Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001 Sistina Software (UK) Limited. | |
d5816876 | 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
1da177e4 LT |
4 | * |
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> | |
e7d2860b | 15 | #include <linux/string.h> |
1da177e4 LT |
16 | #include <linux/slab.h> |
17 | #include <linux/interrupt.h> | |
48c9c27b | 18 | #include <linux/mutex.h> |
d5816876 | 19 | #include <linux/delay.h> |
60063497 | 20 | #include <linux/atomic.h> |
1da177e4 | 21 | |
72d94861 AK |
22 | #define DM_MSG_PREFIX "table" |
23 | ||
1da177e4 LT |
24 | #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 | ||
d5816876 MP |
29 | /* |
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 | ||
1da177e4 | 42 | struct dm_table { |
1134e5ae | 43 | struct mapped_device *md; |
1da177e4 | 44 | atomic_t holders; |
e6ee8c0b | 45 | unsigned type; |
1da177e4 LT |
46 | |
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 | ||
a91a2785 | 57 | unsigned integrity_supported:1; |
3791e2fc | 58 | unsigned singleton:1; |
5ae89a87 | 59 | |
1da177e4 LT |
60 | /* |
61 | * Indicates the rw permissions for the new logical | |
62 | * device. This should be a combination of FMODE_READ | |
63 | * and FMODE_WRITE. | |
64 | */ | |
aeb5d727 | 65 | fmode_t mode; |
1da177e4 LT |
66 | |
67 | /* a list of devices used by this table */ | |
68 | struct list_head devices; | |
69 | ||
1da177e4 LT |
70 | /* events get handed up using this callback */ |
71 | void (*event_fn)(void *); | |
72 | void *event_context; | |
e6ee8c0b KU |
73 | |
74 | struct dm_md_mempools *mempools; | |
9d357b07 N |
75 | |
76 | struct list_head target_callbacks; | |
1da177e4 LT |
77 | }; |
78 | ||
79 | /* | |
80 | * Similar to ceiling(log_size(n)) | |
81 | */ | |
82 | static unsigned int int_log(unsigned int n, unsigned int base) | |
83 | { | |
84 | int result = 0; | |
85 | ||
86 | while (n > 1) { | |
87 | n = dm_div_up(n, base); | |
88 | result++; | |
89 | } | |
90 | ||
91 | return result; | |
92 | } | |
93 | ||
1da177e4 LT |
94 | /* |
95 | * Calculate the index of the child node of the n'th node k'th key. | |
96 | */ | |
97 | static inline unsigned int get_child(unsigned int n, unsigned int k) | |
98 | { | |
99 | return (n * CHILDREN_PER_NODE) + k; | |
100 | } | |
101 | ||
102 | /* | |
103 | * Return the n'th node of level l from table t. | |
104 | */ | |
105 | static inline sector_t *get_node(struct dm_table *t, | |
106 | unsigned int l, unsigned int n) | |
107 | { | |
108 | return t->index[l] + (n * KEYS_PER_NODE); | |
109 | } | |
110 | ||
111 | /* | |
112 | * Return the highest key that you could lookup from the n'th | |
113 | * node on level l of the btree. | |
114 | */ | |
115 | static sector_t high(struct dm_table *t, unsigned int l, unsigned int n) | |
116 | { | |
117 | for (; l < t->depth - 1; l++) | |
118 | n = get_child(n, CHILDREN_PER_NODE - 1); | |
119 | ||
120 | if (n >= t->counts[l]) | |
121 | return (sector_t) - 1; | |
122 | ||
123 | return get_node(t, l, n)[KEYS_PER_NODE - 1]; | |
124 | } | |
125 | ||
126 | /* | |
127 | * Fills in a level of the btree based on the highs of the level | |
128 | * below it. | |
129 | */ | |
130 | static int setup_btree_index(unsigned int l, struct dm_table *t) | |
131 | { | |
132 | unsigned int n, k; | |
133 | sector_t *node; | |
134 | ||
135 | for (n = 0U; n < t->counts[l]; n++) { | |
136 | node = get_node(t, l, n); | |
137 | ||
138 | for (k = 0U; k < KEYS_PER_NODE; k++) | |
139 | node[k] = high(t, l + 1, get_child(n, k)); | |
140 | } | |
141 | ||
142 | return 0; | |
143 | } | |
144 | ||
145 | void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size) | |
146 | { | |
147 | unsigned long size; | |
148 | void *addr; | |
149 | ||
150 | /* | |
151 | * Check that we're not going to overflow. | |
152 | */ | |
153 | if (nmemb > (ULONG_MAX / elem_size)) | |
154 | return NULL; | |
155 | ||
156 | size = nmemb * elem_size; | |
e29e65aa | 157 | addr = vzalloc(size); |
1da177e4 LT |
158 | |
159 | return addr; | |
160 | } | |
08649012 | 161 | EXPORT_SYMBOL(dm_vcalloc); |
1da177e4 LT |
162 | |
163 | /* | |
164 | * highs, and targets are managed as dynamic arrays during a | |
165 | * table load. | |
166 | */ | |
167 | static int alloc_targets(struct dm_table *t, unsigned int num) | |
168 | { | |
169 | sector_t *n_highs; | |
170 | struct dm_target *n_targets; | |
171 | int n = t->num_targets; | |
172 | ||
173 | /* | |
174 | * Allocate both the target array and offset array at once. | |
512875bd JN |
175 | * Append an empty entry to catch sectors beyond the end of |
176 | * the device. | |
1da177e4 | 177 | */ |
512875bd | 178 | n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) + |
1da177e4 LT |
179 | sizeof(sector_t)); |
180 | if (!n_highs) | |
181 | return -ENOMEM; | |
182 | ||
183 | n_targets = (struct dm_target *) (n_highs + num); | |
184 | ||
185 | if (n) { | |
186 | memcpy(n_highs, t->highs, sizeof(*n_highs) * n); | |
187 | memcpy(n_targets, t->targets, sizeof(*n_targets) * n); | |
188 | } | |
189 | ||
190 | memset(n_highs + n, -1, sizeof(*n_highs) * (num - n)); | |
191 | vfree(t->highs); | |
192 | ||
193 | t->num_allocated = num; | |
194 | t->highs = n_highs; | |
195 | t->targets = n_targets; | |
196 | ||
197 | return 0; | |
198 | } | |
199 | ||
aeb5d727 | 200 | int dm_table_create(struct dm_table **result, fmode_t mode, |
1134e5ae | 201 | unsigned num_targets, struct mapped_device *md) |
1da177e4 | 202 | { |
094262db | 203 | struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL); |
1da177e4 LT |
204 | |
205 | if (!t) | |
206 | return -ENOMEM; | |
207 | ||
1da177e4 | 208 | INIT_LIST_HEAD(&t->devices); |
9d357b07 | 209 | INIT_LIST_HEAD(&t->target_callbacks); |
d5816876 | 210 | atomic_set(&t->holders, 0); |
1da177e4 LT |
211 | |
212 | if (!num_targets) | |
213 | num_targets = KEYS_PER_NODE; | |
214 | ||
215 | num_targets = dm_round_up(num_targets, KEYS_PER_NODE); | |
216 | ||
217 | if (alloc_targets(t, num_targets)) { | |
218 | kfree(t); | |
219 | t = NULL; | |
220 | return -ENOMEM; | |
221 | } | |
222 | ||
223 | t->mode = mode; | |
1134e5ae | 224 | t->md = md; |
1da177e4 LT |
225 | *result = t; |
226 | return 0; | |
227 | } | |
228 | ||
229 | static void free_devices(struct list_head *devices) | |
230 | { | |
231 | struct list_head *tmp, *next; | |
232 | ||
afb24528 | 233 | list_for_each_safe(tmp, next, devices) { |
