blob: 2e227fbf1622c075c6fc543f5b2d30fb2bf7dbd7 [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
8#include "dm.h"
9
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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
Alasdair G Kergon72d94862006-06-26 00:27:35 -070022#define DM_MSG_PREFIX "table"
23
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#define MAX_DEPTH 16
25#define NODE_SIZE L1_CACHE_BYTES
26#define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
27#define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
28
Mikulas Patockad5816872009-01-06 03:05:10 +000029/*
30 * The table has always exactly one reference from either mapped_device->map
31 * or hash_cell->new_map. This reference is not counted in table->holders.
32 * A pair of dm_create_table/dm_destroy_table functions is used for table
33 * creation/destruction.
34 *
35 * Temporary references from the other code increase table->holders. A pair
36 * of dm_table_get/dm_table_put functions is used to manipulate it.
37 *
38 * When the table is about to be destroyed, we wait for table->holders to
39 * drop to zero.
40 */
41
Linus Torvalds1da177e2005-04-16 15:20:36 -070042struct dm_table {
Mike Anderson1134e5ae2006-03-27 01:17:54 -080043 struct mapped_device *md;
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 atomic_t holders;
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +010045 unsigned type;
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47 /* btree table */
48 unsigned int depth;
49 unsigned int counts[MAX_DEPTH]; /* in nodes */
50 sector_t *index[MAX_DEPTH];
51
52 unsigned int num_targets;
53 unsigned int num_allocated;
54 sector_t *highs;
55 struct dm_target *targets;
56
Alasdair G Kergon36a04562011-10-31 20:19:04 +000057 struct target_type *immutable_target_type;
Martin K. Petersena91a2782011-03-17 11:11:05 +010058 unsigned integrity_supported:1;
Alasdair G Kergon3791e2f2011-10-31 20:19:00 +000059 unsigned singleton:1;
Mike Snitzer5ae89a82010-08-12 04:14:08 +010060
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 /*
62 * Indicates the rw permissions for the new logical
63 * device. This should be a combination of FMODE_READ
64 * and FMODE_WRITE.
65 */
Al Viroaeb5d722008-09-02 15:28:45 -040066 fmode_t mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
68 /* a list of devices used by this table */
69 struct list_head devices;
70
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 /* events get handed up using this callback */
72 void (*event_fn)(void *);
73 void *event_context;
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +010074
75 struct dm_md_mempools *mempools;
NeilBrown9d357b02011-01-13 20:00:01 +000076
77 struct list_head target_callbacks;
Linus Torvalds1da177e2005-04-16 15:20:36 -070078};
79
80/*
81 * Similar to ceiling(log_size(n))
82 */
83static unsigned int int_log(unsigned int n, unsigned int base)
84{
85 int result = 0;
86
87 while (n > 1) {
88 n = dm_div_up(n, base);
89 result++;
90 }
91
92 return result;
93}
94
95/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 * Calculate the index of the child node of the n'th node k'th key.
97 */
98static inline unsigned int get_child(unsigned int n, unsigned int k)
99{
100 return (n * CHILDREN_PER_NODE) + k;
101}
102
103/*
104 * Return the n'th node of level l from table t.
105 */
106static inline sector_t *get_node(struct dm_table *t,
107 unsigned int l, unsigned int n)
108{
109 return t->index[l] + (n * KEYS_PER_NODE);
110}
111
112/*
113 * Return the highest key that you could lookup from the n'th
114 * node on level l of the btree.
115 */
116static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
117{
118 for (; l < t->depth - 1; l++)
119 n = get_child(n, CHILDREN_PER_NODE - 1);
120
121 if (n >= t->counts[l])
122 return (sector_t) - 1;
123
124 return get_node(t, l, n)[KEYS_PER_NODE - 1];
125}
126
127/*
128 * Fills in a level of the btree based on the highs of the level
129 * below it.
130 */
131static int setup_btree_index(unsigned int l, struct dm_table *t)
132{
133 unsigned int n, k;
134 sector_t *node;
135
136 for (n = 0U; n < t->counts[l]; n++) {
137 node = get_node(t, l, n);
138
139 for (k = 0U; k < KEYS_PER_NODE; k++)
140 node[k] = high(t, l + 1, get_child(n, k));
141 }
142
143 return 0;
144}
145
146void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
147{
148 unsigned long size;
149 void *addr;
150
151 /*
152 * Check that we're not going to overflow.
153 */
154 if (nmemb > (ULONG_MAX / elem_size))
155 return NULL;
156
157 size = nmemb * elem_size;
Joe Perchese29e65a2011-08-02 12:32:02 +0100158 addr = vzalloc(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159
160 return addr;
161}
Mike Snitzer08649012011-08-02 12:32:04 +0100162EXPORT_SYMBOL(dm_vcalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
164/*
165 * highs, and targets are managed as dynamic arrays during a
166 * table load.
167 */
168static int alloc_targets(struct dm_table *t, unsigned int num)
169{
170 sector_t *n_highs;
171 struct dm_target *n_targets;
172 int n = t->num_targets;
173
174 /*
175 * Allocate both the target array and offset array at once.
