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> | |
15 | #include <linux/slab.h> | |
16 | #include <linux/interrupt.h> | |
48c9c27b | 17 | #include <linux/mutex.h> |
d5816876 | 18 | #include <linux/delay.h> |
1da177e4 LT |
19 | #include <asm/atomic.h> |
20 | ||
72d94861 AK |
21 | #define DM_MSG_PREFIX "table" |
22 | ||
1da177e4 LT |
23 | #define MAX_DEPTH 16 |
24 | #define NODE_SIZE L1_CACHE_BYTES | |
25 | #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t)) | |
26 | #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1) | |
27 | ||
d5816876 MP |
28 | /* |
29 | * The table has always exactly one reference from either mapped_device->map | |
30 | * or hash_cell->new_map. This reference is not counted in table->holders. | |
31 | * A pair of dm_create_table/dm_destroy_table functions is used for table | |
32 | * creation/destruction. | |
33 | * | |
34 | * Temporary references from the other code increase table->holders. A pair | |
35 | * of dm_table_get/dm_table_put functions is used to manipulate it. | |
36 | * | |
37 | * When the table is about to be destroyed, we wait for table->holders to | |
38 | * drop to zero. | |
39 | */ | |
40 | ||
1da177e4 | 41 | struct dm_table { |
1134e5ae | 42 | struct mapped_device *md; |
1da177e4 LT |
43 | atomic_t holders; |
44 | ||
45 | /* btree table */ | |
46 | unsigned int depth; | |
47 | unsigned int counts[MAX_DEPTH]; /* in nodes */ | |
48 | sector_t *index[MAX_DEPTH]; | |
49 | ||
50 | unsigned int num_targets; | |
51 | unsigned int num_allocated; | |
52 | sector_t *highs; | |
53 | struct dm_target *targets; | |
54 | ||
55 | /* | |
56 | * Indicates the rw permissions for the new logical | |
57 | * device. This should be a combination of FMODE_READ | |
58 | * and FMODE_WRITE. | |
59 | */ | |
aeb5d727 | 60 | fmode_t mode; |
1da177e4 LT |
61 | |
62 | /* a list of devices used by this table */ | |
63 | struct list_head devices; | |
64 | ||
65 | /* | |
66 | * These are optimistic limits taken from all the | |
67 | * targets, some targets will need smaller limits. | |
68 | */ | |
5ab97588 | 69 | struct queue_limits limits; |
1da177e4 LT |
70 | |
71 | /* events get handed up using this callback */ | |
72 | void (*event_fn)(void *); | |
73 | void *event_context; | |
74 | }; | |
75 | ||
76 | /* | |
77 | * Similar to ceiling(log_size(n)) | |
78 | */ | |
79 | static 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 | ||
1da177e4 LT |
91 | /* |
92 | * Calculate the index of the child node of the n'th node k'th key. | |
93 | */ | |
94 | static 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 | */ | |
102 | static 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 | */ | |
112 | static 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 | */ | |
127 | static 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 | ||
142 | void *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 | */ | |
165 | static 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. | |
512875bd JN |
173 | * Append an empty entry to catch sectors beyond the end of |
174 | * the device. | |
1da177e4 | 175 | */ |
512875bd | 176 | n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) + |
1da177e4 LT |
177 | 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 | ||
aeb5d727 | 198 | int dm_table_create(struct dm_table **result, fmode_t mode, |
1134e5ae | 199 | unsigned num_targets, struct mapped_device *md) |
1da177e4 | 200 | { |
094262db | 201 | struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL); |
1da177e4 LT |
202 | |
203 | if (!t) | |
204 | return -ENOMEM; | |
205 | ||
1da177e4 | 206 | INIT_LIST_HEAD(&t->devices); |
d5816876 | 207 | atomic_set(&t->holders, 0); |
1da177e4 LT |
208 | |
209 | if (!num_targets) | |
210 | num_targets = KEYS_PER_NODE; | |
211 | ||
212 | num_targets = dm_round_up(num_targets, KEYS_PER_NODE); | |
213 | ||
214 | if (alloc_targets(t, num_targets)) { | |
215 | kfree(t); | |
216 | t = NULL; | |
217 | return -ENOMEM; | |
218 | } | |
219 | ||
220 | t->mode = mode; | |
1134e5ae | 221 | t->md = md; |
1da177e4 LT |
222 | *result = t; |
223 | return 0; | |
224 | } | |
