15 14:10:51 +02003026 return r;
3027}
3028
3029static int dm_pr_clear(struct block_device *bdev, u64 key)
3030{
3031 struct mapped_device *md = bdev->bd_disk->private_data;
3032 const struct pr_ops *ops;
Mike Snitzer971888c2018-04-03 15:05:12 -04003033 int r, srcu_idx;
Christoph Hellwig71cdb692015-10-15 14:10:51 +02003034
Mike Snitzer5bd5e8d2018-04-03 16:54:10 -04003035 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
Christoph Hellwig71cdb692015-10-15 14:10:51 +02003036 if (r < 0)
Mike Snitzer971888c2018-04-03 15:05:12 -04003037 goto out;
Christoph Hellwig71cdb692015-10-15 14:10:51 +02003038
3039 ops = bdev->bd_disk->fops->pr_ops;
3040 if (ops && ops->pr_clear)
3041 r = ops->pr_clear(bdev, key);
3042 else
3043 r = -EOPNOTSUPP;
Mike Snitzer971888c2018-04-03 15:05:12 -04003044out:
3045 dm_unprepare_ioctl(md, srcu_idx);
Christoph Hellwig71cdb692015-10-15 14:10:51 +02003046 return r;
3047}
3048
3049static const struct pr_ops dm_pr_ops = {
3050 .pr_register = dm_pr_register,
3051 .pr_reserve = dm_pr_reserve,
3052 .pr_release = dm_pr_release,
3053 .pr_preempt = dm_pr_preempt,
3054 .pr_clear = dm_pr_clear,
3055};
3056
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07003057static const struct block_device_operations dm_blk_dops = {
Christoph Hellwigc62b37d2020-07-01 10:59:43 +02003058 .submit_bio = dm_submit_bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059 .open = dm_blk_open,
3060 .release = dm_blk_close,
Milan Brozaa129a22006-10-03 01:15:15 -07003061 .ioctl = dm_blk_ioctl,
Darrick J. Wong3ac51e72006-03-27 01:17:54 -08003062 .getgeo = dm_blk_getgeo,
Christoph Hellwige76239a2018-10-12 19:08:49 +09003063 .report_zones = dm_blk_report_zones,
Christoph Hellwig71cdb692015-10-15 14:10:51 +02003064 .pr_ops = &dm_pr_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065 .owner = THIS_MODULE
3066};
3067
Mike Snitzer681cc5e2020-10-07 16:41:01 -04003068static const struct block_device_operations dm_rq_blk_dops = {
3069 .open = dm_blk_open,
3070 .release = dm_blk_close,
3071 .ioctl = dm_blk_ioctl,
3072 .getgeo = dm_blk_getgeo,
3073 .pr_ops = &dm_pr_ops,
3074 .owner = THIS_MODULE
3075};
3076
Dan Williamsf26c5712017-04-12 12:35:44 -07003077static const struct dax_operations dm_dax_ops = {
3078 .direct_access = dm_dax_direct_access,
Dan Williams7bf7eac2019-05-16 13:26:29 -07003079 .dax_supported = dm_dax_supported,
Dan Williams7e026c82017-05-29 12:57:56 -07003080 .copy_from_iter = dm_dax_copy_from_iter,
Dan Williamsb3a9a0c2018-05-02 06:46:33 -07003081 .copy_to_iter = dm_dax_copy_to_iter,
Vivek Goyalcdf6cdc2020-02-28 11:34:54 -05003082 .zero_page_range = dm_dax_zero_page_range,
Dan Williamsf26c5712017-04-12 12:35:44 -07003083};
3084
Linus Torvalds1da177e2005-04-16 15:20:36 -07003085/*
3086 * module hooks
3087 */
3088module_init(dm_init);
3089module_exit(dm_exit);
3090
3091module_param(major, uint, 0);
3092MODULE_PARM_DESC(major, "The major number of the device mapper");
Mike Snitzerf4790822013-09-12 18:06:12 -04003093
Mike Snitzere8603132013-09-12 18:06:12 -04003094module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3095MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3096
Mike Snitzer115485e2016-02-22 12:16:21 -05003097module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3098MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3099
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100MODULE_DESCRIPTION(DM_NAME " driver");
3101MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3102MODULE_LICENSE("GPL");