Jun'ichi Nomura512875b2007-12-13 14:15:25 +0000176 * Append an empty entry to catch sectors beyond the end of
177 * the device.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 */
Jun'ichi Nomura512875b2007-12-13 14:15:25 +0000179 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 sizeof(sector_t));
181 if (!n_highs)
182 return -ENOMEM;
183
184 n_targets = (struct dm_target *) (n_highs + num);
185
186 if (n) {
187 memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
188 memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
189 }
190
191 memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
192 vfree(t->highs);
193
194 t->num_allocated = num;
195 t->highs = n_highs;
196 t->targets = n_targets;
197
198 return 0;
199}
200
Al Viroaeb5d722008-09-02 15:28:45 -0400201int dm_table_create(struct dm_table **result, fmode_t mode,
Mike Anderson1134e5ae2006-03-27 01:17:54 -0800202 unsigned num_targets, struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203{
Dmitry Monakhov094262d2007-10-19 22:38:51 +0100204 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205
206 if (!t)
207 return -ENOMEM;
208
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 INIT_LIST_HEAD(&t->devices);
NeilBrown9d357b02011-01-13 20:00:01 +0000210 INIT_LIST_HEAD(&t->target_callbacks);
Mikulas Patockad5816872009-01-06 03:05:10 +0000211 atomic_set(&t->holders, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212
213 if (!num_targets)
214 num_targets = KEYS_PER_NODE;
215
216 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
217
218 if (alloc_targets(t, num_targets)) {
219 kfree(t);
220 t = NULL;
221 return -ENOMEM;
222 }
223
224 t->mode = mode;
Mike Anderson1134e5ae2006-03-27 01:17:54 -0800225 t->md = md;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 *result = t;
227 return 0;
228}
229
230static void free_devices(struct list_head *devices)
231{
232 struct list_head *tmp, *next;
233
Paul Jimenezafb24522008-02-08 02:09:59 +0000234 list_for_each_safe(tmp, next, devices) {
Mikulas Patocka82b15192008-10-10 13:37:09 +0100235 struct dm_dev_internal *dd =
236 list_entry(tmp, struct dm_dev_internal, list);
Jonthan Brassow1b6da752009-06-22 10:12:29 +0100237 DMWARN("dm_table_destroy: dm_put_device call missing for %s",
238 dd->dm_dev.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 kfree(dd);
240 }
241}
242
Mikulas Patockad5816872009-01-06 03:05:10 +0000243void dm_table_destroy(struct dm_table *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244{
245 unsigned int i;
246
Alasdair G Kergona7940152009-12-10 23:52:23 +0000247 if (!t)
248 return;
249
Mikulas Patockad5816872009-01-06 03:05:10 +0000250 while (atomic_read(&t->holders))
251 msleep(1);
252 smp_mb();
253
Will Drewry26803b92010-08-12 04:14:03 +0100254 /* free the indexes */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 if (t->depth >= 2)
256 vfree(t->index[t->depth - 2]);
257
258 /* free the targets */
259 for (i = 0; i < t->num_targets; i++) {
260 struct dm_target *tgt = t->targets + i;
261
262 if (tgt->type->dtr)
263 tgt->type->dtr(tgt);
264
265 dm_put_target_type(tgt->type);
266 }
267
268 vfree(t->highs);
269
270 /* free the device list */
Hannes Reinecke574ce072012-03-28 18:41:24 +0100271 free_devices(&t->devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100273 dm_free_md_mempools(t->mempools);
274
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 kfree(t);
276}
277
278void dm_table_get(struct dm_table *t)
279{
280 atomic_inc(&t->holders);
281}
Mike Snitzer08649012011-08-02 12:32:04 +0100282EXPORT_SYMBOL(dm_table_get);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283
284void dm_table_put(struct dm_table *t)
285{
286 if (!t)
287 return;
288
Mikulas Patockad5816872009-01-06 03:05:10 +0000289 smp_mb__before_atomic_dec();
290 atomic_dec(&t->holders);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291}
Mike Snitzer08649012011-08-02 12:32:04 +0100292EXPORT_SYMBOL(dm_table_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293
294/*
295 * Checks to see if we need to extend highs or targets.
296 */
297static inline int check_space(struct dm_table *t)
298{
299 if (t->num_targets >= t->num_allocated)
300 return alloc_targets(t, t->num_allocated * 2);
301
302 return 0;
303}
304
305/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 * See if we've already got a device in the list.
307 */
Mikulas Patocka82b15192008-10-10 13:37:09 +0100308static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309{
Mikulas Patocka82b15192008-10-10 13:37:09 +0100310 struct dm_dev_internal *dd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311
312 list_for_each_entry (dd, l, list)
Mikulas Patocka82b15192008-10-10 13:37:09 +0100313 if (dd->dm_dev.bdev->bd_dev == dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 return dd;
315
316 return NULL;
317}
318
319/*
320 * Open a device so we can use it as a map destination.
321 */
Mikulas Patocka82b15192008-10-10 13:37:09 +0100322static int open_dev(struct dm_dev_internal *d, dev_t dev,
323 struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324{
325 static char *_claim_ptr = "I belong to device-mapper";
326 struct block_device *bdev;
327
328 int r;
329
Mikulas Patocka82b15192008-10-10 13:37:09 +0100330 BUG_ON(d->dm_dev.bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331
Tejun Heod4d77622010-11-13 11:55:18 +0100332 bdev = blkdev_get_by_dev(dev, d->dm_dev.mode | FMODE_EXCL, _claim_ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 if (IS_ERR(bdev))
334 return PTR_ERR(bdev);
Tejun Heoe09b4572010-11-13 11:55:17 +0100335
Tejun Heoe09b4572010-11-13 11:55:17 +0100336 r = bd_link_disk_holder(bdev, dm_disk(md));
337 if (r) {
Tejun Heoe525fd82010-11-13 11:55:17 +0100338 blkdev_put(bdev, d->dm_dev.mode | FMODE_EXCL);
Tejun Heoe09b4572010-11-13 11:55:17 +0100339 return r;
340 }
341
342 d->dm_dev.bdev = bdev;
343 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344}
345
346/*
347 * Close a device that we've been using.
348 */
Mikulas Patocka82b15192008-10-10 13:37:09 +0100349static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350{
Mikulas Patocka82b15192008-10-10 13:37:09 +0100351 if (!d->dm_dev.bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 return;
353
Tejun Heo49731ba2011-01-14 18:43:57 +0100354 bd_unlink_disk_holder(d->dm_dev.bdev, dm_disk(md));
Tejun Heoe525fd82010-11-13 11:55:17 +0100355 blkdev_put(d->dm_dev.bdev, d->dm_dev.mode | FMODE_EXCL);
Mikulas Patocka82b15192008-10-10 13:37:09 +0100356 d->dm_dev.bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357}
358
359/*
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100360 * If possible, this checks an area of a destination device is invalid.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 */
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100362static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
363 sector_t start, sector_t len, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364{
Milan Brozf4808ca2011-05-29 13:02:52 +0100365 struct request_queue *q;
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100366 struct queue_limits *limits = data;
367 struct block_device *bdev = dev->bdev;
368 sector_t dev_size =
369 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100370 unsigned short logical_block_size_sectors =
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100371 limits->logical_block_size >> SECTOR_SHIFT;
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100372 char b[BDEVNAME_SIZE];
Mike Anderson2cd54d92007-05-09 02:32:57 -0700373
Milan Brozf4808ca2011-05-29 13:02:52 +0100374 /*
375 * Some devices exist without request functions,
376 * such as loop devices not yet bound to backing files.
377 * Forbid the use of such devices.