82b1519b MP |
234 | struct dm_dev_internal *dd = |
235 | list_entry(tmp, struct dm_dev_internal, list); | |
1b6da754 JB |
236 | DMWARN("dm_table_destroy: dm_put_device call missing for %s", |
237 | dd->dm_dev.name); | |
1da177e4 LT |
238 | kfree(dd); |
239 | } | |
240 | } | |
241 | ||
d5816876 | 242 | void dm_table_destroy(struct dm_table *t) |
1da177e4 LT |
243 | { |
244 | unsigned int i; | |
245 | ||
a7940155 AK |
246 | if (!t) |
247 | return; | |
248 | ||
d5816876 MP |
249 | while (atomic_read(&t->holders)) |
250 | msleep(1); | |
251 | smp_mb(); | |
252 | ||
26803b9f | 253 | /* free the indexes */ |
1da177e4 LT |
254 | if (t->depth >= 2) |
255 | vfree(t->index[t->depth - 2]); | |
256 | ||
257 | /* free the targets */ | |
258 | for (i = 0; i < t->num_targets; i++) { | |
259 | struct dm_target *tgt = t->targets + i; | |
260 | ||
261 | if (tgt->type->dtr) | |
262 | tgt->type->dtr(tgt); | |
263 | ||
264 | dm_put_target_type(tgt->type); | |
265 | } | |
266 | ||
267 | vfree(t->highs); | |
268 | ||
269 | /* free the device list */ | |
1b6da754 | 270 | if (t->devices.next != &t->devices) |
1da177e4 | 271 | free_devices(&t->devices); |
1da177e4 | 272 | |
e6ee8c0b KU |
273 | dm_free_md_mempools(t->mempools); |
274 | ||
1da177e4 LT |
275 | kfree(t); |
276 | } | |
277 | ||
278 | void dm_table_get(struct dm_table *t) | |
279 | { | |
280 | atomic_inc(&t->holders); | |
281 | } | |
08649012 | 282 | EXPORT_SYMBOL(dm_table_get); |
1da177e4 LT |
283 | |
284 | void dm_table_put(struct dm_table *t) | |
285 | { | |
286 | if (!t) | |
287 | return; | |
288 | ||
d5816876 MP |
289 | smp_mb__before_atomic_dec(); |
290 | atomic_dec(&t->holders); | |
1da177e4 | 291 | } |
08649012 | 292 | EXPORT_SYMBOL(dm_table_put); |
1da177e4 LT |
293 | |
294 | /* | |
295 | * Checks to see if we need to extend highs or targets. | |
296 | */ | |
297 | static 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 | ||
1da177e4 LT |
305 | /* |
306 | * See if we've already got a device in the list. | |
307 | */ | |
82b1519b | 308 | static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev) |
1da177e4 | 309 | { |
82b1519b | 310 | struct dm_dev_internal *dd; |
1da177e4 LT |
311 | |
312 | list_for_each_entry (dd, l, list) | |
82b1519b | 313 | if (dd->dm_dev.bdev->bd_dev == dev) |
1da177e4 LT |
314 | return dd; |
315 | ||
316 | return NULL; | |
317 | } | |
318 | ||
319 | /* | |
320 | * Open a device so we can use it as a map destination. | |
321 | */ | |
82b1519b MP |
322 | static int open_dev(struct dm_dev_internal *d, dev_t dev, |
323 | struct mapped_device *md) | |
1da177e4 LT |
324 | { |
325 | static char *_claim_ptr = "I belong to device-mapper"; | |
326 | struct block_device *bdev; | |
327 | ||
328 | int r; | |
329 | ||
82b1519b | 330 | BUG_ON(d->dm_dev.bdev); |
1da177e4 | 331 | |
d4d77629 | 332 | bdev = blkdev_get_by_dev(dev, d->dm_dev.mode | FMODE_EXCL, _claim_ptr); |
1da177e4 LT |
333 | if (IS_ERR(bdev)) |
334 | return PTR_ERR(bdev); | |
e09b457b | 335 | |
e09b457b TH |
336 | r = bd_link_disk_holder(bdev, dm_disk(md)); |
337 | if (r) { | |
e525fd89 | 338 | blkdev_put(bdev, d->dm_dev.mode | FMODE_EXCL); |
e09b457b TH |
339 | return r; |
340 | } | |
341 | ||
342 | d->dm_dev.bdev = bdev; | |
343 | return 0; | |
1da177e4 LT |
344 | } |
345 | ||
346 | /* | |
347 | * Close a device that we've been using. | |
348 | */ | |
82b1519b | 349 | static void close_dev(struct dm_dev_internal *d, struct mapped_device *md) |
1da177e4 | 350 | { |
82b1519b | 351 | if (!d->dm_dev.bdev) |
1da177e4 LT |
352 | return; |
353 | ||
49731baa | 354 | bd_unlink_disk_holder(d->dm_dev.bdev, dm_disk(md)); |
e525fd89 | 355 | blkdev_put(d->dm_dev.bdev, d->dm_dev.mode | FMODE_EXCL); |
82b1519b | 356 | d->dm_dev.bdev = NULL; |
1da177e4 LT |
357 | } |
358 | ||
359 | /* | |
f6a1ed10 | 360 | * If possible, this checks an area of a destination device is invalid. |
1da177e4 | 361 | */ |
f6a1ed10 MP |
362 | static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev, |
363 | sector_t start, sector_t len, void *data) | |
1da177e4 | 364 | { |
f4808ca9 | 365 | struct request_queue *q; |
754c5fc7 MS |
366 | 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; | |
02acc3a4 | 370 | unsigned short logical_block_size_sectors = |
754c5fc7 | 371 | limits->logical_block_size >> SECTOR_SHIFT; |
02acc3a4 | 372 | char b[BDEVNAME_SIZE]; |
2cd54d9b | 373 | |
f4808ca9 MB |
374 | /* |
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 | ||
2cd54d9b | 390 | if (!dev_size) |
f6a1ed10 | 391 | return 0; |
2cd54d9b | 392 | |
5dea271b | 393 | if ((start >= dev_size) || (start + len > dev_size)) { |
a963a956 MS |
394 | 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); | |
f6a1ed10 | 400 | return 1; |
02acc3a4 MS |
401 | } |
402 | ||
403 | if (logical_block_size_sectors <= 1) | |
f6a1ed10 | 404 | return 0; |
02acc3a4 MS |
405 | |
406 | if (start & (logical_block_size_sectors - 1)) { | |
407 | DMWARN("%s: start=%llu not aligned to h/w " | |
a963a956 | 408 | "logical block size %u of %s", |
02acc3a4 MS |
409 | dm_device_name(ti->table->md), |
410 | (unsigned long long)start, | |
754c5fc7 | 411 | limits->logical_block_size, bdevname(bdev, b)); |
f6a1ed10 | 412 | return 1; |
02acc3a4 MS |
413 | } |
414 | ||
5dea271b | 415 | if (len & (logical_block_size_sectors - 1)) { |
02acc3a4 | 416 | DMWARN("%s: len=%llu not aligned to h/w " |
a963a956 | 417 | "logical block size %u of %s", |
02acc3a4 | 418 | dm_device_name(ti->table->md), |
5dea271b | 419 | (unsigned long long)len, |
754c5fc7 | 420 | limits->logical_block_size, bdevname(bdev, b)); |
f6a1ed10 | 421 | return 1; |
02acc3a4 MS |
422 | } |
423 | ||
f6a1ed10 | 424 | return 0; |
1da177e4 LT |
425 | } |
426 | ||
427 | /* | |
570b9d96 | 428 | * This upgrades the mode on an already open dm_dev, being |
1da177e4 | 429 | * careful to leave things as they were if we fail to reopen the |
570b9d96 AK |
430 | * device and not to touch the existing bdev field in case |
431 | * it is accessed concurrently inside dm_table_any_congested(). | |
1da177e4 | 432 | */ |
aeb5d727 | 433 | static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode, |
82b1519b | 434 | struct mapped_device *md) |
1da177e4 LT |
435 | { |
436 | int r; | |
570b9d96 | 437 | struct dm_dev_internal dd_new, dd_old; |
1da177e4 | 438 | |
570b9d96 AK |
439 | 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; | |
