]>
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 | ||
ab4c1424 AK |
55 | unsigned barriers_supported:1; |
56 | ||
1da177e4 LT |
57 | /* |
58 | * Indicates the rw permissions for the new logical | |
59 | * device. This should be a combination of FMODE_READ | |
60 | * and FMODE_WRITE. | |
61 | */ | |
aeb5d727 | 62 | fmode_t mode; |
1da177e4 LT |
63 | |
64 | /* a list of devices used by this table */ | |
65 | struct list_head devices; | |
66 | ||
67 | /* | |
68 | * These are optimistic limits taken from all the | |
69 | * targets, some targets will need smaller limits. | |
70 | */ | |
71 | struct io_restrictions limits; | |
72 | ||
73 | /* events get handed up using this callback */ | |
74 | void (*event_fn)(void *); | |
75 | void *event_context; | |
76 | }; | |
77 | ||
78 | /* | |
79 | * Similar to ceiling(log_size(n)) | |
80 | */ | |
81 | static unsigned int int_log(unsigned int n, unsigned int base) | |
82 | { | |
83 | int result = 0; | |
84 | ||
85 | while (n > 1) { | |
86 | n = dm_div_up(n, base); | |
87 | result++; | |
88 | } | |
89 | ||
90 | return result; | |
91 | } | |
92 | ||
93 | /* | |
94 | * Returns the minimum that is _not_ zero, unless both are zero. | |
95 | */ | |
96 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
97 | ||
98 | /* | |
99 | * Combine two io_restrictions, always taking the lower value. | |
100 | */ | |
101 | static void combine_restrictions_low(struct io_restrictions *lhs, | |
102 | struct io_restrictions *rhs) | |
103 | { | |
104 | lhs->max_sectors = | |
105 | min_not_zero(lhs->max_sectors, rhs->max_sectors); | |
106 | ||
107 | lhs->max_phys_segments = | |
108 | min_not_zero(lhs->max_phys_segments, rhs->max_phys_segments); | |
109 | ||
110 | lhs->max_hw_segments = | |
111 | min_not_zero(lhs->max_hw_segments, rhs->max_hw_segments); | |
112 | ||
113 | lhs->hardsect_size = max(lhs->hardsect_size, rhs->hardsect_size); | |
114 | ||
115 | lhs->max_segment_size = | |
116 | min_not_zero(lhs->max_segment_size, rhs->max_segment_size); | |
117 | ||
91212507 NB |
118 | lhs->max_hw_sectors = |
119 | min_not_zero(lhs->max_hw_sectors, rhs->max_hw_sectors); | |
120 | ||
1da177e4 LT |
121 | lhs->seg_boundary_mask = |
122 | min_not_zero(lhs->seg_boundary_mask, rhs->seg_boundary_mask); | |
969429b5 | 123 | |
5ec140e6 VA |
124 | lhs->bounce_pfn = min_not_zero(lhs->bounce_pfn, rhs->bounce_pfn); |
125 | ||
969429b5 | 126 | lhs->no_cluster |= rhs->no_cluster; |
1da177e4 LT |
127 | } |
128 | ||
129 | /* | |
130 | * Calculate the index of the child node of the n'th node k'th key. | |
131 | */ | |
132 | static inline unsigned int get_child(unsigned int n, unsigned int k) | |
133 | { | |
134 | return (n * CHILDREN_PER_NODE) + k; | |
135 | } | |
136 | ||
137 | /* | |
138 | * Return the n'th node of level l from table t. | |
139 | */ | |
140 | static inline sector_t *get_node(struct dm_table *t, | |
141 | unsigned int l, unsigned int n) | |
142 | { | |
143 | return t->index[l] + (n * KEYS_PER_NODE); | |
144 | } | |
145 | ||
146 | /* | |
147 | * Return the highest key that you could lookup from the n'th | |
148 | * node on level l of the btree. | |
149 | */ | |
150 | static sector_t high(struct dm_table *t, unsigned int l, unsigned int n) | |
151 | { | |
152 | for (; l < t->depth - 1; l++) | |
153 | n = get_child(n, CHILDREN_PER_NODE - 1); | |
154 | ||
155 | if (n >= t->counts[l]) | |
156 | return (sector_t) - 1; | |
157 | ||
158 | return get_node(t, l, n)[KEYS_PER_NODE - 1]; | |
159 | } | |
160 | ||
161 | /* | |
162 | * Fills in a level of the btree based on the highs of the level | |
163 | * below it. | |
164 | */ | |
165 | static int setup_btree_index(unsigned int l, struct dm_table *t) | |
166 | { | |
167 | unsigned int n, k; | |
168 | sector_t *node; | |
169 | ||
170 | for (n = 0U; n < t->counts[l]; n++) { | |
171 | node = get_node(t, l, n); | |
172 | ||
173 | for (k = 0U; k < KEYS_PER_NODE; k++) | |
174 | node[k] = high(t, l + 1, get_child(n, k)); | |
175 | } | |