378 */
379 q = bdev_get_queue(bdev);
380 if (!q || !q->make_request_fn) {
381 DMWARN("%s: %s is not yet initialised: "
382 "start=%llu, len=%llu, dev_size=%llu",
383 dm_device_name(ti->table->md), bdevname(bdev, b),
384 (unsigned long long)start,
385 (unsigned long long)len,
386 (unsigned long long)dev_size);
387 return 1;
388 }
389
Mike Anderson2cd54d92007-05-09 02:32:57 -0700390 if (!dev_size)
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100391 return 0;
Mike Anderson2cd54d92007-05-09 02:32:57 -0700392
Mike Snitzer5dea2712009-07-23 20:30:42 +0100393 if ((start >= dev_size) || (start + len > dev_size)) {
Mike Snitzera963a952009-09-04 20:40:24 +0100394 DMWARN("%s: %s too small for target: "
395 "start=%llu, len=%llu, dev_size=%llu",
396 dm_device_name(ti->table->md), bdevname(bdev, b),
397 (unsigned long long)start,
398 (unsigned long long)len,
399 (unsigned long long)dev_size);
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100400 return 1;
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100401 }
402
403 if (logical_block_size_sectors <= 1)
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100404 return 0;
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100405
406 if (start & (logical_block_size_sectors - 1)) {
407 DMWARN("%s: start=%llu not aligned to h/w "
Mike Snitzera963a952009-09-04 20:40:24 +0100408 "logical block size %u of %s",
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100409 dm_device_name(ti->table->md),
410 (unsigned long long)start,
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100411 limits->logical_block_size, bdevname(bdev, b));
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100412 return 1;
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100413 }
414
Mike Snitzer5dea2712009-07-23 20:30:42 +0100415 if (len & (logical_block_size_sectors - 1)) {
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100416 DMWARN("%s: len=%llu not aligned to h/w "
Mike Snitzera963a952009-09-04 20:40:24 +0100417 "logical block size %u of %s",
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100418 dm_device_name(ti->table->md),
Mike Snitzer5dea2712009-07-23 20:30:42 +0100419 (unsigned long long)len,
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100420 limits->logical_block_size, bdevname(bdev, b));
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100421 return 1;
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100422 }
423
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100424 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425}
426
427/*
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100428 * This upgrades the mode on an already open dm_dev, being
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 * careful to leave things as they were if we fail to reopen the
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100430 * device and not to touch the existing bdev field in case
431 * it is accessed concurrently inside dm_table_any_congested().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 */
Al Viroaeb5d722008-09-02 15:28:45 -0400433static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
Mikulas Patocka82b15192008-10-10 13:37:09 +0100434 struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435{
436 int r;
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100437 struct dm_dev_internal dd_new, dd_old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100439 dd_new = dd_old = *dd;
440
441 dd_new.dm_dev.mode |= new_mode;
442 dd_new.dm_dev.bdev = NULL;
443
444 r = open_dev(&dd_new, dd->dm_dev.bdev->bd_dev, md);
445 if (r)
446 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447
Mikulas Patocka82b15192008-10-10 13:37:09 +0100448 dd->dm_dev.mode |= new_mode;
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100449 close_dev(&dd_old, md);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100451 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452}
453
454/*
455 * Add a device to the list, or just increment the usage count if
456 * it's already present.
457 */
Mike Snitzer08649012011-08-02 12:32:04 +0100458int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
459 struct dm_dev **result)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460{
461 int r;
Andrew Morton69a2ce72008-02-08 02:10:14 +0000462 dev_t uninitialized_var(dev);
Mikulas Patocka82b15192008-10-10 13:37:09 +0100463 struct dm_dev_internal *dd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 unsigned int major, minor;
Mike Snitzer08649012011-08-02 12:32:04 +0100465 struct dm_table *t = ti->table;
Mikulas Patocka31998ef2012-03-28 18:41:26 +0100466 char dummy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
Eric Sesterhenn547bc922006-03-26 18:22:50 +0200468 BUG_ON(!t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
Mikulas Patocka31998ef2012-03-28 18:41:26 +0100470 if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 /* Extract the major/minor numbers */
472 dev = MKDEV(major, minor);
473 if (MAJOR(dev) != major || MINOR(dev) != minor)
474 return -EOVERFLOW;
475 } else {
476 /* convert the path to a device */
Christoph Hellwig72e82642008-08-11 00:24:08 +0200477 struct block_device *bdev = lookup_bdev(path);
478
479 if (IS_ERR(bdev))
480 return PTR_ERR(bdev);
481 dev = bdev->bd_dev;
482 bdput(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 }
484
485 dd = find_device(&t->devices, dev);
486 if (!dd) {
487 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
488 if (!dd)
489 return -ENOMEM;
490
Mikulas Patocka82b15192008-10-10 13:37:09 +0100491 dd->dm_dev.mode = mode;
492 dd->dm_dev.bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
Jun'ichi Nomuraf1659212006-03-27 01:17:59 -0800494 if ((r = open_dev(dd, dev, t->md))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 kfree(dd);
496 return r;
497 }
498
Mikulas Patocka82b15192008-10-10 13:37:09 +0100499 format_dev_t(dd->dm_dev.name, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500
501 atomic_set(&dd->count, 0);
502 list_add(&dd->list, &t->devices);
503
Mikulas Patocka82b15192008-10-10 13:37:09 +0100504 } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) {
Jun'ichi Nomuraf1659212006-03-27 01:17:59 -0800505 r = upgrade_mode(dd, mode, t->md);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 if (r)
507 return r;
508 }
509 atomic_inc(&dd->count);
510
Mikulas Patocka82b15192008-10-10 13:37:09 +0100511 *result = &dd->dm_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 return 0;
513}
Mike Snitzer08649012011-08-02 12:32:04 +0100514EXPORT_SYMBOL(dm_get_device);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100516int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
Mike Snitzer5dea2712009-07-23 20:30:42 +0100517 sector_t start, sector_t len, void *data)
Bryn Reeves3cb40212006-10-03 01:15:42 -0700518{
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100519 struct queue_limits *limits = data;
520 struct block_device *bdev = dev->bdev;
Jens Axboe165125e2007-07-24 09:28:11 +0200521 struct request_queue *q = bdev_get_queue(bdev);
Alasdair G Kergon0c2322e2008-10-10 13:37:13 +0100522 char b[BDEVNAME_SIZE];
523
524 if (unlikely(!q)) {
525 DMWARN("%s: Cannot set limits for nonexistent device %s",
526 dm_device_name(ti->table->md), bdevname(bdev, b));
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100527 return 0;
Alasdair G Kergon0c2322e2008-10-10 13:37:13 +0100528 }
Bryn Reeves3cb40212006-10-03 01:15:42 -0700529
Martin K. Petersenb27d7f12010-01-11 03:21:50 -0500530 if (bdev_stack_limits(limits, bdev, start) < 0)
531 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
Mike Snitzera963a952009-09-04 20:40:24 +0100532 "physical_block_size=%u, logical_block_size=%u, "
533 "alignment_offset=%u, start=%llu",
534 dm_device_name(ti->table->md), bdevname(bdev, b),
535 q->limits.physical_block_size,
536 q->limits.logical_block_size,
537 q->limits.alignment_offset,
Martin K. Petersenb27d7f12010-01-11 03:21:50 -0500538 (unsigned long long) start << SECTOR_SHIFT);
Bryn Reeves3cb40212006-10-03 01:15:42 -0700539
Milan Broz9980c632008-07-21 12:00:39 +0100540 /*
541 * Check if merge fn is supported.
542 * If not we'll force DM to use PAGE_SIZE or
543 * smaller I/O, just to be safe.
