blob: 0a2cc197f62b4bc5db79068fe1cef078d0b72900 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright (C) 2001 Sistina Software (UK) Limited.
Mikulas Patockad5816872009-01-06 03:05:10 +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"
Linus Torvalds1da177e2005-04-16 15:20:36 -07009
10#include <linux/module.h>
11#include <linux/vmalloc.h>
12#include <linux/blkdev.h>
13#include <linux/namei.h>
14#include <linux/ctype.h>
André Goddard Rosae7d28602009-12-14 18:01:06 -080015#include <linux/string.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/slab.h>
17#include <linux/interrupt.h>
Arjan van de Ven48c9c272006-03-27 01:18:20 -080018#include <linux/mutex.h>
Mikulas Patockad5816872009-01-06 03:05:10 +000019#include <linux/delay.h>
Arun Sharma600634972011-07-26 16:09:06 -070020#include <linux/atomic.h>
Mike Snitzerbfebd1c2015-03-08 00:51:47 -050021#include <linux/blk-mq.h>
Dan Ehrenberg644bda62015-02-10 15:20:51 -080022#include <linux/mount.h>
Vivek Goyal273752c2017-07-26 09:35:09 -040023#include <linux/dax.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024
Alasdair G Kergon72d94862006-06-26 00:27:35 -070025#define DM_MSG_PREFIX "table"
26
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#define MAX_DEPTH 16
28#define NODE_SIZE L1_CACHE_BYTES
29#define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
30#define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
31
32struct dm_table {
Mike Anderson1134e5ae2006-03-27 01:17:54 -080033 struct mapped_device *md;
Bart Van Assche7e0d5742017-04-27 10:11:23 -070034 enum dm_queue_mode type;
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
36 /* btree table */
37 unsigned int depth;
38 unsigned int counts[MAX_DEPTH]; /* in nodes */
39 sector_t *index[MAX_DEPTH];
40
41 unsigned int num_targets;
42 unsigned int num_allocated;
43 sector_t *highs;
44 struct dm_target *targets;
45
Alasdair G Kergon36a04562011-10-31 20:19:04 +000046 struct target_type *immutable_target_type;
Mike Snitzere83068a2016-05-24 21:16:51 -040047
48 bool integrity_supported:1;
49 bool singleton:1;
Milan Broz9b4b5a72017-01-04 20:23:51 +010050 unsigned integrity_added:1;
Mike Snitzer5ae89a82010-08-12 04:14:08 +010051
Linus Torvalds1da177e2005-04-16 15:20:36 -070052 /*
53 * Indicates the rw permissions for the new logical
54 * device. This should be a combination of FMODE_READ
55 * and FMODE_WRITE.
56 */
Al Viroaeb5d722008-09-02 15:28:45 -040057 fmode_t mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
59 /* a list of devices used by this table */
60 struct list_head devices;
61
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 /* events get handed up using this callback */
63 void (*event_fn)(void *);
64 void *event_context;
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +010065
66 struct dm_md_mempools *mempools;
NeilBrown9d357b02011-01-13 20:00:01 +000067
68 struct list_head target_callbacks;
Linus Torvalds1da177e2005-04-16 15:20:36 -070069};
70
71/*
72 * Similar to ceiling(log_size(n))
73 */
74static unsigned int int_log(unsigned int n, unsigned int base)
75{
76 int result = 0;
77
78 while (n > 1) {
79 n = dm_div_up(n, base);
80 result++;
81 }
82
83 return result;
84}
85
86/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 * Calculate the index of the child node of the n'th node k'th key.
88 */
89static inline unsigned int get_child(unsigned int n, unsigned int k)
90{
91 return (n * CHILDREN_PER_NODE) + k;
92}
93
94/*
95 * Return the n'th node of level l from table t.
96 */
97static inline sector_t *get_node(struct dm_table *t,
98 unsigned int l, unsigned int n)
99{
100 return t->index[l] + (n * KEYS_PER_NODE);
101}
102
103/*
104 * Return the highest key that you could lookup from the n'th
105 * node on level l of the btree.
106 */
107static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
108{
109 for (; l < t->depth - 1; l++)
110 n = get_child(n, CHILDREN_PER_NODE - 1);
111
112 if (n >= t->counts[l])
113 return (sector_t) - 1;
114
115 return get_node(t, l, n)[KEYS_PER_NODE - 1];
116}
117
118/*
119 * Fills in a level of the btree based on the highs of the level
120 * below it.
121 */
122static int setup_btree_index(unsigned int l, struct dm_table *t)
123{
124 unsigned int n, k;
125 sector_t *node;
126
127 for (n = 0U; n < t->counts[l]; n++) {
128 node = get_node(t, l, n);
129
130 for (k = 0U; k < KEYS_PER_NODE; k++)
131 node[k] = high(t, l + 1, get_child(n, k));
132 }
133
134 return 0;
135}
136
137void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
138{
139 unsigned long size;
140 void *addr;
141
142 /*
143 * Check that we're not going to overflow.
144 */
145 if (nmemb > (ULONG_MAX / elem_size))
146 return NULL;
147
148 size = nmemb * elem_size;
Joe Perchese29e65a2011-08-02 12:32:02 +0100149 addr = vzalloc(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
151 return addr;
152}
Mike Snitzer08649012011-08-02 12:32:04 +0100153EXPORT_SYMBOL(dm_vcalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
155/*
156 * highs, and targets are managed as dynamic arrays during a
157 * table load.
158 */
159static int alloc_targets(struct dm_table *t, unsigned int num)
160{
161 sector_t *n_highs;
162 struct dm_target *n_targets;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
164 /*
165 * Allocate both the target array and offset array at once.
166 */
Mikulas Patocka123d87d2019-08-23 09:55:26 -0400167 n_highs = (sector_t *) dm_vcalloc(num, sizeof(struct dm_target) +
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 sizeof(sector_t));
169 if (!n_highs)
170 return -ENOMEM;
171
172 n_targets = (struct dm_target *) (n_highs + num);
173
Mikulas Patocka57a2f232013-11-22 19:51:39 -0500174 memset(n_highs, -1, sizeof(*n_highs) * num);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 vfree(t->highs);
176
177 t->num_allocated = num;
178 t->highs = n_highs;
179 t->targets = n_targets;
180
181 return 0;
182}
183
Al Viroaeb5d722008-09-02 15:28:45 -0400184int dm_table_create(struct dm_table **result, fmode_t mode,
Mike Anderson1134e5ae2006-03-27 01:17:54 -0800185 unsigned num_targets, struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186{
Dmitry Monakhov094262d2007-10-19 22:38:51 +0100187 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
189 if (!t)
190 return -ENOMEM;
191
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 INIT_LIST_HEAD(&t->devices);
NeilBrown9d357b02011-01-13 20:00:01 +0000193 INIT_LIST_HEAD(&t->target_callbacks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
195 if (!num_targets)
196 num_targets = KEYS_PER_NODE;
197
198 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
199
Mikulas Patocka5b2d0652013-11-22 19:52:06 -0500200 if (!num_targets) {
201 kfree(t);
202 return -ENOMEM;
203 }
204
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 if (alloc_targets(t, num_targets)) {
206 kfree(t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 return -ENOMEM;
208 }
209
Mike Snitzere83068a2016-05-24 21:16:51 -0400210 t->type = DM_TYPE_NONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 t->mode = mode;
Mike Anderson1134e5ae2006-03-27 01:17:54 -0800212 t->md = md;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 *result = t;
214 return 0;
215}
216
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500217static void free_devices(struct list_head *devices, struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218{
219 struct list_head *tmp, *next;
220
Paul Jimenezafb24522008-02-08 02:09:59 +0000221 list_for_each_safe(tmp, next, devices) {
Mikulas Patocka82b15192008-10-10 13:37:09 +0100222 struct dm_dev_internal *dd =
223 list_entry(tmp, struct dm_dev_internal, list);
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500224 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
225 dm_device_name(md), dd->dm_dev->name);
226 dm_put_table_device(md, dd->dm_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 kfree(dd);
228 }
229}
230
Mikulas Patockad5816872009-01-06 03:05:10 +0000231void dm_table_destroy(struct dm_table *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232{
233 unsigned int i;
234
Alasdair G Kergona7940152009-12-10 23:52:23 +0000235 if (!t)
236 return;
237
Will Drewry26803b92010-08-12 04:14:03 +0100238 /* free the indexes */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 if (t->depth >= 2)
240 vfree(t->index[t->depth - 2]);
241
242 /* free the targets */
243 for (i = 0; i < t->num_targets; i++) {
244 struct dm_target *tgt = t->targets + i;
245
246 if (tgt->type->dtr)
247 tgt->type->dtr(tgt);
248
249 dm_put_target_type(tgt->type);
250 }
251
252 vfree(t->highs);
253
254 /* free the device list */
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500255 free_devices(&t->devices, t->md);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100257 dm_free_md_mempools(t->mempools);
258
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 kfree(t);
260}
261
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 * See if we've already got a device in the list.
264 */
Mikulas Patocka82b15192008-10-10 13:37:09 +0100265static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266{
Mikulas Patocka82b15192008-10-10 13:37:09 +0100267 struct dm_dev_internal *dd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268
269 list_for_each_entry (dd, l, list)
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500270 if (dd->dm_dev->bdev->bd_dev == dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 return dd;
272
273 return NULL;
274}
275
276/*
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100277 * If possible, this checks an area of a destination device is invalid.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 */
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100279static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
280 sector_t start, sector_t len, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281{
Milan Brozf4808ca2011-05-29 13:02:52 +0100282 struct request_queue *q;
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100283 struct queue_limits *limits = data;
284 struct block_device *bdev = dev->bdev;
285 sector_t dev_size =
286 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100287 unsigned short logical_block_size_sectors =
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100288 limits->logical_block_size >> SECTOR_SHIFT;
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100289 char b[BDEVNAME_SIZE];
Mike Anderson2cd54d92007-05-09 02:32:57 -0700290
Milan Brozf4808ca2011-05-29 13:02:52 +0100291 /*
292 * Some devices exist without request functions,
293 * such as loop devices not yet bound to backing files.
294 * Forbid the use of such devices.
295 */
296 q = bdev_get_queue(bdev);
297 if (!q || !q->make_request_fn) {
298 DMWARN("%s: %s is not yet initialised: "
299 "start=%llu, len=%llu, dev_size=%llu",
300 dm_device_name(ti->table->md), bdevname(bdev, b),
301 (unsigned long long)start,
302 (unsigned long long)len,
303 (unsigned long long)dev_size);
304 return 1;
305 }
306
Mike Anderson2cd54d92007-05-09 02:32:57 -0700307 if (!dev_size)
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100308 return 0;
Mike Anderson2cd54d92007-05-09 02:32:57 -0700309
Mike Snitzer5dea2712009-07-23 20:30:42 +0100310 if ((start >= dev_size) || (start + len > dev_size)) {
Mike Snitzera963a952009-09-04 20:40:24 +0100311 DMWARN("%s: %s too small for target: "
312 "start=%llu, len=%llu, dev_size=%llu",
313 dm_device_name(ti->table->md), bdevname(bdev, b),
314 (unsigned long long)start,
315 (unsigned long long)len,
316 (unsigned long long)dev_size);
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100317 return 1;
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100318 }
319
Damien Le Moaldd88d312017-05-08 16:40:43 -0700320 /*
321 * If the target is mapped to zoned block device(s), check
322 * that the zones are not partially mapped.