176 | ||
177 | return 0; | |
178 | } | |
179 | ||
180 | void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size) | |
181 | { | |
182 | unsigned long size; | |
183 | void *addr; | |
184 | ||
185 | /* | |
186 | * Check that we're not going to overflow. | |
187 | */ | |
188 | if (nmemb > (ULONG_MAX / elem_size)) | |
189 | return NULL; | |
190 | ||
191 | size = nmemb * elem_size; | |
192 | addr = vmalloc(size); | |
193 | if (addr) | |
194 | memset(addr, 0, size); | |
195 | ||
196 | return addr; | |
197 | } | |
198 | ||
199 | /* | |
200 | * highs, and targets are managed as dynamic arrays during a | |
201 | * table load. | |
202 | */ | |
203 | static int alloc_targets(struct dm_table *t, unsigned int num) | |
204 | { | |
205 | sector_t *n_highs; | |
206 | struct dm_target *n_targets; | |
207 | int n = t->num_targets; | |
208 | ||
209 | /* | |
210 | * Allocate both the target array and offset array at once. | |
512875bd JN |
211 | * Append an empty entry to catch sectors beyond the end of |
212 | * the device. | |
1da177e4 | 213 | */ |
512875bd | 214 | n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) + |
1da177e4 LT |
215 | sizeof(sector_t)); |
216 | if (!n_highs) | |
217 | return -ENOMEM; | |
218 | ||
219 | n_targets = (struct dm_target *) (n_highs + num); | |
220 | ||
221 | if (n) { | |
222 | memcpy(n_highs, t->highs, sizeof(*n_highs) * n); | |
223 | memcpy(n_targets, t->targets, sizeof(*n_targets) * n); | |
224 | } | |
225 | ||
226 | memset(n_highs + n, -1, sizeof(*n_highs) * (num - n)); | |
227 | vfree(t->highs); | |
228 | ||
229 | t->num_allocated = num; | |
230 | t->highs = n_highs; | |
231 | t->targets = n_targets; | |
232 | ||
233 | return 0; | |
234 | } | |
235 | ||
aeb5d727 | 236 | int dm_table_create(struct dm_table **result, fmode_t mode, |
1134e5ae | 237 | unsigned num_targets, struct mapped_device *md) |
1da177e4 | 238 | { |
094262db | 239 | struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL); |
1da177e4 LT |
240 | |
241 | if (!t) | |
242 | return -ENOMEM; | |
243 | ||
1da177e4 | 244 | INIT_LIST_HEAD(&t->devices); |
d5816876 | 245 | atomic_set(&t->holders, 0); |
ab4c1424 | 246 | t->barriers_supported = 1; |
1da177e4 LT |
247 | |
248 | if (!num_targets) | |
249 | num_targets = KEYS_PER_NODE; | |
250 | ||
251 | num_targets = dm_round_up(num_targets, KEYS_PER_NODE); | |
252 | ||
253 | if (alloc_targets(t, num_targets)) { | |
254 | kfree(t); | |
255 | t = NULL; | |
256 | return -ENOMEM; | |
257 | } | |
258 | ||
259 | t->mode = mode; | |
1134e5ae | 260 | t->md = md; |
1da177e4 LT |
261 | *result = t; |
262 | return 0; | |
263 | } | |
264 | ||
265 | static void free_devices(struct list_head *devices) | |
266 | { | |
267 | struct list_head *tmp, *next; | |
268 | ||
afb24528 | 269 | list_for_each_safe(tmp, next, devices) { |
82b1519b MP |
270 | struct dm_dev_internal *dd = |
271 | list_entry(tmp, struct dm_dev_internal, list); | |
1da177e4 LT |
272 | kfree(dd); |
273 | } | |
274 | } | |
275 | ||
d5816876 | 276 | void dm_table_destroy(struct dm_table *t) |
1da177e4 LT |
277 | { |
278 | unsigned int i; | |
279 | ||
d5816876 MP |
280 | while (atomic_read(&t->holders)) |
281 | msleep(1); | |
282 | smp_mb(); | |
283 | ||
1da177e4 LT |
284 | /* free the indexes (see dm_table_complete) */ |
285 | if (t->depth >= 2) | |
286 | vfree(t->index[t->depth - 2]); | |
287 | ||
288 | /* free the targets */ | |
289 | for (i = 0; i < t->num_targets; i++) { | |
290 | struct dm_target *tgt = t->targets + i; | |
291 | ||
292 | if (tgt->type->dtr) | |
293 | tgt->type->dtr(tgt); | |
294 | ||
295 | dm_put_target_type(tgt->type); | |
296 | } | |
297 | ||
298 | vfree(t->highs); | |
299 | ||
300 | /* free the device list */ | |
301 | if (t->devices.next != &t->devices) { | |
302 | DMWARN("devices still present during destroy: " | |
303 | "dm_table_remove_device calls missing"); | |
304 | ||
305 | free_devices(&t->devices); | |
306 | } | |
307 | ||
308 | kfree(t); | |
309 | } | |
310 | ||