Bryn Reeves3cb40212006-10-03 01:15:42 -0700544 */
Mikulas Patockad5b9dd02011-08-02 12:32:04 +0100545 if (dm_queue_merge_is_compulsory(q) && !ti->type->merge)
Mike Snitzer72d4cd92010-12-17 08:34:20 +0100546 blk_limits_max_hw_sectors(limits,
547 (unsigned int) (PAGE_SIZE >> 9));
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100548 return 0;
Bryn Reeves3cb40212006-10-03 01:15:42 -0700549}
550EXPORT_SYMBOL_GPL(dm_set_device_limits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552/*
Mike Snitzer08649012011-08-02 12:32:04 +0100553 * Decrement a device's use count and remove it if necessary.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 */
Mikulas Patocka82b15192008-10-10 13:37:09 +0100555void dm_put_device(struct dm_target *ti, struct dm_dev *d)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556{
Mikulas Patocka82b15192008-10-10 13:37:09 +0100557 struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal,
558 dm_dev);
559
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 if (atomic_dec_and_test(&dd->count)) {
Jun'ichi Nomuraf1659212006-03-27 01:17:59 -0800561 close_dev(dd, ti->table->md);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 list_del(&dd->list);
563 kfree(dd);
564 }
565}
Mike Snitzer08649012011-08-02 12:32:04 +0100566EXPORT_SYMBOL(dm_put_device);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
568/*
569 * Checks to see if the target joins onto the end of the table.
570 */
571static int adjoin(struct dm_table *table, struct dm_target *ti)
572{
573 struct dm_target *prev;
574
575 if (!table->num_targets)
576 return !ti->begin;
577
578 prev = &table->targets[table->num_targets - 1];
579 return (ti->begin == (prev->begin + prev->len));
580}
581
582/*
583 * Used to dynamically allocate the arg array.
584 */
585static char **realloc_argv(unsigned *array_size, char **old_argv)
586{
587 char **argv;
588 unsigned new_size;
589
590 new_size = *array_size ? *array_size * 2 : 64;
591 argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
592 if (argv) {
593 memcpy(argv, old_argv, *array_size * sizeof(*argv));
594 *array_size = new_size;
595 }
596
597 kfree(old_argv);
598 return argv;
599}
600
601/*
602 * Destructively splits up the argument list to pass to ctr.
603 */
604int dm_split_args(int *argc, char ***argvp, char *input)
605{
606 char *start, *end = input, *out, **argv = NULL;
607 unsigned array_size = 0;
608
609 *argc = 0;
David Teigland814d6862006-06-26 00:27:31 -0700610
611 if (!input) {
612 *argvp = NULL;
613 return 0;
614 }
615
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 argv = realloc_argv(&array_size, argv);
617 if (!argv)
618 return -ENOMEM;
619
620 while (1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 /* Skip whitespace */
André Goddard Rosae7d28602009-12-14 18:01:06 -0800622 start = skip_spaces(end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623
624 if (!*start)
625 break; /* success, we hit the end */
626
627 /* 'out' is used to remove any back-quotes */
628 end = out = start;
629 while (*end) {
630 /* Everything apart from '\0' can be quoted */
631 if (*end == '\\' && *(end + 1)) {
632 *out++ = *(end + 1);
633 end += 2;
634 continue;
635 }
636
637 if (isspace(*end))
638 break; /* end of token */
639
640 *out++ = *end++;
641 }
642
643 /* have we already filled the array ? */
644 if ((*argc + 1) > array_size) {
645 argv = realloc_argv(&array_size, argv);
646 if (!argv)
647 return -ENOMEM;
648 }
649
650 /* we know this is whitespace */
651 if (*end)
652 end++;
653
654 /* terminate the string and put it in the array */
655 *out = '\0';
656 argv[*argc] = start;
657 (*argc)++;
658 }
659
660 *argvp = argv;
661 return 0;
662}
663
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100664/*
665 * Impose necessary and sufficient conditions on a devices's table such
666 * that any incoming bio which respects its logical_block_size can be
667 * processed successfully. If it falls across the boundary between
668 * two or more targets, the size of each piece it gets split into must
669 * be compatible with the logical_block_size of the target processing it.
670 */
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100671static int validate_hardware_logical_block_alignment(struct dm_table *table,
672 struct queue_limits *limits)
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100673{
674 /*
675 * This function uses arithmetic modulo the logical_block_size
676 * (in units of 512-byte sectors).
677 */
678 unsigned short device_logical_block_size_sects =
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100679 limits->logical_block_size >> SECTOR_SHIFT;
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100680
681 /*
682 * Offset of the start of the next table entry, mod logical_block_size.
683 */
684 unsigned short next_target_start = 0;
685
686 /*
687 * Given an aligned bio that extends beyond the end of a
688 * target, how many sectors must the next target handle?
689 */
690 unsigned short remaining = 0;
691
692 struct dm_target *uninitialized_var(ti);
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100693 struct queue_limits ti_limits;
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100694 unsigned i = 0;
695
696 /*
697 * Check each entry in the table in turn.
698 */
699 while (i < dm_table_get_num_targets(table)) {
700 ti = dm_table_get_target(table, i++);
701
Martin K. Petersenb1bd0552012-01-11 16:27:11 +0100702 blk_set_stacking_limits(&ti_limits);
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100703
704 /* combine all target devices' limits */
705 if (ti->type->iterate_devices)
706 ti->type->iterate_devices(ti, dm_set_device_limits,
707 &ti_limits);
708
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100709 /*
710 * If the remaining sectors fall entirely within this
711 * table entry are they compatible with its logical_block_size?
712 */
713 if (remaining < ti->len &&
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100714 remaining & ((ti_limits.logical_block_size >>
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100715 SECTOR_SHIFT) - 1))
716 break; /* Error */
717
718 next_target_start =
719 (unsigned short) ((next_target_start + ti->len) &
720 (device_logical_block_size_sects - 1));
721 remaining = next_target_start ?
722 device_logical_block_size_sects - next_target_start : 0;
723 }
724
725 if (remaining) {
726 DMWARN("%s: table line %u (start sect %llu len %llu) "
Mike Snitzera963a952009-09-04 20:40:24 +0100727 "not aligned to h/w logical block size %u",
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100728 dm_device_name(table->md), i,
729 (unsigned long long) ti->begin,
730 (unsigned long long) ti->len,
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100731 limits->logical_block_size);
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100732 return -EINVAL;
733 }
734
735 return 0;
736}
737
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738int dm_table_add_target(struct dm_table *t, const char *type,
739 sector_t start, sector_t len, char *params)
740{
741 int r = -EINVAL, argc;
742 char **argv;
743 struct dm_target *tgt;
744
Alasdair G Kergon3791e2f2011-10-31 20:19:00 +0000745 if (t->singleton) {
746 DMERR("%s: target type %s must appear alone in table",
747 dm_device_name(t->md), t->targets->type->name);
748 return -EINVAL;
749 }
750
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 if ((r = check_space(t)))
752 return r;
753
754 tgt = t->targets + t->num_targets;
755 memset(tgt, 0, sizeof(*tgt));
756
757 if (!len) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700758 DMERR("%s: zero-length target", dm_device_name(t->md));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 return -EINVAL;
760 }
761
762 tgt->type = dm_get_target_type(type);
763 if (!tgt->type) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700764 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
765 type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 return -EINVAL;
767 }
768
Alasdair G Kergon3791e2f2011-10-31 20:19:00 +0000769 if (dm_target_needs_singleton(tgt->type)) {
770 if (t->num_targets) {
771 DMERR("%s: target type %s must appear alone in table",
772 dm_device_name(t->md), type);
773 return -EINVAL;
774 }
775 t->singleton = 1;
776 }
777
Alasdair G Kergoncc6cbe12011-10-31 20:19:02 +0000778 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
779 DMERR("%s: target type %s may not be included in read-only tables",
780 dm_device_name(t->md), type);
781 return -EINVAL;
782 }
783
Alasdair G Kergon36a04562011-10-31 20:19:04 +0000784 if (t->immutable_target_type) {
785 if (t->immutable_target_type != tgt->type) {
786 DMERR("%s: immutable target type %s cannot be mixed with other target types",
787 dm_device_name(t->md), t->immutable_target_type->name);
788 return -EINVAL;
789 }
790 } else if (dm_target_is_immutable(tgt->type)) {
791 if (t->num_targets) {
792 DMERR("%s: immutable target type %s cannot be mixed with other target types",
793 dm_device_name(t->md), tgt->type->name);
794 return -EINVAL;
795 }
796 t->immutable_target_type = tgt->type;
797 }
798
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 tgt->table = t;
800 tgt->begin = start;
801 tgt->len = len;
802 tgt->error = "Unknown error";
803
804 /*
805 * Does this target adjoin the previous one ?