323 */
324 if (bdev_zoned_model(bdev) != BLK_ZONED_NONE) {
325 unsigned int zone_sectors = bdev_zone_sectors(bdev);
326
327 if (start & (zone_sectors - 1)) {
328 DMWARN("%s: start=%llu not aligned to h/w zone size %u of %s",
329 dm_device_name(ti->table->md),
330 (unsigned long long)start,
331 zone_sectors, bdevname(bdev, b));
332 return 1;
333 }
334
335 /*
336 * Note: The last zone of a zoned block device may be smaller
337 * than other zones. So for a target mapping the end of a
338 * zoned block device with such a zone, len would not be zone
339 * aligned. We do not allow such last smaller zone to be part
340 * of the mapping here to ensure that mappings with multiple
341 * devices do not end up with a smaller zone in the middle of
342 * the sector range.
343 */
344 if (len & (zone_sectors - 1)) {
345 DMWARN("%s: len=%llu not aligned to h/w zone size %u of %s",
346 dm_device_name(ti->table->md),
347 (unsigned long long)len,
348 zone_sectors, bdevname(bdev, b));
349 return 1;
350 }
351 }
352
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100353 if (logical_block_size_sectors <= 1)
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100354 return 0;
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100355
356 if (start & (logical_block_size_sectors - 1)) {
357 DMWARN("%s: start=%llu not aligned to h/w "
Mike Snitzera963a952009-09-04 20:40:24 +0100358 "logical block size %u of %s",
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100359 dm_device_name(ti->table->md),
360 (unsigned long long)start,
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100361 limits->logical_block_size, bdevname(bdev, b));
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100362 return 1;
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100363 }
364
Mike Snitzer5dea2712009-07-23 20:30:42 +0100365 if (len & (logical_block_size_sectors - 1)) {
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100366 DMWARN("%s: len=%llu not aligned to h/w "
Mike Snitzera963a952009-09-04 20:40:24 +0100367 "logical block size %u of %s",
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100368 dm_device_name(ti->table->md),
Mike Snitzer5dea2712009-07-23 20:30:42 +0100369 (unsigned long long)len,
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100370 limits->logical_block_size, bdevname(bdev, b));
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100371 return 1;
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100372 }
373
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100374 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375}
376
377/*
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100378 * This upgrades the mode on an already open dm_dev, being
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 * careful to leave things as they were if we fail to reopen the
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100380 * device and not to touch the existing bdev field in case
381 * it is accessed concurrently inside dm_table_any_congested().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 */
Al Viroaeb5d722008-09-02 15:28:45 -0400383static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
Mikulas Patocka82b15192008-10-10 13:37:09 +0100384 struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385{
386 int r;
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500387 struct dm_dev *old_dev, *new_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500389 old_dev = dd->dm_dev;
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100390
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500391 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
392 dd->dm_dev->mode | new_mode, &new_dev);
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100393 if (r)
394 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500396 dd->dm_dev = new_dev;
397 dm_put_table_device(md, old_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100399 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400}
401
402/*
DingXiang4df2bf42016-02-02 12:29:18 +0800403 * Convert the path to a device
404 */
405dev_t dm_get_dev_t(const char *path)
406{
Mikulas Patocka3c120162017-04-18 16:51:46 -0400407 dev_t dev;
DingXiang4df2bf42016-02-02 12:29:18 +0800408 struct block_device *bdev;
409
410 bdev = lookup_bdev(path);
411 if (IS_ERR(bdev))
412 dev = name_to_dev_t(path);
413 else {
414 dev = bdev->bd_dev;
415 bdput(bdev);
416 }
417
418 return dev;
419}
420EXPORT_SYMBOL_GPL(dm_get_dev_t);
421
422/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 * Add a device to the list, or just increment the usage count if
424 * it's already present.
425 */
Mike Snitzer08649012011-08-02 12:32:04 +0100426int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
427 struct dm_dev **result)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428{
429 int r;
DingXiang4df2bf42016-02-02 12:29:18 +0800430 dev_t dev;
Mikulas Patocka82b15192008-10-10 13:37:09 +0100431 struct dm_dev_internal *dd;
Mike Snitzer08649012011-08-02 12:32:04 +0100432 struct dm_table *t = ti->table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433
Eric Sesterhenn547bc922006-03-26 18:22:50 +0200434 BUG_ON(!t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
DingXiang4df2bf42016-02-02 12:29:18 +0800436 dev = dm_get_dev_t(path);
437 if (!dev)
438 return -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439
440 dd = find_device(&t->devices, dev);
441 if (!dd) {
442 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
443 if (!dd)
444 return -ENOMEM;
445
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500446 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 kfree(dd);
448 return r;
449 }
450
Elena Reshetova2a0b4682017-10-20 10:37:38 +0300451 refcount_set(&dd->count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 list_add(&dd->list, &t->devices);
Mike Snitzerafc567a2017-11-25 00:27:26 -0500453 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500455 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
Jun'ichi Nomuraf1659212006-03-27 01:17:59 -0800456 r = upgrade_mode(dd, mode, t->md);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 if (r)
458 return r;
459 }
Mike Snitzerafc567a2017-11-25 00:27:26 -0500460 refcount_inc(&dd->count);
461out:
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500462 *result = dd->dm_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 return 0;
464}
Mike Snitzer08649012011-08-02 12:32:04 +0100465EXPORT_SYMBOL(dm_get_device);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
Mike Snitzer11f04312014-06-03 10:30:28 -0400467static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
468 sector_t start, sector_t len, void *data)
Bryn Reeves3cb40212006-10-03 01:15:42 -0700469{
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100470 struct queue_limits *limits = data;
471 struct block_device *bdev = dev->bdev;
Jens Axboe165125e2007-07-24 09:28:11 +0200472 struct request_queue *q = bdev_get_queue(bdev);
Alasdair G Kergon0c2322e2008-10-10 13:37:13 +0100473 char b[BDEVNAME_SIZE];
474
475 if (unlikely(!q)) {
476 DMWARN("%s: Cannot set limits for nonexistent device %s",
477 dm_device_name(ti->table->md), bdevname(bdev, b));
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100478 return 0;
Alasdair G Kergon0c2322e2008-10-10 13:37:13 +0100479 }
Bryn Reeves3cb40212006-10-03 01:15:42 -0700480
Martin K. Petersenb27d7f12010-01-11 03:21:50 -0500481 if (bdev_stack_limits(limits, bdev, start) < 0)
482 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
Mike Snitzera963a952009-09-04 20:40:24 +0100483 "physical_block_size=%u, logical_block_size=%u, "
484 "alignment_offset=%u, start=%llu",
485 dm_device_name(ti->table->md), bdevname(bdev, b),
486 q->limits.physical_block_size,
487 q->limits.logical_block_size,
488 q->limits.alignment_offset,
Martin K. Petersenb27d7f12010-01-11 03:21:50 -0500489 (unsigned long long) start << SECTOR_SHIFT);
Bryn Reeves3cb40212006-10-03 01:15:42 -0700490
Damien Le Moaldd88d312017-05-08 16:40:43 -0700491 limits->zoned = blk_queue_zoned_model(q);
492
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100493 return 0;
Bryn Reeves3cb40212006-10-03 01:15:42 -0700494}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496/*
Mike Snitzer08649012011-08-02 12:32:04 +0100497 * Decrement a device's use count and remove it if necessary.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 */
Mikulas Patocka82b15192008-10-10 13:37:09 +0100499void dm_put_device(struct dm_target *ti, struct dm_dev *d)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500{
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500501 int found = 0;
502 struct list_head *devices = &ti->table->devices;
503 struct dm_dev_internal *dd;
Mikulas Patocka82b15192008-10-10 13:37:09 +0100504
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500505 list_for_each_entry(dd, devices, list) {
506 if (dd->dm_dev == d) {
507 found = 1;
508 break;
509 }
510 }
511 if (!found) {
512 DMWARN("%s: device %s not in table devices list",
513 dm_device_name(ti->table->md), d->name);
514 return;
515 }
Elena Reshetova2a0b4682017-10-20 10:37:38 +0300516 if (refcount_dec_and_test(&dd->count)) {
Benjamin Marzinski86f11522014-08-13 13:53:43 -0500517 dm_put_table_device(ti->table->md, d);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 list_del(&dd->list);
519 kfree(dd);
520 }
521}
Mike Snitzer08649012011-08-02 12:32:04 +0100522EXPORT_SYMBOL(dm_put_device);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523
524/*
525 * Checks to see if the target joins onto the end of the table.
526 */
527static int adjoin(struct dm_table *table, struct dm_target *ti)
528{
529 struct dm_target *prev;
530
531 if (!table->num_targets)
532 return !ti->begin;
533
534 prev = &table->targets[table->num_targets - 1];
535 return (ti->begin == (prev->begin + prev->len));
536}
537
538/*
539 * Used to dynamically allocate the arg array.
Mikulas Patockaf36afb32013-10-31 13:55:45 -0400540 *
541 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
542 * process messages even if some device is suspended. These messages have a
543 * small fixed number of arguments.
544 *
545 * On the other hand, dm-switch needs to process bulk data using messages and
546 * excessive use of GFP_NOIO could cause trouble.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 */
Kees Cook610b15c2018-05-07 16:47:02 -0700548static char **realloc_argv(unsigned *size, char **old_argv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549{
550 char **argv;
551 unsigned new_size;
Mikulas Patockaf36afb32013-10-31 13:55:45 -0400552 gfp_t gfp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553
Kees Cook610b15c2018-05-07 16:47:02 -0700554 if (*size) {
555 new_size = *size * 2;
Mikulas Patockaf36afb32013-10-31 13:55:45 -0400556 gfp = GFP_KERNEL;
557 } else {
558 new_size = 8;
559 gfp = GFP_NOIO;
560 }
Kees Cook6da2ec52018-06-12 13:55:00 -0700561 argv = kmalloc_array(new_size, sizeof(*argv), gfp);
Jerome Marchanda0651922019-06-12 18:22:26 +0200562 if (argv && old_argv) {
Kees Cook610b15c2018-05-07 16:47:02 -0700563 memcpy(argv, old_argv, *size * sizeof(*argv));
564 *size = new_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 }
566
567 kfree(old_argv);
568 return argv;
569}
570
571/*
572 * Destructively splits up the argument list to pass to ctr.
573 */
574int dm_split_args(int *argc, char ***argvp, char *input)
575{
576 char *start, *end = input, *out, **argv = NULL;
577 unsigned array_size = 0;
578
579 *argc = 0;
David Teigland814d6862006-06-26 00:27:31 -0700580
581 if (!input) {
582 *argvp = NULL;
583 return 0;
584 }
585
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 argv = realloc_argv(&array_size, argv);
587 if (!argv)
588 return -ENOMEM;
589
590 while (1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 /* Skip whitespace */
André Goddard Rosae7d28602009-12-14 18:01:06 -0800592 start = skip_spaces(end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
594 if (!*start)
595 break; /* success, we hit the end */
596
597 /* 'out' is used to remove any back-quotes */
598 end = out = start;
599 while (*end) {
600 /* Everything apart from '\0' can be quoted */
601 if (*end == '\\' && *(end + 1)) {
602 *out++ = *(end + 1);
603 end += 2;
604 continue;
605 }
606
607 if (isspace(*end))
608 break; /* end of token */
609
610 *out++ = *end++;
611 }
612
613 /* have we already filled the array ? */
614 if ((*argc + 1) > array_size) {
615 argv = realloc_argv(&array_size, argv);
616 if (!argv)
617 return -ENOMEM;
618 }
619
620 /* we know this is whitespace */
621 if (*end)
622 end++;
623
624 /* terminate the string and put it in the array */
625 *out = '\0';
626 argv[*argc] = start;
627 (*argc)++;
628 }
629
630 *argvp = argv;
631 return 0;
632}
633
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100634/*
635 * Impose necessary and sufficient conditions on a devices's table such
636 * that any incoming bio which respects its logical_block_size can be
637 * processed successfully. If it falls across the boundary between
638 * two or more targets, the size of each piece it gets split into must
639 * be compatible with the logical_block_size of the target processing it.
