blob: f9fc07d7a4b91f9db6aeb688cb5346bf604d1ac4 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <asm/atomic.h>
21
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 Anderson1134e5a2006-03-27 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
Mike Snitzer5ae89a82010-08-12 04:14:08 +010057 unsigned discards_supported:1;
58
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 /*
60 * Indicates the rw permissions for the new logical
61 * device. This should be a combination of FMODE_READ
62 * and FMODE_WRITE.
63 */
Al Viroaeb5d722008-09-02 15:28:45 -040064 fmode_t mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
66 /* a list of devices used by this table */
67 struct list_head devices;
68
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 /* events get handed up using this callback */
70 void (*event_fn)(void *);
71 void *event_context;
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +010072
73 struct dm_md_mempools *mempools;
Linus Torvalds1da177e2005-04-16 15:20:36 -070074};
75
76/*
77 * Similar to ceiling(log_size(n))
78 */
79static unsigned int int_log(unsigned int n, unsigned int base)
80{
81 int result = 0;
82
83 while (n > 1) {
84 n = dm_div_up(n, base);
85 result++;
86 }
87
88 return result;
89}
90
91/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 * Calculate the index of the child node of the n'th node k'th key.
93 */
94static inline unsigned int get_child(unsigned int n, unsigned int k)
95{
96 return (n * CHILDREN_PER_NODE) + k;
97}
98
99/*
100 * Return the n'th node of level l from table t.
101 */
102static inline sector_t *get_node(struct dm_table *t,
103 unsigned int l, unsigned int n)
104{
105 return t->index[l] + (n * KEYS_PER_NODE);
106}
107
108/*
109 * Return the highest key that you could lookup from the n'th
110 * node on level l of the btree.
111 */
112static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
113{
114 for (; l < t->depth - 1; l++)
115 n = get_child(n, CHILDREN_PER_NODE - 1);
116
117 if (n >= t->counts[l])
118 return (sector_t) - 1;
119
120 return get_node(t, l, n)[KEYS_PER_NODE - 1];
121}
122
123/*
124 * Fills in a level of the btree based on the highs of the level
125 * below it.
126 */
127static int setup_btree_index(unsigned int l, struct dm_table *t)
128{
129 unsigned int n, k;
130 sector_t *node;
131
132 for (n = 0U; n < t->counts[l]; n++) {
133 node = get_node(t, l, n);
134
135 for (k = 0U; k < KEYS_PER_NODE; k++)
136 node[k] = high(t, l + 1, get_child(n, k));
137 }
138
139 return 0;
140}
141
142void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
143{
144 unsigned long size;
145 void *addr;
146
147 /*
148 * Check that we're not going to overflow.
149 */
150 if (nmemb > (ULONG_MAX / elem_size))
151 return NULL;
152
153 size = nmemb * elem_size;
154 addr = vmalloc(size);
155 if (addr)
156 memset(addr, 0, size);
157
158 return addr;
159}
160
161/*
162 * highs, and targets are managed as dynamic arrays during a
163 * table load.
164 */
165static int alloc_targets(struct dm_table *t, unsigned int num)
166{
167 sector_t *n_highs;
168 struct dm_target *n_targets;
169 int n = t->num_targets;
170
171 /*
172 * Allocate both the target array and offset array at once.
Jun'ichi Nomura512875b2007-12-13 14:15:25 +0000173 * Append an empty entry to catch sectors beyond the end of
174 * the device.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 */
Jun'ichi Nomura512875b2007-12-13 14:15:25 +0000176 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 sizeof(sector_t));
178 if (!n_highs)
179 return -ENOMEM;
180
181 n_targets = (struct dm_target *) (n_highs + num);
182
183 if (n) {
184 memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
185 memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
186 }
187
188 memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
189 vfree(t->highs);
190
191 t->num_allocated = num;
192 t->highs = n_highs;
193 t->targets = n_targets;
194
195 return 0;
196}
197
Al Viroaeb5d722008-09-02 15:28:45 -0400198int dm_table_create(struct dm_table **result, fmode_t mode,
Mike Anderson1134e5a2006-03-27 01:17:54 -0800199 unsigned num_targets, struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200{
Dmitry Monakhov094262d2007-10-19 22:38:51 +0100201 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202
203 if (!t)
204 return -ENOMEM;
205
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 INIT_LIST_HEAD(&t->devices);
Mikulas Patockad5816872009-01-06 03:05:10 +0000207 atomic_set(&t->holders, 0);
Mike Snitzer5ae89a82010-08-12 04:14:08 +0100208 t->discards_supported = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
210 if (!num_targets)
211 num_targets = KEYS_PER_NODE;
212
213 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
214
215 if (alloc_targets(t, num_targets)) {
216 kfree(t);
217 t = NULL;
218 return -ENOMEM;
219 }
220
221 t->mode = mode;
Mike Anderson1134e5a2006-03-27 01:17:54 -0800222 t->md = md;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 *result = t;
224 return 0;
225}
226
227static void free_devices(struct list_head *devices)
228{
229 struct list_head *tmp, *next;
230
Paul Jimenezafb24522008-02-08 02:09:59 +0000231 list_for_each_safe(tmp, next, devices) {
Mikulas Patocka82b15192008-10-10 13:37:09 +0100232 struct dm_dev_internal *dd =
233 list_entry(tmp, struct dm_dev_internal, list);
Jonthan Brassow1b6da752009-06-22 10:12:29 +0100234 DMWARN("dm_table_destroy: dm_put_device call missing for %s",
235 dd->dm_dev.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 kfree(dd);
237 }
238}
239
Mikulas Patockad5816872009-01-06 03:05:10 +0000240void dm_table_destroy(struct dm_table *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241{
242 unsigned int i;
243
Alasdair G Kergona7940152009-12-10 23:52:23 +0000244 if (!t)
245 return;
246
Mikulas Patockad5816872009-01-06 03:05:10 +0000247 while (atomic_read(&t->holders))
248 msleep(1);
249 smp_mb();
250
Will Drewry26803b92010-08-12 04:14:03 +0100251 /* free the indexes */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 if (t->depth >= 2)
253 vfree(t->index[t->depth - 2]);
254
255 /* free the targets */
256 for (i = 0; i < t->num_targets; i++) {
257 struct dm_target *tgt = t->targets + i;
258
259 if (tgt->type->dtr)
260 tgt->type->dtr(tgt);
261
262 dm_put_target_type(tgt->type);
263 }
264
265 vfree(t->highs);
266
267 /* free the device list */
Jonthan Brassow1b6da752009-06-22 10:12:29 +0100268 if (t->devices.next != &t->devices)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 free_devices(&t->devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100271 dm_free_md_mempools(t->mempools);
272
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 kfree(t);
274}
275
276void dm_table_get(struct dm_table *t)
277{
278 atomic_inc(&t->holders);
279}
280
281void dm_table_put(struct dm_table *t)
282{
283 if (!t)
284 return;
285
Mikulas Patockad5816872009-01-06 03:05:10 +0000286 smp_mb__before_atomic_dec();
287 atomic_dec(&t->holders);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288}
289
290/*
291 * Checks to see if we need to extend highs or targets.