1da177e4 | 447 | |
82b1519b | 448 | dd->dm_dev.mode |= new_mode; |
570b9d96 | 449 | close_dev(&dd_old, md); |
1da177e4 | 450 | |
570b9d96 | 451 | return 0; |
1da177e4 LT |
452 | } |
453 | ||
454 | /* | |
455 | * Add a device to the list, or just increment the usage count if | |
456 | * it's already present. | |
457 | */ | |
08649012 MS |
458 | int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode, |
459 | struct dm_dev **result) | |
1da177e4 LT |
460 | { |
461 | int r; | |
69a2ce72 | 462 | dev_t uninitialized_var(dev); |
82b1519b | 463 | struct dm_dev_internal *dd; |
1da177e4 | 464 | unsigned int major, minor; |
08649012 | 465 | struct dm_table *t = ti->table; |
1da177e4 | 466 | |
547bc926 | 467 | BUG_ON(!t); |
1da177e4 LT |
468 | |
469 | if (sscanf(path, "%u:%u", &major, &minor) == 2) { | |
470 | /* Extract the major/minor numbers */ | |
471 | dev = MKDEV(major, minor); | |
472 | if (MAJOR(dev) != major || MINOR(dev) != minor) | |
473 | return -EOVERFLOW; | |
474 | } else { | |
475 | /* convert the path to a device */ | |
72e8264e CH |
476 | struct block_device *bdev = lookup_bdev(path); |
477 | ||
478 | if (IS_ERR(bdev)) | |
479 | return PTR_ERR(bdev); | |
480 | dev = bdev->bd_dev; | |
481 | bdput(bdev); | |
1da177e4 LT |
482 | } |
483 | ||
484 | dd = find_device(&t->devices, dev); | |
485 | if (!dd) { | |
486 | dd = kmalloc(sizeof(*dd), GFP_KERNEL); | |
487 | if (!dd) | |
488 | return -ENOMEM; | |
489 | ||
82b1519b MP |
490 | dd->dm_dev.mode = mode; |
491 | dd->dm_dev.bdev = NULL; | |
1da177e4 | 492 | |
f165921d | 493 | if ((r = open_dev(dd, dev, t->md))) { |
1da177e4 LT |
494 | kfree(dd); |
495 | return r; | |
496 | } | |
497 | ||
82b1519b | 498 | format_dev_t(dd->dm_dev.name, dev); |
1da177e4 LT |
499 | |
500 | atomic_set(&dd->count, 0); | |
501 | list_add(&dd->list, &t->devices); | |
502 | ||
82b1519b | 503 | } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) { |
f165921d | 504 | r = upgrade_mode(dd, mode, t->md); |
1da177e4 LT |
505 | if (r) |
506 | return r; | |
507 | } | |
508 | atomic_inc(&dd->count); | |
509 | ||
82b1519b | 510 | *result = &dd->dm_dev; |
1da177e4 LT |
511 | return 0; |
512 | } | |
08649012 | 513 | EXPORT_SYMBOL(dm_get_device); |
1da177e4 | 514 | |
754c5fc7 | 515 | int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev, |
5dea271b | 516 | sector_t start, sector_t len, void *data) |
1da177e4 | 517 | { |
754c5fc7 MS |
518 | struct queue_limits *limits = data; |
519 | struct block_device *bdev = dev->bdev; | |
165125e1 | 520 | struct request_queue *q = bdev_get_queue(bdev); |
0c2322e4 AK |
521 | char b[BDEVNAME_SIZE]; |
522 | ||
523 | if (unlikely(!q)) { | |
524 | DMWARN("%s: Cannot set limits for nonexistent device %s", | |
525 | dm_device_name(ti->table->md), bdevname(bdev, b)); | |
754c5fc7 | 526 | return 0; |
0c2322e4 | 527 | } |
3cb40214 | 528 | |
b27d7f16 MP |
529 | if (bdev_stack_limits(limits, bdev, start) < 0) |
530 | DMWARN("%s: adding target device %s caused an alignment inconsistency: " | |
a963a956 MS |
531 | "physical_block_size=%u, logical_block_size=%u, " |
532 | "alignment_offset=%u, start=%llu", | |
533 | dm_device_name(ti->table->md), bdevname(bdev, b), | |
534 | q->limits.physical_block_size, | |
535 | q->limits.logical_block_size, | |
536 | q->limits.alignment_offset, | |
b27d7f16 | 537 | (unsigned long long) start << SECTOR_SHIFT); |
3cb40214 | 538 | |
9980c638 MB |
539 | /* |
540 | * Check if merge fn is supported. | |
541 | * If not we'll force DM to use PAGE_SIZE or | |
542 | * smaller I/O, just to be safe. | |
3cb40214 | 543 | */ |
d5b9dd04 | 544 | if (dm_queue_merge_is_compulsory(q) && !ti->type->merge) |
72d4cd9f MS |
545 | blk_limits_max_hw_sectors(limits, |
546 | (unsigned int) (PAGE_SIZE >> 9)); | |
754c5fc7 | 547 | return 0; |
3cb40214 BR |
548 | } |
549 | EXPORT_SYMBOL_GPL(dm_set_device_limits); | |
969429b5 | 550 | |
1da177e4 | 551 | /* |
08649012 | 552 | * Decrement a device's use count and remove it if necessary. |
1da177e4 | 553 | */ |
82b1519b | 554 | void dm_put_device(struct dm_target *ti, struct dm_dev *d) |
1da177e4 | 555 | { |
82b1519b MP |
556 | struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal, |
557 | dm_dev); | |
558 | ||
1da177e4 | 559 | if (atomic_dec_and_test(&dd->count)) { |
f165921d | 560 | close_dev(dd, ti->table->md); |
1da177e4 LT |
561 | list_del(&dd->list); |
562 | kfree(dd); | |
563 | } | |
564 | } | |
08649012 | 565 | EXPORT_SYMBOL(dm_put_device); |
1da177e4 LT |
566 | |
567 | /* | |
568 | * Checks to see if the target joins onto the end of the table. | |
569 | */ | |
570 | static int adjoin(struct dm_table *table, struct dm_target *ti) | |
571 | { | |
572 | struct dm_target *prev; | |
573 | ||
574 | if (!table->num_targets) | |
575 | return !ti->begin; | |
576 | ||
577 | prev = &table->targets[table->num_targets - 1]; | |
578 | return (ti->begin == (prev->begin + prev->len)); | |
579 | } | |
580 | ||
581 | /* | |
582 | * Used to dynamically allocate the arg array. | |
583 | */ | |
584 | static char **realloc_argv(unsigned *array_size, char **old_argv) | |
585 | { | |
586 | char **argv; | |
587 | unsigned new_size; | |
588 | ||
589 | new_size = *array_size ? *array_size * 2 : 64; | |
590 | argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL); | |
591 | if (argv) { | |
592 | memcpy(argv, old_argv, *array_size * sizeof(*argv)); | |
593 | *array_size = new_size; | |
594 | } | |
595 | ||
596 | kfree(old_argv); | |
597 | return argv; | |
598 | } | |
599 | ||
600 | /* | |
601 | * Destructively splits up the argument list to pass to ctr. | |
602 | */ | |
603 | int dm_split_args(int *argc, char ***argvp, char *input) | |
604 | { | |
605 | char *start, *end = input, *out, **argv = NULL; | |
606 | unsigned array_size = 0; | |
607 | ||
608 | *argc = 0; | |
814d6862 DT |
609 | |
610 | if (!input) { | |
611 | *argvp = NULL; | |
612 | return 0; | |
613 | } | |
614 | ||
1da177e4 LT |
615 | argv = realloc_argv(&array_size, argv); |
616 | if (!argv) | |
617 | return -ENOMEM; | |
618 | ||
619 | while (1) { | |
1da177e4 | 620 | /* Skip whitespace */ |
e7d2860b | 621 | start = skip_spaces(end); |
1da177e4 LT |
622 | |
623 | if (!*start) | |
624 | break; /* success, we hit the end */ | |
625 | ||
626 | /* 'out' is used to remove any back-quotes */ | |
627 | end = out = start; | |
628 | while (*end) { | |