311 | void dm_table_get(struct dm_table *t) | |
312 | { | |
313 | atomic_inc(&t->holders); | |
314 | } | |
315 | ||
316 | void dm_table_put(struct dm_table *t) | |
317 | { | |
318 | if (!t) | |
319 | return; | |
320 | ||
d5816876 MP |
321 | smp_mb__before_atomic_dec(); |
322 | atomic_dec(&t->holders); | |
1da177e4 LT |
323 | } |
324 | ||
325 | /* | |
326 | * Checks to see if we need to extend highs or targets. | |
327 | */ | |
328 | static inline int check_space(struct dm_table *t) | |
329 | { | |
330 | if (t->num_targets >= t->num_allocated) | |
331 | return alloc_targets(t, t->num_allocated * 2); | |
332 | ||
333 | return 0; | |
334 | } | |
335 | ||
1da177e4 LT |
336 | /* |
337 | * See if we've already got a device in the list. | |
338 | */ | |
82b1519b | 339 | static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev) |
1da177e4 | 340 | { |
82b1519b | 341 | struct dm_dev_internal *dd; |
1da177e4 LT |
342 | |
343 | list_for_each_entry (dd, l, list) | |
82b1519b | 344 | if (dd->dm_dev.bdev->bd_dev == dev) |
1da177e4 LT |
345 | return dd; |
346 | ||
347 | return NULL; | |
348 | } | |
349 | ||
350 | /* | |
351 | * Open a device so we can use it as a map destination. | |
352 | */ | |
82b1519b MP |
353 | static int open_dev(struct dm_dev_internal *d, dev_t dev, |
354 | struct mapped_device *md) | |
1da177e4 LT |
355 | { |
356 | static char *_claim_ptr = "I belong to device-mapper"; | |
357 | struct block_device *bdev; | |
358 | ||
359 | int r; | |
360 | ||
82b1519b | 361 | BUG_ON(d->dm_dev.bdev); |
1da177e4 | 362 | |
82b1519b | 363 | bdev = open_by_devnum(dev, d->dm_dev.mode); |
1da177e4 LT |
364 | if (IS_ERR(bdev)) |
365 | return PTR_ERR(bdev); | |
f165921d | 366 | r = bd_claim_by_disk(bdev, _claim_ptr, dm_disk(md)); |
1da177e4 | 367 | if (r) |
9a1c3542 | 368 | blkdev_put(bdev, d->dm_dev.mode); |
1da177e4 | 369 | else |
82b1519b | 370 | d->dm_dev.bdev = bdev; |
1da177e4 LT |
371 | return r; |
372 | } | |
373 | ||
374 | /* | |
375 | * Close a device that we've been using. | |
376 | */ | |
82b1519b | 377 | static void close_dev(struct dm_dev_internal *d, struct mapped_device *md) |
1da177e4 | 378 | { |
82b1519b | 379 | if (!d->dm_dev.bdev) |
1da177e4 LT |
380 | return; |
381 | ||
82b1519b | 382 | bd_release_from_disk(d->dm_dev.bdev, dm_disk(md)); |
9a1c3542 | 383 | blkdev_put(d->dm_dev.bdev, d->dm_dev.mode); |
82b1519b | 384 | d->dm_dev.bdev = NULL; |
1da177e4 LT |
385 | } |
386 | ||
387 | /* | |
2cd54d9b | 388 | * If possible, this checks an area of a destination device is valid. |
1da177e4 | 389 | */ |
82b1519b MP |
390 | static int check_device_area(struct dm_dev_internal *dd, sector_t start, |
391 | sector_t len) | |
1da177e4 | 392 | { |
82b1519b | 393 | sector_t dev_size = dd->dm_dev.bdev->bd_inode->i_size >> SECTOR_SHIFT; |
2cd54d9b MA |
394 | |
395 | if (!dev_size) | |
396 | return 1; | |
397 | ||
1da177e4 LT |
398 | return ((start < dev_size) && (len <= (dev_size - start))); |
399 | } | |
400 | ||
401 | /* | |
402 | * This upgrades the mode on an already open dm_dev. Being | |
403 | * careful to leave things as they were if we fail to reopen the | |
404 | * device. | |
405 | */ | |
aeb5d727 | 406 | static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode, |
82b1519b | 407 | struct mapped_device *md) |
1da177e4 LT |
408 | { |
409 | int r; | |
82b1519b MP |
410 | struct dm_dev_internal dd_copy; |
411 | dev_t dev = dd->dm_dev.bdev->bd_dev; | |
1da177e4 LT |
412 | |
413 | dd_copy = *dd; | |
414 | ||
82b1519b MP |
415 | dd->dm_dev.mode |= new_mode; |
416 | dd->dm_dev.bdev = NULL; | |
f165921d | 417 | r = open_dev(dd, dev, md); |
1da177e4 | 418 | if (!r) |
f165921d | 419 | close_dev(&dd_copy, md); |
1da177e4 LT |
420 | else |
421 | *dd = dd_copy; | |
422 | ||
423 | return r; | |
424 | } | |
425 | ||
426 | /* | |
427 | * Add a device to the list, or just increment the usage count if | |
428 | * it's already present. | |
429 | */ | |
430 | static int __table_get_device(struct dm_table *t, struct dm_target *ti, | |