292 */
293static inline int check_space(struct dm_table *t)
294{
295 if (t->num_targets >= t->num_allocated)
296 return alloc_targets(t, t->num_allocated * 2);
297
298 return 0;
299}
300
301/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 * See if we've already got a device in the list.
303 */
Mikulas Patocka82b15192008-10-10 13:37:09 +0100304static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305{
Mikulas Patocka82b15192008-10-10 13:37:09 +0100306 struct dm_dev_internal *dd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307
308 list_for_each_entry (dd, l, list)
Mikulas Patocka82b15192008-10-10 13:37:09 +0100309 if (dd->dm_dev.bdev->bd_dev == dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 return dd;
311
312 return NULL;
313}
314
315/*
316 * Open a device so we can use it as a map destination.
317 */
Mikulas Patocka82b15192008-10-10 13:37:09 +0100318static int open_dev(struct dm_dev_internal *d, dev_t dev,
319 struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320{
321 static char *_claim_ptr = "I belong to device-mapper";
322 struct block_device *bdev;
323
324 int r;
325
Mikulas Patocka82b15192008-10-10 13:37:09 +0100326 BUG_ON(d->dm_dev.bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327
Mikulas Patocka82b15192008-10-10 13:37:09 +0100328 bdev = open_by_devnum(dev, d->dm_dev.mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 if (IS_ERR(bdev))
330 return PTR_ERR(bdev);
Jun'ichi Nomuraf1659212006-03-27 01:17:59 -0800331 r = bd_claim_by_disk(bdev, _claim_ptr, dm_disk(md));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 if (r)
Al Viro9a1c3542008-02-22 20:40:24 -0500333 blkdev_put(bdev, d->dm_dev.mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 else
Mikulas Patocka82b15192008-10-10 13:37:09 +0100335 d->dm_dev.bdev = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 return r;
337}
338
339/*
340 * Close a device that we've been using.
341 */
Mikulas Patocka82b15192008-10-10 13:37:09 +0100342static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343{
Mikulas Patocka82b15192008-10-10 13:37:09 +0100344 if (!d->dm_dev.bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 return;
346
Mikulas Patocka82b15192008-10-10 13:37:09 +0100347 bd_release_from_disk(d->dm_dev.bdev, dm_disk(md));
Al Viro9a1c3542008-02-22 20:40:24 -0500348 blkdev_put(d->dm_dev.bdev, d->dm_dev.mode);
Mikulas Patocka82b15192008-10-10 13:37:09 +0100349 d->dm_dev.bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350}
351
352/*
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100353 * If possible, this checks an area of a destination device is invalid.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 */
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100355static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
356 sector_t start, sector_t len, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357{
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100358 struct queue_limits *limits = data;
359 struct block_device *bdev = dev->bdev;
360 sector_t dev_size =
361 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100362 unsigned short logical_block_size_sectors =
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100363 limits->logical_block_size >> SECTOR_SHIFT;
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100364 char b[BDEVNAME_SIZE];
Mike Anderson2cd54d92007-05-09 02:32:57 -0700365
366 if (!dev_size)
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100367 return 0;
Mike Anderson2cd54d92007-05-09 02:32:57 -0700368
Mike Snitzer5dea2712009-07-23 20:30:42 +0100369 if ((start >= dev_size) || (start + len > dev_size)) {
Mike Snitzera963a952009-09-04 20:40:24 +0100370 DMWARN("%s: %s too small for target: "
371 "start=%llu, len=%llu, dev_size=%llu",
372 dm_device_name(ti->table->md), bdevname(bdev, b),
373 (unsigned long long)start,
374 (unsigned long long)len,
375 (unsigned long long)dev_size);
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100376 return 1;
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100377 }
378
379 if (logical_block_size_sectors <= 1)
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100380 return 0;
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100381
382 if (start & (logical_block_size_sectors - 1)) {
383 DMWARN("%s: start=%llu not aligned to h/w "
Mike Snitzera963a952009-09-04 20:40:24 +0100384 "logical block size %u of %s",
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100385 dm_device_name(ti->table->md),
386 (unsigned long long)start,
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100387 limits->logical_block_size, bdevname(bdev, b));
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100388 return 1;
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100389 }
390
Mike Snitzer5dea2712009-07-23 20:30:42 +0100391 if (len & (logical_block_size_sectors - 1)) {
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100392 DMWARN("%s: len=%llu not aligned to h/w "
Mike Snitzera963a952009-09-04 20:40:24 +0100393 "logical block size %u of %s",
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100394 dm_device_name(ti->table->md),
Mike Snitzer5dea2712009-07-23 20:30:42 +0100395 (unsigned long long)len,
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100396 limits->logical_block_size, bdevname(bdev, b));
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100397 return 1;
Mike Snitzer02acc3a2009-06-22 10:12:30 +0100398 }
399
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +0100400 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401}