806 */
807 if (!adjoin(t, tgt)) {
808 tgt->error = "Gap in table";
809 r = -EINVAL;
810 goto bad;
811 }
812
813 r = dm_split_args(&argc, &argv, params);
814 if (r) {
815 tgt->error = "couldn't split parameters (insufficient memory)";
816 goto bad;
817 }
818
819 r = tgt->type->ctr(tgt, argc, argv);
820 kfree(argv);
821 if (r)
822 goto bad;
823
824 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
825
Mike Snitzer936688d2011-08-02 12:32:01 +0100826 if (!tgt->num_discard_requests && tgt->discards_supported)
827 DMWARN("%s: %s: ignoring discards_supported because num_discard_requests is zero.",
828 dm_device_name(t->md), type);
Mike Snitzer5ae89a82010-08-12 04:14:08 +0100829
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 return 0;
831
832 bad:
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700833 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 dm_put_target_type(tgt->type);
835 return r;
836}
837
Mike Snitzer498f0102011-08-02 12:32:04 +0100838/*
839 * Target argument parsing helpers.
840 */
841static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
842 unsigned *value, char **error, unsigned grouped)
843{
844 const char *arg_str = dm_shift_arg(arg_set);
Mikulas Patocka31998ef2012-03-28 18:41:26 +0100845 char dummy;
Mike Snitzer498f0102011-08-02 12:32:04 +0100846
847 if (!arg_str ||
Mikulas Patocka31998ef2012-03-28 18:41:26 +0100848 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
Mike Snitzer498f0102011-08-02 12:32:04 +0100849 (*value < arg->min) ||
850 (*value > arg->max) ||
851 (grouped && arg_set->argc < *value)) {
852 *error = arg->error;
853 return -EINVAL;
854 }
855
856 return 0;
857}
858
859int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
860 unsigned *value, char **error)
861{
862 return validate_next_arg(arg, arg_set, value, error, 0);
863}
864EXPORT_SYMBOL(dm_read_arg);
865
866int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
867 unsigned *value, char **error)
868{
869 return validate_next_arg(arg, arg_set, value, error, 1);
870}
871EXPORT_SYMBOL(dm_read_arg_group);
872
873const char *dm_shift_arg(struct dm_arg_set *as)
874{
875 char *r;
876
877 if (as->argc) {
878 as->argc--;
879 r = *as->argv;
880 as->argv++;
881 return r;
882 }
883
884 return NULL;
885}
886EXPORT_SYMBOL(dm_shift_arg);
887
888void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
889{
890 BUG_ON(as->argc < num_args);
891 as->argc -= num_args;
892 as->argv += num_args;
893}
894EXPORT_SYMBOL(dm_consume_args);
895
Will Drewry26803b92010-08-12 04:14:03 +0100896static int dm_table_set_type(struct dm_table *t)
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100897{
898 unsigned i;
899 unsigned bio_based = 0, request_based = 0;
900 struct dm_target *tgt;
901 struct dm_dev_internal *dd;
902 struct list_head *devices;
903
904 for (i = 0; i < t->num_targets; i++) {
905 tgt = t->targets + i;
906 if (dm_target_request_based(tgt))
907 request_based = 1;
908 else
909 bio_based = 1;
910
911 if (bio_based && request_based) {
912 DMWARN("Inconsistent table: different target types"
913 " can't be mixed up");
914 return -EINVAL;
915 }
916 }
917
918 if (bio_based) {
919 /* We must use this table as bio-based */
920 t->type = DM_TYPE_BIO_BASED;
921 return 0;
922 }
923
924 BUG_ON(!request_based); /* No targets in this table */
925
926 /* Non-request-stackable devices can't be used for request-based dm */
927 devices = dm_table_get_devices(t);
928 list_for_each_entry(dd, devices, list) {
929 if (!blk_queue_stackable(bdev_get_queue(dd->dm_dev.bdev))) {
930 DMWARN("table load rejected: including"
931 " non-request-stackable devices");
932 return -EINVAL;
933 }
934 }
935
936 /*
937 * Request-based dm supports only tables that have a single target now.
938 * To support multiple targets, request splitting support is needed,
939 * and that needs lots of changes in the block-layer.
940 * (e.g. request completion process for partial completion.)
941 */
942 if (t->num_targets > 1) {
943 DMWARN("Request-based dm doesn't support multiple targets yet");
944 return -EINVAL;
945 }
946
947 t->type = DM_TYPE_REQUEST_BASED;
948
949 return 0;
950}
951
952unsigned dm_table_get_type(struct dm_table *t)
953{
954 return t->type;
955}
956
Alasdair G Kergon36a04562011-10-31 20:19:04 +0000957struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
958{
959 return t->immutable_target_type;
960}
961
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100962bool dm_table_request_based(struct dm_table *t)
963{
964 return dm_table_get_type(t) == DM_TYPE_REQUEST_BASED;
965}
966
967int dm_table_alloc_md_mempools(struct dm_table *t)
968{
969 unsigned type = dm_table_get_type(t);
970
971 if (unlikely(type == DM_TYPE_NONE)) {
972 DMWARN("no table type is set, can't allocate mempools");
973 return -EINVAL;
974 }
975
Martin K. Petersena91a2782011-03-17 11:11:05 +0100976 t->mempools = dm_alloc_md_mempools(type, t->integrity_supported);
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100977 if (!t->mempools)
978 return -ENOMEM;
979
980 return 0;
981}
982
983void dm_table_free_md_mempools(struct dm_table *t)
984{
985 dm_free_md_mempools(t->mempools);
986 t->mempools = NULL;
987}
988
989struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
990{
991 return t->mempools;
992}
993
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994static int setup_indexes(struct dm_table *t)
995{
996 int i;
997 unsigned int total = 0;
998 sector_t *indexes;
999
1000 /* allocate the space for *all* the indexes */
1001 for (i = t->depth - 2; i >= 0; i--) {
1002 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1003 total += t->counts[i];
1004 }
1005
1006 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1007 if (!indexes)
1008 return -ENOMEM;
1009
1010 /* set up internal nodes, bottom-up */
Jun'ichi Nomura82d601dc2008-02-08 02:10:04 +00001011 for (i = t->depth - 2; i >= 0; i--) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 t->index[i] = indexes;
1013 indexes += (KEYS_PER_NODE * t->counts[i]);
1014 setup_btree_index(i, t);
1015 }
1016
1017 return 0;
1018}
1019
1020/*
1021 * Builds the btree to index the map.
