]>
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 | ||
4cc96131 | 8 | #include "dm-core.h" |
1da177e4 LT |
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> |
bfebd1cd | 21 | #include <linux/blk-mq.h> |
644bda6f | 22 | #include <linux/mount.h> |
273752c9 | 23 | #include <linux/dax.h> |
1da177e4 | 24 | |
72d94861 AK |
25 | #define DM_MSG_PREFIX "table" |
26 | ||
1da177e4 LT |
27 | #define MAX_DEPTH 16 |
28 | #define NODE_SIZE L1_CACHE_BYTES | |
29 | #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t)) | |
30 | #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1) | |
31 | ||
32 | struct dm_table { | |
1134e5ae | 33 | struct mapped_device *md; |
7e0d574f | 34 | enum dm_queue_mode type; |
1da177e4 LT |
35 | |
36 | /* btree table */ | |
37 | unsigned int depth; | |
38 | unsigned int counts[MAX_DEPTH]; /* in nodes */ | |
39 | sector_t *index[MAX_DEPTH]; | |
40 | ||
41 | unsigned int num_targets; | |
42 | unsigned int num_allocated; | |
43 | sector_t *highs; | |
44 | struct dm_target *targets; | |
45 | ||
36a0456f | 46 | struct target_type *immutable_target_type; |
e83068a5 MS |
47 | |
48 | bool integrity_supported:1; | |
49 | bool singleton:1; | |
50 | bool all_blk_mq:1; | |
9b4b5a79 | 51 | unsigned integrity_added:1; |
5ae89a87 | 52 | |
1da177e4 LT |
53 | /* |
54 | * Indicates the rw permissions for the new logical | |
55 | * device. This should be a combination of FMODE_READ | |
56 | * and FMODE_WRITE. | |
57 | */ | |
aeb5d727 | 58 | fmode_t mode; |
1da177e4 LT |
59 | |
60 | /* a list of devices used by this table */ | |
61 | struct list_head devices; | |
62 | ||
1da177e4 LT |
63 | /* events get handed up using this callback */ |
64 | void (*event_fn)(void *); | |
65 | void *event_context; | |
e6ee8c0b KU |
66 | |
67 | struct dm_md_mempools *mempools; | |
9d357b07 N |
68 | |
69 | struct list_head target_callbacks; | |
1da177e4 LT |
70 | }; |
71 | ||
72 | /* | |
73 | * Similar to ceiling(log_size(n)) | |
74 | */ | |
75 | static unsigned int int_log(unsigned int n, unsigned int base) | |
76 | { | |
77 | int result = 0; | |
78 | ||
79 | while (n > 1) { | |
80 | n = dm_div_up(n, base); | |
81 | result++; | |
82 | } | |
83 | ||
84 | return result; | |
85 | } | |
86 | ||
1da177e4 LT |
87 | /* |
88 | * Calculate the index of the child node of the n'th node k'th key. | |
89 | */ | |
90 | static inline unsigned int get_child(unsigned int n, unsigned int k) | |
91 | { | |
92 | return (n * CHILDREN_PER_NODE) + k; | |
93 | } | |
94 | ||
95 | /* | |
96 | * Return the n'th node of level l from table t. | |
97 | */ | |
98 | static inline sector_t *get_node(struct dm_table *t, | |
99 | unsigned int l, unsigned int n) | |
100 | { | |
101 | return t->index[l] + (n * KEYS_PER_NODE); | |
102 | } | |
103 | ||
104 | /* | |
105 | * Return the highest key that you could lookup from the n'th | |
106 | * node on level l of the btree. | |
107 | */ | |
108 | static sector_t high(struct dm_table *t, unsigned int l, unsigned int n) | |
109 | { | |
110 | for (; l < t->depth - 1; l++) | |
111 | n = get_child(n, CHILDREN_PER_NODE - 1); | |
112 | ||
113 | if (n >= t->counts[l]) | |
114 | return (sector_t) - 1; | |
115 | ||
116 | return get_node(t, l, n)[KEYS_PER_NODE - 1]; | |
117 | } | |
118 | ||
119 | /* | |
120 | * Fills in a level of the btree based on the highs of the level | |
121 | * below it. | |
122 | */ | |
123 | static int setup_btree_index(unsigned int l, struct dm_table *t) | |
124 | { | |
125 | unsigned int n, k; | |
126 | sector_t *node; | |
127 | ||
128 | for (n = 0U; n < t->counts[l]; n++) { | |
129 | node = get_node(t, l, n); | |
130 | ||
131 | for (k = 0U; k < KEYS_PER_NODE; k++) | |
132 | node[k] = high(t, l + 1, get_child(n, k)); | |
133 | } | |
134 | ||
135 | return 0; | |
136 | } | |
137 | ||