640 */
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100641static int validate_hardware_logical_block_alignment(struct dm_table *table,
642 struct queue_limits *limits)
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100643{
644 /*
645 * This function uses arithmetic modulo the logical_block_size
646 * (in units of 512-byte sectors).
647 */
648 unsigned short device_logical_block_size_sects =
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100649 limits->logical_block_size >> SECTOR_SHIFT;
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100650
651 /*
652 * Offset of the start of the next table entry, mod logical_block_size.
653 */
654 unsigned short next_target_start = 0;
655
656 /*
657 * Given an aligned bio that extends beyond the end of a
658 * target, how many sectors must the next target handle?
659 */
660 unsigned short remaining = 0;
661
662 struct dm_target *uninitialized_var(ti);
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100663 struct queue_limits ti_limits;
Mikulas Patocka3c120162017-04-18 16:51:46 -0400664 unsigned i;
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100665
666 /*
667 * Check each entry in the table in turn.
668 */
Mikulas Patocka3c120162017-04-18 16:51:46 -0400669 for (i = 0; i < dm_table_get_num_targets(table); i++) {
670 ti = dm_table_get_target(table, i);
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100671
Martin K. Petersenb1bd0552012-01-11 16:27:11 +0100672 blk_set_stacking_limits(&ti_limits);
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100673
674 /* combine all target devices' limits */
675 if (ti->type->iterate_devices)
676 ti->type->iterate_devices(ti, dm_set_device_limits,
677 &ti_limits);
678
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100679 /*
680 * If the remaining sectors fall entirely within this
681 * table entry are they compatible with its logical_block_size?
682 */
683 if (remaining < ti->len &&
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100684 remaining & ((ti_limits.logical_block_size >>
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100685 SECTOR_SHIFT) - 1))
686 break; /* Error */
687
688 next_target_start =
689 (unsigned short) ((next_target_start + ti->len) &
690 (device_logical_block_size_sects - 1));
691 remaining = next_target_start ?
692 device_logical_block_size_sects - next_target_start : 0;
693 }
694
695 if (remaining) {
696 DMWARN("%s: table line %u (start sect %llu len %llu) "
Mike Snitzera963a952009-09-04 20:40:24 +0100697 "not aligned to h/w logical block size %u",
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100698 dm_device_name(table->md), i,
699 (unsigned long long) ti->begin,
700 (unsigned long long) ti->len,
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100701 limits->logical_block_size);
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100702 return -EINVAL;
703 }
704
705 return 0;
706}
707
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708int dm_table_add_target(struct dm_table *t, const char *type,
709 sector_t start, sector_t len, char *params)
710{
711 int r = -EINVAL, argc;
712 char **argv;
713 struct dm_target *tgt;
714
Alasdair G Kergon3791e2f2011-10-31 20:19:00 +0000715 if (t->singleton) {
716 DMERR("%s: target type %s must appear alone in table",
717 dm_device_name(t->md), t->targets->type->name);
718 return -EINVAL;
719 }
720
Mikulas Patocka57a2f232013-11-22 19:51:39 -0500721 BUG_ON(t->num_targets >= t->num_allocated);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
723 tgt = t->targets + t->num_targets;
724 memset(tgt, 0, sizeof(*tgt));
725
726 if (!len) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700727 DMERR("%s: zero-length target", dm_device_name(t->md));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 return -EINVAL;
729 }
730
731 tgt->type = dm_get_target_type(type);
732 if (!tgt->type) {
tang.junhuidafa7242016-10-21 09:35:32 +0800733 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 return -EINVAL;
735 }
736
Alasdair G Kergon3791e2f2011-10-31 20:19:00 +0000737 if (dm_target_needs_singleton(tgt->type)) {
738 if (t->num_targets) {
tang.junhuidafa7242016-10-21 09:35:32 +0800739 tgt->error = "singleton target type must appear alone in table";
740 goto bad;
Alasdair G Kergon3791e2f2011-10-31 20:19:00 +0000741 }
Mike Snitzere83068a2016-05-24 21:16:51 -0400742 t->singleton = true;
Alasdair G Kergon3791e2f2011-10-31 20:19:00 +0000743 }
744
Alasdair G Kergoncc6cbe12011-10-31 20:19:02 +0000745 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
tang.junhuidafa7242016-10-21 09:35:32 +0800746 tgt->error = "target type may not be included in a read-only table";
747 goto bad;
Alasdair G Kergoncc6cbe12011-10-31 20:19:02 +0000748 }
749
Alasdair G Kergon36a04562011-10-31 20:19:04 +0000750 if (t->immutable_target_type) {
751 if (t->immutable_target_type != tgt->type) {
tang.junhuidafa7242016-10-21 09:35:32 +0800752 tgt->error = "immutable target type cannot be mixed with other target types";
753 goto bad;
Alasdair G Kergon36a04562011-10-31 20:19:04 +0000754 }
755 } else if (dm_target_is_immutable(tgt->type)) {
756 if (t->num_targets) {
tang.junhuidafa7242016-10-21 09:35:32 +0800757 tgt->error = "immutable target type cannot be mixed with other target types";
758 goto bad;
Alasdair G Kergon36a04562011-10-31 20:19:04 +0000759 }
760 t->immutable_target_type = tgt->type;
761 }
762
Milan Broz9b4b5a72017-01-04 20:23:51 +0100763 if (dm_target_has_integrity(tgt->type))
764 t->integrity_added = 1;
765
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 tgt->table = t;
767 tgt->begin = start;
768 tgt->len = len;
769 tgt->error = "Unknown error";
770
771 /*
772 * Does this target adjoin the previous one ?
773 */
774 if (!adjoin(t, tgt)) {
775 tgt->error = "Gap in table";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 goto bad;
777 }
778
779 r = dm_split_args(&argc, &argv, params);
780 if (r) {
781 tgt->error = "couldn't split parameters (insufficient memory)";
782 goto bad;
783 }
784
785 r = tgt->type->ctr(tgt, argc, argv);
786 kfree(argv);
787 if (r)
788 goto bad;
789
790 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
791
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +0000792 if (!tgt->num_discard_bios && tgt->discards_supported)
793 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
Mike Snitzer936688d2011-08-02 12:32:01 +0100794 dm_device_name(t->md), type);
Mike Snitzer5ae89a82010-08-12 04:14:08 +0100795
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 return 0;
797
798 bad:
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700799 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 dm_put_target_type(tgt->type);
801 return r;
802}
803
Mike Snitzer498f0102011-08-02 12:32:04 +0100804/*
805 * Target argument parsing helpers.
806 */
Eric Biggers5916a222017-06-22 11:32:45 -0700807static int validate_next_arg(const struct dm_arg *arg,
808 struct dm_arg_set *arg_set,
Mike Snitzer498f0102011-08-02 12:32:04 +0100809 unsigned *value, char **error, unsigned grouped)
810{
811 const char *arg_str = dm_shift_arg(arg_set);
Mikulas Patocka31998ef2012-03-28 18:41:26 +0100812 char dummy;
Mike Snitzer498f0102011-08-02 12:32:04 +0100813
814 if (!arg_str ||
Mikulas Patocka31998ef2012-03-28 18:41:26 +0100815 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
Mike Snitzer498f0102011-08-02 12:32:04 +0100816 (*value < arg->min) ||
817 (*value > arg->max) ||
818 (grouped && arg_set->argc < *value)) {
819 *error = arg->error;
820 return -EINVAL;
821 }
822
823 return 0;
824}
825
Eric Biggers5916a222017-06-22 11:32:45 -0700826int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
Mike Snitzer498f0102011-08-02 12:32:04 +0100827 unsigned *value, char **error)
828{
829 return validate_next_arg(arg, arg_set, value, error, 0);
830}
831EXPORT_SYMBOL(dm_read_arg);
832
Eric Biggers5916a222017-06-22 11:32:45 -0700833int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
Mike Snitzer498f0102011-08-02 12:32:04 +0100834 unsigned *value, char **error)
835{
836 return validate_next_arg(arg, arg_set, value, error, 1);
837}
838EXPORT_SYMBOL(dm_read_arg_group);
839
840const char *dm_shift_arg(struct dm_arg_set *as)
841{
842 char *r;
843
844 if (as->argc) {
845 as->argc--;
846 r = *as->argv;
847 as->argv++;
848 return r;
849 }
850
851 return NULL;
852}
853EXPORT_SYMBOL(dm_shift_arg);
854
855void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
856{
857 BUG_ON(as->argc < num_args);
858 as->argc -= num_args;
859 as->argv += num_args;
860}
861EXPORT_SYMBOL(dm_consume_args);
862
Bart Van Assche7e0d5742017-04-27 10:11:23 -0700863static bool __table_type_bio_based(enum dm_queue_mode table_type)
Toshi Kani545ed202016-06-22 17:54:53 -0600864{
865 return (table_type == DM_TYPE_BIO_BASED ||
Mike Snitzer22c11852017-12-04 21:07:37 -0500866 table_type == DM_TYPE_DAX_BIO_BASED ||
867 table_type == DM_TYPE_NVME_BIO_BASED);
Toshi Kani545ed202016-06-22 17:54:53 -0600868}
869
Bart Van Assche7e0d5742017-04-27 10:11:23 -0700870static bool __table_type_request_based(enum dm_queue_mode table_type)
Junichi Nomura15b94a62015-05-29 08:51:03 +0000871{
Mike Snitzer953923c2018-10-11 11:06:29 -0400872 return table_type == DM_TYPE_REQUEST_BASED;
Junichi Nomura15b94a62015-05-29 08:51:03 +0000873}
874
Bart Van Assche7e0d5742017-04-27 10:11:23 -0700875void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
Mike Snitzere83068a2016-05-24 21:16:51 -0400876{
877 t->type = type;
878}
879EXPORT_SYMBOL_GPL(dm_table_set_type);
880
Dan Williams7bf7eac2019-05-16 13:26:29 -0700881/* validate the dax capability of the target device span */
Pankaj Gupta2e9ee092019-07-05 19:33:25 +0530882int device_supports_dax(struct dm_target *ti, struct dm_dev *dev,
Mike Snitzer9c50a982019-07-30 14:39:43 -0400883 sector_t start, sector_t len, void *data)
Toshi Kani545ed202016-06-22 17:54:53 -0600884{
Dan Williams7bf7eac2019-05-16 13:26:29 -0700885 int blocksize = *(int *) data;
886
887 return generic_fsdax_supported(dev->dax_dev, dev->bdev, blocksize,
Mike Snitzer9c50a982019-07-30 14:39:43 -0400888 start, len);
Toshi Kani545ed202016-06-22 17:54:53 -0600889}
890
Pankaj Gupta2e9ee092019-07-05 19:33:25 +0530891/* Check devices support synchronous DAX */
Mike Snitzer9c50a982019-07-30 14:39:43 -0400892static int device_dax_synchronous(struct dm_target *ti, struct dm_dev *dev,
893 sector_t start, sector_t len, void *data)
Pankaj Gupta2e9ee092019-07-05 19:33:25 +0530894{
Pankaj Gupta5348deb2019-07-30 17:07:08 +0530895 return dev->dax_dev && dax_synchronous(dev->dax_dev);
Pankaj Gupta2e9ee092019-07-05 19:33:25 +0530896}
897
898bool dm_table_supports_dax(struct dm_table *t,
Mike Snitzer9c50a982019-07-30 14:39:43 -0400899 iterate_devices_callout_fn iterate_fn, int *blocksize)
Toshi Kani545ed202016-06-22 17:54:53 -0600900{
901 struct dm_target *ti;
Mikulas Patocka3c120162017-04-18 16:51:46 -0400902 unsigned i;
Toshi Kani545ed202016-06-22 17:54:53 -0600903
904 /* Ensure that all targets support DAX. */
Mikulas Patocka3c120162017-04-18 16:51:46 -0400905 for (i = 0; i < dm_table_get_num_targets(t); i++) {
906 ti = dm_table_get_target(t, i);
Toshi Kani545ed202016-06-22 17:54:53 -0600907
908 if (!ti->type->direct_access)