431 | const char *path, sector_t start, sector_t len, | |
aeb5d727 | 432 | fmode_t mode, struct dm_dev **result) |
1da177e4 LT |
433 | { |
434 | int r; | |
69a2ce72 | 435 | dev_t uninitialized_var(dev); |
82b1519b | 436 | struct dm_dev_internal *dd; |
1da177e4 LT |
437 | unsigned int major, minor; |
438 | ||
547bc926 | 439 | BUG_ON(!t); |
1da177e4 LT |
440 | |
441 | if (sscanf(path, "%u:%u", &major, &minor) == 2) { | |
442 | /* Extract the major/minor numbers */ | |
443 | dev = MKDEV(major, minor); | |
444 | if (MAJOR(dev) != major || MINOR(dev) != minor) | |
445 | return -EOVERFLOW; | |
446 | } else { | |
447 | /* convert the path to a device */ | |
72e8264e CH |
448 | struct block_device *bdev = lookup_bdev(path); |
449 | ||
450 | if (IS_ERR(bdev)) | |
451 | return PTR_ERR(bdev); | |
452 | dev = bdev->bd_dev; | |
453 | bdput(bdev); | |
1da177e4 LT |
454 | } |
455 | ||
456 | dd = find_device(&t->devices, dev); | |
457 | if (!dd) { | |
458 | dd = kmalloc(sizeof(*dd), GFP_KERNEL); | |
459 | if (!dd) | |
460 | return -ENOMEM; | |
461 | ||
82b1519b MP |
462 | dd->dm_dev.mode = mode; |
463 | dd->dm_dev.bdev = NULL; | |
1da177e4 | 464 | |
f165921d | 465 | if ((r = open_dev(dd, dev, t->md))) { |
1da177e4 LT |
466 | kfree(dd); |
467 | return r; | |
468 | } | |
469 | ||
82b1519b | 470 | format_dev_t(dd->dm_dev.name, dev); |
1da177e4 LT |
471 | |
472 | atomic_set(&dd->count, 0); | |
473 | list_add(&dd->list, &t->devices); | |
474 | ||
82b1519b | 475 | } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) { |
f165921d | 476 | r = upgrade_mode(dd, mode, t->md); |
1da177e4 LT |
477 | if (r) |
478 | return r; | |
479 | } | |
480 | atomic_inc(&dd->count); | |
481 | ||
482 | if (!check_device_area(dd, start, len)) { | |
483 | DMWARN("device %s too small for target", path); | |
82b1519b | 484 | dm_put_device(ti, &dd->dm_dev); |
1da177e4 LT |
485 | return -EINVAL; |
486 | } | |
487 | ||
82b1519b | 488 | *result = &dd->dm_dev; |
1da177e4 LT |
489 | |
490 | return 0; | |
491 | } | |
492 | ||
3cb40214 | 493 | void dm_set_device_limits(struct dm_target *ti, struct block_device *bdev) |
1da177e4 | 494 | { |
165125e1 | 495 | struct request_queue *q = bdev_get_queue(bdev); |
3cb40214 | 496 | struct io_restrictions *rs = &ti->limits; |
0c2322e4 AK |
497 | char b[BDEVNAME_SIZE]; |
498 | ||
499 | if (unlikely(!q)) { | |
500 | DMWARN("%s: Cannot set limits for nonexistent device %s", | |
501 | dm_device_name(ti->table->md), bdevname(bdev, b)); | |
502 | return; | |
503 | } | |
3cb40214 BR |
504 | |
505 | /* | |
506 | * Combine the device limits low. | |
507 | * | |
508 | * FIXME: if we move an io_restriction struct | |
509 | * into q this would just be a call to | |
510 | * combine_restrictions_low() | |
511 | */ | |
512 | rs->max_sectors = | |
513 | min_not_zero(rs->max_sectors, q->max_sectors); | |
514 | ||
9980c638 MB |
515 | /* |
516 | * Check if merge fn is supported. | |
517 | * If not we'll force DM to use PAGE_SIZE or | |
518 | * smaller I/O, just to be safe. | |
3cb40214 | 519 | */ |
9980c638 MB |
520 | |
521 | if (q->merge_bvec_fn && !ti->type->merge) | |
1da177e4 | 522 | rs->max_sectors = |
3cb40214 BR |
523 | min_not_zero(rs->max_sectors, |
524 | (unsigned int) (PAGE_SIZE >> 9)); | |
1da177e4 | 525 | |
3cb40214 BR |
526 | rs->max_phys_segments = |
527 | min_not_zero(rs->max_phys_segments, | |
528 | q->max_phys_segments); | |
1da177e4 | 529 | |
3cb40214 BR |
530 | rs->max_hw_segments = |
531 | min_not_zero(rs->max_hw_segments, q->max_hw_segments); | |
1da177e4 | 532 | |
3cb40214 | 533 | rs->hardsect_size = max(rs->hardsect_size, q->hardsect_size); |
1da177e4 | 534 | |
3cb40214 BR |
535 | rs->max_segment_size = |
536 | min_not_zero(rs->max_segment_size, q->max_segment_size); | |
1da177e4 | 537 | |
91212507 NB |
538 | rs->max_hw_sectors = |
539 | min_not_zero(rs->max_hw_sectors, q->max_hw_sectors); | |
540 | ||
3cb40214 BR |
541 | rs->seg_boundary_mask = |