138 | void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size) | |
139 | { | |
140 | unsigned long size; | |
141 | void *addr; | |
142 | ||
143 | /* | |
144 | * Check that we're not going to overflow. | |
145 | */ | |
146 | if (nmemb > (ULONG_MAX / elem_size)) | |
147 | return NULL; | |
148 | ||
149 | size = nmemb * elem_size; | |
e29e65aa | 150 | addr = vzalloc(size); |
1da177e4 LT |
151 | |
152 | return addr; | |
153 | } | |
08649012 | 154 | EXPORT_SYMBOL(dm_vcalloc); |
1da177e4 LT |
155 | |
156 | /* | |
157 | * highs, and targets are managed as dynamic arrays during a | |
158 | * table load. | |
159 | */ | |
160 | static int alloc_targets(struct dm_table *t, unsigned int num) | |
161 | { | |
162 | sector_t *n_highs; | |
163 | struct dm_target *n_targets; | |
1da177e4 LT |
164 | |
165 | /* | |
166 | * Allocate both the target array and offset array at once. | |
512875bd JN |
167 | * Append an empty entry to catch sectors beyond the end of |
168 | * the device. | |
1da177e4 | 169 | */ |
512875bd | 170 | n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) + |
1da177e4 LT |
171 | sizeof(sector_t)); |
172 | if (!n_highs) | |
173 | return -ENOMEM; | |
174 | ||
175 | n_targets = (struct dm_target *) (n_highs + num); | |
176 | ||
57a2f238 | 177 | memset(n_highs, -1, sizeof(*n_highs) * num); |
1da177e4 LT |
178 | vfree(t->highs); |
179 | ||
180 | t->num_allocated = num; | |
181 | t->highs = n_highs; | |
182 | t->targets = n_targets; | |
183 | ||
184 | return 0; | |
185 | } | |
186 | ||
aeb5d727 | 187 | int dm_table_create(struct dm_table **result, fmode_t mode, |
1134e5ae | 188 | unsigned num_targets, struct mapped_device *md) |
1da177e4 | 189 | { |
094262db | 190 | struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL); |
1da177e4 LT |
191 | |
192 | if (!t) | |
193 | return -ENOMEM; | |
194 | ||
1da177e4 | 195 | INIT_LIST_HEAD(&t->devices); |
9d357b07 | 196 | INIT_LIST_HEAD(&t->target_callbacks); |
1da177e4 LT |
197 | |
198 | if (!num_targets) | |
199 | num_targets = KEYS_PER_NODE; | |
200 | ||
201 | num_targets = dm_round_up(num_targets, KEYS_PER_NODE); | |
202 | ||
5b2d0657 MP |
203 | if (!num_targets) { |
204 | kfree(t); | |
205 | return -ENOMEM; | |
206 | } | |
207 | ||
1da177e4 LT |
208 | if (alloc_targets(t, num_targets)) { |
209 | kfree(t); | |
1da177e4 LT |
210 | return -ENOMEM; |
211 | } | |
212 | ||
e83068a5 | 213 | t->type = DM_TYPE_NONE; |
1da177e4 | 214 | t->mode = mode; |
1134e5ae | 215 | t->md = md; |
1da177e4 LT |
216 | *result = t; |
217 | return 0; | |
218 | } | |
219 | ||
86f1152b | 220 | static void free_devices(struct list_head *devices, struct mapped_device *md) |
1da177e4 LT |
221 | { |
222 | struct list_head *tmp, *next; | |
223 | ||
afb24528 | 224 | list_for_each_safe(tmp, next, devices) { |
82b1519b MP |
225 | struct dm_dev_internal *dd = |
226 | list_entry(tmp, struct dm_dev_internal, list); | |
86f1152b BM |
227 | DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s", |
228 | dm_device_name(md), dd->dm_dev->name); | |
229 | dm_put_table_device(md, dd->dm_dev); | |
1da177e4 LT |
230 | kfree(dd); |
231 | } | |
232 | } | |
233 | ||
d5816876 | 234 | void dm_table_destroy(struct dm_table *t) |
1da177e4 LT |
235 | { |
236 | unsigned int i; | |
237 | ||
a7940155 AK |
238 | if (!t) |
239 | return; | |
240 | ||
26803b9f | 241 | /* free the indexes */ |
1da177e4 LT |
242 | if (t->depth >= 2) |
243 | vfree(t->index[t->depth - 2]); | |
244 | ||
245 | /* free the targets */ | |
246 | for (i = 0; i < t->num_targets; i++) { | |
247 | struct dm_target *tgt = t->targets + i; | |
248 | ||
249 | if (tgt->type->dtr) | |
250 | tgt->type->dtr(tgt); | |
251 | ||
252 | dm_put_target_type(tgt->type); | |
253 | } | |
254 | ||
255 | vfree(t->highs); | |
256 | ||
257 | /* free the device list */ | |
86f1152b | 258 | free_devices(&t->devices, t->md); |