225 | ||
226 | static void free_devices(struct list_head *devices) | |
227 | { | |
228 | struct list_head *tmp, *next; | |
229 | ||
afb24528 | 230 | list_for_each_safe(tmp, next, devices) { |
82b1519b MP |
231 | struct dm_dev_internal *dd = |
232 | list_entry(tmp, struct dm_dev_internal, list); | |
1b6da754 JB |
233 | DMWARN("dm_table_destroy: dm_put_device call missing for %s", |
234 | dd->dm_dev.name); | |
1da177e4 LT |
235 | kfree(dd); |
236 | } | |
237 | } | |
238 | ||
d5816876 | 239 | void dm_table_destroy(struct dm_table *t) |
1da177e4 LT |
240 | { |
241 | unsigned int i; | |
242 | ||
d5816876 MP |
243 | while (atomic_read(&t->holders)) |
244 | msleep(1); | |
245 | smp_mb(); | |
246 | ||
1da177e4 LT |
247 | /* free the indexes (see dm_table_complete) */ |
248 | if (t->depth >= 2) | |
249 | vfree(t->index[t->depth - 2]); | |
250 | ||
251 | /* free the targets */ | |
252 | for (i = 0; i < t->num_targets; i++) { | |
253 | struct dm_target *tgt = t->targets + i; | |
254 | ||
255 | if (tgt->type->dtr) | |
256 | tgt->type->dtr(tgt); | |
257 | ||
258 | dm_put_target_type(tgt->type); | |
259 | } | |
260 | ||
261 | vfree(t->highs); | |
262 | ||
263 | /* free the device list */ | |
1b6da754 | 264 | if (t->devices.next != &t->devices) |
1da177e4 | 265 | free_devices(&t->devices); |
1da177e4 LT |
266 | |
267 | kfree(t); | |
268 | } | |
269 | ||
270 | void dm_table_get(struct dm_table *t) | |
271 | { | |
272 | atomic_inc(&t->holders); | |
273 | } | |
274 | ||
275 | void dm_table_put(struct dm_table *t) | |
276 | { | |
277 | if (!t) | |
278 | return; | |
279 | ||
d5816876 MP |
280 | smp_mb__before_atomic_dec(); |
281 | atomic_dec(&t->holders); | |
1da177e4 LT |
282 | } |
283 | ||
284 | /* | |
285 | * Checks to see if we need to extend highs or targets. | |
286 | */ | |
287 | static inline int check_space(struct dm_table *t) | |
288 | { | |
289 | if (t->num_targets >= t->num_allocated) | |
290 | return alloc_targets(t, t->num_allocated * 2); | |
291 | ||
292 | return 0; | |
293 | } | |
294 | ||
1da177e4 LT |
295 | /* |
296 | * See if we've already got a device in the list. | |
297 | */ | |
82b1519b | 298 | static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev) |
1da177e4 | 299 | { |
82b1519b | 300 | struct dm_dev_internal *dd; |
1da177e4 LT |
301 | |
302 | list_for_each_entry (dd, l, list) | |
82b1519b | 303 | if (dd->dm_dev.bdev->bd_dev == dev) |
1da177e4 LT |
304 | return dd; |
305 | ||
306 | return NULL; | |
307 | } | |
308 | ||
309 | /* | |
310 | * Open a device so we can use it as a map destination. | |
311 | */ | |
82b1519b MP |
312 | static int open_dev(struct dm_dev_internal *d, dev_t dev, |
313 | struct mapped_device *md) | |
1da177e4 LT |
314 | { |
315 | static char *_claim_ptr = "I belong to device-mapper"; | |
316 | struct block_device *bdev; | |
317 | ||
318 | int r; | |
319 | ||
82b1519b | 320 | BUG_ON(d->dm_dev.bdev); |
1da177e4 | 321 | |
82b1519b | 322 | bdev = open_by_devnum(dev, d->dm_dev.mode); |
1da177e4 LT |
323 | if (IS_ERR(bdev)) |
324 | return PTR_ERR(bdev); | |
f165921d | 325 | r = bd_claim_by_disk(bdev, _claim_ptr, dm_disk(md)); |
1da177e4 | 326 | if (r) |
9a1c3542 | 327 | blkdev_put(bdev, d->dm_dev.mode); |
1da177e4 | 328 | else |
82b1519b | 329 | d->dm_dev.bdev = bdev; |
1da177e4 LT |
330 | return r; |
331 | } | |
332 | ||
333 | /* | |
334 | * Close a device that we've been using. | |
335 | */ | |
82b1519b | 336 | static void close_dev(struct dm_dev_internal *d, struct mapped_device *md) |
1da177e4 | 337 | { |
82b1519b | 338 | if (!d->dm_dev.bdev) |
1da177e4 LT |
339 | return; |
340 | ||
82b1519b | 341 | bd_release_from_disk(d->dm_dev.bdev, dm_disk(md)); |
9a1c3542 | 342 | blkdev_put(d->dm_dev.bdev, d->dm_dev.mode); |
82b1519b | 343 | d->dm_dev.bdev = NULL; |
1da177e4 LT |
344 | } |
345 | ||
346 | /* | |
2cd54d9b | 347 | * If possible, this checks an area of a destination device is valid. |
1da177e4 | 348 | */ |
02acc3a4 MS |
349 | static int device_area_is_valid(struct dm_target *ti, struct block_device *bdev, |
350 | sector_t start, sector_t len) | |
1da177e4 | 351 | { |
02acc3a4 MS |