909 return false;
910
911 if (!ti->type->iterate_devices ||
Mike Snitzer9c50a982019-07-30 14:39:43 -0400912 !ti->type->iterate_devices(ti, iterate_fn, blocksize))
Toshi Kani545ed202016-06-22 17:54:53 -0600913 return false;
914 }
915
916 return true;
917}
918
Mike Snitzer22c11852017-12-04 21:07:37 -0500919static bool dm_table_does_not_support_partial_completion(struct dm_table *t);
920
Mike Snitzer6ba01df2019-11-05 10:43:44 -0500921static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev,
922 sector_t start, sector_t len, void *data)
Mike Snitzereaa160e2018-01-13 14:33:30 -0500923{
Mike Snitzer6ba01df2019-11-05 10:43:44 -0500924 struct block_device *bdev = dev->bdev;
925 struct request_queue *q = bdev_get_queue(bdev);
Mike Snitzereaa160e2018-01-13 14:33:30 -0500926
Mike Snitzer6ba01df2019-11-05 10:43:44 -0500927 /* request-based cannot stack on partitions! */
928 if (bdev != bdev->bd_contains)
929 return false;
Mike Snitzereaa160e2018-01-13 14:33:30 -0500930
Jens Axboe344e9ff2018-11-15 12:22:51 -0700931 return queue_is_mq(q);
Mike Snitzereaa160e2018-01-13 14:33:30 -0500932}
933
Mike Snitzere83068a2016-05-24 21:16:51 -0400934static int dm_table_determine_type(struct dm_table *t)
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100935{
936 unsigned i;
Mike Snitzer169e2cc2013-08-22 18:21:38 -0400937 unsigned bio_based = 0, request_based = 0, hybrid = 0;
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100938 struct dm_target *tgt;
Mike Snitzere83068a2016-05-24 21:16:51 -0400939 struct list_head *devices = dm_table_get_devices(t);
Bart Van Assche7e0d5742017-04-27 10:11:23 -0700940 enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
Pankaj Gupta2e9ee092019-07-05 19:33:25 +0530941 int page_size = PAGE_SIZE;
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100942
Mike Snitzere83068a2016-05-24 21:16:51 -0400943 if (t->type != DM_TYPE_NONE) {
944 /* target already set the table's type */
Mike Snitzerc934eda2018-03-05 15:26:06 -0500945 if (t->type == DM_TYPE_BIO_BASED) {
946 /* possibly upgrade to a variant of bio-based */
947 goto verify_bio_based;
Mike Snitzer22c11852017-12-04 21:07:37 -0500948 }
Toshi Kani545ed202016-06-22 17:54:53 -0600949 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
Mike Snitzerc934eda2018-03-05 15:26:06 -0500950 BUG_ON(t->type == DM_TYPE_NVME_BIO_BASED);
Mike Snitzere83068a2016-05-24 21:16:51 -0400951 goto verify_rq_based;
952 }
953
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100954 for (i = 0; i < t->num_targets; i++) {
955 tgt = t->targets + i;
Mike Snitzer169e2cc2013-08-22 18:21:38 -0400956 if (dm_target_hybrid(tgt))
957 hybrid = 1;
958 else if (dm_target_request_based(tgt))
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100959 request_based = 1;
960 else
961 bio_based = 1;
962
963 if (bio_based && request_based) {
Mike Snitzer22c11852017-12-04 21:07:37 -0500964 DMERR("Inconsistent table: different target types"
965 " can't be mixed up");
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100966 return -EINVAL;
967 }
968 }
969
Mike Snitzer169e2cc2013-08-22 18:21:38 -0400970 if (hybrid && !bio_based && !request_based) {
971 /*
972 * The targets can work either way.
973 * Determine the type from the live device.
974 * Default to bio-based if device is new.
975 */
Junichi Nomura15b94a62015-05-29 08:51:03 +0000976 if (__table_type_request_based(live_md_type))
Mike Snitzer169e2cc2013-08-22 18:21:38 -0400977 request_based = 1;
978 else
979 bio_based = 1;
980 }
981
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100982 if (bio_based) {
Mike Snitzerc934eda2018-03-05 15:26:06 -0500983verify_bio_based:
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100984 /* We must use this table as bio-based */
985 t->type = DM_TYPE_BIO_BASED;
Pankaj Gupta2e9ee092019-07-05 19:33:25 +0530986 if (dm_table_supports_dax(t, device_supports_dax, &page_size) ||
Mike Snitzer22c11852017-12-04 21:07:37 -0500987 (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
Toshi Kani545ed202016-06-22 17:54:53 -0600988 t->type = DM_TYPE_DAX_BIO_BASED;
Mike Snitzereaa160e2018-01-13 14:33:30 -0500989 } else {
990 /* Check if upgrading to NVMe bio-based is valid or required */
991 tgt = dm_table_get_immutable_target(t);
992 if (tgt && !tgt->max_io_len && dm_table_does_not_support_partial_completion(t)) {
993 t->type = DM_TYPE_NVME_BIO_BASED;
994 goto verify_rq_based; /* must be stacked directly on NVMe (blk-mq) */
995 } else if (list_empty(devices) && live_md_type == DM_TYPE_NVME_BIO_BASED) {
996 t->type = DM_TYPE_NVME_BIO_BASED;
997 }
Mike Snitzer22c11852017-12-04 21:07:37 -0500998 }
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100999 return 0;
1000 }
1001
1002 BUG_ON(!request_based); /* No targets in this table */
1003
Mike Snitzere83068a2016-05-24 21:16:51 -04001004 t->type = DM_TYPE_REQUEST_BASED;
1005
1006verify_rq_based:
Mike Snitzer65803c22014-12-18 16:26:47 -05001007 /*
1008 * Request-based dm supports only tables that have a single target now.
1009 * To support multiple targets, request splitting support is needed,
1010 * and that needs lots of changes in the block-layer.
1011 * (e.g. request completion process for partial completion.)
1012 */
1013 if (t->num_targets > 1) {
Mike Snitzer22c11852017-12-04 21:07:37 -05001014 DMERR("%s DM doesn't support multiple targets",
1015 t->type == DM_TYPE_NVME_BIO_BASED ? "nvme bio-based" : "request-based");
Mike Snitzer65803c22014-12-18 16:26:47 -05001016 return -EINVAL;
1017 }
1018
Mike Snitzer6936c122016-11-23 13:51:09 -05001019 if (list_empty(devices)) {
1020 int srcu_idx;
1021 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
1022
Jens Axboe6a23e052018-10-10 20:49:26 -06001023 /* inherit live table's type */
1024 if (live_table)
Mike Snitzer6936c122016-11-23 13:51:09 -05001025 t->type = live_table->type;
Mike Snitzer6936c122016-11-23 13:51:09 -05001026 dm_put_live_table(t->md, srcu_idx);
1027 return 0;
1028 }
1029
Mike Snitzer22c11852017-12-04 21:07:37 -05001030 tgt = dm_table_get_immutable_target(t);
1031 if (!tgt) {
1032 DMERR("table load rejected: immutable target is required");
1033 return -EINVAL;
1034 } else if (tgt->max_io_len) {
1035 DMERR("table load rejected: immutable target that splits IO is not supported");
1036 return -EINVAL;
1037 }
1038
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001039 /* Non-request-stackable devices can't be used for request-based dm */
Mike Snitzereaa160e2018-01-13 14:33:30 -05001040 if (!tgt->type->iterate_devices ||
Mike Snitzer6ba01df2019-11-05 10:43:44 -05001041 !tgt->type->iterate_devices(tgt, device_is_rq_stackable, NULL)) {
Mike Snitzereaa160e2018-01-13 14:33:30 -05001042 DMERR("table load rejected: including non-request-stackable devices");
1043 return -EINVAL;
Mike Snitzere5863d92014-12-17 21:08:12 -05001044 }
Bart Van Assche301fc3f2016-12-07 16:56:06 -08001045
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001046 return 0;
1047}
1048
Bart Van Assche7e0d5742017-04-27 10:11:23 -07001049enum dm_queue_mode dm_table_get_type(struct dm_table *t)
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001050{
1051 return t->type;
1052}
1053
Alasdair G Kergon36a04562011-10-31 20:19:04 +00001054struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
1055{
1056 return t->immutable_target_type;
1057}
1058
Mike Snitzer16f12262016-01-31 17:22:27 -05001059struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
1060{
1061 /* Immutable target is implicitly a singleton */
1062 if (t->num_targets > 1 ||
1063 !dm_target_is_immutable(t->targets[0].type))
1064 return NULL;
1065
1066 return t->targets;
1067}
1068
Mike Snitzerf083b092016-02-06 18:38:46 -05001069struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1070{
Mikulas Patocka3c120162017-04-18 16:51:46 -04001071 struct dm_target *ti;
1072 unsigned i;
Mike Snitzerf083b092016-02-06 18:38:46 -05001073
Mikulas Patocka3c120162017-04-18 16:51:46 -04001074 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1075 ti = dm_table_get_target(t, i);
Mike Snitzerf083b092016-02-06 18:38:46 -05001076 if (dm_target_is_wildcard(ti->type))
1077 return ti;
1078 }
1079
1080 return NULL;
1081}
1082
Toshi Kani545ed202016-06-22 17:54:53 -06001083bool dm_table_bio_based(struct dm_table *t)
1084{
1085 return __table_type_bio_based(dm_table_get_type(t));
1086}
1087
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001088bool dm_table_request_based(struct dm_table *t)
1089{
Junichi Nomura15b94a62015-05-29 08:51:03 +00001090 return __table_type_request_based(dm_table_get_type(t));
Mike Snitzere5863d92014-12-17 21:08:12 -05001091}
1092
Mike Snitzer17e149b2015-03-11 15:01:09 -04001093static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001094{
Bart Van Assche7e0d5742017-04-27 10:11:23 -07001095 enum dm_queue_mode type = dm_table_get_type(t);
Mike Snitzer30187e12016-01-31 13:28:26 -05001096 unsigned per_io_data_size = 0;
Mike Snitzer0776aa02017-12-08 14:40:52 -05001097 unsigned min_pool_size = 0;
1098 struct dm_target *ti;
Mikulas Patockac0820cf2012-12-21 20:23:38 +00001099 unsigned i;
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001100
Mike Snitzer78d8e582015-06-26 10:01:13 -04001101 if (unlikely(type == DM_TYPE_NONE)) {
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001102 DMWARN("no table type is set, can't allocate mempools");
1103 return -EINVAL;
1104 }
1105
Toshi Kani545ed202016-06-22 17:54:53 -06001106 if (__table_type_bio_based(type))
Mike Snitzer78d8e582015-06-26 10:01:13 -04001107 for (i = 0; i < t->num_targets; i++) {
Mike Snitzer0776aa02017-12-08 14:40:52 -05001108 ti = t->targets + i;
1109 per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1110 min_pool_size = max(min_pool_size, ti->num_flush_bios);
Mike Snitzer78d8e582015-06-26 10:01:13 -04001111 }
1112
Mike Snitzer0776aa02017-12-08 14:40:52 -05001113 t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported,
1114 per_io_data_size, min_pool_size);
Mike Snitzer4e6e36c2015-06-26 09:42:57 -04001115 if (!t->mempools)
1116 return -ENOMEM;
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001117
1118 return 0;
1119}
1120
1121void dm_table_free_md_mempools(struct dm_table *t)
1122{
1123 dm_free_md_mempools(t->mempools);
1124 t->mempools = NULL;
1125}
1126
1127struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1128{
1129 return t->mempools;
1130}
1131
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132static int setup_indexes(struct dm_table *t)
1133{
1134 int i;
1135 unsigned int total = 0;
1136 sector_t *indexes;
1137
1138 /* allocate the space for *all* the indexes */
1139 for (i = t->depth - 2; i >= 0; i--) {
1140 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1141 total += t->counts[i];
1142 }
1143
1144 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1145 if (!indexes)
1146 return -ENOMEM;
1147
1148 /* set up internal nodes, bottom-up */
Jun'ichi Nomura82d601dc2008-02-08 02:10:04 +00001149 for (i = t->depth - 2; i >= 0; i--) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 t->index[i] = indexes;
1151 indexes += (KEYS_PER_NODE * t->counts[i]);
1152 setup_btree_index(i, t);
1153 }
1154
1155 return 0;
1156}
1157
1158/*
1159 * Builds the btree to index the map.