629 | /* Everything apart from '\0' can be quoted */ | |
630 | if (*end == '\\' && *(end + 1)) { | |
631 | *out++ = *(end + 1); | |
632 | end += 2; | |
633 | continue; | |
634 | } | |
635 | ||
636 | if (isspace(*end)) | |
637 | break; /* end of token */ | |
638 | ||
639 | *out++ = *end++; | |
640 | } | |
641 | ||
642 | /* have we already filled the array ? */ | |
643 | if ((*argc + 1) > array_size) { | |
644 | argv = realloc_argv(&array_size, argv); | |
645 | if (!argv) | |
646 | return -ENOMEM; | |
647 | } | |
648 | ||
649 | /* we know this is whitespace */ | |
650 | if (*end) | |
651 | end++; | |
652 | ||
653 | /* terminate the string and put it in the array */ | |
654 | *out = '\0'; | |
655 | argv[*argc] = start; | |
656 | (*argc)++; | |
657 | } | |
658 | ||
659 | *argvp = argv; | |
660 | return 0; | |
661 | } | |
662 | ||
be6d4305 MS |
663 | /* |
664 | * Impose necessary and sufficient conditions on a devices's table such | |
665 | * that any incoming bio which respects its logical_block_size can be | |
666 | * processed successfully. If it falls across the boundary between | |
667 | * two or more targets, the size of each piece it gets split into must | |
668 | * be compatible with the logical_block_size of the target processing it. | |
669 | */ | |
754c5fc7 MS |
670 | static int validate_hardware_logical_block_alignment(struct dm_table *table, |
671 | struct queue_limits *limits) | |
be6d4305 MS |
672 | { |
673 | /* | |
674 | * This function uses arithmetic modulo the logical_block_size | |
675 | * (in units of 512-byte sectors). | |
676 | */ | |
677 | unsigned short device_logical_block_size_sects = | |
754c5fc7 | 678 | limits->logical_block_size >> SECTOR_SHIFT; |
be6d4305 MS |
679 | |
680 | /* | |
681 | * Offset of the start of the next table entry, mod logical_block_size. | |
682 | */ | |
683 | unsigned short next_target_start = 0; | |
684 | ||
685 | /* | |
686 | * Given an aligned bio that extends beyond the end of a | |
687 | * target, how many sectors must the next target handle? | |
688 | */ | |
689 | unsigned short remaining = 0; | |
690 | ||
691 | struct dm_target *uninitialized_var(ti); | |
754c5fc7 | 692 | struct queue_limits ti_limits; |
be6d4305 MS |
693 | unsigned i = 0; |
694 | ||
695 | /* | |
696 | * Check each entry in the table in turn. | |
697 | */ | |
698 | while (i < dm_table_get_num_targets(table)) { | |
699 | ti = dm_table_get_target(table, i++); | |
700 | ||
754c5fc7 MS |
701 | blk_set_default_limits(&ti_limits); |
702 | ||
703 | /* combine all target devices' limits */ | |
704 | if (ti->type->iterate_devices) | |
705 | ti->type->iterate_devices(ti, dm_set_device_limits, | |
706 | &ti_limits); | |
707 | ||
be6d4305 MS |
708 | /* |
709 | * If the remaining sectors fall entirely within this | |
710 | * table entry are they compatible with its logical_block_size? | |
711 | */ | |
712 | if (remaining < ti->len && | |
754c5fc7 | 713 | remaining & ((ti_limits.logical_block_size >> |
be6d4305 MS |
714 | SECTOR_SHIFT) - 1)) |
715 | break; /* Error */ | |
716 | ||
717 | next_target_start = | |
718 | (unsigned short) ((next_target_start + ti->len) & | |
719 | (device_logical_block_size_sects - 1)); | |
720 | remaining = next_target_start ? | |
721 | device_logical_block_size_sects - next_target_start : 0; | |
722 | } | |
723 | ||
724 | if (remaining) { | |
725 | DMWARN("%s: table line %u (start sect %llu len %llu) " | |
a963a956 | 726 | "not aligned to h/w logical block size %u", |
be6d4305 MS |
727 | dm_device_name(table->md), i, |
728 | (unsigned long long) ti->begin, | |
729 | (unsigned long long) ti->len, | |
754c5fc7 | 730 | limits->logical_block_size); |
be6d4305 MS |
731 | return -EINVAL; |
732 | } | |
733 | ||
734 | return 0; | |
735 | } | |
736 | ||
1da177e4 LT |
737 | int dm_table_add_target(struct dm_table *t, const char *type, |
738 | sector_t start, sector_t len, char *params) | |
739 | { | |
740 | int r = -EINVAL, argc; | |
741 | char **argv; | |
742 | struct dm_target *tgt; | |
743 | ||
3791e2fc AK |
744 | if (t->singleton) { |
745 | DMERR("%s: target type %s must appear alone in table", | |
746 | dm_device_name(t->md), t->targets->type->name); | |
747 | return -EINVAL; | |
748 | } | |
749 | ||
1da177e4 LT |
750 | if ((r = check_space(t))) |
751 | return r; | |
752 | ||
753 | tgt = t->targets + t->num_targets; | |
754 | memset(tgt, 0, sizeof(*tgt)); | |
755 | ||
756 | if (!len) { | |
72d94861 | 757 | DMERR("%s: zero-length target", dm_device_name(t->md)); |
1da177e4 LT |
758 | return -EINVAL; |
759 | } | |
760 | ||
761 | tgt->type = dm_get_target_type(type); | |
762 | if (!tgt->type) { | |
72d94861 AK |
763 | DMERR("%s: %s: unknown target type", dm_device_name(t->md), |
764 | type); | |
1da177e4 LT |
765 | return -EINVAL; |
766 | } | |
767 | ||
3791e2fc AK |
768 | if (dm_target_needs_singleton(tgt->type)) { |
769 | if (t->num_targets) { | |
770 | DMERR("%s: target type %s must appear alone in table", | |
771 | dm_device_name(t->md), type); | |
772 | return -EINVAL; | |
773 | } | |
774 | t->singleton = 1; | |
775 | } | |
776 | ||
1da177e4 LT |
777 | tgt->table = t; |
778 | tgt->begin = start; | |
779 | tgt->len = len; | |
780 | tgt->error = "Unknown error"; | |
781 | ||
782 | /* | |
783 | * Does this target adjoin the previous one ? | |
784 | */ | |
785 | if (!adjoin(t, tgt)) { | |
786 | tgt->error = "Gap in table"; | |
787 | r = -EINVAL; | |
788 | goto bad; | |
789 | } | |
790 | ||
791 | r = dm_split_args(&argc, &argv, params); | |
792 | if (r) { | |
793 | tgt->error = "couldn't split parameters (insufficient memory)"; | |
794 | goto bad; | |
795 | } | |
796 | ||
797 | r = tgt->type->ctr(tgt, argc, argv); | |
798 | kfree(argv); | |
799 | if (r) | |
800 | goto bad; | |
801 | ||
802 | t->highs[t->num_targets++] = tgt->begin + tgt->len - 1; | |
803 | ||
936688d7 MS |
804 | if (!tgt->num_discard_requests && tgt->discards_supported) |
805 | DMWARN("%s: %s: ignoring discards_supported because num_discard_requests is zero.", | |
806 | dm_device_name(t->md), type); | |
5ae89a87 | 807 | |
1da177e4 LT |
808 | return 0; |
809 | ||
810 | bad: | |
72d94861 | 811 | DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error); |
1da177e4 LT |
812 | dm_put_target_type(tgt->type); |
813 | return r; | |
814 | } | |
815 | ||
498f0103 MS |
816 | /* |
817 | * Target argument parsing helpers. | |