1022 */
Will Drewry26803b92010-08-12 04:14:03 +01001023static int dm_table_build_index(struct dm_table *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024{
1025 int r = 0;
1026 unsigned int leaf_nodes;
1027
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 /* how many indexes will the btree have ? */
1029 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1030 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1031
1032 /* leaf layer has already been set up */
1033 t->counts[t->depth - 1] = leaf_nodes;
1034 t->index[t->depth - 1] = t->highs;
1035
1036 if (t->depth >= 2)
1037 r = setup_indexes(t);
1038
1039 return r;
1040}
1041
Will Drewry26803b92010-08-12 04:14:03 +01001042/*
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001043 * Get a disk whose integrity profile reflects the table's profile.
1044 * If %match_all is true, all devices' profiles must match.
1045 * If %match_all is false, all devices must at least have an
1046 * allocated integrity profile; but uninitialized is ok.
1047 * Returns NULL if integrity support was inconsistent or unavailable.
1048 */
1049static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t,
1050 bool match_all)
1051{
1052 struct list_head *devices = dm_table_get_devices(t);
1053 struct dm_dev_internal *dd = NULL;
1054 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1055
1056 list_for_each_entry(dd, devices, list) {
1057 template_disk = dd->dm_dev.bdev->bd_disk;
1058 if (!blk_get_integrity(template_disk))
1059 goto no_integrity;
1060 if (!match_all && !blk_integrity_is_initialized(template_disk))
1061 continue; /* skip uninitialized profiles */
1062 else if (prev_disk &&
1063 blk_integrity_compare(prev_disk, template_disk) < 0)
1064 goto no_integrity;
1065 prev_disk = template_disk;
1066 }
1067
1068 return template_disk;
1069
1070no_integrity:
1071 if (prev_disk)
1072 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1073 dm_device_name(t->md),
1074 prev_disk->disk_name,
1075 template_disk->disk_name);
1076 return NULL;
1077}
1078
1079/*
Will Drewry26803b92010-08-12 04:14:03 +01001080 * Register the mapped device for blk_integrity support if
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001081 * the underlying devices have an integrity profile. But all devices
1082 * may not have matching profiles (checking all devices isn't reliable
1083 * during table load because this table may use other DM device(s) which
1084 * must be resumed before they will have an initialized integity profile).
1085 * Stacked DM devices force a 2 stage integrity profile validation:
1086 * 1 - during load, validate all initialized integrity profiles match
1087 * 2 - during resume, validate all integrity profiles match
Will Drewry26803b92010-08-12 04:14:03 +01001088 */
1089static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md)
1090{
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001091 struct gendisk *template_disk = NULL;
Will Drewry26803b92010-08-12 04:14:03 +01001092
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001093 template_disk = dm_table_get_integrity_disk(t, false);
1094 if (!template_disk)
1095 return 0;
Will Drewry26803b92010-08-12 04:14:03 +01001096
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001097 if (!blk_integrity_is_initialized(dm_disk(md))) {
1098 t->integrity_supported = 1;
1099 return blk_integrity_register(dm_disk(md), NULL);
1100 }
1101
1102 /*
1103 * If DM device already has an initalized integrity
1104 * profile the new profile should not conflict.
1105 */
1106 if (blk_integrity_is_initialized(template_disk) &&
1107 blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1108 DMWARN("%s: conflict with existing integrity profile: "
1109 "%s profile mismatch",
1110 dm_device_name(t->md),
1111 template_disk->disk_name);
1112 return 1;
1113 }
1114
1115 /* Preserve existing initialized integrity profile */
1116 t->integrity_supported = 1;
Will Drewry26803b92010-08-12 04:14:03 +01001117 return 0;
1118}
1119
1120/*
1121 * Prepares the table for use by building the indices,
1122 * setting the type, and allocating mempools.
1123 */
1124int dm_table_complete(struct dm_table *t)
1125{
1126 int r;
1127
1128 r = dm_table_set_type(t);
1129 if (r) {
1130 DMERR("unable to set table type");
1131 return r;
1132 }
1133
1134 r = dm_table_build_index(t);
1135 if (r) {
1136 DMERR("unable to build btrees");
1137 return r;
1138 }
1139
1140 r = dm_table_prealloc_integrity(t, t->md);
1141 if (r) {
1142 DMERR("could not register integrity profile.");
1143 return r;
1144 }
1145
1146 r = dm_table_alloc_md_mempools(t);
1147 if (r)
1148 DMERR("unable to allocate mempools");
1149
1150 return r;
1151}
1152
Arjan van de Ven48c9c272006-03-27 01:18:20 -08001153static DEFINE_MUTEX(_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154void dm_table_event_callback(struct dm_table *t,
1155 void (*fn)(void *), void *context)
1156{
Arjan van de Ven48c9c272006-03-27 01:18:20 -08001157 mutex_lock(&_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 t->event_fn = fn;
1159 t->event_context = context;
Arjan van de Ven48c9c272006-03-27 01:18:20 -08001160 mutex_unlock(&_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161}
1162
1163void dm_table_event(struct dm_table *t)
1164{
1165 /*
1166 * You can no longer call dm_table_event() from interrupt
1167 * context, use a bottom half instead.
1168 */
1169 BUG_ON(in_interrupt());
1170
Arjan van de Ven48c9c272006-03-27 01:18:20 -08001171 mutex_lock(&_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 if (t->event_fn)
1173 t->event_fn(t->event_context);
Arjan van de Ven48c9c272006-03-27 01:18:20 -08001174 mutex_unlock(&_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175}
Mike Snitzer08649012011-08-02 12:32:04 +01001176EXPORT_SYMBOL(dm_table_event);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177
1178sector_t dm_table_get_size(struct dm_table *t)
1179{
1180 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1181}
Mike Snitzer08649012011-08-02 12:32:04 +01001182EXPORT_SYMBOL(dm_table_get_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
1184struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1185{
Milan Broz14353532006-06-26 00:27:27 -07001186 if (index >= t->num_targets)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 return NULL;
1188
1189 return t->targets + index;
1190}
1191
1192/*
1193 * Search the btree for the correct target.