402
403/*
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100404 * This upgrades the mode on an already open dm_dev, being
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 * careful to leave things as they were if we fail to reopen the
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100406 * device and not to touch the existing bdev field in case
407 * it is accessed concurrently inside dm_table_any_congested().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 */
Al Viroaeb5d722008-09-02 15:28:45 -0400409static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
Mikulas Patocka82b15192008-10-10 13:37:09 +0100410 struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411{
412 int r;
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100413 struct dm_dev_internal dd_new, dd_old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100415 dd_new = dd_old = *dd;
416
417 dd_new.dm_dev.mode |= new_mode;
418 dd_new.dm_dev.bdev = NULL;
419
420 r = open_dev(&dd_new, dd->dm_dev.bdev->bd_dev, md);
421 if (r)
422 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423
Mikulas Patocka82b15192008-10-10 13:37:09 +0100424 dd->dm_dev.mode |= new_mode;
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100425 close_dev(&dd_old, md);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100427 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428}
429
430/*
431 * Add a device to the list, or just increment the usage count if
432 * it's already present.
433 */
434static int __table_get_device(struct dm_table *t, struct dm_target *ti,
Nikanth Karthikesan8215d6e2010-03-06 02:32:27 +0000435 const char *path, fmode_t mode, struct dm_dev **result)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436{
437 int r;
Andrew Morton69a2ce72008-02-08 02:10:14 +0000438 dev_t uninitialized_var(dev);
Mikulas Patocka82b15192008-10-10 13:37:09 +0100439 struct dm_dev_internal *dd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 unsigned int major, minor;
441
Eric Sesterhenn547bc922006-03-26 18:22:50 +0200442 BUG_ON(!t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443
444 if (sscanf(path, "%u:%u", &major, &minor) == 2) {
445 /* Extract the major/minor numbers */
446 dev = MKDEV(major, minor);
447 if (MAJOR(dev) != major || MINOR(dev) != minor)
448 return -EOVERFLOW;
449 } else {
450 /* convert the path to a device */
Christoph Hellwig72e82642008-08-11 00:24:08 +0200451 struct block_device *bdev = lookup_bdev(path);
452
453 if (IS_ERR(bdev))
454 return PTR_ERR(bdev);
455 dev = bdev->bd_dev;
456 bdput(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 }
458
459 dd = find_device(&t->devices, dev);
460 if (!dd) {
461 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
462 if (!dd)
463 return -ENOMEM;
464
Mikulas Patocka82b15192008-10-10 13:37:09 +0100465 dd->dm_dev.mode = mode;
466 dd->dm_dev.bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
Jun'ichi Nomuraf1659212006-03-27 01:17:59 -0800468 if ((r = open_dev(dd, dev, t->md))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 kfree(dd);
470 return r;
471 }
472
Mikulas Patocka82b15192008-10-10 13:37:09 +0100473 format_dev_t(dd->dm_dev.name, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474
475 atomic_set(&dd->count, 0);
476 list_add(&dd->list, &t->devices);
477
Mikulas Patocka82b15192008-10-10 13:37:09 +0100478 } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) {
Jun'ichi Nomuraf1659212006-03-27 01:17:59 -0800479 r = upgrade_mode(dd, mode, t->md);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 if (r)
481 return r;
482 }
483 atomic_inc(&dd->count);
484
Mikulas Patocka82b15192008-10-10 13:37:09 +0100485 *result = &dd->dm_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 return 0;
487}
488
Mike Snitzer5ab97582009-06-22 10:12:32 +0100489/*
490 * Returns the minimum that is _not_ zero, unless both are zero.
491 */
492#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
493
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100494int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
Mike Snitzer5dea2712009-07-23 20:30:42 +0100495 sector_t start, sector_t len, void *data)
Bryn Reeves3cb40212006-10-03 01:15:42 -0700496{
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100497 struct queue_limits *limits = data;
498 struct block_device *bdev = dev->bdev;
Jens Axboe165125e2007-07-24 09:28:11 +0200499 struct request_queue *q = bdev_get_queue(bdev);
Alasdair G Kergon0c2322e2008-10-10 13:37:13 +0100500 char b[BDEVNAME_SIZE];
501
502 if (unlikely(!q)) {
503 DMWARN("%s: Cannot set limits for nonexistent device %s",
504 dm_device_name(ti->table->md), bdevname(bdev, b));
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100505 return 0;
Alasdair G Kergon0c2322e2008-10-10 13:37:13 +0100506 }
Bryn Reeves3cb40212006-10-03 01:15:42 -0700507
Martin K. Petersenb27d7f12010-01-11 03:21:50 -0500508 if (bdev_stack_limits(limits, bdev, start) < 0)
509 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
Mike Snitzera963a952009-09-04 20:40:24 +0100510 "physical_block_size=%u, logical_block_size=%u, "
511 "alignment_offset=%u, start=%llu",
512 dm_device_name(ti->table->md), bdevname(bdev, b),
513 q->limits.physical_block_size,
514 q->limits.logical_block_size,
515 q->limits.alignment_offset,
Martin K. Petersenb27d7f12010-01-11 03:21:50 -0500516 (unsigned long long) start << SECTOR_SHIFT);
Bryn Reeves3cb40212006-10-03 01:15:42 -0700517
Milan Broz9980c632008-07-21 12:00:39 +0100518 /*
519 * Check if merge fn is supported.