352 | sector_t dev_size = i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; |
353 | unsigned short logical_block_size_sectors = | |
354 | ti->limits.logical_block_size >> SECTOR_SHIFT; | |
355 | char b[BDEVNAME_SIZE]; | |
2cd54d9b MA |
356 | |
357 | if (!dev_size) | |
358 | return 1; | |
359 | ||
02acc3a4 MS |
360 | if ((start >= dev_size) || (start + len > dev_size)) { |
361 | DMWARN("%s: %s too small for target", | |
362 | dm_device_name(ti->table->md), bdevname(bdev, b)); | |
363 | return 0; | |
364 | } | |
365 | ||
366 | if (logical_block_size_sectors <= 1) | |
367 | return 1; | |
368 | ||
369 | if (start & (logical_block_size_sectors - 1)) { | |
370 | DMWARN("%s: start=%llu not aligned to h/w " | |
371 | "logical block size %hu of %s", | |
372 | dm_device_name(ti->table->md), | |
373 | (unsigned long long)start, | |
374 | ti->limits.logical_block_size, bdevname(bdev, b)); | |
375 | return 0; | |
376 | } | |
377 | ||
378 | if (len & (logical_block_size_sectors - 1)) { | |
379 | DMWARN("%s: len=%llu not aligned to h/w " | |
380 | "logical block size %hu of %s", | |
381 | dm_device_name(ti->table->md), | |
382 | (unsigned long long)len, | |
383 | ti->limits.logical_block_size, bdevname(bdev, b)); | |
384 | return 0; | |
385 | } | |
386 | ||
387 | return 1; | |
1da177e4 LT |
388 | } |
389 | ||
390 | /* | |
570b9d96 | 391 | * This upgrades the mode on an already open dm_dev, being |
1da177e4 | 392 | * careful to leave things as they were if we fail to reopen the |
570b9d96 AK |
393 | * device and not to touch the existing bdev field in case |
394 | * it is accessed concurrently inside dm_table_any_congested(). | |
1da177e4 | 395 | */ |
aeb5d727 | 396 | static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode, |
82b1519b | 397 | struct mapped_device *md) |
1da177e4 LT |
398 | { |
399 | int r; | |
570b9d96 | 400 | struct dm_dev_internal dd_new, dd_old; |
1da177e4 | 401 | |
570b9d96 AK |
402 | dd_new = dd_old = *dd; |
403 | ||
404 | dd_new.dm_dev.mode |= new_mode; | |
405 | dd_new.dm_dev.bdev = NULL; | |
406 | ||
407 | r = open_dev(&dd_new, dd->dm_dev.bdev->bd_dev, md); | |
408 | if (r) | |
409 | return r; | |
1da177e4 | 410 | |
82b1519b | 411 | dd->dm_dev.mode |= new_mode; |
570b9d96 | 412 | close_dev(&dd_old, md); |
1da177e4 | 413 | |
570b9d96 | 414 | return 0; |
1da177e4 LT |
415 | } |
416 | ||
417 | /* | |
418 | * Add a device to the list, or just increment the usage count if | |
419 | * it's already present. | |
420 | */ | |
421 | static int __table_get_device(struct dm_table *t, struct dm_target *ti, | |
422 | const char *path, sector_t start, sector_t len, | |
aeb5d727 | 423 | fmode_t mode, struct dm_dev **result) |
1da177e4 LT |
424 | { |
425 | int r; | |
69a2ce72 | 426 | dev_t uninitialized_var(dev); |
82b1519b | 427 | struct dm_dev_internal *dd; |
1da177e4 LT |
428 | unsigned int major, minor; |
429 | ||
547bc926 | 430 | BUG_ON(!t); |
1da177e4 LT |
431 | |
432 | if (sscanf(path, "%u:%u", &major, &minor) == 2) { | |
433 | /* Extract the major/minor numbers */ | |
434 | dev = MKDEV(major, minor); | |
435 | if (MAJOR(dev) != major || MINOR(dev) != minor) | |
436 | return -EOVERFLOW; | |
437 | } else { | |
438 | /* convert the path to a device */ | |
72e8264e CH |
439 | struct block_device *bdev = lookup_bdev(path); |
440 | ||
441 | if (IS_ERR(bdev)) | |
442 | return PTR_ERR(bdev); | |
443 | dev = bdev->bd_dev; | |
444 | bdput(bdev); | |
1da177e4 LT |
445 | } |
446 | ||
447 | dd = find_device(&t->devices, dev); | |
448 | if (!dd) { | |
449 | dd = kmalloc(sizeof(*dd), GFP_KERNEL); | |
450 | if (!dd) | |
451 | return -ENOMEM; | |
452 | ||
82b1519b MP |
453 | dd->dm_dev.mode = mode; |
454 | dd->dm_dev.bdev = NULL; | |
1da177e4 | 455 | |
f165921d | 456 | if ((r = open_dev(dd, dev, t->md))) { |
1da177e4 LT |
457 | kfree(dd); |
458 | return r; | |
459 | } | |
460 | ||
82b1519b | 461 | format_dev_t(dd->dm_dev.name, dev); |
1da177e4 LT |
462 | |
463 | atomic_set(&dd->count, 0); | |
464 | list_add(&dd->list, &t->devices); | |
465 | ||
82b1519b | 466 | } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) { |