818 | */ | |
819 | static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set, | |
820 | unsigned *value, char **error, unsigned grouped) | |
821 | { | |
822 | const char *arg_str = dm_shift_arg(arg_set); | |
823 | ||
824 | if (!arg_str || | |
825 | (sscanf(arg_str, "%u", value) != 1) || | |
826 | (*value < arg->min) || | |
827 | (*value > arg->max) || | |
828 | (grouped && arg_set->argc < *value)) { | |
829 | *error = arg->error; | |
830 | return -EINVAL; | |
831 | } | |
832 | ||
833 | return 0; | |
834 | } | |
835 | ||
836 | int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set, | |
837 | unsigned *value, char **error) | |
838 | { | |
839 | return validate_next_arg(arg, arg_set, value, error, 0); | |
840 | } | |
841 | EXPORT_SYMBOL(dm_read_arg); | |
842 | ||
843 | int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set, | |
844 | unsigned *value, char **error) | |
845 | { | |
846 | return validate_next_arg(arg, arg_set, value, error, 1); | |
847 | } | |
848 | EXPORT_SYMBOL(dm_read_arg_group); | |
849 | ||
850 | const char *dm_shift_arg(struct dm_arg_set *as) | |
851 | { | |
852 | char *r; | |
853 | ||
854 | if (as->argc) { | |
855 | as->argc--; | |
856 | r = *as->argv; | |
857 | as->argv++; | |
858 | return r; | |
859 | } | |
860 | ||
861 | return NULL; | |
862 | } | |
863 | EXPORT_SYMBOL(dm_shift_arg); | |
864 | ||
865 | void dm_consume_args(struct dm_arg_set *as, unsigned num_args) | |
866 | { | |
867 | BUG_ON(as->argc < num_args); | |
868 | as->argc -= num_args; | |
869 | as->argv += num_args; | |
870 | } | |
871 | EXPORT_SYMBOL(dm_consume_args); | |
872 | ||
26803b9f | 873 | static int dm_table_set_type(struct dm_table *t) |
e6ee8c0b KU |
874 | { |
875 | unsigned i; | |
876 | unsigned bio_based = 0, request_based = 0; | |
877 | struct dm_target *tgt; | |
878 | struct dm_dev_internal *dd; | |
879 | struct list_head *devices; | |
880 | ||
881 | for (i = 0; i < t->num_targets; i++) { | |
882 | tgt = t->targets + i; | |
883 | if (dm_target_request_based(tgt)) | |
884 | request_based = 1; | |
885 | else | |
886 | bio_based = 1; | |
887 | ||
888 | if (bio_based && request_based) { | |
889 | DMWARN("Inconsistent table: different target types" | |
890 | " can't be mixed up"); | |
891 | return -EINVAL; | |
892 | } | |
893 | } | |
894 | ||
895 | if (bio_based) { | |
896 | /* We must use this table as bio-based */ | |
897 | t->type = DM_TYPE_BIO_BASED; | |
898 | return 0; | |
899 | } | |
900 | ||
901 | BUG_ON(!request_based); /* No targets in this table */ | |
902 | ||
903 | /* Non-request-stackable devices can't be used for request-based dm */ | |
904 | devices = dm_table_get_devices(t); | |
905 | list_for_each_entry(dd, devices, list) { | |
906 | if (!blk_queue_stackable(bdev_get_queue(dd->dm_dev.bdev))) { | |
907 | DMWARN("table load rejected: including" | |
908 | " non-request-stackable devices"); | |
909 | return -EINVAL; | |
910 | } | |
911 | } | |
912 | ||
913 | /* | |
914 | * Request-based dm supports only tables that have a single target now. | |
915 | * To support multiple targets, request splitting support is needed, | |
916 | * and that needs lots of changes in the block-layer. | |
917 | * (e.g. request completion process for partial completion.) | |
918 | */ | |
919 | if (t->num_targets > 1) { | |
920 | DMWARN("Request-based dm doesn't support multiple targets yet"); | |
921 | return -EINVAL; | |
922 | } | |
923 | ||
924 | t->type = DM_TYPE_REQUEST_BASED; | |
925 | ||
926 | return 0; | |
927 | } | |
928 | ||
929 | unsigned dm_table_get_type(struct dm_table *t) | |
930 | { | |
931 | return t->type; | |
932 | } | |
933 | ||
934 | bool dm_table_request_based(struct dm_table *t) | |
935 | { | |
936 | return dm_table_get_type(t) == DM_TYPE_REQUEST_BASED; | |
937 | } | |
938 | ||
939 | int dm_table_alloc_md_mempools(struct dm_table *t) | |
940 | { | |
941 | unsigned type = dm_table_get_type(t); | |
942 | ||
943 | if (unlikely(type == DM_TYPE_NONE)) { | |
944 | DMWARN("no table type is set, can't allocate mempools"); | |
945 | return -EINVAL; | |
946 | } | |
947 | ||
a91a2785 | 948 | t->mempools = dm_alloc_md_mempools(type, t->integrity_supported); |
e6ee8c0b KU |
949 | if (!t->mempools) |
950 | return -ENOMEM; | |
951 | ||
952 | return 0; | |
953 | } | |
954 | ||
955 | void dm_table_free_md_mempools(struct dm_table *t) | |
956 | { | |
957 | dm_free_md_mempools(t->mempools); | |
958 | t->mempools = NULL; | |
959 | } | |
960 | ||
961 | struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t) | |
962 | { | |
963 | return t->mempools; | |
964 | } | |
965 | ||
1da177e4 LT |
966 | static int setup_indexes(struct dm_table *t) |
967 | { | |
968 | int i; | |
969 | unsigned int total = 0; | |
970 | sector_t *indexes; | |
971 | ||
972 | /* allocate the space for *all* the indexes */ | |
973 | for (i = t->depth - 2; i >= 0; i--) { | |
974 | t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE); | |
975 | total += t->counts[i]; | |
976 | } | |
977 | ||
978 | indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE); | |
979 | if (!indexes) | |
980 | return -ENOMEM; | |
981 | ||
982 | /* set up internal nodes, bottom-up */ | |
82d601dc | 983 | for (i = t->depth - 2; i >= 0; i--) { |
1da177e4 LT |
984 | t->index[i] = indexes; |
985 | indexes += (KEYS_PER_NODE * t->counts[i]); | |
986 | setup_btree_index(i, t); | |
987 | } | |
988 | ||
989 | return 0; | |
990 | } | |
991 | ||
992 | /* | |
993 | * Builds the btree to index the map. | |
994 | */ | |
26803b9f | 995 | static int dm_table_build_index(struct dm_table *t) |
1da177e4 LT |
996 | { |
997 | int r = 0; | |
998 | unsigned int leaf_nodes; | |
999 | ||
1da177e4 LT |
1000 | /* how many indexes will the btree have ? */ |
1001 | leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE); | |
1002 | t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE); | |
1003 | ||
1004 | /* leaf layer has already been set up */ | |
1005 | t->counts[t->depth - 1] = leaf_nodes; | |
1006 | t->index[t->depth - 1] = t->highs; | |
1007 | ||
1008 | if (t->depth >= 2) | |
1009 | r = setup_indexes(t); | |
1010 | ||
1011 | return r; | |
1012 | } | |
1013 | ||
a63a5cf8 MS |
1014 | /* |
1015 | * Get a disk whose integrity profile reflects the table's profile. | |
1016 | * If %match_all is true, all devices' profiles must match. | |
1017 | * If %match_all is false, all devices must at least have an | |