542 | min_not_zero(rs->seg_boundary_mask, | |
543 | q->seg_boundary_mask); | |
1da177e4 | 544 | |
5ec140e6 VA |
545 | rs->bounce_pfn = min_not_zero(rs->bounce_pfn, q->bounce_pfn); |
546 | ||
3cb40214 BR |
547 | rs->no_cluster |= !test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags); |
548 | } | |
549 | EXPORT_SYMBOL_GPL(dm_set_device_limits); | |
969429b5 | 550 | |
3cb40214 | 551 | int dm_get_device(struct dm_target *ti, const char *path, sector_t start, |
aeb5d727 | 552 | sector_t len, fmode_t mode, struct dm_dev **result) |
3cb40214 BR |
553 | { |
554 | int r = __table_get_device(ti->table, ti, path, | |
555 | start, len, mode, result); | |
556 | ||
557 | if (!r) | |
558 | dm_set_device_limits(ti, (*result)->bdev); | |
1da177e4 LT |
559 | |
560 | return r; | |
561 | } | |
562 | ||
563 | /* | |
564 | * Decrement a devices use count and remove it if necessary. | |
565 | */ | |
82b1519b | 566 | void dm_put_device(struct dm_target *ti, struct dm_dev *d) |
1da177e4 | 567 | { |
82b1519b MP |
568 | struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal, |
569 | dm_dev); | |
570 | ||
1da177e4 | 571 | if (atomic_dec_and_test(&dd->count)) { |
f165921d | 572 | close_dev(dd, ti->table->md); |
1da177e4 LT |
573 | list_del(&dd->list); |
574 | kfree(dd); | |
575 | } | |
576 | } | |
577 | ||
578 | /* | |
579 | * Checks to see if the target joins onto the end of the table. | |
580 | */ | |
581 | static int adjoin(struct dm_table *table, struct dm_target *ti) | |
582 | { | |
583 | struct dm_target *prev; | |
584 | ||
585 | if (!table->num_targets) | |
586 | return !ti->begin; | |
587 | ||
588 | prev = &table->targets[table->num_targets - 1]; | |
589 | return (ti->begin == (prev->begin + prev->len)); | |
590 | } | |
591 | ||
592 | /* | |
593 | * Used to dynamically allocate the arg array. | |
594 | */ | |
595 | static char **realloc_argv(unsigned *array_size, char **old_argv) | |
596 | { | |
597 | char **argv; | |
598 | unsigned new_size; | |
599 | ||
600 | new_size = *array_size ? *array_size * 2 : 64; | |
601 | argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL); | |
602 | if (argv) { | |
603 | memcpy(argv, old_argv, *array_size * sizeof(*argv)); | |
604 | *array_size = new_size; | |
605 | } | |
606 | ||
607 | kfree(old_argv); | |
608 | return argv; | |
609 | } | |
610 | ||
611 | /* | |
612 | * Destructively splits up the argument list to pass to ctr. | |
613 | */ | |
614 | int dm_split_args(int *argc, char ***argvp, char *input) | |
615 | { | |
616 | char *start, *end = input, *out, **argv = NULL; | |
617 | unsigned array_size = 0; | |
618 | ||
619 | *argc = 0; | |
814d6862 DT |
620 | |
621 | if (!input) { | |
622 | *argvp = NULL; | |
623 | return 0; | |
624 | } | |
625 | ||
1da177e4 LT |
626 | argv = realloc_argv(&array_size, argv); |
627 | if (!argv) | |
628 | return -ENOMEM; | |
629 | ||
630 | while (1) { | |
631 | start = end; | |
632 | ||
633 | /* Skip whitespace */ | |
634 | while (*start && isspace(*start)) | |
635 | start++; | |
636 | ||
637 | if (!*start) | |
638 | break; /* success, we hit the end */ | |
639 | ||
640 | /* 'out' is used to remove any back-quotes */ | |
641 | end = out = start; | |
642 | while (*end) { | |
643 | /* Everything apart from '\0' can be quoted */ | |
644 | if (*end == '\\' && *(end + 1)) { | |
645 | *out++ = *(end + 1); | |
646 | end += 2; | |
647 | continue; | |
648 | } | |
649 | ||
650 | if (isspace(*end)) | |
651 | break; /* end of token */ | |
652 | ||
653 | *out++ = *end++; | |
654 | } | |
655 | ||
656 | /* have we already filled the array ? */ | |
657 | if ((*argc + 1) > array_size) { | |
658 | argv = realloc_argv(&array_size, argv); | |
659 | if (!argv) | |
660 | return -ENOMEM; | |
661 | } | |
662 | ||
663 | /* we know this is whitespace */ | |
664 | if (*end) | |
665 | end++; | |
666 | ||
667 | /* terminate the string and put it in the array */ | |
668 | *out = '\0'; | |
669 | argv[*argc] = start; | |
670 | (*argc)++; | |
671 | } | |