f165921d | 467 | r = upgrade_mode(dd, mode, t->md); |
1da177e4 LT |
468 | if (r) |
469 | return r; | |
470 | } | |
471 | atomic_inc(&dd->count); | |
472 | ||
82b1519b | 473 | *result = &dd->dm_dev; |
1da177e4 LT |
474 | return 0; |
475 | } | |
476 | ||
5ab97588 MS |
477 | /* |
478 | * Returns the minimum that is _not_ zero, unless both are zero. | |
479 | */ | |
480 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
481 | ||
3cb40214 | 482 | void dm_set_device_limits(struct dm_target *ti, struct block_device *bdev) |
1da177e4 | 483 | { |
165125e1 | 484 | struct request_queue *q = bdev_get_queue(bdev); |
0c2322e4 AK |
485 | char b[BDEVNAME_SIZE]; |
486 | ||
487 | if (unlikely(!q)) { | |
488 | DMWARN("%s: Cannot set limits for nonexistent device %s", | |
489 | dm_device_name(ti->table->md), bdevname(bdev, b)); | |
490 | return; | |
491 | } | |
3cb40214 | 492 | |
5ab97588 MS |
493 | if (blk_stack_limits(&ti->limits, &q->limits, 0) < 0) |
494 | DMWARN("%s: target device %s is misaligned", | |
495 | dm_device_name(ti->table->md), bdevname(bdev, b)); | |
3cb40214 | 496 | |
9980c638 MB |
497 | /* |
498 | * Check if merge fn is supported. | |
499 | * If not we'll force DM to use PAGE_SIZE or | |
500 | * smaller I/O, just to be safe. | |
3cb40214 | 501 | */ |
9980c638 MB |
502 | |
503 | if (q->merge_bvec_fn && !ti->type->merge) | |
5ab97588 MS |
504 | ti->limits.max_sectors = |
505 | min_not_zero(ti->limits.max_sectors, | |
3cb40214 | 506 | (unsigned int) (PAGE_SIZE >> 9)); |
3cb40214 BR |
507 | } |
508 | EXPORT_SYMBOL_GPL(dm_set_device_limits); | |
969429b5 | 509 | |
3cb40214 | 510 | int dm_get_device(struct dm_target *ti, const char *path, sector_t start, |
aeb5d727 | 511 | sector_t len, fmode_t mode, struct dm_dev **result) |
3cb40214 BR |
512 | { |
513 | int r = __table_get_device(ti->table, ti, path, | |
514 | start, len, mode, result); | |
515 | ||
02acc3a4 MS |
516 | if (r) |
517 | return r; | |
518 | ||
519 | dm_set_device_limits(ti, (*result)->bdev); | |
520 | ||
521 | if (!device_area_is_valid(ti, (*result)->bdev, start, len)) { | |
522 | dm_put_device(ti, *result); | |
523 | *result = NULL; | |
524 | return -EINVAL; | |
525 | } | |
1da177e4 LT |
526 | |
527 | return r; | |
528 | } | |
529 | ||
530 | /* | |
531 | * Decrement a devices use count and remove it if necessary. | |
532 | */ | |
82b1519b | 533 | void dm_put_device(struct dm_target *ti, struct dm_dev *d) |
1da177e4 | 534 | { |
82b1519b MP |
535 | struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal, |
536 | dm_dev); | |
537 | ||
1da177e4 | 538 | if (atomic_dec_and_test(&dd->count)) { |
f165921d | 539 | close_dev(dd, ti->table->md); |
1da177e4 LT |
540 | list_del(&dd->list); |
541 | kfree(dd); | |
542 | } | |
543 | } | |
544 | ||
545 | /* | |
546 | * Checks to see if the target joins onto the end of the table. | |
547 | */ | |
548 | static int adjoin(struct dm_table *table, struct dm_target *ti) | |
549 | { | |
550 | struct dm_target *prev; | |
551 | ||
552 | if (!table->num_targets) | |
553 | return !ti->begin; | |
554 | ||
555 | prev = &table->targets[table->num_targets - 1]; | |
556 | return (ti->begin == (prev->begin + prev->len)); | |
557 | } | |
558 | ||
559 | /* | |
560 | * Used to dynamically allocate the arg array. | |
561 | */ | |
562 | static char **realloc_argv(unsigned *array_size, char **old_argv) | |
563 | { | |
564 | char **argv; | |
565 | unsigned new_size; | |
566 | ||
567 | new_size = *array_size ? *array_size * 2 : 64; | |
568 | argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL); | |
569 | if (argv) { | |
570 | memcpy(argv, old_argv, *array_size * sizeof(*argv)); | |
571 | *array_size = new_size; | |
572 | } | |
573 | ||
574 | kfree(old_argv); | |
575 | return argv; | |
576 | } | |
577 | ||
578 | /* | |
579 | * Destructively splits up the argument list to pass to ctr. | |
580 | */ | |
581 | int dm_split_args(int *argc, char ***argvp, char *input) | |
582 | { | |
583 | char *start, *end = input, *out, **argv = NULL; | |
584 | unsigned array_size = 0; | |
585 | ||
586 | *argc = 0; | |
814d6862 DT |
587 | |