1160 */
Will Drewry26803b92010-08-12 04:14:03 +01001161static int dm_table_build_index(struct dm_table *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162{
1163 int r = 0;
1164 unsigned int leaf_nodes;
1165
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 /* how many indexes will the btree have ? */
1167 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1168 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1169
1170 /* leaf layer has already been set up */
1171 t->counts[t->depth - 1] = leaf_nodes;
1172 t->index[t->depth - 1] = t->highs;
1173
1174 if (t->depth >= 2)
1175 r = setup_indexes(t);
1176
1177 return r;
1178}
1179
Martin K. Petersen25520d52015-10-21 13:19:49 -04001180static bool integrity_profile_exists(struct gendisk *disk)
1181{
1182 return !!blk_get_integrity(disk);
1183}
1184
Will Drewry26803b92010-08-12 04:14:03 +01001185/*
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001186 * Get a disk whose integrity profile reflects the table's profile.
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001187 * Returns NULL if integrity support was inconsistent or unavailable.
1188 */
Martin K. Petersen25520d52015-10-21 13:19:49 -04001189static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001190{
1191 struct list_head *devices = dm_table_get_devices(t);
1192 struct dm_dev_internal *dd = NULL;
1193 struct gendisk *prev_disk = NULL, *template_disk = NULL;
Mikulas Patockae2460f22017-04-18 16:51:48 -04001194 unsigned i;
1195
1196 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1197 struct dm_target *ti = dm_table_get_target(t, i);
1198 if (!dm_target_passes_integrity(ti->type))
1199 goto no_integrity;
1200 }
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001201
1202 list_for_each_entry(dd, devices, list) {
Benjamin Marzinski86f11522014-08-13 13:53:43 -05001203 template_disk = dd->dm_dev->bdev->bd_disk;
Martin K. Petersen25520d52015-10-21 13:19:49 -04001204 if (!integrity_profile_exists(template_disk))
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001205 goto no_integrity;
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001206 else if (prev_disk &&
1207 blk_integrity_compare(prev_disk, template_disk) < 0)
1208 goto no_integrity;
1209 prev_disk = template_disk;
1210 }
1211
1212 return template_disk;
1213
1214no_integrity:
1215 if (prev_disk)
1216 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1217 dm_device_name(t->md),
1218 prev_disk->disk_name,
1219 template_disk->disk_name);
1220 return NULL;
1221}
1222
1223/*
Martin K. Petersen25520d52015-10-21 13:19:49 -04001224 * Register the mapped device for blk_integrity support if the
1225 * underlying devices have an integrity profile. But all devices may
1226 * not have matching profiles (checking all devices isn't reliable
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001227 * during table load because this table may use other DM device(s) which
Martin K. Petersen25520d52015-10-21 13:19:49 -04001228 * must be resumed before they will have an initialized integity
1229 * profile). Consequently, stacked DM devices force a 2 stage integrity
1230 * profile validation: First pass during table load, final pass during
1231 * resume.
Will Drewry26803b92010-08-12 04:14:03 +01001232 */
Martin K. Petersen25520d52015-10-21 13:19:49 -04001233static int dm_table_register_integrity(struct dm_table *t)
Will Drewry26803b92010-08-12 04:14:03 +01001234{
Martin K. Petersen25520d52015-10-21 13:19:49 -04001235 struct mapped_device *md = t->md;
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001236 struct gendisk *template_disk = NULL;
Will Drewry26803b92010-08-12 04:14:03 +01001237
Milan Broz9b4b5a72017-01-04 20:23:51 +01001238 /* If target handles integrity itself do not register it here. */
1239 if (t->integrity_added)
1240 return 0;
1241
Martin K. Petersen25520d52015-10-21 13:19:49 -04001242 template_disk = dm_table_get_integrity_disk(t);
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001243 if (!template_disk)
1244 return 0;
Will Drewry26803b92010-08-12 04:14:03 +01001245
Martin K. Petersen25520d52015-10-21 13:19:49 -04001246 if (!integrity_profile_exists(dm_disk(md))) {
Mike Snitzere83068a2016-05-24 21:16:51 -04001247 t->integrity_supported = true;
Martin K. Petersen25520d52015-10-21 13:19:49 -04001248 /*
1249 * Register integrity profile during table load; we can do
1250 * this because the final profile must match during resume.
1251 */
1252 blk_integrity_register(dm_disk(md),
1253 blk_get_integrity(template_disk));
1254 return 0;
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001255 }
1256
1257 /*
Martin K. Petersen25520d52015-10-21 13:19:49 -04001258 * If DM device already has an initialized integrity
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001259 * profile the new profile should not conflict.
1260 */
Martin K. Petersen25520d52015-10-21 13:19:49 -04001261 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001262 DMWARN("%s: conflict with existing integrity profile: "
1263 "%s profile mismatch",
1264 dm_device_name(t->md),
1265 template_disk->disk_name);
1266 return 1;
1267 }
1268
Martin K. Petersen25520d52015-10-21 13:19:49 -04001269 /* Preserve existing integrity profile */
Mike Snitzere83068a2016-05-24 21:16:51 -04001270 t->integrity_supported = true;
Will Drewry26803b92010-08-12 04:14:03 +01001271 return 0;
1272}
1273
1274/*
1275 * Prepares the table for use by building the indices,
1276 * setting the type, and allocating mempools.
1277 */
1278int dm_table_complete(struct dm_table *t)
1279{
1280 int r;
1281
Mike Snitzere83068a2016-05-24 21:16:51 -04001282 r = dm_table_determine_type(t);
Will Drewry26803b92010-08-12 04:14:03 +01001283 if (r) {
Mike Snitzere83068a2016-05-24 21:16:51 -04001284 DMERR("unable to determine table type");
Will Drewry26803b92010-08-12 04:14:03 +01001285 return r;
1286 }
1287
1288 r = dm_table_build_index(t);
1289 if (r) {
1290 DMERR("unable to build btrees");
1291 return r;
1292 }
1293
Martin K. Petersen25520d52015-10-21 13:19:49 -04001294 r = dm_table_register_integrity(t);
Will Drewry26803b92010-08-12 04:14:03 +01001295 if (r) {
1296 DMERR("could not register integrity profile.");
1297 return r;
1298 }
1299
Mike Snitzer17e149b2015-03-11 15:01:09 -04001300 r = dm_table_alloc_md_mempools(t, t->md);
Will Drewry26803b92010-08-12 04:14:03 +01001301 if (r)
1302 DMERR("unable to allocate mempools");
1303
1304 return r;
1305}
1306
Arjan van de Ven48c9c272006-03-27 01:18:20 -08001307static DEFINE_MUTEX(_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308void dm_table_event_callback(struct dm_table *t,
1309 void (*fn)(void *), void *context)
1310{
Arjan van de Ven48c9c272006-03-27 01:18:20 -08001311 mutex_lock(&_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 t->event_fn = fn;
1313 t->event_context = context;
Arjan van de Ven48c9c272006-03-27 01:18:20 -08001314 mutex_unlock(&_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315}
1316
1317void dm_table_event(struct dm_table *t)
1318{
1319 /*
1320 * You can no longer call dm_table_event() from interrupt
1321 * context, use a bottom half instead.
1322 */
1323 BUG_ON(in_interrupt());
1324
Arjan van de Ven48c9c272006-03-27 01:18:20 -08001325 mutex_lock(&_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 if (t->event_fn)
1327 t->event_fn(t->event_context);
Arjan van de Ven48c9c272006-03-27 01:18:20 -08001328 mutex_unlock(&_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329}
Mike Snitzer08649012011-08-02 12:32:04 +01001330EXPORT_SYMBOL(dm_table_event);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331
Mikulas Patocka1cfd5d32019-08-23 09:54:09 -04001332inline sector_t dm_table_get_size(struct dm_table *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333{
1334 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1335}
Mike Snitzer08649012011-08-02 12:32:04 +01001336EXPORT_SYMBOL(dm_table_get_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337
1338struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1339{
Milan Broz14353532006-06-26 00:27:27 -07001340 if (index >= t->num_targets)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 return NULL;
1342
1343 return t->targets + index;
1344}
1345
1346/*
1347 * Search the btree for the correct target.
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001348 *
Mikulas Patocka123d87d2019-08-23 09:55:26 -04001349 * Caller should check returned pointer for NULL
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001350 * to trap I/O beyond end of device.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 */
1352struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1353{
1354 unsigned int l, n = 0, k = 0;
1355 sector_t *node;
1356
Mikulas Patocka1cfd5d32019-08-23 09:54:09 -04001357 if (unlikely(sector >= dm_table_get_size(t)))
Mikulas Patocka123d87d2019-08-23 09:55:26 -04001358 return NULL;
Mikulas Patocka1cfd5d32019-08-23 09:54:09 -04001359
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 for (l = 0; l < t->depth; l++) {
1361 n = get_child(n, k);
1362 node = get_node(t, l, n);
1363
1364 for (k = 0; k < KEYS_PER_NODE; k++)
1365 if (node[k] >= sector)
1366 break;
1367 }
1368
1369 return &t->targets[(KEYS_PER_NODE * n) + k];
1370}
1371
Mike Snitzer3ae70652012-09-26 23:45:45 +01001372static int count_device(struct dm_target *ti, struct dm_dev *dev,
1373 sector_t start, sector_t len, void *data)
1374{
1375 unsigned *num_devices = data;
1376
1377 (*num_devices)++;
1378
1379 return 0;
1380}
1381
1382/*
1383 * Check whether a table has no data devices attached using each
1384 * target's iterate_devices method.