520 * If not we'll force DM to use PAGE_SIZE or
521 * smaller I/O, just to be safe.
Bryn Reeves3cb40212006-10-03 01:15:42 -0700522 */
Milan Broz9980c632008-07-21 12:00:39 +0100523
524 if (q->merge_bvec_fn && !ti->type->merge)
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100525 limits->max_sectors =
526 min_not_zero(limits->max_sectors,
Bryn Reeves3cb40212006-10-03 01:15:42 -0700527 (unsigned int) (PAGE_SIZE >> 9));
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100528 return 0;
Bryn Reeves3cb40212006-10-03 01:15:42 -0700529}
530EXPORT_SYMBOL_GPL(dm_set_device_limits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531
Nikanth Karthikesan8215d6e2010-03-06 02:32:27 +0000532int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
533 struct dm_dev **result)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534{
Nikanth Karthikesan8215d6e2010-03-06 02:32:27 +0000535 return __table_get_device(ti->table, ti, path, mode, result);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536}
537
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100538
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539/*
540 * Decrement a devices use count and remove it if necessary.
541 */
Mikulas Patocka82b15192008-10-10 13:37:09 +0100542void dm_put_device(struct dm_target *ti, struct dm_dev *d)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543{
Mikulas Patocka82b15192008-10-10 13:37:09 +0100544 struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal,
545 dm_dev);
546
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 if (atomic_dec_and_test(&dd->count)) {
Jun'ichi Nomuraf1659212006-03-27 01:17:59 -0800548 close_dev(dd, ti->table->md);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 list_del(&dd->list);
550 kfree(dd);
551 }
552}
553
554/*
555 * Checks to see if the target joins onto the end of the table.
556 */
557static int adjoin(struct dm_table *table, struct dm_target *ti)
558{
559 struct dm_target *prev;
560
561 if (!table->num_targets)
562 return !ti->begin;
563
564 prev = &table->targets[table->num_targets - 1];
565 return (ti->begin == (prev->begin + prev->len));
566}
567
568/*
569 * Used to dynamically allocate the arg array.
570 */
571static char **realloc_argv(unsigned *array_size, char **old_argv)
572{
573 char **argv;
574 unsigned new_size;
575
576 new_size = *array_size ? *array_size * 2 : 64;
577 argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
578 if (argv) {
579 memcpy(argv, old_argv, *array_size * sizeof(*argv));
580 *array_size = new_size;
581 }
582
583 kfree(old_argv);
584 return argv;
585}
586
587/*
588 * Destructively splits up the argument list to pass to ctr.
589 */
590int dm_split_args(int *argc, char ***argvp, char *input)
591{
592 char *start, *end = input, *out, **argv = NULL;
593 unsigned array_size = 0;
594
595 *argc = 0;
David Teigland814d6862006-06-26 00:27:31 -0700596
597 if (!input) {
598 *argvp = NULL;
599 return 0;
600 }
601
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 argv = realloc_argv(&array_size, argv);
603 if (!argv)
604 return -ENOMEM;
605
606 while (1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 /* Skip whitespace */
André Goddard Rosae7d28602009-12-14 18:01:06 -0800608 start = skip_spaces(end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
610 if (!*start)
611 break; /* success, we hit the end */
612
613 /* 'out' is used to remove any back-quotes */
614 end = out = start;
615 while (*end) {
616 /* Everything apart from '\0' can be quoted */
617 if (*end == '\\' && *(end + 1)) {
618 *out++ = *(end + 1);
619 end += 2;
620 continue;
621 }
622
623 if (isspace(*end))
624 break; /* end of token */
625
626 *out++ = *end++;
627 }
628
629 /* have we already filled the array ? */
630 if ((*argc + 1) > array_size) {
631 argv = realloc_argv(&array_size, argv);
632 if (!argv)
633 return -ENOMEM;
634 }
635
636 /* we know this is whitespace */
637 if (*end)
638 end++;
639
640 /* terminate the string and put it in the array */
641 *out = '\0';
642 argv[*argc] = start;
643 (*argc)++;
644 }
645
646 *argvp = argv;
647 return 0;
648}
649
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100650/*
651 * Impose necessary and sufficient conditions on a devices's table such
652 * that any incoming bio which respects its logical_block_size can be
653 * processed successfully. If it falls across the boundary between
654 * two or more targets, the size of each piece it gets split into must
655 * be compatible with the logical_block_size of the target processing it.
656 */
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100657static int validate_hardware_logical_block_alignment(struct dm_table *table,
658 struct queue_limits *limits)
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100659{
660 /*
661 * This function uses arithmetic modulo the logical_block_size
662 * (in units of 512-byte sectors).
663 */
664 unsigned short device_logical_block_size_sects =
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100665 limits->logical_block_size >> SECTOR_SHIFT;
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100666
667 /*
668 * Offset of the start of the next table entry, mod logical_block_size.