588 | if (!input) { | |
589 | *argvp = NULL; | |
590 | return 0; | |
591 | } | |
592 | ||
1da177e4 LT |
593 | argv = realloc_argv(&array_size, argv); |
594 | if (!argv) | |
595 | return -ENOMEM; | |
596 | ||
597 | while (1) { | |
598 | start = end; | |
599 | ||
600 | /* Skip whitespace */ | |
601 | while (*start && isspace(*start)) | |
602 | start++; | |
603 | ||
604 | if (!*start) | |
605 | break; /* success, we hit the end */ | |
606 | ||
607 | /* 'out' is used to remove any back-quotes */ | |
608 | end = out = start; | |
609 | while (*end) { | |
610 | /* Everything apart from '\0' can be quoted */ | |
611 | if (*end == '\\' && *(end + 1)) { | |
612 | *out++ = *(end + 1); | |
613 | end += 2; | |
614 | continue; | |
615 | } | |
616 | ||
617 | if (isspace(*end)) | |
618 | break; /* end of token */ | |
619 | ||
620 | *out++ = *end++; | |
621 | } | |
622 | ||
623 | /* have we already filled the array ? */ | |
624 | if ((*argc + 1) > array_size) { | |
625 | argv = realloc_argv(&array_size, argv); | |
626 | if (!argv) | |
627 | return -ENOMEM; | |
628 | } | |
629 | ||
630 | /* we know this is whitespace */ | |
631 | if (*end) | |
632 | end++; | |
633 | ||
634 | /* terminate the string and put it in the array */ | |
635 | *out = '\0'; | |
636 | argv[*argc] = start; | |
637 | (*argc)++; | |
638 | } | |
639 | ||
640 | *argvp = argv; | |
641 | return 0; | |
642 | } | |
643 | ||
5ab97588 MS |
644 | static void init_valid_queue_limits(struct queue_limits *limits) |
645 | { | |
646 | if (!limits->max_sectors) | |
647 | limits->max_sectors = SAFE_MAX_SECTORS; | |
648 | if (!limits->max_hw_sectors) | |
649 | limits->max_hw_sectors = SAFE_MAX_SECTORS; | |
650 | if (!limits->max_phys_segments) | |
651 | limits->max_phys_segments = MAX_PHYS_SEGMENTS; | |
652 | if (!limits->max_hw_segments) | |
653 | limits->max_hw_segments = MAX_HW_SEGMENTS; | |
654 | if (!limits->logical_block_size) | |
655 | limits->logical_block_size = 1 << SECTOR_SHIFT; | |
656 | if (!limits->physical_block_size) | |
657 | limits->physical_block_size = 1 << SECTOR_SHIFT; | |
658 | if (!limits->io_min) | |
659 | limits->io_min = 1 << SECTOR_SHIFT; | |
660 | if (!limits->max_segment_size) | |
661 | limits->max_segment_size = MAX_SEGMENT_SIZE; | |
662 | if (!limits->seg_boundary_mask) | |
663 | limits->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK; | |
664 | if (!limits->bounce_pfn) | |
665 | limits->bounce_pfn = -1; | |
666 | /* | |
667 | * The other fields (alignment_offset, io_opt, misaligned) | |
668 | * hold 0 from the kzalloc(). | |
669 | */ | |
1da177e4 LT |
670 | } |
671 | ||
be6d4305 MS |
672 | /* |
673 | * Impose necessary and sufficient conditions on a devices's table such | |
674 | * that any incoming bio which respects its logical_block_size can be | |
675 | * processed successfully. If it falls across the boundary between | |
676 | * two or more targets, the size of each piece it gets split into must | |
677 | * be compatible with the logical_block_size of the target processing it. | |
678 | */ | |
679 | static int validate_hardware_logical_block_alignment(struct dm_table *table) | |
680 | { | |
681 | /* | |
682 | * This function uses arithmetic modulo the logical_block_size | |
683 | * (in units of 512-byte sectors). | |
684 | */ | |
685 | unsigned short device_logical_block_size_sects = | |
686 | table->limits.logical_block_size >> SECTOR_SHIFT; | |
687 | ||
688 | /* | |
689 | * Offset of the start of the next table entry, mod logical_block_size. | |
690 | */ | |
691 | unsigned short next_target_start = 0; | |
692 | ||
693 | /* | |
694 | * Given an aligned bio that extends beyond the end of a | |
695 | * target, how many sectors must the next target handle? | |
696 | */ | |
697 | unsigned short remaining = 0; | |
698 | ||
699 | struct dm_target *uninitialized_var(ti); | |
700 | unsigned i = 0; | |
701 | ||
702 | /* | |
703 | * Check each entry in the table in turn. | |
704 | */ | |
705 | while (i < dm_table_get_num_targets(table)) { | |
706 | ti = dm_table_get_target(table, i++); | |
707 | ||
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 && | |