1018 | * allocated integrity profile; but uninitialized is ok. | |
1019 | * Returns NULL if integrity support was inconsistent or unavailable. | |
1020 | */ | |
1021 | static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t, | |
1022 | bool match_all) | |
1023 | { | |
1024 | struct list_head *devices = dm_table_get_devices(t); | |
1025 | struct dm_dev_internal *dd = NULL; | |
1026 | struct gendisk *prev_disk = NULL, *template_disk = NULL; | |
1027 | ||
1028 | list_for_each_entry(dd, devices, list) { | |
1029 | template_disk = dd->dm_dev.bdev->bd_disk; | |
1030 | if (!blk_get_integrity(template_disk)) | |
1031 | goto no_integrity; | |
1032 | if (!match_all && !blk_integrity_is_initialized(template_disk)) | |
1033 | continue; /* skip uninitialized profiles */ | |
1034 | else if (prev_disk && | |
1035 | blk_integrity_compare(prev_disk, template_disk) < 0) | |
1036 | goto no_integrity; | |
1037 | prev_disk = template_disk; | |
1038 | } | |
1039 | ||
1040 | return template_disk; | |
1041 | ||
1042 | no_integrity: | |
1043 | if (prev_disk) | |
1044 | DMWARN("%s: integrity not set: %s and %s profile mismatch", | |
1045 | dm_device_name(t->md), | |
1046 | prev_disk->disk_name, | |
1047 | template_disk->disk_name); | |
1048 | return NULL; | |
1049 | } | |
1050 | ||
26803b9f WD |
1051 | /* |
1052 | * Register the mapped device for blk_integrity support if | |
a63a5cf8 MS |
1053 | * the underlying devices have an integrity profile. But all devices |
1054 | * may not have matching profiles (checking all devices isn't reliable | |
1055 | * during table load because this table may use other DM device(s) which | |
1056 | * must be resumed before they will have an initialized integity profile). | |
1057 | * Stacked DM devices force a 2 stage integrity profile validation: | |
1058 | * 1 - during load, validate all initialized integrity profiles match | |
1059 | * 2 - during resume, validate all integrity profiles match | |
26803b9f WD |
1060 | */ |
1061 | static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md) | |
1062 | { | |
a63a5cf8 | 1063 | struct gendisk *template_disk = NULL; |
26803b9f | 1064 | |
a63a5cf8 MS |
1065 | template_disk = dm_table_get_integrity_disk(t, false); |
1066 | if (!template_disk) | |
1067 | return 0; | |
26803b9f | 1068 | |
a63a5cf8 MS |
1069 | if (!blk_integrity_is_initialized(dm_disk(md))) { |
1070 | t->integrity_supported = 1; | |
1071 | return blk_integrity_register(dm_disk(md), NULL); | |
1072 | } | |
1073 | ||
1074 | /* | |
1075 | * If DM device already has an initalized integrity | |
1076 | * profile the new profile should not conflict. | |
1077 | */ | |
1078 | if (blk_integrity_is_initialized(template_disk) && | |
1079 | blk_integrity_compare(dm_disk(md), template_disk) < 0) { | |
1080 | DMWARN("%s: conflict with existing integrity profile: " | |
1081 | "%s profile mismatch", | |
1082 | dm_device_name(t->md), | |
1083 | template_disk->disk_name); | |
1084 | return 1; | |
1085 | } | |
1086 | ||
1087 | /* Preserve existing initialized integrity profile */ | |
1088 | t->integrity_supported = 1; | |
26803b9f WD |
1089 | return 0; |
1090 | } | |
1091 | ||
1092 | /* | |
1093 | * Prepares the table for use by building the indices, | |
1094 | * setting the type, and allocating mempools. | |
1095 | */ | |
1096 | int dm_table_complete(struct dm_table *t) | |
1097 | { | |
1098 | int r; | |
1099 | ||
1100 | r = dm_table_set_type(t); | |
1101 | if (r) { | |
1102 | DMERR("unable to set table type"); | |
1103 | return r; | |
1104 | } | |
1105 | ||
1106 | r = dm_table_build_index(t); | |
1107 | if (r) { | |
1108 | DMERR("unable to build btrees"); | |
1109 | return r; | |
1110 | } | |
1111 | ||
1112 | r = dm_table_prealloc_integrity(t, t->md); | |
1113 | if (r) { | |
1114 | DMERR("could not register integrity profile."); | |
1115 | return r; | |
1116 | } | |
1117 | ||
1118 | r = dm_table_alloc_md_mempools(t); | |
1119 | if (r) | |
1120 | DMERR("unable to allocate mempools"); | |
1121 | ||
1122 | return r; | |
1123 | } | |
1124 | ||
48c9c27b | 1125 | static DEFINE_MUTEX(_event_lock); |
1da177e4 LT |
1126 | void dm_table_event_callback(struct dm_table *t, |
1127 | void (*fn)(void *), void *context) | |
1128 | { | |
48c9c27b | 1129 | mutex_lock(&_event_lock); |
1da177e4 LT |
1130 | t->event_fn = fn; |
1131 | t->event_context = context; | |
48c9c27b | 1132 | mutex_unlock(&_event_lock); |
1da177e4 LT |
1133 | } |
1134 | ||
1135 | void dm_table_event(struct dm_table *t) | |
1136 | { | |
1137 | /* | |
1138 | * You can no longer call dm_table_event() from interrupt | |
1139 | * context, use a bottom half instead. | |
1140 | */ | |
1141 | BUG_ON(in_interrupt()); | |
1142 | ||
48c9c27b | 1143 | mutex_lock(&_event_lock); |
1da177e4 LT |
1144 | if (t->event_fn) |
1145 | t->event_fn(t->event_context); | |
48c9c27b | 1146 | mutex_unlock(&_event_lock); |
1da177e4 | 1147 | } |
08649012 | 1148 | EXPORT_SYMBOL(dm_table_event); |
1da177e4 LT |
1149 | |
1150 | sector_t dm_table_get_size(struct dm_table *t) | |
1151 | { | |
1152 | return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0; | |
1153 | } | |
08649012 | 1154 | EXPORT_SYMBOL(dm_table_get_size); |
1da177e4 LT |
1155 | |
1156 | struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index) | |
1157 | { | |
14353539 | 1158 | if (index >= t->num_targets) |
1da177e4 LT |
1159 | return NULL; |
1160 | ||
1161 | return t->targets + index; | |
1162 | } | |
1163 | ||
1164 | /* | |
1165 | * Search the btree for the correct target. | |
512875bd JN |
1166 | * |
1167 | * Caller should check returned pointer with dm_target_is_valid() | |
1168 | * to trap I/O beyond end of device. | |
1da177e4 LT |
1169 | */ |
1170 | struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector) | |
1171 | { | |
1172 | unsigned int l, n = 0, k = 0; | |
1173 | sector_t *node; | |
1174 | ||
1175 | for (l = 0; l < t->depth; l++) { | |
1176 | n = get_child(n, k); | |
1177 | node = get_node(t, l, n); | |
1178 | ||
1179 | for (k = 0; k < KEYS_PER_NODE; k++) | |
1180 | if (node[k] >= sector) | |
1181 | break; | |
1182 | } | |
1183 | ||
1184 | return &t->targets[(KEYS_PER_NODE * n) + k]; | |
1185 | } | |
1186 | ||
754c5fc7 MS |
1187 | /* |
1188 | * Establish the new table's queue_limits and validate them. | |
1189 | */ | |
1190 | int dm_calculate_queue_limits(struct dm_table *table, | |