669 */
670 unsigned short next_target_start = 0;
671
672 /*
673 * Given an aligned bio that extends beyond the end of a
674 * target, how many sectors must the next target handle?
675 */
676 unsigned short remaining = 0;
677
678 struct dm_target *uninitialized_var(ti);
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100679 struct queue_limits ti_limits;
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100680 unsigned i = 0;
681
682 /*
683 * Check each entry in the table in turn.
684 */
685 while (i < dm_table_get_num_targets(table)) {
686 ti = dm_table_get_target(table, i++);
687
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100688 blk_set_default_limits(&ti_limits);
689
690 /* combine all target devices' limits */
691 if (ti->type->iterate_devices)
692 ti->type->iterate_devices(ti, dm_set_device_limits,
693 &ti_limits);
694
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100695 /*
696 * If the remaining sectors fall entirely within this
697 * table entry are they compatible with its logical_block_size?
698 */
699 if (remaining < ti->len &&
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100700 remaining & ((ti_limits.logical_block_size >>
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100701 SECTOR_SHIFT) - 1))
702 break; /* Error */
703
704 next_target_start =
705 (unsigned short) ((next_target_start + ti->len) &
706 (device_logical_block_size_sects - 1));
707 remaining = next_target_start ?
708 device_logical_block_size_sects - next_target_start : 0;
709 }
710
711 if (remaining) {
712 DMWARN("%s: table line %u (start sect %llu len %llu) "
Mike Snitzera963a952009-09-04 20:40:24 +0100713 "not aligned to h/w logical block size %u",
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100714 dm_device_name(table->md), i,
715 (unsigned long long) ti->begin,
716 (unsigned long long) ti->len,
Mike Snitzer754c5fc2009-06-22 10:12:34 +0100717 limits->logical_block_size);
Mike Snitzerbe6d4302009-06-22 10:12:31 +0100718 return -EINVAL;
719 }
720
721 return 0;
722}
723
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724int dm_table_add_target(struct dm_table *t, const char *type,
725 sector_t start, sector_t len, char *params)
726{
727 int r = -EINVAL, argc;
728 char **argv;
729 struct dm_target *tgt;
730
731 if ((r = check_space(t)))
732 return r;
733
734 tgt = t->targets + t->num_targets;
735 memset(tgt, 0, sizeof(*tgt));
736
737 if (!len) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700738 DMERR("%s: zero-length target", dm_device_name(t->md));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 return -EINVAL;
740 }
741
742 tgt->type = dm_get_target_type(type);
743 if (!tgt->type) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700744 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
745 type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 return -EINVAL;
747 }
748
749 tgt->table = t;
750 tgt->begin = start;
751 tgt->len = len;
752 tgt->error = "Unknown error";
753
754 /*
755 * Does this target adjoin the previous one ?
756 */
757 if (!adjoin(t, tgt)) {
758 tgt->error = "Gap in table";
759 r = -EINVAL;
760 goto bad;
761 }
762
763 r = dm_split_args(&argc, &argv, params);
764 if (r) {
765 tgt->error = "couldn't split parameters (insufficient memory)";
766 goto bad;
767 }
768
769 r = tgt->type->ctr(tgt, argc, argv);
770 kfree(argv);
771 if (r)
772 goto bad;
773
774 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
775
Mike Snitzer5ae89a82010-08-12 04:14:08 +0100776 if (!tgt->num_discard_requests)
777 t->discards_supported = 0;
778
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 return 0;
780
781 bad:
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700782 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 dm_put_target_type(tgt->type);
784 return r;
785}
786
Will Drewry26803b92010-08-12 04:14:03 +0100787static int dm_table_set_type(struct dm_table *t)
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +0100788{
789 unsigned i;
790 unsigned bio_based = 0, request_based = 0;
791 struct dm_target *tgt;
792 struct dm_dev_internal *dd;
793 struct list_head *devices;
794
795 for (i = 0; i < t->num_targets; i++) {
796 tgt = t->targets + i;
797 if (dm_target_request_based(tgt))
798 request_based = 1;
799 else
800 bio_based = 1;
801
802 if (bio_based && request_based) {
803 DMWARN("Inconsistent table: different target types"
804 " can't be mixed up");
805 return -EINVAL;
806 }
807 }
808
809 if (bio_based) {
810 /* We must use this table as bio-based */
811 t->type = DM_TYPE_BIO_BASED;
812 return 0;
813 }
814
815 BUG_ON(!request_based); /* No targets in this table */
816
817 /* Non-request-stackable devices can't be used for request-based dm */
818 devices = dm_table_get_devices(t);
819 list_for_each_entry(dd, devices, list) {
820 if (!blk_queue_stackable(bdev_get_queue(dd->dm_dev.bdev))) {
821 DMWARN("table load rejected: including"
822 " non-request-stackable devices");
823 return -EINVAL;
824 }
825 }
826
827 /*
828 * Request-based dm supports only tables that have a single target now.
829 * To support multiple targets, request splitting support is needed,
830 * and that needs lots of changes in the block-layer.
831 * (e.g. request completion process for partial completion.)