713 | remaining & ((ti->limits.logical_block_size >> | |
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) " | |
726 | "not aligned to hardware logical block size %hu", | |
727 | dm_device_name(table->md), i, | |
728 | (unsigned long long) ti->begin, | |
729 | (unsigned long long) ti->len, | |
730 | table->limits.logical_block_size); | |
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 | ||
744 | if ((r = check_space(t))) | |
745 | return r; | |
746 | ||
747 | tgt = t->targets + t->num_targets; | |
748 | memset(tgt, 0, sizeof(*tgt)); | |
749 | ||
750 | if (!len) { | |
72d94861 | 751 | DMERR("%s: zero-length target", dm_device_name(t->md)); |
1da177e4 LT |
752 | return -EINVAL; |
753 | } | |
754 | ||
755 | tgt->type = dm_get_target_type(type); | |
756 | if (!tgt->type) { | |
72d94861 AK |
757 | DMERR("%s: %s: unknown target type", dm_device_name(t->md), |
758 | type); | |
1da177e4 LT |
759 | return -EINVAL; |
760 | } | |
761 | ||
762 | tgt->table = t; | |
763 | tgt->begin = start; | |
764 | tgt->len = len; | |
765 | tgt->error = "Unknown error"; | |
766 | ||
767 | /* | |
768 | * Does this target adjoin the previous one ? | |
769 | */ | |
770 | if (!adjoin(t, tgt)) { | |
771 | tgt->error = "Gap in table"; | |
772 | r = -EINVAL; | |
773 | goto bad; | |
774 | } | |
775 | ||
776 | r = dm_split_args(&argc, &argv, params); | |
777 | if (r) { | |
778 | tgt->error = "couldn't split parameters (insufficient memory)"; | |
779 | goto bad; | |
780 | } | |
781 | ||
782 | r = tgt->type->ctr(tgt, argc, argv); | |
783 | kfree(argv); | |
784 | if (r) | |
785 | goto bad; | |
786 | ||
787 | t->highs[t->num_targets++] = tgt->begin + tgt->len - 1; | |
788 | ||
5ab97588 MS |
789 | if (blk_stack_limits(&t->limits, &tgt->limits, 0) < 0) |
790 | DMWARN("%s: target device (start sect %llu len %llu) " | |
791 | "is misaligned", | |
792 | dm_device_name(t->md), | |
793 | (unsigned long long) tgt->begin, | |
794 | (unsigned long long) tgt->len); | |
1da177e4 LT |
795 | return 0; |
796 | ||
797 | bad: | |
72d94861 | 798 | DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error); |
1da177e4 LT |
799 | dm_put_target_type(tgt->type); |
800 | return r; | |
801 | } | |
802 | ||
803 | static int setup_indexes(struct dm_table *t) | |
804 | { | |
805 | int i; | |
806 | unsigned int total = 0; | |
807 | sector_t *indexes; | |
808 | ||
809 | /* allocate the space for *all* the indexes */ | |
810 | for (i = t->depth - 2; i >= 0; i--) { | |
811 | t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE); | |
812 | total += t->counts[i]; | |
813 | } | |
814 | ||
815 | indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE); | |
816 | if (!indexes) | |
817 | return -ENOMEM; | |
818 | ||
819 | /* set up internal nodes, bottom-up */ | |
82d601dc | 820 | for (i = t->depth - 2; i >= 0; i--) { |
1da177e4 LT |
821 | t->index[i] = indexes; |
822 | indexes += (KEYS_PER_NODE * t->counts[i]); | |
823 | setup_btree_index(i, t); | |
824 | } | |
825 | ||
826 | return 0; | |
827 | } | |
828 | ||
829 | /* | |
830 | * Builds the btree to index the map. | |
831 | */ | |
832 | int dm_table_complete(struct dm_table *t) | |
833 | { | |
834 | int r = 0; | |
835 | unsigned int leaf_nodes; | |
836 | ||
5ab97588 | 837 | init_valid_queue_limits(&t->limits); |
1da177e4 | 838 | |
be6d4305 MS |
839 | r = validate_hardware_logical_block_alignment(t); |
840 | if (r) | |
841 | return r; | |
842 | ||
1da177e4 LT |
843 | /* how many indexes will the btree have ? */ |
844 | leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE); | |
845 | t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE); | |
846 | ||
847 | /* leaf layer has already been set up */ | |
848 | t->counts[t->depth - 1] = leaf_nodes; | |
849 | t->index[t->depth - 1] = t->highs; | |
850 | ||
851 | if (t->depth >= 2) | |
852 | r = setup_indexes(t); | |
853 | ||
854 | return r; | |
855 | } | |
856 | ||
48c9c27b | 857 | static DEFINE_MUTEX(_event_lock); |
1da177e4 LT |
858 | void dm_table_event_callback(struct dm_table *t, |
859 | void (*fn)(void *), void *context) | |
860 | { | |