1da177e4 | 259 | |
e6ee8c0b KU |
260 | dm_free_md_mempools(t->mempools); |
261 | ||
1da177e4 LT |
262 | kfree(t); |
263 | } | |
264 | ||
1da177e4 LT |
265 | /* |
266 | * See if we've already got a device in the list. | |
267 | */ | |
82b1519b | 268 | static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev) |
1da177e4 | 269 | { |
82b1519b | 270 | struct dm_dev_internal *dd; |
1da177e4 LT |
271 | |
272 | list_for_each_entry (dd, l, list) | |
86f1152b | 273 | if (dd->dm_dev->bdev->bd_dev == dev) |
1da177e4 LT |
274 | return dd; |
275 | ||
276 | return NULL; | |
277 | } | |
278 | ||
1da177e4 | 279 | /* |
f6a1ed10 | 280 | * If possible, this checks an area of a destination device is invalid. |
1da177e4 | 281 | */ |
f6a1ed10 MP |
282 | static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev, |
283 | sector_t start, sector_t len, void *data) | |
1da177e4 | 284 | { |
f4808ca9 | 285 | struct request_queue *q; |
754c5fc7 MS |
286 | struct queue_limits *limits = data; |
287 | struct block_device *bdev = dev->bdev; | |
288 | sector_t dev_size = | |
289 | i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; | |
02acc3a4 | 290 | unsigned short logical_block_size_sectors = |
754c5fc7 | 291 | limits->logical_block_size >> SECTOR_SHIFT; |
02acc3a4 | 292 | char b[BDEVNAME_SIZE]; |
2cd54d9b | 293 | |
f4808ca9 MB |
294 | /* |
295 | * Some devices exist without request functions, | |
296 | * such as loop devices not yet bound to backing files. | |
297 | * Forbid the use of such devices. | |
298 | */ | |
299 | q = bdev_get_queue(bdev); | |
300 | if (!q || !q->make_request_fn) { | |
301 | DMWARN("%s: %s is not yet initialised: " | |
302 | "start=%llu, len=%llu, dev_size=%llu", | |
303 | dm_device_name(ti->table->md), bdevname(bdev, b), | |
304 | (unsigned long long)start, | |
305 | (unsigned long long)len, | |
306 | (unsigned long long)dev_size); | |
307 | return 1; | |
308 | } | |
309 | ||
2cd54d9b | 310 | if (!dev_size) |
f6a1ed10 | 311 | return 0; |
2cd54d9b | 312 | |
5dea271b | 313 | if ((start >= dev_size) || (start + len > dev_size)) { |
a963a956 MS |
314 | DMWARN("%s: %s too small for target: " |
315 | "start=%llu, len=%llu, dev_size=%llu", | |
316 | dm_device_name(ti->table->md), bdevname(bdev, b), | |
317 | (unsigned long long)start, | |
318 | (unsigned long long)len, | |
319 | (unsigned long long)dev_size); | |
f6a1ed10 | 320 | return 1; |
02acc3a4 MS |
321 | } |
322 | ||
dd88d313 DLM |
323 | /* |
324 | * If the target is mapped to zoned block device(s), check | |
325 | * that the zones are not partially mapped. | |
326 | */ | |
327 | if (bdev_zoned_model(bdev) != BLK_ZONED_NONE) { | |
328 | unsigned int zone_sectors = bdev_zone_sectors(bdev); | |
329 | ||
330 | if (start & (zone_sectors - 1)) { | |
331 | DMWARN("%s: start=%llu not aligned to h/w zone size %u of %s", | |
332 | dm_device_name(ti->table->md), | |
333 | (unsigned long long)start, | |
334 | zone_sectors, bdevname(bdev, b)); | |
335 | return 1; | |
336 | } | |
337 | ||
338 | /* | |
339 | * Note: The last zone of a zoned block device may be smaller | |
340 | * than other zones. So for a target mapping the end of a | |
341 | * zoned block device with such a zone, len would not be zone | |
342 | * aligned. We do not allow such last smaller zone to be part | |
343 | * of the mapping here to ensure that mappings with multiple | |
344 | * devices do not end up with a smaller zone in the middle of | |
345 | * the sector range. | |
346 | */ | |
347 | if (len & (zone_sectors - 1)) { | |
348 | DMWARN("%s: len=%llu not aligned to h/w zone size %u of %s", | |
349 | dm_device_name(ti->table->md), | |
350 | (unsigned long long)len, | |
351 | zone_sectors, bdevname(bdev, b)); | |
352 | return 1; | |
353 | } | |
354 | } | |
355 | ||
02acc3a4 | 356 | if (logical_block_size_sectors <= 1) |