1385 * Returns false if the result is unknown because a target doesn't
1386 * support iterate_devices.
1387 */
1388bool dm_table_has_no_data_devices(struct dm_table *table)
1389{
Mikulas Patocka3c120162017-04-18 16:51:46 -04001390 struct dm_target *ti;
1391 unsigned i, num_devices;
Mike Snitzer3ae70652012-09-26 23:45:45 +01001392
Mikulas Patocka3c120162017-04-18 16:51:46 -04001393 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1394 ti = dm_table_get_target(table, i);
Mike Snitzer3ae70652012-09-26 23:45:45 +01001395
1396 if (!ti->type->iterate_devices)
1397 return false;
1398
Mikulas Patocka3c120162017-04-18 16:51:46 -04001399 num_devices = 0;
Mike Snitzer3ae70652012-09-26 23:45:45 +01001400 ti->type->iterate_devices(ti, count_device, &num_devices);
1401 if (num_devices)
1402 return false;
1403 }
1404
1405 return true;
1406}
1407
Damien Le Moaldd88d312017-05-08 16:40:43 -07001408static int device_is_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1409 sector_t start, sector_t len, void *data)
1410{
1411 struct request_queue *q = bdev_get_queue(dev->bdev);
1412 enum blk_zoned_model *zoned_model = data;
1413
1414 return q && blk_queue_zoned_model(q) == *zoned_model;
1415}
1416
1417static bool dm_table_supports_zoned_model(struct dm_table *t,
1418 enum blk_zoned_model zoned_model)
1419{
1420 struct dm_target *ti;
1421 unsigned i;
1422
1423 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1424 ti = dm_table_get_target(t, i);
1425
1426 if (zoned_model == BLK_ZONED_HM &&
1427 !dm_target_supports_zoned_hm(ti->type))
1428 return false;
1429
1430 if (!ti->type->iterate_devices ||
1431 !ti->type->iterate_devices(ti, device_is_zoned_model, &zoned_model))
1432 return false;
1433 }
1434
1435 return true;
1436}
1437
1438static int device_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1439 sector_t start, sector_t len, void *data)
1440{
1441 struct request_queue *q = bdev_get_queue(dev->bdev);
1442 unsigned int *zone_sectors = data;
1443
1444 return q && blk_queue_zone_sectors(q) == *zone_sectors;
1445}
1446
1447static bool dm_table_matches_zone_sectors(struct dm_table *t,
1448 unsigned int zone_sectors)
1449{
1450 struct dm_target *ti;
1451 unsigned i;
1452
1453 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1454 ti = dm_table_get_target(t, i);
1455
1456 if (!ti->type->iterate_devices ||
1457 !ti->type->iterate_devices(ti, device_matches_zone_sectors, &zone_sectors))
1458 return false;
1459 }
1460
1461 return true;
1462}
1463
1464static int validate_hardware_zoned_model(struct dm_table *table,
1465 enum blk_zoned_model zoned_model,
1466 unsigned int zone_sectors)
1467{
1468 if (zoned_model == BLK_ZONED_NONE)
1469 return 0;
1470
1471 if (!dm_table_supports_zoned_model(table, zoned_model)) {
1472 DMERR("%s: zoned model is not consistent across all devices",
1473 dm_device_name(table->md));
1474 return -EINVAL;
1475 }
1476
1477 /* Check zone size validity and compatibility */
1478 if (!zone_sectors || !is_power_of_2(zone_sectors))
1479 return -EINVAL;
1480
1481 if (!dm_table_matches_zone_sectors(table, zone_sectors)) {
1482 DMERR("%s: zone sectors is not consistent across all devices",
1483 dm_device_name(table->md));
1484 return -EINVAL;
1485 }
1486
1487 return 0;
1488}
1489
Martin K. Petersen9c470082009-04-09 00:27:12 +01001490/*
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001491 * Establish the new table's queue_limits and validate them.
1492 */
1493int dm_calculate_queue_limits(struct dm_table *table,
1494 struct queue_limits *limits)
1495{
Mikulas Patocka3c120162017-04-18 16:51:46 -04001496 struct dm_target *ti;
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001497 struct queue_limits ti_limits;
Mikulas Patocka3c120162017-04-18 16:51:46 -04001498 unsigned i;
Damien Le Moaldd88d312017-05-08 16:40:43 -07001499 enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1500 unsigned int zone_sectors = 0;
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001501
Martin K. Petersenb1bd0552012-01-11 16:27:11 +01001502 blk_set_stacking_limits(limits);
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001503
Mikulas Patocka3c120162017-04-18 16:51:46 -04001504 for (i = 0; i < dm_table_get_num_targets(table); i++) {
Martin K. Petersenb1bd0552012-01-11 16:27:11 +01001505 blk_set_stacking_limits(&ti_limits);
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001506
Mikulas Patocka3c120162017-04-18 16:51:46 -04001507 ti = dm_table_get_target(table, i);
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001508
1509 if (!ti->type->iterate_devices)
1510 goto combine_limits;
1511
1512 /*
1513 * Combine queue limits of all the devices this target uses.
1514 */
1515 ti->type->iterate_devices(ti, dm_set_device_limits,
1516 &ti_limits);
1517
Damien Le Moaldd88d312017-05-08 16:40:43 -07001518 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1519 /*
1520 * After stacking all limits, validate all devices
1521 * in table support this zoned model and zone sectors.
1522 */
1523 zoned_model = ti_limits.zoned;
1524 zone_sectors = ti_limits.chunk_sectors;
1525 }
1526
Mike Snitzer40bea432009-09-04 20:40:25 +01001527 /* Set I/O hints portion of queue limits */
1528 if (ti->type->io_hints)
1529 ti->type->io_hints(ti, &ti_limits);
1530
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001531 /*
1532 * Check each device area is consistent with the target's
1533 * overall queue limits.
1534 */
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +01001535 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1536 &ti_limits))
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001537 return -EINVAL;
1538
1539combine_limits:
1540 /*
1541 * Merge this target's queue limits into the overall limits
1542 * for the table.
1543 */
1544 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
Martin K. Petersenb27d7f12010-01-11 03:21:50 -05001545 DMWARN("%s: adding target device "
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001546 "(start sect %llu len %llu) "
Martin K. Petersenb27d7f12010-01-11 03:21:50 -05001547 "caused an alignment inconsistency",
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001548 dm_device_name(table->md),
1549 (unsigned long long) ti->begin,
1550 (unsigned long long) ti->len);
Damien Le Moaldd88d312017-05-08 16:40:43 -07001551
1552 /*
1553 * FIXME: this should likely be moved to blk_stack_limits(), would
1554 * also eliminate limits->zoned stacking hack in dm_set_device_limits()
1555 */
1556 if (limits->zoned == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1557 /*
1558 * By default, the stacked limits zoned model is set to
1559 * BLK_ZONED_NONE in blk_set_stacking_limits(). Update
1560 * this model using the first target model reported
1561 * that is not BLK_ZONED_NONE. This will be either the
1562 * first target device zoned model or the model reported
1563 * by the target .io_hints.
1564 */
1565 limits->zoned = ti_limits.zoned;
1566 }
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001567 }
1568
Damien Le Moaldd88d312017-05-08 16:40:43 -07001569 /*
1570 * Verify that the zoned model and zone sectors, as determined before
1571 * any .io_hints override, are the same across all devices in the table.
1572 * - this is especially relevant if .io_hints is emulating a disk-managed
1573 * zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1574 * BUT...
1575 */
1576 if (limits->zoned != BLK_ZONED_NONE) {
1577 /*
1578 * ...IF the above limits stacking determined a zoned model
1579 * validate that all of the table's devices conform to it.
1580 */
1581 zoned_model = limits->zoned;
1582 zone_sectors = limits->chunk_sectors;
1583 }
1584 if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1585 return -EINVAL;
1586
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001587 return validate_hardware_logical_block_alignment(table, limits);
1588}
1589
1590/*
Martin K. Petersen25520d52015-10-21 13:19:49 -04001591 * Verify that all devices have an integrity profile that matches the
1592 * DM device's registered integrity profile. If the profiles don't
1593 * match then unregister the DM device's integrity profile.
Martin K. Petersen9c470082009-04-09 00:27:12 +01001594 */
Martin K. Petersen25520d52015-10-21 13:19:49 -04001595static void dm_table_verify_integrity(struct dm_table *t)
Martin K. Petersen9c470082009-04-09 00:27:12 +01001596{
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001597 struct gendisk *template_disk = NULL;
Martin K. Petersen9c470082009-04-09 00:27:12 +01001598
Milan Broz9b4b5a72017-01-04 20:23:51 +01001599 if (t->integrity_added)
1600 return;
1601
Martin K. Petersen25520d52015-10-21 13:19:49 -04001602 if (t->integrity_supported) {
1603 /*
1604 * Verify that the original integrity profile
1605 * matches all the devices in this table.
1606 */
1607 template_disk = dm_table_get_integrity_disk(t);
1608 if (template_disk &&
1609 blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1610 return;
1611 }
Martin K. Petersen9c470082009-04-09 00:27:12 +01001612
Martin K. Petersen25520d52015-10-21 13:19:49 -04001613 if (integrity_profile_exists(dm_disk(t->md))) {
Mike Snitzer876fbba2011-09-25 23:26:17 +01001614 DMWARN("%s: unable to establish an integrity profile",
1615 dm_device_name(t->md));
Martin K. Petersen25520d52015-10-21 13:19:49 -04001616 blk_integrity_unregister(dm_disk(t->md));
1617 }
Martin K. Petersen9c470082009-04-09 00:27:12 +01001618}
1619
Mike Snitzered8b7522011-08-02 12:32:08 +01001620static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1621 sector_t start, sector_t len, void *data)
1622{
Jens Axboec888a8f2016-04-13 13:33:19 -06001623 unsigned long flush = (unsigned long) data;
Mike Snitzered8b7522011-08-02 12:32:08 +01001624 struct request_queue *q = bdev_get_queue(dev->bdev);
1625
Jens Axboec888a8f2016-04-13 13:33:19 -06001626 return q && (q->queue_flags & flush);
Mike Snitzered8b7522011-08-02 12:32:08 +01001627}
1628
Jens Axboec888a8f2016-04-13 13:33:19 -06001629static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
Mike Snitzered8b7522011-08-02 12:32:08 +01001630{
1631 struct dm_target *ti;
Mikulas Patocka3c120162017-04-18 16:51:46 -04001632 unsigned i;
Mike Snitzered8b7522011-08-02 12:32:08 +01001633
1634 /*
1635 * Require at least one underlying device to support flushes.
1636 * t->devices includes internal dm devices such as mirror logs
1637 * so we need to use iterate_devices here, which targets
1638 * supporting flushes must provide.