48c9c27b | 861 | mutex_lock(&_event_lock); |
1da177e4 LT |
862 | t->event_fn = fn; |
863 | t->event_context = context; | |
48c9c27b | 864 | mutex_unlock(&_event_lock); |
1da177e4 LT |
865 | } |
866 | ||
867 | void dm_table_event(struct dm_table *t) | |
868 | { | |
869 | /* | |
870 | * You can no longer call dm_table_event() from interrupt | |
871 | * context, use a bottom half instead. | |
872 | */ | |
873 | BUG_ON(in_interrupt()); | |
874 | ||
48c9c27b | 875 | mutex_lock(&_event_lock); |
1da177e4 LT |
876 | if (t->event_fn) |
877 | t->event_fn(t->event_context); | |
48c9c27b | 878 | mutex_unlock(&_event_lock); |
1da177e4 LT |
879 | } |
880 | ||
881 | sector_t dm_table_get_size(struct dm_table *t) | |
882 | { | |
883 | return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0; | |
884 | } | |
885 | ||
886 | struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index) | |
887 | { | |
14353539 | 888 | if (index >= t->num_targets) |
1da177e4 LT |
889 | return NULL; |
890 | ||
891 | return t->targets + index; | |
892 | } | |
893 | ||
894 | /* | |
895 | * Search the btree for the correct target. | |
512875bd JN |
896 | * |
897 | * Caller should check returned pointer with dm_target_is_valid() | |
898 | * to trap I/O beyond end of device. | |
1da177e4 LT |
899 | */ |
900 | struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector) | |
901 | { | |
902 | unsigned int l, n = 0, k = 0; | |
903 | sector_t *node; | |
904 | ||
905 | for (l = 0; l < t->depth; l++) { | |
906 | n = get_child(n, k); | |
907 | node = get_node(t, l, n); | |
908 | ||
909 | for (k = 0; k < KEYS_PER_NODE; k++) | |
910 | if (node[k] >= sector) | |
911 | break; | |
912 | } | |
913 | ||
914 | return &t->targets[(KEYS_PER_NODE * n) + k]; | |
915 | } | |
916 | ||
9c47008d MP |
917 | /* |
918 | * Set the integrity profile for this device if all devices used have | |
919 | * matching profiles. | |
920 | */ | |
921 | static void dm_table_set_integrity(struct dm_table *t) | |
922 | { | |
923 | struct list_head *devices = dm_table_get_devices(t); | |
924 | struct dm_dev_internal *prev = NULL, *dd = NULL; | |
925 | ||
926 | if (!blk_get_integrity(dm_disk(t->md))) | |
927 | return; | |
928 | ||
929 | list_for_each_entry(dd, devices, list) { | |
930 | if (prev && | |
931 | blk_integrity_compare(prev->dm_dev.bdev->bd_disk, | |
932 | dd->dm_dev.bdev->bd_disk) < 0) { | |
933 | DMWARN("%s: integrity not set: %s and %s mismatch", | |
934 | dm_device_name(t->md), | |
935 | prev->dm_dev.bdev->bd_disk->disk_name, | |
936 | dd->dm_dev.bdev->bd_disk->disk_name); | |
937 | goto no_integrity; | |
938 | } | |
939 | prev = dd; | |
940 | } | |
941 | ||
942 | if (!prev || !bdev_get_integrity(prev->dm_dev.bdev)) | |
943 | goto no_integrity; | |
944 | ||
945 | blk_integrity_register(dm_disk(t->md), | |
946 | bdev_get_integrity(prev->dm_dev.bdev)); | |
947 | ||
948 | return; | |
949 | ||
950 | no_integrity: | |
951 | blk_integrity_register(dm_disk(t->md), NULL); | |
952 | ||
953 | return; | |
954 | } | |
955 | ||
1da177e4 LT |
956 | void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q) |
957 | { | |
958 | /* | |
959 | * Make sure we obey the optimistic sub devices | |
960 | * restrictions. | |
961 | */ | |
962 | blk_queue_max_sectors(q, t->limits.max_sectors); | |
ae03bf63 MP |
963 | blk_queue_max_phys_segments(q, t->limits.max_phys_segments); |
964 | blk_queue_max_hw_segments(q, t->limits.max_hw_segments); | |
965 | blk_queue_logical_block_size(q, t->limits.logical_block_size); | |
966 | blk_queue_max_segment_size(q, t->limits.max_segment_size); | |
967 | blk_queue_max_hw_sectors(q, t->limits.max_hw_sectors); | |
968 | blk_queue_segment_boundary(q, t->limits.seg_boundary_mask); | |
9df1bb9b | 969 | blk_queue_bounce_limit(q, t->limits.bounce_pfn); |
c9a3f6d6 | 970 | |
969429b5 | 971 | if (t->limits.no_cluster) |
c9a3f6d6 | 972 | queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q); |
969429b5 | 973 | else |
c9a3f6d6 | 974 | queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, q); |
969429b5 | 975 | |
9c47008d | 976 | dm_table_set_integrity(t); |
1da177e4 LT |
977 | } |