672 | ||
673 | *argvp = argv; | |
674 | return 0; | |
675 | } | |
676 | ||
677 | static void check_for_valid_limits(struct io_restrictions *rs) | |
678 | { | |
679 | if (!rs->max_sectors) | |
defd94b7 | 680 | rs->max_sectors = SAFE_MAX_SECTORS; |
91212507 NB |
681 | if (!rs->max_hw_sectors) |
682 | rs->max_hw_sectors = SAFE_MAX_SECTORS; | |
1da177e4 LT |
683 | if (!rs->max_phys_segments) |
684 | rs->max_phys_segments = MAX_PHYS_SEGMENTS; | |
685 | if (!rs->max_hw_segments) | |
686 | rs->max_hw_segments = MAX_HW_SEGMENTS; | |
687 | if (!rs->hardsect_size) | |
688 | rs->hardsect_size = 1 << SECTOR_SHIFT; | |
689 | if (!rs->max_segment_size) | |
690 | rs->max_segment_size = MAX_SEGMENT_SIZE; | |
691 | if (!rs->seg_boundary_mask) | |
0e435ac2 | 692 | rs->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK; |
5ec140e6 VA |
693 | if (!rs->bounce_pfn) |
694 | rs->bounce_pfn = -1; | |
1da177e4 LT |
695 | } |
696 | ||
697 | int dm_table_add_target(struct dm_table *t, const char *type, | |
698 | sector_t start, sector_t len, char *params) | |
699 | { | |
700 | int r = -EINVAL, argc; | |
701 | char **argv; | |
702 | struct dm_target *tgt; | |
703 | ||
704 | if ((r = check_space(t))) | |
705 | return r; | |
706 | ||
707 | tgt = t->targets + t->num_targets; | |
708 | memset(tgt, 0, sizeof(*tgt)); | |
709 | ||
710 | if (!len) { | |
72d94861 | 711 | DMERR("%s: zero-length target", dm_device_name(t->md)); |
1da177e4 LT |
712 | return -EINVAL; |
713 | } | |
714 | ||
715 | tgt->type = dm_get_target_type(type); | |
716 | if (!tgt->type) { | |
72d94861 AK |
717 | DMERR("%s: %s: unknown target type", dm_device_name(t->md), |
718 | type); | |
1da177e4 LT |
719 | return -EINVAL; |
720 | } | |
721 | ||
722 | tgt->table = t; | |
723 | tgt->begin = start; | |
724 | tgt->len = len; | |
725 | tgt->error = "Unknown error"; | |
726 | ||
727 | /* | |
728 | * Does this target adjoin the previous one ? | |
729 | */ | |
730 | if (!adjoin(t, tgt)) { | |
731 | tgt->error = "Gap in table"; | |
732 | r = -EINVAL; | |
733 | goto bad; | |
734 | } | |
735 | ||
736 | r = dm_split_args(&argc, &argv, params); | |
737 | if (r) { | |
738 | tgt->error = "couldn't split parameters (insufficient memory)"; | |
739 | goto bad; | |
740 | } | |
741 | ||
742 | r = tgt->type->ctr(tgt, argc, argv); | |
743 | kfree(argv); | |
744 | if (r) | |
745 | goto bad; | |
746 | ||
747 | t->highs[t->num_targets++] = tgt->begin + tgt->len - 1; | |
748 | ||
749 | /* FIXME: the plan is to combine high here and then have | |
750 | * the merge fn apply the target level restrictions. */ | |
751 | combine_restrictions_low(&t->limits, &tgt->limits); | |
ab4c1424 AK |
752 | |
753 | if (!(tgt->type->features & DM_TARGET_SUPPORTS_BARRIERS)) | |
754 | t->barriers_supported = 0; | |
755 | ||
1da177e4 LT |
756 | return 0; |
757 | ||
758 | bad: | |
72d94861 | 759 | DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error); |
1da177e4 LT |
760 | dm_put_target_type(tgt->type); |
761 | return r; | |
762 | } | |
763 | ||
764 | static int setup_indexes(struct dm_table *t) | |
765 | { | |
766 | int i; | |
767 | unsigned int total = 0; | |
768 | sector_t *indexes; | |
769 | ||
770 | /* allocate the space for *all* the indexes */ | |
771 | for (i = t->depth - 2; i >= 0; i--) { | |
772 | t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE); | |
773 | total += t->counts[i]; | |
774 | } | |
775 | ||
776 | indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE); | |
777 | if (!indexes) | |
778 | return -ENOMEM; | |
779 | ||
780 | /* set up internal nodes, bottom-up */ | |
82d601dc | 781 | for (i = t->depth - 2; i >= 0; i--) { |
1da177e4 LT |
782 | t->index[i] = indexes; |
783 | indexes += (KEYS_PER_NODE * t->counts[i]); | |
784 | setup_btree_index(i, t); | |
785 | } | |
786 | ||
787 | return 0; | |
788 | } | |
789 | ||
790 | /* | |
791 | * Builds the btree to index the map. | |
792 | */ | |
793 | int dm_table_complete(struct dm_table *t) | |
794 | { | |