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001194 *
1195 * Caller should check returned pointer with dm_target_is_valid()
1196 * to trap I/O beyond end of device.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 */
1198struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1199{
1200 unsigned int l, n = 0, k = 0;
1201 sector_t *node;
1202
1203 for (l = 0; l < t->depth; l++) {
1204 n = get_child(n, k);
1205 node = get_node(t, l, n);
1206
1207 for (k = 0; k < KEYS_PER_NODE; k++)
1208 if (node[k] >= sector)
1209 break;
1210 }
1211
1212 return &t->targets[(KEYS_PER_NODE * n) + k];
1213}
1214
Martin K. Petersen9c470082009-04-09 00:27:12 +01001215/*
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001216 * Establish the new table's queue_limits and validate them.
1217 */
1218int dm_calculate_queue_limits(struct dm_table *table,
1219 struct queue_limits *limits)
1220{
1221 struct dm_target *uninitialized_var(ti);
1222 struct queue_limits ti_limits;
1223 unsigned i = 0;
1224
Martin K. Petersenb1bd0552012-01-11 16:27:11 +01001225 blk_set_stacking_limits(limits);
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001226
1227 while (i < dm_table_get_num_targets(table)) {
Martin K. Petersenb1bd0552012-01-11 16:27:11 +01001228 blk_set_stacking_limits(&ti_limits);
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001229
1230 ti = dm_table_get_target(table, i++);
1231
1232 if (!ti->type->iterate_devices)
1233 goto combine_limits;
1234
1235 /*
1236 * Combine queue limits of all the devices this target uses.
1237 */
1238 ti->type->iterate_devices(ti, dm_set_device_limits,
1239 &ti_limits);
1240
Mike Snitzer40bea432009-09-04 20:40:25 +01001241 /* Set I/O hints portion of queue limits */
1242 if (ti->type->io_hints)
1243 ti->type->io_hints(ti, &ti_limits);
1244
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001245 /*
1246 * Check each device area is consistent with the target's
1247 * overall queue limits.
1248 */
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +01001249 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1250 &ti_limits))
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001251 return -EINVAL;
1252
1253combine_limits:
1254 /*
1255 * Merge this target's queue limits into the overall limits
1256 * for the table.
1257 */
1258 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
Martin K. Petersenb27d7f12010-01-11 03:21:50 -05001259 DMWARN("%s: adding target device "
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001260 "(start sect %llu len %llu) "
Martin K. Petersenb27d7f12010-01-11 03:21:50 -05001261 "caused an alignment inconsistency",
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001262 dm_device_name(table->md),
1263 (unsigned long long) ti->begin,
1264 (unsigned long long) ti->len);
1265 }
1266
1267 return validate_hardware_logical_block_alignment(table, limits);
1268}
1269
1270/*
Martin K. Petersen9c470082009-04-09 00:27:12 +01001271 * Set the integrity profile for this device if all devices used have
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001272 * matching profiles. We're quite deep in the resume path but still
1273 * don't know if all devices (particularly DM devices this device
1274 * may be stacked on) have matching profiles. Even if the profiles
1275 * don't match we have no way to fail (to resume) at this point.
Martin K. Petersen9c470082009-04-09 00:27:12 +01001276 */
1277static void dm_table_set_integrity(struct dm_table *t)
1278{
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001279 struct gendisk *template_disk = NULL;
Martin K. Petersen9c470082009-04-09 00:27:12 +01001280
1281 if (!blk_get_integrity(dm_disk(t->md)))
1282 return;
1283
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001284 template_disk = dm_table_get_integrity_disk(t, true);
Mike Snitzer876fbba2011-09-25 23:26:17 +01001285 if (template_disk)
1286 blk_integrity_register(dm_disk(t->md),
1287 blk_get_integrity(template_disk));
1288 else if (blk_integrity_is_initialized(dm_disk(t->md)))
Mike Snitzera63a5cf2011-04-01 21:02:31 +02001289 DMWARN("%s: device no longer has a valid integrity profile",
1290 dm_device_name(t->md));
Mike Snitzer876fbba2011-09-25 23:26:17 +01001291 else
1292 DMWARN("%s: unable to establish an integrity profile",
1293 dm_device_name(t->md));
Martin K. Petersen9c470082009-04-09 00:27:12 +01001294}
1295
Mike Snitzered8b7522011-08-02 12:32:08 +01001296static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1297 sector_t start, sector_t len, void *data)
1298{
1299 unsigned flush = (*(unsigned *)data);
1300 struct request_queue *q = bdev_get_queue(dev->bdev);
1301
1302 return q && (q->flush_flags & flush);
1303}
1304
1305static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1306{
1307 struct dm_target *ti;
1308 unsigned i = 0;
1309
1310 /*
1311 * Require at least one underlying device to support flushes.
1312 * t->devices includes internal dm devices such as mirror logs
1313 * so we need to use iterate_devices here, which targets
1314 * supporting flushes must provide.
1315 */
1316 while (i < dm_table_get_num_targets(t)) {
1317 ti = dm_table_get_target(t, i++);
1318
1319 if (!ti->num_flush_requests)
1320 continue;
1321
1322 if (ti->type->iterate_devices &&
1323 ti->type->iterate_devices(ti, device_flush_capable, &flush))
1324 return 1;
1325 }
1326
1327 return 0;
1328}
1329
Milan Broz983c7db2011-09-25 23:26:21 +01001330static bool dm_table_discard_zeroes_data(struct dm_table *t)
1331{
1332 struct dm_target *ti;
1333 unsigned i = 0;
1334
1335 /* Ensure that all targets supports discard_zeroes_data. */
1336 while (i < dm_table_get_num_targets(t)) {
1337 ti = dm_table_get_target(t, i++);
1338
1339 if (ti->discard_zeroes_data_unsupported)
1340 return 0;
1341 }
1342
1343 return 1;
1344}
1345
Mandeep Singh Baines4693c9662011-10-31 20:18:50 +00001346static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1347 sector_t start, sector_t len, void *data)
1348{
1349 struct request_queue *q = bdev_get_queue(dev->bdev);
1350
1351 return q && blk_queue_nonrot(q);
1352}
1353
1354static bool dm_table_is_nonrot(struct dm_table *t)
1355{
1356 struct dm_target *ti;
1357 unsigned i = 0;
1358
1359 /* Ensure that all underlying device are non-rotational. */
1360 while (i < dm_table_get_num_targets(t)) {
1361 ti = dm_table_get_target(t, i++);
1362
1363 if (!ti->type->iterate_devices ||
1364 !ti->type->iterate_devices(ti, device_is_nonrot, NULL))
1365 return 0;
1366 }
1367
1368 return 1;
1369}
1370
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001371void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1372 struct queue_limits *limits)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373{
Mike Snitzered8b7522011-08-02 12:32:08 +01001374 unsigned flush = 0;
1375
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 /*
Mike Snitzer11977642009-06-22 10:12:32 +01001377 * Copy table's limits to the DM device's request_queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 */
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001379 q->limits = *limits;
Jens Axboec9a3f6d2008-04-29 19:12:35 +02001380
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001381 if (!dm_table_supports_discards(t))
1382 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1383 else
1384 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1385
Mike Snitzered8b7522011-08-02 12:32:08 +01001386 if (dm_table_supports_flush(t, REQ_FLUSH)) {
1387 flush |= REQ_FLUSH;
1388 if (dm_table_supports_flush(t, REQ_FUA))
1389 flush |= REQ_FUA;
1390 }
1391 blk_queue_flush(q, flush);
1392
Milan Broz983c7db2011-09-25 23:26:21 +01001393 if (!dm_table_discard_zeroes_data(t))
1394 q->limits.discard_zeroes_data = 0;
1395
Mandeep Singh Baines4693c9662011-10-31 20:18:50 +00001396 if (dm_table_is_nonrot(t))
1397 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1398 else
1399 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1400
Martin K. Petersen9c470082009-04-09 00:27:12 +01001401 dm_table_set_integrity(t);
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001402
1403 /*
1404 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1405 * visible to other CPUs because, once the flag is set, incoming bios
1406 * are processed by request-based dm, which refers to the queue
1407 * settings.