1639 */
Mikulas Patocka3c120162017-04-18 16:51:46 -04001640 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1641 ti = dm_table_get_target(t, i);
Mike Snitzered8b7522011-08-02 12:32:08 +01001642
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00001643 if (!ti->num_flush_bios)
Mike Snitzered8b7522011-08-02 12:32:08 +01001644 continue;
1645
Joe Thornber0e9c24e2012-07-27 15:08:07 +01001646 if (ti->flush_supported)
Joe Perches7f61f5a2015-03-30 10:43:18 -07001647 return true;
Joe Thornber0e9c24e2012-07-27 15:08:07 +01001648
Mike Snitzered8b7522011-08-02 12:32:08 +01001649 if (ti->type->iterate_devices &&
Jens Axboec888a8f2016-04-13 13:33:19 -06001650 ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
Joe Perches7f61f5a2015-03-30 10:43:18 -07001651 return true;
Mike Snitzered8b7522011-08-02 12:32:08 +01001652 }
1653
Joe Perches7f61f5a2015-03-30 10:43:18 -07001654 return false;
Mike Snitzered8b7522011-08-02 12:32:08 +01001655}
1656
Vivek Goyal273752c2017-07-26 09:35:09 -04001657static int device_dax_write_cache_enabled(struct dm_target *ti,
1658 struct dm_dev *dev, sector_t start,
1659 sector_t len, void *data)
1660{
1661 struct dax_device *dax_dev = dev->dax_dev;
1662
1663 if (!dax_dev)
1664 return false;
1665
1666 if (dax_write_cache_enabled(dax_dev))
1667 return true;
1668 return false;
1669}
1670
1671static int dm_table_supports_dax_write_cache(struct dm_table *t)
1672{
1673 struct dm_target *ti;
1674 unsigned i;
1675
1676 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1677 ti = dm_table_get_target(t, i);
1678
1679 if (ti->type->iterate_devices &&
1680 ti->type->iterate_devices(ti,
1681 device_dax_write_cache_enabled, NULL))
1682 return true;
1683 }
1684
1685 return false;
1686}
1687
Mandeep Singh Baines4693c9662011-10-31 20:18:50 +00001688static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1689 sector_t start, sector_t len, void *data)
1690{
1691 struct request_queue *q = bdev_get_queue(dev->bdev);
1692
1693 return q && blk_queue_nonrot(q);
1694}
1695
Milan Brozc3c45552012-09-26 23:45:43 +01001696static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1697 sector_t start, sector_t len, void *data)
1698{
1699 struct request_queue *q = bdev_get_queue(dev->bdev);
1700
1701 return q && !blk_queue_add_random(q);
1702}
1703
1704static bool dm_table_all_devices_attribute(struct dm_table *t,
1705 iterate_devices_callout_fn func)
Mandeep Singh Baines4693c9662011-10-31 20:18:50 +00001706{
1707 struct dm_target *ti;
Mikulas Patocka3c120162017-04-18 16:51:46 -04001708 unsigned i;
Mandeep Singh Baines4693c9662011-10-31 20:18:50 +00001709
Mikulas Patocka3c120162017-04-18 16:51:46 -04001710 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1711 ti = dm_table_get_target(t, i);
Mandeep Singh Baines4693c9662011-10-31 20:18:50 +00001712
1713 if (!ti->type->iterate_devices ||
Milan Brozc3c45552012-09-26 23:45:43 +01001714 !ti->type->iterate_devices(ti, func, NULL))
Joe Perches7f61f5a2015-03-30 10:43:18 -07001715 return false;
Mandeep Singh Baines4693c9662011-10-31 20:18:50 +00001716 }
1717
Joe Perches7f61f5a2015-03-30 10:43:18 -07001718 return true;
Mandeep Singh Baines4693c9662011-10-31 20:18:50 +00001719}
1720
Mike Snitzer22c11852017-12-04 21:07:37 -05001721static int device_no_partial_completion(struct dm_target *ti, struct dm_dev *dev,
1722 sector_t start, sector_t len, void *data)
1723{
1724 char b[BDEVNAME_SIZE];
1725
1726 /* For now, NVMe devices are the only devices of this class */
Mikulas Patocka99243b92018-02-26 15:22:32 -05001727 return (strncmp(bdevname(dev->bdev, b), "nvme", 4) == 0);
Mike Snitzer22c11852017-12-04 21:07:37 -05001728}
1729
1730static bool dm_table_does_not_support_partial_completion(struct dm_table *t)
1731{
1732 return dm_table_all_devices_attribute(t, device_no_partial_completion);
1733}
1734
Mike Snitzerd54eaa52012-12-21 20:23:36 +00001735static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1736 sector_t start, sector_t len, void *data)
1737{
1738 struct request_queue *q = bdev_get_queue(dev->bdev);
1739
1740 return q && !q->limits.max_write_same_sectors;
1741}
1742
1743static bool dm_table_supports_write_same(struct dm_table *t)
1744{
1745 struct dm_target *ti;
Mikulas Patocka3c120162017-04-18 16:51:46 -04001746 unsigned i;
Mike Snitzerd54eaa52012-12-21 20:23:36 +00001747
Mikulas Patocka3c120162017-04-18 16:51:46 -04001748 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1749 ti = dm_table_get_target(t, i);
Mike Snitzerd54eaa52012-12-21 20:23:36 +00001750
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00001751 if (!ti->num_write_same_bios)
Mike Snitzerd54eaa52012-12-21 20:23:36 +00001752 return false;
1753
1754 if (!ti->type->iterate_devices ||
Mike Snitzerdc019b22013-05-10 14:37:16 +01001755 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
Mike Snitzerd54eaa52012-12-21 20:23:36 +00001756 return false;
1757 }
1758
1759 return true;
1760}
1761
Christoph Hellwigac62d622017-04-05 19:21:05 +02001762static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1763 sector_t start, sector_t len, void *data)
1764{
1765 struct request_queue *q = bdev_get_queue(dev->bdev);
1766
1767 return q && !q->limits.max_write_zeroes_sectors;
1768}
1769
1770static bool dm_table_supports_write_zeroes(struct dm_table *t)
1771{
1772 struct dm_target *ti;
1773 unsigned i = 0;
1774
1775 while (i < dm_table_get_num_targets(t)) {
1776 ti = dm_table_get_target(t, i++);
1777
1778 if (!ti->num_write_zeroes_bios)
1779 return false;
1780
1781 if (!ti->type->iterate_devices ||
1782 ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1783 return false;
1784 }
1785
1786 return true;
1787}
1788
Mike Snitzer8a74d292017-11-14 15:40:52 -05001789static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1790 sector_t start, sector_t len, void *data)
Mikulas Patockaa7ffb6a2014-07-10 12:23:07 -04001791{
1792 struct request_queue *q = bdev_get_queue(dev->bdev);
1793
Mike Snitzer8a74d292017-11-14 15:40:52 -05001794 return q && !blk_queue_discard(q);
Mikulas Patockaa7ffb6a2014-07-10 12:23:07 -04001795}
1796
1797static bool dm_table_supports_discards(struct dm_table *t)
1798{
1799 struct dm_target *ti;
Mikulas Patocka3c120162017-04-18 16:51:46 -04001800 unsigned i;
Mikulas Patockaa7ffb6a2014-07-10 12:23:07 -04001801
Mikulas Patocka3c120162017-04-18 16:51:46 -04001802 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1803 ti = dm_table_get_target(t, i);
Mikulas Patockaa7ffb6a2014-07-10 12:23:07 -04001804
1805 if (!ti->num_discard_bios)
Mike Snitzer8a74d292017-11-14 15:40:52 -05001806 return false;
Mikulas Patockaa7ffb6a2014-07-10 12:23:07 -04001807
Mike Snitzer8a74d292017-11-14 15:40:52 -05001808 /*
1809 * Either the target provides discard support (as implied by setting
1810 * 'discards_supported') or it relies on _all_ data devices having
1811 * discard support.
1812 */
1813 if (!ti->discards_supported &&
1814 (!ti->type->iterate_devices ||
1815 ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1816 return false;
Mikulas Patockaa7ffb6a2014-07-10 12:23:07 -04001817 }
1818
Mike Snitzer8a74d292017-11-14 15:40:52 -05001819 return true;
Mikulas Patockaa7ffb6a2014-07-10 12:23:07 -04001820}
1821
Denis Semakin00716542018-03-13 13:23:45 +04001822static int device_not_secure_erase_capable(struct dm_target *ti,
1823 struct dm_dev *dev, sector_t start,
1824 sector_t len, void *data)
1825{
1826 struct request_queue *q = bdev_get_queue(dev->bdev);
1827
1828 return q && !blk_queue_secure_erase(q);
1829}
1830
1831static bool dm_table_supports_secure_erase(struct dm_table *t)
1832{
1833 struct dm_target *ti;
1834 unsigned int i;
1835
1836 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1837 ti = dm_table_get_target(t, i);
1838
1839 if (!ti->num_secure_erase_bios)
1840 return false;
1841
1842 if (!ti->type->iterate_devices ||
1843 ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1844 return false;
1845 }
1846
1847 return true;
1848}
1849
Ilya Dryomoveb40c0a2019-03-26 20:20:58 +01001850static int device_requires_stable_pages(struct dm_target *ti,
1851 struct dm_dev *dev, sector_t start,
1852 sector_t len, void *data)
1853{
1854 struct request_queue *q = bdev_get_queue(dev->bdev);
1855
1856 return q && bdi_cap_stable_pages_required(q->backing_dev_info);
1857}
1858
1859/*
1860 * If any underlying device requires stable pages, a table must require
1861 * them as well. Only targets that support iterate_devices are considered:
1862 * don't want error, zero, etc to require stable pages.