1191 | struct queue_limits *limits) | |
1192 | { | |
1193 | struct dm_target *uninitialized_var(ti); | |
1194 | struct queue_limits ti_limits; | |
1195 | unsigned i = 0; | |
1196 | ||
1197 | blk_set_default_limits(limits); | |
1198 | ||
1199 | while (i < dm_table_get_num_targets(table)) { | |
1200 | blk_set_default_limits(&ti_limits); | |
1201 | ||
1202 | ti = dm_table_get_target(table, i++); | |
1203 | ||
1204 | if (!ti->type->iterate_devices) | |
1205 | goto combine_limits; | |
1206 | ||
1207 | /* | |
1208 | * Combine queue limits of all the devices this target uses. | |
1209 | */ | |
1210 | ti->type->iterate_devices(ti, dm_set_device_limits, | |
1211 | &ti_limits); | |
1212 | ||
40bea431 MS |
1213 | /* Set I/O hints portion of queue limits */ |
1214 | if (ti->type->io_hints) | |
1215 | ti->type->io_hints(ti, &ti_limits); | |
1216 | ||
754c5fc7 MS |
1217 | /* |
1218 | * Check each device area is consistent with the target's | |
1219 | * overall queue limits. | |
1220 | */ | |
f6a1ed10 MP |
1221 | if (ti->type->iterate_devices(ti, device_area_is_invalid, |
1222 | &ti_limits)) | |
754c5fc7 MS |
1223 | return -EINVAL; |
1224 | ||
1225 | combine_limits: | |
1226 | /* | |
1227 | * Merge this target's queue limits into the overall limits | |
1228 | * for the table. | |
1229 | */ | |
1230 | if (blk_stack_limits(limits, &ti_limits, 0) < 0) | |
b27d7f16 | 1231 | DMWARN("%s: adding target device " |
754c5fc7 | 1232 | "(start sect %llu len %llu) " |
b27d7f16 | 1233 | "caused an alignment inconsistency", |
754c5fc7 MS |
1234 | dm_device_name(table->md), |
1235 | (unsigned long long) ti->begin, | |
1236 | (unsigned long long) ti->len); | |
1237 | } | |
1238 | ||
1239 | return validate_hardware_logical_block_alignment(table, limits); | |
1240 | } | |
1241 | ||
9c47008d MP |
1242 | /* |
1243 | * Set the integrity profile for this device if all devices used have | |
a63a5cf8 MS |
1244 | * matching profiles. We're quite deep in the resume path but still |
1245 | * don't know if all devices (particularly DM devices this device | |
1246 | * may be stacked on) have matching profiles. Even if the profiles | |
1247 | * don't match we have no way to fail (to resume) at this point. | |
9c47008d MP |
1248 | */ |
1249 | static void dm_table_set_integrity(struct dm_table *t) | |
1250 | { | |
a63a5cf8 | 1251 | struct gendisk *template_disk = NULL; |
9c47008d MP |
1252 | |
1253 | if (!blk_get_integrity(dm_disk(t->md))) | |
1254 | return; | |
1255 | ||
a63a5cf8 | 1256 | template_disk = dm_table_get_integrity_disk(t, true); |
876fbba1 MS |
1257 | if (template_disk) |
1258 | blk_integrity_register(dm_disk(t->md), | |
1259 | blk_get_integrity(template_disk)); | |
1260 | else if (blk_integrity_is_initialized(dm_disk(t->md))) | |
a63a5cf8 MS |
1261 | DMWARN("%s: device no longer has a valid integrity profile", |
1262 | dm_device_name(t->md)); | |
876fbba1 MS |
1263 | else |
1264 | DMWARN("%s: unable to establish an integrity profile", | |
1265 | dm_device_name(t->md)); | |
9c47008d MP |
1266 | } |
1267 | ||
ed8b752b MS |
1268 | static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev, |
1269 | sector_t start, sector_t len, void *data) | |
1270 | { | |
1271 | unsigned flush = (*(unsigned *)data); | |
1272 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
1273 | ||
1274 | return q && (q->flush_flags & flush); | |
1275 | } | |
1276 | ||
1277 | static bool dm_table_supports_flush(struct dm_table *t, unsigned flush) | |
1278 | { | |
1279 | struct dm_target *ti; | |
1280 | unsigned i = 0; | |
1281 | ||
1282 | /* | |
1283 | * Require at least one underlying device to support flushes. | |
1284 | * t->devices includes internal dm devices such as mirror logs | |
1285 | * so we need to use iterate_devices here, which targets | |
1286 | * supporting flushes must provide. | |
1287 | */ | |
1288 | while (i < dm_table_get_num_targets(t)) { | |
1289 | ti = dm_table_get_target(t, i++); | |
1290 | ||
1291 | if (!ti->num_flush_requests) | |
1292 | continue; | |
1293 | ||
1294 | if (ti->type->iterate_devices && | |
1295 | ti->type->iterate_devices(ti, device_flush_capable, &flush)) | |
1296 | return 1; | |
1297 | } | |
1298 | ||
1299 | return 0; | |
1300 | } | |
1301 | ||
983c7db3 MB |
1302 | static bool dm_table_discard_zeroes_data(struct dm_table *t) |
1303 | { | |
1304 | struct dm_target *ti; | |
1305 | unsigned i = 0; | |
1306 | ||
1307 | /* Ensure that all targets supports discard_zeroes_data. */ | |
1308 | while (i < dm_table_get_num_targets(t)) { | |
1309 | ti = dm_table_get_target(t, i++); | |
1310 | ||
1311 | if (ti->discard_zeroes_data_unsupported) | |
1312 | return 0; | |
1313 | } | |
1314 | ||
1315 | return 1; | |
1316 | } | |
1317 | ||
4693c966 MSB |
1318 | static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev, |
1319 | sector_t start, sector_t len, void *data) | |
1320 | { | |
1321 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
1322 | ||
1323 | return q && blk_queue_nonrot(q); | |
1324 | } | |
1325 | ||
1326 | static bool dm_table_is_nonrot(struct dm_table *t) | |
1327 | { | |
1328 | struct dm_target *ti; | |
1329 | unsigned i = 0; | |
1330 | ||
1331 | /* Ensure that all underlying device are non-rotational. */ | |
1332 | while (i < dm_table_get_num_targets(t)) { | |
1333 | ti = dm_table_get_target(t, i++); | |
1334 | ||
1335 | if (!ti->type->iterate_devices || | |
1336 | !ti->type->iterate_devices(ti, device_is_nonrot, NULL)) | |
1337 | return 0; | |
1338 | } | |
1339 | ||
1340 | return 1; | |
1341 | } | |
1342 | ||
754c5fc7 MS |
1343 | void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q, |
1344 | struct queue_limits *limits) | |
1da177e4 | 1345 | { |
ed8b752b MS |
1346 | unsigned flush = 0; |
1347 | ||
1da177e4 | 1348 | /* |
1197764e | 1349 | * Copy table's limits to the DM device's request_queue |
1da177e4 | 1350 | */ |
754c5fc7 | 1351 | q->limits = *limits; |
c9a3f6d6 | 1352 | |
5ae89a87 MS |
1353 | if (!dm_table_supports_discards(t)) |
1354 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); | |
1355 | else | |
1356 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); | |
1357 | ||
ed8b752b MS |
1358 | if (dm_table_supports_flush(t, REQ_FLUSH)) { |
1359 | flush |= REQ_FLUSH; | |
1360 | if (dm_table_supports_flush(t, REQ_FUA)) | |
1361 | flush |= REQ_FUA; | |
1362 | } | |
1363 | blk_queue_flush(q, flush); | |
1364 | ||