795 | int r = 0; | |
796 | unsigned int leaf_nodes; | |
797 | ||
798 | check_for_valid_limits(&t->limits); | |
799 | ||
ab4c1424 AK |
800 | /* |
801 | * We only support barriers if there is exactly one underlying device. | |
802 | */ | |
803 | if (!list_is_singular(&t->devices)) | |
804 | t->barriers_supported = 0; | |
805 | ||
1da177e4 LT |
806 | /* how many indexes will the btree have ? */ |
807 | leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE); | |
808 | t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE); | |
809 | ||
810 | /* leaf layer has already been set up */ | |
811 | t->counts[t->depth - 1] = leaf_nodes; | |
812 | t->index[t->depth - 1] = t->highs; | |
813 | ||
814 | if (t->depth >= 2) | |
815 | r = setup_indexes(t); | |
816 | ||
817 | return r; | |
818 | } | |
819 | ||
48c9c27b | 820 | static DEFINE_MUTEX(_event_lock); |
1da177e4 LT |
821 | void dm_table_event_callback(struct dm_table *t, |
822 | void (*fn)(void *), void *context) | |
823 | { | |
48c9c27b | 824 | mutex_lock(&_event_lock); |
1da177e4 LT |
825 | t->event_fn = fn; |
826 | t->event_context = context; | |
48c9c27b | 827 | mutex_unlock(&_event_lock); |
1da177e4 LT |
828 | } |
829 | ||
830 | void dm_table_event(struct dm_table *t) | |
831 | { | |
832 | /* | |
833 | * You can no longer call dm_table_event() from interrupt | |
834 | * context, use a bottom half instead. | |
835 | */ | |
836 | BUG_ON(in_interrupt()); | |
837 | ||
48c9c27b | 838 | mutex_lock(&_event_lock); |
1da177e4 LT |
839 | if (t->event_fn) |
840 | t->event_fn(t->event_context); | |
48c9c27b | 841 | mutex_unlock(&_event_lock); |
1da177e4 LT |
842 | } |
843 | ||
844 | sector_t dm_table_get_size(struct dm_table *t) | |
845 | { | |
846 | return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0; | |
847 | } | |
848 | ||
849 | struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index) | |
850 | { | |
14353539 | 851 | if (index >= t->num_targets) |
1da177e4 LT |
852 | return NULL; |
853 | ||
854 | return t->targets + index; | |
855 | } | |
856 | ||
857 | /* | |
858 | * Search the btree for the correct target. | |
512875bd JN |
859 | * |
860 | * Caller should check returned pointer with dm_target_is_valid() | |
861 | * to trap I/O beyond end of device. | |
1da177e4 LT |
862 | */ |
863 | struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector) | |
864 | { | |
865 | unsigned int l, n = 0, k = 0; | |
866 | sector_t *node; | |
867 | ||
868 | for (l = 0; l < t->depth; l++) { | |
869 | n = get_child(n, k); | |
870 | node = get_node(t, l, n); | |
871 | ||
872 | for (k = 0; k < KEYS_PER_NODE; k++) | |
873 | if (node[k] >= sector) | |
874 | break; | |
875 | } | |
876 | ||
877 | return &t->targets[(KEYS_PER_NODE * n) + k]; | |
878 | } | |
879 | ||
880 | void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q) | |
881 | { | |
882 | /* | |
883 | * Make sure we obey the optimistic sub devices | |
884 | * restrictions. | |
885 | */ | |
886 | blk_queue_max_sectors(q, t->limits.max_sectors); | |
887 | q->max_phys_segments = t->limits.max_phys_segments; | |
888 | q->max_hw_segments = t->limits.max_hw_segments; | |
889 | q->hardsect_size = t->limits.hardsect_size; | |
890 | q->max_segment_size = t->limits.max_segment_size; | |
91212507 | 891 | q->max_hw_sectors = t->limits.max_hw_sectors; |
1da177e4 | 892 | q->seg_boundary_mask = t->limits.seg_boundary_mask; |
5ec140e6 | 893 | q->bounce_pfn = t->limits.bounce_pfn; |
c9a3f6d6 | 894 | |
969429b5 | 895 | if (t->limits.no_cluster) |
c9a3f6d6 | 896 | queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q); |
969429b5 | 897 | else |
c9a3f6d6 | 898 | queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, q); |
969429b5 | 899 | |
1da177e4 LT |
900 | } |
901 | ||
902 | unsigned int dm_table_get_num_targets(struct dm_table *t) | |
903 | { | |
904 | return t->num_targets; | |
905 | } | |
906 | ||
907 | struct list_head *dm_table_get_devices(struct dm_table *t) | |
908 | { | |
909 | return &t->devices; | |