1863 */
1864static bool dm_table_requires_stable_pages(struct dm_table *t)
1865{
1866 struct dm_target *ti;
1867 unsigned i;
1868
1869 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1870 ti = dm_table_get_target(t, i);
1871
1872 if (ti->type->iterate_devices &&
1873 ti->type->iterate_devices(ti, device_requires_stable_pages, NULL))
1874 return true;
1875 }
1876
1877 return false;
1878}
1879
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001880void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1881 struct queue_limits *limits)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882{
Jens Axboe519a7e12016-03-30 10:14:14 -06001883 bool wc = false, fua = false;
Pankaj Gupta2e9ee092019-07-05 19:33:25 +05301884 int page_size = PAGE_SIZE;
Mike Snitzered8b7522011-08-02 12:32:08 +01001885
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 /*
Mike Snitzer11977642009-06-22 10:12:32 +01001887 * Copy table's limits to the DM device's request_queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 */
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001889 q->limits = *limits;
Jens Axboec9a3f6d2008-04-29 19:12:35 +02001890
Mike Snitzer5d47c892017-11-16 15:42:26 -05001891 if (!dm_table_supports_discards(t)) {
Bart Van Assche8b904b52018-03-07 17:10:10 -08001892 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
Mike Snitzer5d47c892017-11-16 15:42:26 -05001893 /* Must also clear discard limits... */
1894 q->limits.max_discard_sectors = 0;
1895 q->limits.max_hw_discard_sectors = 0;
1896 q->limits.discard_granularity = 0;
1897 q->limits.discard_alignment = 0;
1898 q->limits.discard_misaligned = 0;
1899 } else
Bart Van Assche8b904b52018-03-07 17:10:10 -08001900 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001901
Denis Semakin00716542018-03-13 13:23:45 +04001902 if (dm_table_supports_secure_erase(t))
Linus Torvalds83c7c182018-04-06 11:50:19 -07001903 blk_queue_flag_set(QUEUE_FLAG_SECERASE, q);
Denis Semakin00716542018-03-13 13:23:45 +04001904
Jens Axboec888a8f2016-04-13 13:33:19 -06001905 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
Jens Axboe519a7e12016-03-30 10:14:14 -06001906 wc = true;
Jens Axboec888a8f2016-04-13 13:33:19 -06001907 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
Jens Axboe519a7e12016-03-30 10:14:14 -06001908 fua = true;
Mike Snitzered8b7522011-08-02 12:32:08 +01001909 }
Jens Axboe519a7e12016-03-30 10:14:14 -06001910 blk_queue_write_cache(q, wc, fua);
Mike Snitzered8b7522011-08-02 12:32:08 +01001911
Pankaj Gupta2e9ee092019-07-05 19:33:25 +05301912 if (dm_table_supports_dax(t, device_supports_dax, &page_size)) {
Bart Van Assche8b904b52018-03-07 17:10:10 -08001913 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
Mike Snitzer9c50a982019-07-30 14:39:43 -04001914 if (dm_table_supports_dax(t, device_dax_synchronous, NULL))
Pankaj Gupta2e9ee092019-07-05 19:33:25 +05301915 set_dax_synchronous(t->md->dax_dev);
1916 }
Ross Zwislerdbc62652018-06-26 16:30:41 -06001917 else
1918 blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
1919
Vivek Goyal273752c2017-07-26 09:35:09 -04001920 if (dm_table_supports_dax_write_cache(t))
1921 dax_write_cache(t->md->dax_dev, true);
1922
Milan Brozc3c45552012-09-26 23:45:43 +01001923 /* Ensure that all underlying devices are non-rotational. */
1924 if (dm_table_all_devices_attribute(t, device_is_nonrot))
Bart Van Assche8b904b52018-03-07 17:10:10 -08001925 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
Mandeep Singh Baines4693c9662011-10-31 20:18:50 +00001926 else
Bart Van Assche8b904b52018-03-07 17:10:10 -08001927 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
Mandeep Singh Baines4693c9662011-10-31 20:18:50 +00001928
Mike Snitzerd54eaa52012-12-21 20:23:36 +00001929 if (!dm_table_supports_write_same(t))
1930 q->limits.max_write_same_sectors = 0;
Christoph Hellwigac62d622017-04-05 19:21:05 +02001931 if (!dm_table_supports_write_zeroes(t))
1932 q->limits.max_write_zeroes_sectors = 0;
Mike Snitzerc1a94672012-12-21 20:23:30 +00001933
Martin K. Petersen25520d52015-10-21 13:19:49 -04001934 dm_table_verify_integrity(t);
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001935
1936 /*
Ilya Dryomoveb40c0a2019-03-26 20:20:58 +01001937 * Some devices don't use blk_integrity but still want stable pages
1938 * because they do their own checksumming.
1939 */
1940 if (dm_table_requires_stable_pages(t))
1941 q->backing_dev_info->capabilities |= BDI_CAP_STABLE_WRITES;
1942 else
1943 q->backing_dev_info->capabilities &= ~BDI_CAP_STABLE_WRITES;
1944
1945 /*
Milan Brozc3c45552012-09-26 23:45:43 +01001946 * Determine whether or not this queue's I/O timings contribute
1947 * to the entropy pool, Only request-based targets use this.
1948 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1949 * have it set.
1950 */
1951 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
Bart Van Assche8b904b52018-03-07 17:10:10 -08001952 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
Damien Le Moalbf505452018-10-12 19:08:50 +09001953
1954 /*
1955 * For a zoned target, the number of zones should be updated for the
1956 * correct value to be exposed in sysfs queue/nr_zones. For a BIO based
Christoph Hellwigae589542019-12-03 10:39:07 +01001957 * target, this is all that is needed.
Damien Le Moalbf505452018-10-12 19:08:50 +09001958 */
Christoph Hellwigae589542019-12-03 10:39:07 +01001959#ifdef CONFIG_BLK_DEV_ZONED
1960 if (blk_queue_is_zoned(q)) {
1961 WARN_ON_ONCE(queue_is_mq(q));
1962 q->nr_zones = blkdev_nr_zones(t->md->disk);
1963 }
1964#endif
Jaegeuk Kimc6d6e9b2018-12-18 09:25:37 -08001965
1966 /* Allow reads to exceed readahead limits */
1967 q->backing_dev_info->io_pages = limits->max_sectors >> (PAGE_SHIFT - 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968}
1969
1970unsigned int dm_table_get_num_targets(struct dm_table *t)
1971{
1972 return t->num_targets;
1973}
1974
1975struct list_head *dm_table_get_devices(struct dm_table *t)
1976{
1977 return &t->devices;
1978}
1979
Al Viroaeb5d722008-09-02 15:28:45 -04001980fmode_t dm_table_get_mode(struct dm_table *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981{
1982 return t->mode;
1983}
Mike Snitzer08649012011-08-02 12:32:04 +01001984EXPORT_SYMBOL(dm_table_get_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985
Mike Snitzerd67ee2132014-10-28 20:13:31 -04001986enum suspend_mode {
1987 PRESUSPEND,
1988 PRESUSPEND_UNDO,
1989 POSTSUSPEND,
1990};
1991
1992static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993{
1994 int i = t->num_targets;
1995 struct dm_target *ti = t->targets;
1996
Bart Van Assche1ea0654e2017-04-27 10:11:21 -07001997 lockdep_assert_held(&t->md->suspend_lock);
1998
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 while (i--) {
Mike Snitzerd67ee2132014-10-28 20:13:31 -04002000 switch (mode) {
2001 case PRESUSPEND:
2002 if (ti->type->presuspend)
2003 ti->type->presuspend(ti);
2004 break;
2005 case PRESUSPEND_UNDO:
2006 if (ti->type->presuspend_undo)
2007 ti->type->presuspend_undo(ti);
2008 break;
2009 case POSTSUSPEND:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 if (ti->type->postsuspend)
2011 ti->type->postsuspend(ti);
Mike Snitzerd67ee2132014-10-28 20:13:31 -04002012 break;
2013 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 ti++;
2015 }
2016}
2017
2018void dm_table_presuspend_targets(struct dm_table *t)
2019{
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002020 if (!t)
2021 return;
2022
Mike Snitzerd67ee2132014-10-28 20:13:31 -04002023 suspend_targets(t, PRESUSPEND);
2024}
2025
2026void dm_table_presuspend_undo_targets(struct dm_table *t)
2027{
2028 if (!t)
2029 return;
2030
2031 suspend_targets(t, PRESUSPEND_UNDO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032}
2033
2034void dm_table_postsuspend_targets(struct dm_table *t)
2035{
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07002036 if (!t)
2037 return;
2038
Mike Snitzerd67ee2132014-10-28 20:13:31 -04002039 suspend_targets(t, POSTSUSPEND);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040}
2041
Milan Broz8757b772006-10-03 01:15:36 -07002042int dm_table_resume_targets(struct dm_table *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043{
Milan Broz8757b772006-10-03 01:15:36 -07002044 int i, r = 0;
2045
Bart Van Assche1ea0654e2017-04-27 10:11:21 -07002046 lockdep_assert_held(&t->md->suspend_lock);
2047
Milan Broz8757b772006-10-03 01:15:36 -07002048 for (i = 0; i < t->num_targets; i++) {
2049 struct dm_target *ti = t->targets + i;
2050
2051 if (!ti->type->preresume)
2052 continue;
2053
2054 r = ti->type->preresume(ti);
Mike Snitzer7833b082013-10-24 14:10:29 -04002055 if (r) {
2056 DMERR("%s: %s: preresume failed, error = %d",
2057 dm_device_name(t->md), ti->type->name, r);
Milan Broz8757b772006-10-03 01:15:36 -07002058 return r;
Mike Snitzer7833b082013-10-24 14:10:29 -04002059 }
Milan Broz8757b772006-10-03 01:15:36 -07002060 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061
2062 for (i = 0; i < t->num_targets; i++) {
2063 struct dm_target *ti = t->targets + i;
2064
2065 if (ti->type->resume)
2066 ti->type->resume(ti);
2067 }
Milan Broz8757b772006-10-03 01:15:36 -07002068
2069 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070}
2071
NeilBrown9d357b02011-01-13 20:00:01 +00002072void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
2073{
2074 list_add(&cb->list, &t->target_callbacks);
2075}
2076EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
2077
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078int dm_table_any_congested(struct dm_table *t, int bdi_bits)
2079{
Mikulas Patocka82b15192008-10-10 13:37:09 +01002080 struct dm_dev_internal *dd;
Paul Jimenezafb24522008-02-08 02:09:59 +00002081 struct list_head *devices = dm_table_get_devices(t);
NeilBrown9d357b02011-01-13 20:00:01 +00002082 struct dm_target_callbacks *cb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 int r = 0;
2084
Paul Jimenezafb24522008-02-08 02:09:59 +00002085 list_for_each_entry(dd, devices, list) {
Benjamin Marzinski86f11522014-08-13 13:53:43 -05002086 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
Alasdair G Kergon0c2322e2008-10-10 13:37:13 +01002087 char b[BDEVNAME_SIZE];
2088
2089 if (likely(q))
Jan Karadc3b17c2017-02-02 15:56:50 +01002090 r |= bdi_congested(q->backing_dev_info, bdi_bits);
Alasdair G Kergon0c2322e2008-10-10 13:37:13 +01002091 else
2092 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
2093 dm_device_name(t->md),
Benjamin Marzinski86f11522014-08-13 13:53:43 -05002094 bdevname(dd->dm_dev->bdev, b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 }
2096
NeilBrown9d357b02011-01-13 20:00:01 +00002097 list_for_each_entry(cb, &t->target_callbacks, list)
2098 if (cb->congested_fn)
2099 r |= cb->congested_fn(cb, bdi_bits);
2100
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 return r;
2102}
2103
Mike Anderson1134e5ae2006-03-27 01:17:54 -08002104struct mapped_device *dm_table_get_md(struct dm_table *t)
2105{
Mike Anderson1134e5ae2006-03-27 01:17:54 -08002106 return t->md;
2107}
Mike Snitzer08649012011-08-02 12:32:04 +01002108EXPORT_SYMBOL(dm_table_get_md);
Mike Anderson1134e5ae2006-03-27 01:17:54 -08002109
Michał Mirosławf349b0a2018-10-09 22:13:42 +02002110const char *dm_table_device_name(struct dm_table *t)
2111{
2112 return dm_device_name(t->md);
2113}
2114EXPORT_SYMBOL_GPL(dm_table_device_name);
2115
Mike Snitzer9974fa22014-02-28 15:33:43 +01002116void dm_table_run_md_queue_async(struct dm_table *t)
2117{
2118 struct mapped_device *md;
2119 struct request_queue *queue;
Mike Snitzer9974fa22014-02-28 15:33:43 +01002120
2121 if (!dm_table_request_based(t))
2122 return;
2123
2124 md = dm_table_get_md(t);
2125 queue = dm_get_md_queue(md);
Jens Axboe6a23e052018-10-10 20:49:26 -06002126 if (queue)
2127 blk_mq_run_hw_queues(queue, true);
Mike Snitzer9974fa22014-02-28 15:33:43 +01002128}
2129EXPORT_SYMBOL(dm_table_run_md_queue_async);
2130