832 */
833 if (t->num_targets > 1) {
834 DMWARN("Request-based dm doesn't support multiple targets yet");
835 return -EINVAL;
836 }
837
838 t->type = DM_TYPE_REQUEST_BASED;
839
840 return 0;
841}
842
843unsigned dm_table_get_type(struct dm_table *t)
844{
845 return t->type;
846}
847
848bool dm_table_request_based(struct dm_table *t)
849{
850 return dm_table_get_type(t) == DM_TYPE_REQUEST_BASED;
851}
852
853int dm_table_alloc_md_mempools(struct dm_table *t)
854{
855 unsigned type = dm_table_get_type(t);
856
857 if (unlikely(type == DM_TYPE_NONE)) {
858 DMWARN("no table type is set, can't allocate mempools");
859 return -EINVAL;
860 }
861
862 t->mempools = dm_alloc_md_mempools(type);
863 if (!t->mempools)
864 return -ENOMEM;
865
866 return 0;
867}
868
869void dm_table_free_md_mempools(struct dm_table *t)
870{
871 dm_free_md_mempools(t->mempools);
872 t->mempools = NULL;
873}
874
875struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
876{
877 return t->mempools;
878}
879
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880static int setup_indexes(struct dm_table *t)
881{
882 int i;
883 unsigned int total = 0;
884 sector_t *indexes;
885
886 /* allocate the space for *all* the indexes */
887 for (i = t->depth - 2; i >= 0; i--) {
888 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
889 total += t->counts[i];
890 }
891
892 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
893 if (!indexes)
894 return -ENOMEM;
895
896 /* set up internal nodes, bottom-up */
Jun'ichi Nomura82d601d2008-02-08 02:10:04 +0000897 for (i = t->depth - 2; i >= 0; i--) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 t->index[i] = indexes;
899 indexes += (KEYS_PER_NODE * t->counts[i]);
900 setup_btree_index(i, t);
901 }
902
903 return 0;
904}
905
906/*
907 * Builds the btree to index the map.
908 */
Will Drewry26803b92010-08-12 04:14:03 +0100909static int dm_table_build_index(struct dm_table *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910{
911 int r = 0;
912 unsigned int leaf_nodes;
913
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 /* how many indexes will the btree have ? */
915 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
916 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
917
918 /* leaf layer has already been set up */
919 t->counts[t->depth - 1] = leaf_nodes;
920 t->index[t->depth - 1] = t->highs;
921
922 if (t->depth >= 2)
923 r = setup_indexes(t);
924
925 return r;
926}
927
Will Drewry26803b92010-08-12 04:14:03 +0100928/*
929 * Register the mapped device for blk_integrity support if
930 * the underlying devices support it.
931 */
932static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md)
933{
934 struct list_head *devices = dm_table_get_devices(t);
935 struct dm_dev_internal *dd;
936
937 list_for_each_entry(dd, devices, list)
938 if (bdev_get_integrity(dd->dm_dev.bdev))
939 return blk_integrity_register(dm_disk(md), NULL);
940
941 return 0;
942}
943
944/*
945 * Prepares the table for use by building the indices,
946 * setting the type, and allocating mempools.
947 */
948int dm_table_complete(struct dm_table *t)
949{
950 int r;
951
952 r = dm_table_set_type(t);
953 if (r) {
954 DMERR("unable to set table type");
955 return r;
956 }
957
958 r = dm_table_build_index(t);
959 if (r) {
960 DMERR("unable to build btrees");
961 return r;
962 }
963
964 r = dm_table_prealloc_integrity(t, t->md);
965 if (r) {
966 DMERR("could not register integrity profile.");
967 return r;
968 }
969
970 r = dm_table_alloc_md_mempools(t);
971 if (r)
972 DMERR("unable to allocate mempools");
973
974 return r;
975}
976
Arjan van de Ven48c9c272006-03-27 01:18:20 -0800977static DEFINE_MUTEX(_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978void dm_table_event_callback(struct dm_table *t,
979 void (*fn)(void *), void *context)
980{
Arjan van de Ven48c9c272006-03-27 01:18:20 -0800981 mutex_lock(&_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 t->event_fn = fn;
983 t->event_context = context;
Arjan van de Ven48c9c272006-03-27 01:18:20 -0800984 mutex_unlock(&_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985}
986
987void dm_table_event(struct dm_table *t)
988{
989 /*
990 * You can no longer call dm_table_event() from interrupt
991 * context, use a bottom half instead.
992 */
993 BUG_ON(in_interrupt());
994
Arjan van de Ven48c9c272006-03-27 01:18:20 -0800995 mutex_lock(&_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 if (t->event_fn)
997 t->event_fn(t->event_context);
Arjan van de Ven48c9c272006-03-27 01:18:20 -0800998 mutex_unlock(&_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999}
1000
1001sector_t dm_table_get_size(struct dm_table *t)
1002{
1003 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1004}
1005
1006struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1007{
Milan Broz14353532006-06-26 00:27:27 -07001008 if (index >= t->num_targets)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 return NULL;
1010
1011 return t->targets + index;
1012}
1013
1014/*
1015 * Search the btree for the correct target.
Jun'ichi Nomura512875b2007-12-13 14:15:25 +00001016 *
1017 * Caller should check returned pointer with dm_target_is_valid()
1018 * to trap I/O beyond end of device.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 */
1020struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1021{
1022 unsigned int l, n = 0, k = 0;
1023 sector_t *node;
1024
1025 for (l = 0; l < t->depth; l++) {
1026 n = get_child(n, k);
1027 node = get_node(t, l, n);
1028
1029 for (k = 0; k < KEYS_PER_NODE; k++)
1030 if (node[k] >= sector)
1031 break;
1032 }
1033
1034 return &t->targets[(KEYS_PER_NODE * n) + k];
1035}
1036
Martin K. Petersen9c470082009-04-09 00:27:12 +01001037/*
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001038 * Establish the new table's queue_limits and validate them.