f6a1ed10 | 357 | return 0; |
02acc3a4 MS |
358 | |
359 | if (start & (logical_block_size_sectors - 1)) { | |
360 | DMWARN("%s: start=%llu not aligned to h/w " | |
a963a956 | 361 | "logical block size %u of %s", |
02acc3a4 MS |
362 | dm_device_name(ti->table->md), |
363 | (unsigned long long)start, | |
754c5fc7 | 364 | limits->logical_block_size, bdevname(bdev, b)); |
f6a1ed10 | 365 | return 1; |
02acc3a4 MS |
366 | } |
367 | ||
5dea271b | 368 | if (len & (logical_block_size_sectors - 1)) { |
02acc3a4 | 369 | DMWARN("%s: len=%llu not aligned to h/w " |
a963a956 | 370 | "logical block size %u of %s", |
02acc3a4 | 371 | dm_device_name(ti->table->md), |
5dea271b | 372 | (unsigned long long)len, |
754c5fc7 | 373 | limits->logical_block_size, bdevname(bdev, b)); |
f6a1ed10 | 374 | return 1; |
02acc3a4 MS |
375 | } |
376 | ||
f6a1ed10 | 377 | return 0; |
1da177e4 LT |
378 | } |
379 | ||
380 | /* | |
570b9d96 | 381 | * This upgrades the mode on an already open dm_dev, being |
1da177e4 | 382 | * careful to leave things as they were if we fail to reopen the |
570b9d96 AK |
383 | * device and not to touch the existing bdev field in case |
384 | * it is accessed concurrently inside dm_table_any_congested(). | |
1da177e4 | 385 | */ |
aeb5d727 | 386 | static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode, |
82b1519b | 387 | struct mapped_device *md) |
1da177e4 LT |
388 | { |
389 | int r; | |
86f1152b | 390 | struct dm_dev *old_dev, *new_dev; |
1da177e4 | 391 | |
86f1152b | 392 | old_dev = dd->dm_dev; |
570b9d96 | 393 | |
86f1152b BM |
394 | r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev, |
395 | dd->dm_dev->mode | new_mode, &new_dev); | |
570b9d96 AK |
396 | if (r) |
397 | return r; | |
1da177e4 | 398 | |
86f1152b BM |
399 | dd->dm_dev = new_dev; |
400 | dm_put_table_device(md, old_dev); | |
1da177e4 | 401 | |
570b9d96 | 402 | return 0; |
1da177e4 LT |
403 | } |
404 | ||
4df2bf46 D |
405 | /* |
406 | * Convert the path to a device | |
407 | */ | |
408 | dev_t dm_get_dev_t(const char *path) | |
409 | { | |
3c120169 | 410 | dev_t dev; |
4df2bf46 D |
411 | struct block_device *bdev; |
412 | ||
413 | bdev = lookup_bdev(path); | |
414 | if (IS_ERR(bdev)) | |
415 | dev = name_to_dev_t(path); | |
416 | else { | |
417 | dev = bdev->bd_dev; | |
418 | bdput(bdev); | |
419 | } | |
420 | ||
421 | return dev; | |
422 | } | |
423 | EXPORT_SYMBOL_GPL(dm_get_dev_t); | |
424 | ||
1da177e4 LT |
425 | /* |
426 | * Add a device to the list, or just increment the usage count if | |
427 | * it's already present. | |
428 | */ | |
08649012 MS |
429 | int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode, |
430 | struct dm_dev **result) | |
1da177e4 LT |
431 | { |
432 | int r; | |
4df2bf46 | 433 | dev_t dev; |
82b1519b | 434 | struct dm_dev_internal *dd; |
08649012 | 435 | struct dm_table *t = ti->table; |
1da177e4 | 436 | |
547bc926 | 437 | BUG_ON(!t); |
1da177e4 | 438 | |
4df2bf46 D |
439 | dev = dm_get_dev_t(path); |
440 | if (!dev) | |
441 | return -ENODEV; | |
1da177e4 LT |
442 | |
443 | dd = find_device(&t->devices, dev); | |
444 | if (!dd) { | |
445 | dd = kmalloc(sizeof(*dd), GFP_KERNEL); | |
446 | if (!dd) | |
447 | return -ENOMEM; | |
448 | ||
86f1152b | 449 | if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) { |
1da177e4 LT |
450 | kfree(dd); |
451 | return r; | |
452 | } | |
453 | ||
1da177e4 LT |
454 | atomic_set(&dd->count, 0); |
455 | list_add(&dd->list, &t->devices); | |
456 | ||
86f1152b | 457 | } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) { |
f165921d | 458 | r = upgrade_mode(dd, mode, t->md); |
1da177e4 LT |
459 | if (r) |
460 | return r; | |
461 | } | |
462 | atomic_inc(&dd->count); | |
463 | ||
86f1152b | 464 | *result = dd->dm_dev; |
1da177e4 LT |
465 | return 0; |
466 | } | |
08649012 | 467 | EXPORT_SYMBOL(dm_get_device); |