910 | } | |
911 | ||
aeb5d727 | 912 | fmode_t dm_table_get_mode(struct dm_table *t) |
1da177e4 LT |
913 | { |
914 | return t->mode; | |
915 | } | |
916 | ||
917 | static void suspend_targets(struct dm_table *t, unsigned postsuspend) | |
918 | { | |
919 | int i = t->num_targets; | |
920 | struct dm_target *ti = t->targets; | |
921 | ||
922 | while (i--) { | |
923 | if (postsuspend) { | |
924 | if (ti->type->postsuspend) | |
925 | ti->type->postsuspend(ti); | |
926 | } else if (ti->type->presuspend) | |
927 | ti->type->presuspend(ti); | |
928 | ||
929 | ti++; | |
930 | } | |
931 | } | |
932 | ||
933 | void dm_table_presuspend_targets(struct dm_table *t) | |
934 | { | |
cf222b37 AK |
935 | if (!t) |
936 | return; | |
937 | ||
e8488d08 | 938 | suspend_targets(t, 0); |
1da177e4 LT |
939 | } |
940 | ||
941 | void dm_table_postsuspend_targets(struct dm_table *t) | |
942 | { | |
cf222b37 AK |
943 | if (!t) |
944 | return; | |
945 | ||
e8488d08 | 946 | suspend_targets(t, 1); |
1da177e4 LT |
947 | } |
948 | ||
8757b776 | 949 | int dm_table_resume_targets(struct dm_table *t) |
1da177e4 | 950 | { |
8757b776 MB |
951 | int i, r = 0; |
952 | ||
953 | for (i = 0; i < t->num_targets; i++) { | |
954 | struct dm_target *ti = t->targets + i; | |
955 | ||
956 | if (!ti->type->preresume) | |
957 | continue; | |
958 | ||
959 | r = ti->type->preresume(ti); | |
960 | if (r) | |
961 | return r; | |
962 | } | |
1da177e4 LT |
963 | |
964 | for (i = 0; i < t->num_targets; i++) { | |
965 | struct dm_target *ti = t->targets + i; | |
966 | ||
967 | if (ti->type->resume) | |
968 | ti->type->resume(ti); | |
969 | } | |
8757b776 MB |
970 | |
971 | return 0; | |
1da177e4 LT |
972 | } |
973 | ||
974 | int dm_table_any_congested(struct dm_table *t, int bdi_bits) | |
975 | { | |
82b1519b | 976 | struct dm_dev_internal *dd; |
afb24528 | 977 | struct list_head *devices = dm_table_get_devices(t); |
1da177e4 LT |
978 | int r = 0; |
979 | ||
afb24528 | 980 | list_for_each_entry(dd, devices, list) { |
82b1519b | 981 | struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev); |
0c2322e4 AK |
982 | char b[BDEVNAME_SIZE]; |
983 | ||
984 | if (likely(q)) | |
985 | r |= bdi_congested(&q->backing_dev_info, bdi_bits); | |
986 | else | |
987 | DMWARN_LIMIT("%s: any_congested: nonexistent device %s", | |
988 | dm_device_name(t->md), | |
989 | bdevname(dd->dm_dev.bdev, b)); | |
1da177e4 LT |
990 | } |
991 | ||
992 | return r; | |
993 | } | |
994 | ||
995 | void dm_table_unplug_all(struct dm_table *t) | |
996 | { | |
82b1519b | 997 | struct dm_dev_internal *dd; |
afb24528 | 998 | struct list_head *devices = dm_table_get_devices(t); |
1da177e4 | 999 | |
afb24528 | 1000 | list_for_each_entry(dd, devices, list) { |
82b1519b | 1001 | struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev); |
0c2322e4 AK |
1002 | char b[BDEVNAME_SIZE]; |
1003 | ||
1004 | if (likely(q)) | |
1005 | blk_unplug(q); | |
1006 | else | |
1007 | DMWARN_LIMIT("%s: Cannot unplug nonexistent device %s", | |
1008 | dm_device_name(t->md), | |
1009 | bdevname(dd->dm_dev.bdev, b)); | |
1da177e4 LT |
1010 | } |
1011 | } | |
1012 | ||
1134e5ae MA |
1013 | struct mapped_device *dm_table_get_md(struct dm_table *t) |
1014 | { | |
1015 | dm_get(t->md); | |
1016 | ||
1017 | return t->md; | |
1018 | } | |
1019 | ||
ab4c1424 AK |
1020 | int dm_table_barrier_ok(struct dm_table *t) |
1021 | { | |
1022 | return t->barriers_supported; | |
1023 | } | |
1024 | EXPORT_SYMBOL(dm_table_barrier_ok); | |
1025 | ||
1da177e4 LT |
1026 | EXPORT_SYMBOL(dm_vcalloc); |
1027 | EXPORT_SYMBOL(dm_get_device); | |
1028 | EXPORT_SYMBOL(dm_put_device); | |
1029 | EXPORT_SYMBOL(dm_table_event); | |
d5e404c1 | 1030 | EXPORT_SYMBOL(dm_table_get_size); |
1da177e4 | 1031 | EXPORT_SYMBOL(dm_table_get_mode); |
1134e5ae | 1032 | EXPORT_SYMBOL(dm_table_get_md); |
1da177e4 LT |
1033 | EXPORT_SYMBOL(dm_table_put); |
1034 | EXPORT_SYMBOL(dm_table_get); | |
1035 | EXPORT_SYMBOL(dm_table_unplug_all); |