1039 */
1040int dm_calculate_queue_limits(struct dm_table *table,
1041 struct queue_limits *limits)
1042{
1043 struct dm_target *uninitialized_var(ti);
1044 struct queue_limits ti_limits;
1045 unsigned i = 0;
1046
1047 blk_set_default_limits(limits);
1048
1049 while (i < dm_table_get_num_targets(table)) {
1050 blk_set_default_limits(&ti_limits);
1051
1052 ti = dm_table_get_target(table, i++);
1053
1054 if (!ti->type->iterate_devices)
1055 goto combine_limits;
1056
1057 /*
1058 * Combine queue limits of all the devices this target uses.
1059 */
1060 ti->type->iterate_devices(ti, dm_set_device_limits,
1061 &ti_limits);
1062
Mike Snitzer40bea432009-09-04 20:40:25 +01001063 /* Set I/O hints portion of queue limits */
1064 if (ti->type->io_hints)
1065 ti->type->io_hints(ti, &ti_limits);
1066
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001067 /*
1068 * Check each device area is consistent with the target's
1069 * overall queue limits.
1070 */
Mikulas Patockaf6a1ed12009-09-04 20:40:22 +01001071 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1072 &ti_limits))
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001073 return -EINVAL;
1074
1075combine_limits:
1076 /*
1077 * Merge this target's queue limits into the overall limits
1078 * for the table.
1079 */
1080 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
Martin K. Petersenb27d7f12010-01-11 03:21:50 -05001081 DMWARN("%s: adding target device "
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001082 "(start sect %llu len %llu) "
Martin K. Petersenb27d7f12010-01-11 03:21:50 -05001083 "caused an alignment inconsistency",
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001084 dm_device_name(table->md),
1085 (unsigned long long) ti->begin,
1086 (unsigned long long) ti->len);
1087 }
1088
1089 return validate_hardware_logical_block_alignment(table, limits);
1090}
1091
1092/*
Martin K. Petersen9c470082009-04-09 00:27:12 +01001093 * Set the integrity profile for this device if all devices used have
1094 * matching profiles.
1095 */
1096static void dm_table_set_integrity(struct dm_table *t)
1097{
1098 struct list_head *devices = dm_table_get_devices(t);
1099 struct dm_dev_internal *prev = NULL, *dd = NULL;
1100
1101 if (!blk_get_integrity(dm_disk(t->md)))
1102 return;
1103
1104 list_for_each_entry(dd, devices, list) {
1105 if (prev &&
1106 blk_integrity_compare(prev->dm_dev.bdev->bd_disk,
1107 dd->dm_dev.bdev->bd_disk) < 0) {
1108 DMWARN("%s: integrity not set: %s and %s mismatch",
1109 dm_device_name(t->md),
1110 prev->dm_dev.bdev->bd_disk->disk_name,
1111 dd->dm_dev.bdev->bd_disk->disk_name);
1112 goto no_integrity;
1113 }
1114 prev = dd;
1115 }
1116
1117 if (!prev || !bdev_get_integrity(prev->dm_dev.bdev))
1118 goto no_integrity;
1119
1120 blk_integrity_register(dm_disk(t->md),
1121 bdev_get_integrity(prev->dm_dev.bdev));
1122
1123 return;
1124
1125no_integrity:
1126 blk_integrity_register(dm_disk(t->md), NULL);
1127
1128 return;
1129}
1130
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001131void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1132 struct queue_limits *limits)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133{
1134 /*
Mike Snitzer11977642009-06-22 10:12:32 +01001135 * Copy table's limits to the DM device's request_queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 */
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001137 q->limits = *limits;
Jens Axboec9a3f6d2008-04-29 19:12:35 +02001138
Mike Snitzer754c5fc2009-06-22 10:12:34 +01001139 if (limits->no_cluster)
Jens Axboec9a3f6d2008-04-29 19:12:35 +02001140 queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q);
NeilBrown969429b2006-03-27 01:17:49 -08001141 else
Jens Axboec9a3f6d2008-04-29 19:12:35 +02001142 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, q);
NeilBrown969429b2006-03-27 01:17:49 -08001143
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001144 if (!dm_table_supports_discards(t))
1145 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1146 else
1147 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1148
Martin K. Petersen9c470082009-04-09 00:27:12 +01001149 dm_table_set_integrity(t);
Kiyoshi Uedae6ee8c02009-06-22 10:12:36 +01001150
1151 /*
1152 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1153 * visible to other CPUs because, once the flag is set, incoming bios
1154 * are processed by request-based dm, which refers to the queue
1155 * settings.
1156 * Until the flag set, bios are passed to bio-based dm and queued to
1157 * md->deferred where queue settings are not needed yet.
1158 * Those bios are passed to request-based dm at the resume time.