1da177e4 | 468 | |
11f0431b MS |
469 | static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev, |
470 | sector_t start, sector_t len, void *data) | |
1da177e4 | 471 | { |
754c5fc7 MS |
472 | struct queue_limits *limits = data; |
473 | struct block_device *bdev = dev->bdev; | |
165125e1 | 474 | struct request_queue *q = bdev_get_queue(bdev); |
0c2322e4 AK |
475 | char b[BDEVNAME_SIZE]; |
476 | ||
477 | if (unlikely(!q)) { | |
478 | DMWARN("%s: Cannot set limits for nonexistent device %s", | |
479 | dm_device_name(ti->table->md), bdevname(bdev, b)); | |
754c5fc7 | 480 | return 0; |
0c2322e4 | 481 | } |
3cb40214 | 482 | |
b27d7f16 MP |
483 | if (bdev_stack_limits(limits, bdev, start) < 0) |
484 | DMWARN("%s: adding target device %s caused an alignment inconsistency: " | |
a963a956 MS |
485 | "physical_block_size=%u, logical_block_size=%u, " |
486 | "alignment_offset=%u, start=%llu", | |
487 | dm_device_name(ti->table->md), bdevname(bdev, b), | |
488 | q->limits.physical_block_size, | |
489 | q->limits.logical_block_size, | |
490 | q->limits.alignment_offset, | |
b27d7f16 | 491 | (unsigned long long) start << SECTOR_SHIFT); |
3cb40214 | 492 | |
dd88d313 DLM |
493 | limits->zoned = blk_queue_zoned_model(q); |
494 | ||
754c5fc7 | 495 | return 0; |
3cb40214 | 496 | } |
969429b5 | 497 | |
1da177e4 | 498 | /* |
08649012 | 499 | * Decrement a device's use count and remove it if necessary. |
1da177e4 | 500 | */ |
82b1519b | 501 | void dm_put_device(struct dm_target *ti, struct dm_dev *d) |
1da177e4 | 502 | { |
86f1152b BM |
503 | int found = 0; |
504 | struct list_head *devices = &ti->table->devices; | |
505 | struct dm_dev_internal *dd; | |
82b1519b | 506 | |
86f1152b BM |
507 | list_for_each_entry(dd, devices, list) { |
508 | if (dd->dm_dev == d) { | |
509 | found = 1; | |
510 | break; | |
511 | } | |
512 | } | |
513 | if (!found) { | |
514 | DMWARN("%s: device %s not in table devices list", | |
515 | dm_device_name(ti->table->md), d->name); | |
516 | return; | |
517 | } | |
1da177e4 | 518 | if (atomic_dec_and_test(&dd->count)) { |
86f1152b | 519 | dm_put_table_device(ti->table->md, d); |
1da177e4 LT |
520 | list_del(&dd->list); |
521 | kfree(dd); | |
522 | } | |
523 | } | |
08649012 | 524 | EXPORT_SYMBOL(dm_put_device); |
1da177e4 LT |
525 | |
526 | /* | |
527 | * Checks to see if the target joins onto the end of the table. | |
528 | */ | |
529 | static int adjoin(struct dm_table *table, struct dm_target *ti) | |
530 | { | |
531 | struct dm_target *prev; | |
532 | ||
533 | if (!table->num_targets) | |
534 | return !ti->begin; | |
535 | ||
536 | prev = &table->targets[table->num_targets - 1]; | |
537 | return (ti->begin == (prev->begin + prev->len)); | |
538 | } | |
539 | ||
540 | /* | |
541 | * Used to dynamically allocate the arg array. | |
f36afb39 MP |
542 | * |
543 | * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must | |
544 | * process messages even if some device is suspended. These messages have a | |
545 | * small fixed number of arguments. | |
546 | * | |
547 | * On the other hand, dm-switch needs to process bulk data using messages and | |
548 | * excessive use of GFP_NOIO could cause trouble. | |
1da177e4 LT |
549 | */ |
550 | static char **realloc_argv(unsigned *array_size, char **old_argv) | |
551 | { | |
552 | char **argv; | |
553 | unsigned new_size; | |
f36afb39 | 554 | gfp_t gfp; |
1da177e4 | 555 | |
f36afb39 MP |
556 | if (*array_size) { |
557 | new_size = *array_size * 2; | |
558 | gfp = GFP_KERNEL; | |
559 | } else { | |
560 | new_size = 8; | |
561 | gfp = GFP_NOIO; | |
562 | } | |
563 | argv = kmalloc(new_size * sizeof(*argv), gfp); | |
1da177e4 LT |
564 | if (argv) { |
565 | memcpy(argv, old_argv, *array_size * sizeof(*argv)); | |
566 | *array_size = new_size; | |
567 | } | |
568 | ||