983c7db3 MB |
1365 | if (!dm_table_discard_zeroes_data(t)) |
1366 | q->limits.discard_zeroes_data = 0; | |
1367 | ||
4693c966 MSB |
1368 | if (dm_table_is_nonrot(t)) |
1369 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q); | |
1370 | else | |
1371 | queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q); | |
1372 | ||
9c47008d | 1373 | dm_table_set_integrity(t); |
e6ee8c0b KU |
1374 | |
1375 | /* | |
1376 | * QUEUE_FLAG_STACKABLE must be set after all queue settings are | |
1377 | * visible to other CPUs because, once the flag is set, incoming bios | |
1378 | * are processed by request-based dm, which refers to the queue | |
1379 | * settings. | |
1380 | * Until the flag set, bios are passed to bio-based dm and queued to | |
1381 | * md->deferred where queue settings are not needed yet. | |
1382 | * Those bios are passed to request-based dm at the resume time. | |
1383 | */ | |
1384 | smp_mb(); | |
1385 | if (dm_table_request_based(t)) | |
1386 | queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q); | |
1da177e4 LT |
1387 | } |
1388 | ||
1389 | unsigned int dm_table_get_num_targets(struct dm_table *t) | |
1390 | { | |
1391 | return t->num_targets; | |
1392 | } | |
1393 | ||
1394 | struct list_head *dm_table_get_devices(struct dm_table *t) | |
1395 | { | |
1396 | return &t->devices; | |
1397 | } | |
1398 | ||
aeb5d727 | 1399 | fmode_t dm_table_get_mode(struct dm_table *t) |
1da177e4 LT |
1400 | { |
1401 | return t->mode; | |
1402 | } | |
08649012 | 1403 | EXPORT_SYMBOL(dm_table_get_mode); |
1da177e4 LT |
1404 | |
1405 | static void suspend_targets(struct dm_table *t, unsigned postsuspend) | |
1406 | { | |
1407 | int i = t->num_targets; | |
1408 | struct dm_target *ti = t->targets; | |
1409 | ||
1410 | while (i--) { | |
1411 | if (postsuspend) { | |
1412 | if (ti->type->postsuspend) | |
1413 | ti->type->postsuspend(ti); | |
1414 | } else if (ti->type->presuspend) | |
1415 | ti->type->presuspend(ti); | |
1416 | ||
1417 | ti++; | |
1418 | } | |
1419 | } | |
1420 | ||
1421 | void dm_table_presuspend_targets(struct dm_table *t) | |
1422 | { | |
cf222b37 AK |
1423 | if (!t) |
1424 | return; | |
1425 | ||
e8488d08 | 1426 | suspend_targets(t, 0); |
1da177e4 LT |
1427 | } |
1428 | ||
1429 | void dm_table_postsuspend_targets(struct dm_table *t) | |
1430 | { | |
cf222b37 AK |
1431 | if (!t) |
1432 | return; | |
1433 | ||
e8488d08 | 1434 | suspend_targets(t, 1); |
1da177e4 LT |
1435 | } |
1436 | ||
8757b776 | 1437 | int dm_table_resume_targets(struct dm_table *t) |
1da177e4 | 1438 | { |
8757b776 MB |
1439 | int i, r = 0; |
1440 | ||
1441 | for (i = 0; i < t->num_targets; i++) { | |
1442 | struct dm_target *ti = t->targets + i; | |
1443 | ||
1444 | if (!ti->type->preresume) | |
1445 | continue; | |
1446 | ||
1447 | r = ti->type->preresume(ti); | |
1448 | if (r) | |
1449 | return r; | |
1450 | } | |
1da177e4 LT |
1451 | |
1452 | for (i = 0; i < t->num_targets; i++) { | |
1453 | struct dm_target *ti = t->targets + i; | |
1454 | ||
1455 | if (ti->type->resume) | |
1456 | ti->type->resume(ti); | |
1457 | } | |
8757b776 MB |
1458 | |
1459 | return 0; | |
1da177e4 LT |
1460 | } |
1461 | ||
9d357b07 N |
1462 | void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb) |
1463 | { | |
1464 | list_add(&cb->list, &t->target_callbacks); | |
1465 | } | |
1466 | EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks); | |
1467 | ||
1da177e4 LT |
1468 | int dm_table_any_congested(struct dm_table *t, int bdi_bits) |
1469 | { | |
82b1519b | 1470 | struct dm_dev_internal *dd; |
afb24528 | 1471 | struct list_head *devices = dm_table_get_devices(t); |
9d357b07 | 1472 | struct dm_target_callbacks *cb; |
1da177e4 LT |
1473 | int r = 0; |
1474 | ||
afb24528 | 1475 | list_for_each_entry(dd, devices, list) { |
82b1519b | 1476 | struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev); |
0c2322e4 AK |
1477 | char b[BDEVNAME_SIZE]; |
1478 | ||
1479 | if (likely(q)) | |
1480 | r |= bdi_congested(&q->backing_dev_info, bdi_bits); | |
1481 | else | |
1482 | DMWARN_LIMIT("%s: any_congested: nonexistent device %s", | |
1483 | dm_device_name(t->md), | |
1484 | bdevname(dd->dm_dev.bdev, b)); | |
1da177e4 LT |
1485 | } |
1486 | ||
9d357b07 N |
1487 | list_for_each_entry(cb, &t->target_callbacks, list) |
1488 | if (cb->congested_fn) | |
1489 | r |= cb->congested_fn(cb, bdi_bits); | |
1490 | ||
1da177e4 LT |
1491 | return r; |
1492 | } | |
1493 | ||
cec47e3d KU |
1494 | int dm_table_any_busy_target(struct dm_table *t) |
1495 | { | |
1496 | unsigned i; | |
1497 | struct dm_target *ti; | |
1498 | ||
1499 | for (i = 0; i < t->num_targets; i++) { | |
1500 | ti = t->targets + i; | |
1501 | if (ti->type->busy && ti->type->busy(ti)) | |
1502 | return 1; | |
1503 | } | |
1504 | ||
1505 | return 0; | |
1506 | } | |
1507 | ||
1134e5ae MA |
1508 | struct mapped_device *dm_table_get_md(struct dm_table *t) |
1509 | { | |
1134e5ae MA |
1510 | return t->md; |
1511 | } | |
08649012 | 1512 | EXPORT_SYMBOL(dm_table_get_md); |
1134e5ae | 1513 | |
5ae89a87 MS |
1514 | static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev, |
1515 | sector_t start, sector_t len, void *data) | |
1516 | { | |
1517 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
1518 | ||
1519 | return q && blk_queue_discard(q); | |
1520 | } | |
1521 | ||
1522 | bool dm_table_supports_discards(struct dm_table *t) | |
1523 | { | |
1524 | struct dm_target *ti; | |
1525 | unsigned i = 0; | |
1526 | ||
5ae89a87 | 1527 | /* |
4c259327 MS |
1528 | * Unless any target used by the table set discards_supported, |
1529 | * require at least one underlying device to support discards. | |
5ae89a87 MS |
1530 | * t->devices includes internal dm devices such as mirror logs |
1531 | * so we need to use iterate_devices here, which targets | |
936688d7 | 1532 | * supporting discard selectively must provide. |
5ae89a87 MS |
1533 | */ |
1534 | while (i < dm_table_get_num_targets(t)) { | |
1535 | ti = dm_table_get_target(t, i++); | |
1536 | ||
936688d7 MS |
1537 | if (!ti->num_discard_requests) |
1538 | continue; | |
1539 | ||
4c259327 MS |
1540 | if (ti->discards_supported) |
1541 | return 1; | |
1542 | ||
5ae89a87 MS |
1543 | if (ti->type->iterate_devices && |
1544 | ti->type->iterate_devices(ti, device_discard_capable, NULL)) | |
1545 | return 1; | |
1546 | } | |
1547 | ||
1548 | return 0; | |
1549 | } |