1159 */
1160 smp_mb();
1161 if (dm_table_request_based(t))
1162 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163}
1164
1165unsigned int dm_table_get_num_targets(struct dm_table *t)
1166{
1167 return t->num_targets;
1168}
1169
1170struct list_head *dm_table_get_devices(struct dm_table *t)
1171{
1172 return &t->devices;
1173}
1174
Al Viroaeb5d722008-09-02 15:28:45 -04001175fmode_t dm_table_get_mode(struct dm_table *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176{
1177 return t->mode;
1178}
1179
1180static void suspend_targets(struct dm_table *t, unsigned postsuspend)
1181{
1182 int i = t->num_targets;
1183 struct dm_target *ti = t->targets;
1184
1185 while (i--) {
1186 if (postsuspend) {
1187 if (ti->type->postsuspend)
1188 ti->type->postsuspend(ti);
1189 } else if (ti->type->presuspend)
1190 ti->type->presuspend(ti);
1191
1192 ti++;
1193 }
1194}
1195
1196void dm_table_presuspend_targets(struct dm_table *t)
1197{
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001198 if (!t)
1199 return;
1200
Adrian Bunke8488d02008-04-24 22:10:51 +01001201 suspend_targets(t, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202}
1203
1204void dm_table_postsuspend_targets(struct dm_table *t)
1205{
Alasdair G Kergoncf222b32005-07-28 21:15:57 -07001206 if (!t)
1207 return;
1208
Adrian Bunke8488d02008-04-24 22:10:51 +01001209 suspend_targets(t, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210}
1211
Milan Broz8757b772006-10-03 01:15:36 -07001212int dm_table_resume_targets(struct dm_table *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213{
Milan Broz8757b772006-10-03 01:15:36 -07001214 int i, r = 0;
1215
1216 for (i = 0; i < t->num_targets; i++) {
1217 struct dm_target *ti = t->targets + i;
1218
1219 if (!ti->type->preresume)
1220 continue;
1221
1222 r = ti->type->preresume(ti);
1223 if (r)
1224 return r;
1225 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226
1227 for (i = 0; i < t->num_targets; i++) {
1228 struct dm_target *ti = t->targets + i;
1229
1230 if (ti->type->resume)
1231 ti->type->resume(ti);
1232 }
Milan Broz8757b772006-10-03 01:15:36 -07001233
1234 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235}
1236
1237int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1238{
Mikulas Patocka82b15192008-10-10 13:37:09 +01001239 struct dm_dev_internal *dd;
Paul Jimenezafb24522008-02-08 02:09:59 +00001240 struct list_head *devices = dm_table_get_devices(t);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 int r = 0;
1242
Paul Jimenezafb24522008-02-08 02:09:59 +00001243 list_for_each_entry(dd, devices, list) {
Mikulas Patocka82b15192008-10-10 13:37:09 +01001244 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
Alasdair G Kergon0c2322e2008-10-10 13:37:13 +01001245 char b[BDEVNAME_SIZE];
1246
1247 if (likely(q))
1248 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1249 else
1250 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1251 dm_device_name(t->md),
1252 bdevname(dd->dm_dev.bdev, b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 }
1254
1255 return r;
1256}
1257
Kiyoshi Uedacec47e32009-06-22 10:12:35 +01001258int dm_table_any_busy_target(struct dm_table *t)
1259{
1260 unsigned i;
1261 struct dm_target *ti;
1262
1263 for (i = 0; i < t->num_targets; i++) {
1264 ti = t->targets + i;
1265 if (ti->type->busy && ti->type->busy(ti))
1266 return 1;
1267 }
1268
1269 return 0;
1270}
1271
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272void dm_table_unplug_all(struct dm_table *t)
1273{
Mikulas Patocka82b15192008-10-10 13:37:09 +01001274 struct dm_dev_internal *dd;
Paul Jimenezafb24522008-02-08 02:09:59 +00001275 struct list_head *devices = dm_table_get_devices(t);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276
Paul Jimenezafb24522008-02-08 02:09:59 +00001277 list_for_each_entry(dd, devices, list) {
Mikulas Patocka82b15192008-10-10 13:37:09 +01001278 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
Alasdair G Kergon0c2322e2008-10-10 13:37:13 +01001279 char b[BDEVNAME_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280
Alasdair G Kergon0c2322e2008-10-10 13:37:13 +01001281 if (likely(q))
1282 blk_unplug(q);
1283 else
1284 DMWARN_LIMIT("%s: Cannot unplug nonexistent device %s",
1285 dm_device_name(t->md),
1286 bdevname(dd->dm_dev.bdev, b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 }
1288}
1289
Mike Anderson1134e5a2006-03-27 01:17:54 -08001290struct mapped_device *dm_table_get_md(struct dm_table *t)
1291{
Mike Anderson1134e5a2006-03-27 01:17:54 -08001292 return t->md;
1293}
1294
Mike Snitzer5ae89a82010-08-12 04:14:08 +01001295static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1296 sector_t start, sector_t len, void *data)
1297{
1298 struct request_queue *q = bdev_get_queue(dev->bdev);
1299
1300 return q && blk_queue_discard(q);
1301}
1302
1303bool dm_table_supports_discards(struct dm_table *t)
1304{
1305 struct dm_target *ti;
1306 unsigned i = 0;
1307
1308 if (!t->discards_supported)
1309 return 0;
1310
1311 /*
1312 * Ensure that at least one underlying device supports discards.
1313 * t->devices includes internal dm devices such as mirror logs
1314 * so we need to use iterate_devices here, which targets
1315 * supporting discard must provide.
1316 */
1317 while (i < dm_table_get_num_targets(t)) {
1318 ti = dm_table_get_target(t, i++);
1319
1320 if (ti->type->iterate_devices &&
1321 ti->type->iterate_devices(ti, device_discard_capable, NULL))
1322 return 1;
1323 }
1324
1325 return 0;
1326}
1327
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328EXPORT_SYMBOL(dm_vcalloc);
1329EXPORT_SYMBOL(dm_get_device);
1330EXPORT_SYMBOL(dm_put_device);
1331EXPORT_SYMBOL(dm_table_event);
Alasdair G Kergond5e404c2005-07-12 15:53:05 -07001332EXPORT_SYMBOL(dm_table_get_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333EXPORT_SYMBOL(dm_table_get_mode);
Mike Anderson1134e5a2006-03-27 01:17:54 -08001334EXPORT_SYMBOL(dm_table_get_md);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335EXPORT_SYMBOL(dm_table_put);
1336EXPORT_SYMBOL(dm_table_get);
1337EXPORT_SYMBOL(dm_table_unplug_all);