569 | kfree(old_argv); | |
570 | return argv; | |
571 | } | |
572 | ||
573 | /* | |
574 | * Destructively splits up the argument list to pass to ctr. | |
575 | */ | |
576 | int dm_split_args(int *argc, char ***argvp, char *input) | |
577 | { | |
578 | char *start, *end = input, *out, **argv = NULL; | |
579 | unsigned array_size = 0; | |
580 | ||
581 | *argc = 0; | |
814d6862 DT |
582 | |
583 | if (!input) { | |
584 | *argvp = NULL; | |
585 | return 0; | |
586 | } | |
587 | ||
1da177e4 LT |
588 | argv = realloc_argv(&array_size, argv); |
589 | if (!argv) | |
590 | return -ENOMEM; | |
591 | ||
592 | while (1) { | |
1da177e4 | 593 | /* Skip whitespace */ |
e7d2860b | 594 | start = skip_spaces(end); |
1da177e4 LT |
595 | |
596 | if (!*start) | |
597 | break; /* success, we hit the end */ | |
598 | ||
599 | /* 'out' is used to remove any back-quotes */ | |
600 | end = out = start; | |
601 | while (*end) { | |
602 | /* Everything apart from '\0' can be quoted */ | |
603 | if (*end == '\\' && *(end + 1)) { | |
604 | *out++ = *(end + 1); | |
605 | end += 2; | |
606 | continue; | |
607 | } | |
608 | ||
609 | if (isspace(*end)) | |
610 | break; /* end of token */ | |
611 | ||
612 | *out++ = *end++; | |
613 | } | |
614 | ||
615 | /* have we already filled the array ? */ | |
616 | if ((*argc + 1) > array_size) { | |
617 | argv = realloc_argv(&array_size, argv); | |
618 | if (!argv) | |
619 | return -ENOMEM; | |
620 | } | |
621 | ||
622 | /* we know this is whitespace */ | |
623 | if (*end) | |
624 | end++; | |
625 | ||
626 | /* terminate the string and put it in the array */ | |
627 | *out = '\0'; | |
628 | argv[*argc] = start; | |
629 | (*argc)++; | |
630 | } | |
631 | ||
632 | *argvp = argv; | |
633 | return 0; | |
634 | } | |
635 | ||
be6d4305 MS |
636 | /* |
637 | * Impose necessary and sufficient conditions on a devices's table such | |
638 | * that any incoming bio which respects its logical_block_size can be | |
639 | * processed successfully. If it falls across the boundary between | |
640 | * two or more targets, the size of each piece it gets split into must | |
641 | * be compatible with the logical_block_size of the target processing it. | |
642 | */ | |
754c5fc7 MS |
643 | static int validate_hardware_logical_block_alignment(struct dm_table *table, |
644 | struct queue_limits *limits) | |
be6d4305 MS |
645 | { |
646 | /* | |
647 | * This function uses arithmetic modulo the logical_block_size | |
648 | * (in units of 512-byte sectors). | |
649 | */ | |
650 | unsigned short device_logical_block_size_sects = | |
754c5fc7 | 651 | limits->logical_block_size >> SECTOR_SHIFT; |
be6d4305 MS |
652 | |
653 | /* | |
654 | * Offset of the start of the next table entry, mod logical_block_size. | |
655 | */ | |
656 | unsigned short next_target_start = 0; | |
657 | ||
658 | /* | |
659 | * Given an aligned bio that extends beyond the end of a | |
660 | * target, how many sectors must the next target handle? | |
661 | */ | |
662 | unsigned short remaining = 0; | |
663 | ||
664 | struct dm_target *uninitialized_var(ti); | |
754c5fc7 | 665 | struct queue_limits ti_limits; |
3c120169 | 666 | unsigned i; |
be6d4305 MS |
667 | |
668 | /* | |
669 | * Check each entry in the table in turn. | |
670 | */ | |
3c120169 MP |
671 | for (i = 0; i < dm_table_get_num_targets(table); i++) { |
672 | ti = dm_table_get_target(table, i); | |
be6d4305 | 673 | |
b1bd055d | 674 | blk_set_stacking_limits(&ti_limits); |
754c5fc7 MS |
675 | |
676 | /* combine all target devices' limits */ | |
677 | if (ti->type->iterate_devices) | |
678 | ti->type->iterate_devices(ti, dm_set_device_limits, | |
679 | &ti_limits); | |
680 | ||
be6d4305 MS |
681 | /* |
682 | * If the remaining sectors fall entirely within this | |
683 | * table entry are they compatible with its logical_block_size? | |