978 | ||
979 | unsigned int dm_table_get_num_targets(struct dm_table *t) | |
980 | { | |
981 | return t->num_targets; | |
982 | } | |
983 | ||
984 | struct list_head *dm_table_get_devices(struct dm_table *t) | |
985 | { | |
986 | return &t->devices; | |
987 | } | |
988 | ||
aeb5d727 | 989 | fmode_t dm_table_get_mode(struct dm_table *t) |
1da177e4 LT |
990 | { |
991 | return t->mode; | |
992 | } | |
993 | ||
994 | static void suspend_targets(struct dm_table *t, unsigned postsuspend) | |
995 | { | |
996 | int i = t->num_targets; | |
997 | struct dm_target *ti = t->targets; | |
998 | ||
999 | while (i--) { | |
1000 | if (postsuspend) { | |
1001 | if (ti->type->postsuspend) | |
1002 | ti->type->postsuspend(ti); | |
1003 | } else if (ti->type->presuspend) | |
1004 | ti->type->presuspend(ti); | |
1005 | ||
1006 | ti++; | |
1007 | } | |
1008 | } | |
1009 | ||
1010 | void dm_table_presuspend_targets(struct dm_table *t) | |
1011 | { | |
cf222b37 AK |
1012 | if (!t) |
1013 | return; | |
1014 | ||
e8488d08 | 1015 | suspend_targets(t, 0); |
1da177e4 LT |
1016 | } |
1017 | ||
1018 | void dm_table_postsuspend_targets(struct dm_table *t) | |
1019 | { | |
cf222b37 AK |
1020 | if (!t) |
1021 | return; | |
1022 | ||
e8488d08 | 1023 | suspend_targets(t, 1); |
1da177e4 LT |
1024 | } |
1025 | ||
8757b776 | 1026 | int dm_table_resume_targets(struct dm_table *t) |
1da177e4 | 1027 | { |
8757b776 MB |
1028 | int i, r = 0; |
1029 | ||
1030 | for (i = 0; i < t->num_targets; i++) { | |
1031 | struct dm_target *ti = t->targets + i; | |
1032 | ||
1033 | if (!ti->type->preresume) | |
1034 | continue; | |
1035 | ||
1036 | r = ti->type->preresume(ti); | |
1037 | if (r) | |
1038 | return r; | |
1039 | } | |
1da177e4 LT |
1040 | |
1041 | for (i = 0; i < t->num_targets; i++) { | |
1042 | struct dm_target *ti = t->targets + i; | |
1043 | ||
1044 | if (ti->type->resume) | |
1045 | ti->type->resume(ti); | |
1046 | } | |
8757b776 MB |
1047 | |
1048 | return 0; | |
1da177e4 LT |
1049 | } |
1050 | ||
1051 | int dm_table_any_congested(struct dm_table *t, int bdi_bits) | |
1052 | { | |
82b1519b | 1053 | struct dm_dev_internal *dd; |
afb24528 | 1054 | struct list_head *devices = dm_table_get_devices(t); |
1da177e4 LT |
1055 | int r = 0; |
1056 | ||
afb24528 | 1057 | list_for_each_entry(dd, devices, list) { |
82b1519b | 1058 | struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev); |
0c2322e4 AK |
1059 | char b[BDEVNAME_SIZE]; |
1060 | ||
1061 | if (likely(q)) | |
1062 | r |= bdi_congested(&q->backing_dev_info, bdi_bits); | |
1063 | else | |
1064 | DMWARN_LIMIT("%s: any_congested: nonexistent device %s", | |
1065 | dm_device_name(t->md), | |
1066 | bdevname(dd->dm_dev.bdev, b)); | |
1da177e4 LT |
1067 | } |
1068 | ||
1069 | return r; | |
1070 | } | |
1071 | ||
1072 | void dm_table_unplug_all(struct dm_table *t) | |
1073 | { | |
82b1519b | 1074 | struct dm_dev_internal *dd; |
afb24528 | 1075 | struct list_head *devices = dm_table_get_devices(t); |
1da177e4 | 1076 | |
afb24528 | 1077 | list_for_each_entry(dd, devices, list) { |
82b1519b | 1078 | struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev); |
0c2322e4 AK |
1079 | char b[BDEVNAME_SIZE]; |
1080 | ||
1081 | if (likely(q)) | |
1082 | blk_unplug(q); | |
1083 | else | |
1084 | DMWARN_LIMIT("%s: Cannot unplug nonexistent device %s", | |
1085 | dm_device_name(t->md), | |
1086 | bdevname(dd->dm_dev.bdev, b)); | |
1da177e4 LT |
1087 | } |
1088 | } | |
1089 | ||
1134e5ae MA |
1090 | struct mapped_device *dm_table_get_md(struct dm_table *t) |
1091 | { | |
1092 | dm_get(t->md); | |
1093 | ||
1094 | return t->md; | |
1095 | } | |
1096 | ||
1da177e4 LT |
1097 | EXPORT_SYMBOL(dm_vcalloc); |
1098 | EXPORT_SYMBOL(dm_get_device); | |
1099 | EXPORT_SYMBOL(dm_put_device); | |
1100 | EXPORT_SYMBOL(dm_table_event); | |
d5e404c1 | 1101 | EXPORT_SYMBOL(dm_table_get_size); |
1da177e4 | 1102 | EXPORT_SYMBOL(dm_table_get_mode); |
1134e5ae | 1103 | EXPORT_SYMBOL(dm_table_get_md); |
1da177e4 LT |
1104 | EXPORT_SYMBOL(dm_table_put); |
1105 | EXPORT_SYMBOL(dm_table_get); | |
1106 | EXPORT_SYMBOL(dm_table_unplug_all); |