1408 * Until the flag set, bios are passed to bio-based dm and queued to
1409 * md->deferred where queue settings are not needed yet.
1410 * Those bios are passed to request-based dm at the resume time.
1411 */
1412 smp_mb();
1413 if (dm_table_request_based(t))
1414 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415}
1416
1417unsigned int dm_table_get_num_targets(struct dm_table *t)
1418{
1419 return t->num_targets;
1420}
1421
1422struct list_head *dm_table_get_devices(struct dm_table *t)
1423{
1424 return &t->devices;
1425}
1426
Al Viroaeb5d722008-09-02 15:28:45 -04001427fmode_t dm_table_get_mode(struct dm_table *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428{
1429 return t->mode;
1430}
Mike Snitzer08649012011-08-02 12:32:04 +01001431EXPORT_SYMBOL(dm_table_get_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
1433static void suspend_targets(struct dm_table *t, unsigned postsuspend)
1434{
1435 int i = t->num_targets;
1436 struct dm_target *ti = t->targets;
1437
1438 while (i--) {
1439 if (postsuspend) {
1440 if (ti->type->postsuspend)
1441 ti->type->postsuspend(ti);
1442 } else if (ti->type->presuspend)
1443 ti->type->presuspend(ti);
1444
1445 ti++;
1446 }
1447}
1448
1449void dm_table_presuspend_targets(struct dm_table *t)
1450{
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001451 if (!t)
1452 return;
1453
Adrian Bunke8488d02008-04-24 22:10:51 +01001454 suspend_targets(t, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455}
1456
1457void dm_table_postsuspend_targets(struct dm_table *t)
1458{
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001459 if (!t)
1460 return;
1461
Adrian Bunke8488d02008-04-24 22:10:51 +01001462 suspend_targets(t, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463}
1464
Milan Broz8757b772006-10-03 01:15:36 -07001465int dm_table_resume_targets(struct dm_table *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466{
Milan Broz8757b772006-10-03 01:15:36 -07001467 int i, r = 0;
1468
1469 for (i = 0; i < t->num_targets; i++) {
1470 struct dm_target *ti = t->targets + i;
1471
1472 if (!ti->type->preresume)
1473 continue;
1474
1475 r = ti->type->preresume(ti);
1476 if (r)
1477 return r;
1478 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479
1480 for (i = 0; i < t->num_targets; i++) {
1481 struct dm_target *ti = t->targets + i;
1482
1483 if (ti->type->resume)
1484 ti->type->resume(ti);
1485 }
Milan Broz8757b772006-10-03 01:15:36 -07001486
1487 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488}
1489
NeilBrown9d357b02011-01-13 20:00:01 +00001490void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1491{
1492 list_add(&cb->list, &t->target_callbacks);
1493}
1494EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1495
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1497{
Mikulas Patocka82b15192008-10-10 13:37:09 +01001498 struct dm_dev_internal *dd;
Paul Jimenezafb24522008-02-08 02:09:59 +00001499 struct list_head *devices = dm_table_get_devices(t);
NeilBrown9d357b02011-01-13 20:00:01 +00001500 struct dm_target_callbacks *cb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 int r = 0;
1502
Paul Jimenezafb24522008-02-08 02:09:59 +00001503 list_for_each_entry(dd, devices, list) {
Mikulas Patocka82b15192008-10-10 13:37:09 +01001504 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
Alasdair G Kergon0c2322e2008-10-10 13:37:13 +01001505 char b[BDEVNAME_SIZE];
1506
1507 if (likely(q))
1508 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1509 else
1510 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1511 dm_device_name(t->md),
1512 bdevname(dd->dm_dev.bdev, b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 }
1514
NeilBrown9d357b02011-01-13 20:00:01 +00001515 list_for_each_entry(cb, &t->target_callbacks, list)
1516 if (cb->congested_fn)
1517 r |= cb->congested_fn(cb, bdi_bits);
1518
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 return r;
1520}
1521
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001522int dm_table_any_busy_target(struct dm_table *t)
1523{
1524 unsigned i;
1525 struct dm_target *ti;
1526
1527 for (i = 0; i < t->num_targets; i++) {
1528 ti = t->targets + i;
1529 if (ti->type->busy && ti->type->busy(ti))
1530 return 1;
1531 }
1532
1533 return 0;
1534}
1535
Mike Anderson1134e5ae2006-03-27 01:17:54 -08001536struct mapped_device *dm_table_get_md(struct dm_table *t)
1537{
Mike Anderson1134e5ae2006-03-27 01:17:54 -08001538 return t->md;
1539}
Mike Snitzer08649012011-08-02 12:32:04 +01001540EXPORT_SYMBOL(dm_table_get_md);
Mike Anderson1134e5ae2006-03-27 01:17:54 -08001541
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001542static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1543 sector_t start, sector_t len, void *data)
1544{
1545 struct request_queue *q = bdev_get_queue(dev->bdev);
1546
1547 return q && blk_queue_discard(q);
1548}
1549
1550bool dm_table_supports_discards(struct dm_table *t)
1551{
1552 struct dm_target *ti;
1553 unsigned i = 0;
1554
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001555 /*
Mike Snitzer4c259322011-05-29 12:52:55 +01001556 * Unless any target used by the table set discards_supported,
1557 * require at least one underlying device to support discards.
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001558 * t->devices includes internal dm devices such as mirror logs
1559 * so we need to use iterate_devices here, which targets
Mike Snitzer936688d2011-08-02 12:32:01 +01001560 * supporting discard selectively must provide.
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001561 */
1562 while (i < dm_table_get_num_targets(t)) {
1563 ti = dm_table_get_target(t, i++);
1564
Mike Snitzer936688d2011-08-02 12:32:01 +01001565 if (!ti->num_discard_requests)
1566 continue;
1567
Mike Snitzer4c259322011-05-29 12:52:55 +01001568 if (ti->discards_supported)
1569 return 1;
1570
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001571 if (ti->type->iterate_devices &&
1572 ti->type->iterate_devices(ti, device_discard_capable, NULL))
1573 return 1;
1574 }
1575
1576 return 0;
1577}