684 | */ | |
685 | if (remaining < ti->len && | |
754c5fc7 | 686 | remaining & ((ti_limits.logical_block_size >> |
be6d4305 MS |
687 | SECTOR_SHIFT) - 1)) |
688 | break; /* Error */ | |
689 | ||
690 | next_target_start = | |
691 | (unsigned short) ((next_target_start + ti->len) & | |
692 | (device_logical_block_size_sects - 1)); | |
693 | remaining = next_target_start ? | |
694 | device_logical_block_size_sects - next_target_start : 0; | |
695 | } | |
696 | ||
697 | if (remaining) { | |
698 | DMWARN("%s: table line %u (start sect %llu len %llu) " | |
a963a956 | 699 | "not aligned to h/w logical block size %u", |
be6d4305 MS |
700 | dm_device_name(table->md), i, |
701 | (unsigned long long) ti->begin, | |
702 | (unsigned long long) ti->len, | |
754c5fc7 | 703 | limits->logical_block_size); |
be6d4305 MS |
704 | return -EINVAL; |
705 | } | |
706 | ||
707 | return 0; | |
708 | } | |
709 | ||
1da177e4 LT |
710 | int dm_table_add_target(struct dm_table *t, const char *type, |
711 | sector_t start, sector_t len, char *params) | |
712 | { | |
713 | int r = -EINVAL, argc; | |
714 | char **argv; | |
715 | struct dm_target *tgt; | |
716 | ||
3791e2fc AK |
717 | if (t->singleton) { |
718 | DMERR("%s: target type %s must appear alone in table", | |
719 | dm_device_name(t->md), t->targets->type->name); | |
720 | return -EINVAL; | |
721 | } | |
722 | ||
57a2f238 | 723 | BUG_ON(t->num_targets >= t->num_allocated); |
1da177e4 LT |
724 | |
725 | tgt = t->targets + t->num_targets; | |
726 | memset(tgt, 0, sizeof(*tgt)); | |
727 | ||
728 | if (!len) { | |
72d94861 | 729 | DMERR("%s: zero-length target", dm_device_name(t->md)); |
1da177e4 LT |
730 | return -EINVAL; |
731 | } | |
732 | ||
733 | tgt->type = dm_get_target_type(type); | |
734 | if (!tgt->type) { | |
dafa724b | 735 | DMERR("%s: %s: unknown target type", dm_device_name(t->md), type); |
1da177e4 LT |
736 | return -EINVAL; |
737 | } | |
738 | ||
3791e2fc AK |
739 | if (dm_target_needs_singleton(tgt->type)) { |
740 | if (t->num_targets) { | |
dafa724b | 741 | tgt->error = "singleton target type must appear alone in table"; |
742 | goto bad; | |
3791e2fc | 743 | } |
e83068a5 | 744 | t->singleton = true; |
3791e2fc AK |
745 | } |
746 | ||
cc6cbe14 | 747 | if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) { |
dafa724b | 748 | tgt->error = "target type may not be included in a read-only table"; |
749 | goto bad; | |
cc6cbe14 AK |
750 | } |
751 | ||
36a0456f AK |
752 | if (t->immutable_target_type) { |
753 | if (t->immutable_target_type != tgt->type) { | |
dafa724b | 754 | tgt->error = "immutable target type cannot be mixed with other target types"; |
755 | goto bad; | |
36a0456f AK |
756 | } |
757 | } else if (dm_target_is_immutable(tgt->type)) { | |
758 | if (t->num_targets) { | |
dafa724b | 759 | tgt->error = "immutable target type cannot be mixed with other target types"; |
760 | goto bad; | |
36a0456f AK |
761 | } |
762 | t->immutable_target_type = tgt->type; | |
763 | } | |
764 | ||
9b4b5a79 MB |
765 | if (dm_target_has_integrity(tgt->type)) |
766 | t->integrity_added = 1; | |
767 | ||
1da177e4 LT |
768 | tgt->table = t; |
769 | tgt->begin = start; | |
770 | tgt->len = len; | |
771 | tgt->error = "Unknown error"; | |
772 | ||
773 | /* | |
774 | * Does this target adjoin the previous one ? | |
775 | */ | |
776 | if (!adjoin(t, tgt)) { | |
777 | tgt->error = "Gap in table"; | |
1da177e4 LT |
778 | goto bad; |
779 | } | |
780 | ||
781 | r = dm_split_args(&argc, &argv, params); | |
782 | if (r) { | |
783 | tgt->error = "couldn't split parameters (insufficient memory)"; | |
784 | goto bad; | |
785 | } | |
786 | ||
787 | r = tgt->type->ctr(tgt, argc, argv); | |
788 | kfree(argv); | |
789 | if (r) | |
790 | goto bad; | |
791 | ||
792 | t->highs[t->num_targets++] = tgt->begin + tgt->len - 1; | |
793 | ||
55a62eef AK |
794 | if (!tgt->num_discard_bios && tgt->discards_supported) |
795 | DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.", | |
936688d7 | 796 | dm_device_name(t->md), type); |
5ae89a87 | 797 | |
1da177e4 LT |
798 | return 0; |
799 | ||
800 | bad: | |
72d94861 | 801 | DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error); |
1da177e4 LT |
802 | dm_put_target_type(tgt->type); |
803 | return r; | |
804 | } | |
805 | ||
498f0103 MS |
806 | /* |
807 | * Target argument parsing helpers. | |
808 | */ | |
809 | static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set, | |
810 | unsigned *value, char **error, unsigned grouped) | |
811 | { | |
812 | const char *arg_str = dm_shift_arg(arg_set); | |
31998ef1 | 813 | char dummy; |
498f0103 MS |
814 | |
815 | if (!arg_str || | |
31998ef1 | 816 | (sscanf(arg_str, "%u%c", value, &dummy) != 1) || |
498f0103 MS |
817 | (*value < arg->min) || |
818 | (*value > arg->max) || | |
819 | (grouped && arg_set->argc < *value)) { | |
820 | *error = arg->error; | |
821 | return -EINVAL; | |
822 | } | |
823 | ||
824 | return 0; | |
825 | } | |
826 | ||
827 | int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set, | |
828 | unsigned *value, char **error) | |
829 | { | |
830 | return validate_next_arg(arg, arg_set, value, error, 0); | |
831 | } | |
832 | EXPORT_SYMBOL(dm_read_arg); | |
833 | ||
834 | int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set, | |
835 | unsigned *value, char **error) | |
836 | { | |
837 | return validate_next_arg(arg, arg_set, value, error, 1); | |
838 | } | |
839 | EXPORT_SYMBOL(dm_read_arg_group); | |
840 | ||
841 | const char *dm_shift_arg(struct dm_arg_set *as) | |
842 | { | |
843 | char *r; | |
844 | ||
845 | if (as->argc) { | |
846 | as->argc--; | |
847 | r = *as->argv; | |
848 | as->argv++; | |
849 | return r; | |
850 | } | |
851 | ||
852 | return NULL; | |
853 | } | |
854 | EXPORT_SYMBOL(dm_shift_arg); | |
855 | ||
856 | void dm_consume_args(struct dm_arg_set *as, unsigned num_args) | |
857 | { | |
858 | BUG_ON(as->argc < num_args); | |
859 | as->argc -= num_args; | |
860 | as->argv += num_args; | |
861 | } | |
862 | EXPORT_SYMBOL(dm_consume_args); | |
863 | ||
7e0d574f | 864 | static bool __table_type_bio_based(enum dm_queue_mode table_type) |
545ed20e TK |
865 | { |
866 | return (table_type == DM_TYPE_BIO_BASED || | |
867 | table_type == DM_TYPE_DAX_BIO_BASED); | |
868 | } | |
869 | ||
7e0d574f | 870 | static bool __table_type_request_based(enum dm_queue_mode table_type) |
15b94a69 JN |
871 | { |
872 | return (table_type == DM_TYPE_REQUEST_BASED || | |
873 | table_type == DM_TYPE_MQ_REQUEST_BASED); | |
874 | } | |
875 | ||
7e0d574f | 876 | void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type) |
e83068a5 MS |
877 | { |
878 | t->type = type; | |
879 | } | |
880 | EXPORT_SYMBOL_GPL(dm_table_set_type); | |
881 | ||
545ed20e TK |
882 | static int device_supports_dax(struct dm_target *ti, struct dm_dev *dev, |
883 | sector_t start, sector_t len, void *data) | |
884 | { | |
885 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
886 | ||
887 | return q && blk_queue_dax(q); | |
888 | } | |
889 | ||
890 | static bool dm_table_supports_dax(struct dm_table *t) | |
891 | { | |
892 | struct dm_target *ti; | |
3c120169 | 893 | unsigned i; |
545ed20e TK |
894 | |
895 | /* Ensure that all targets support DAX. */ | |
3c120169 MP |
896 | for (i = 0; i < dm_table_get_num_targets(t); i++) { |
897 | ti = dm_table_get_target(t, i); | |
545ed20e TK |
898 | |
899 | if (!ti->type->direct_access) | |
900 | return false; | |
901 | ||
902 | if (!ti->type->iterate_devices || | |
903 | !ti->type->iterate_devices(ti, device_supports_dax, NULL)) | |
904 | return false; | |
905 | } | |
906 | ||
907 | return true; | |
908 | } | |
909 | ||
e83068a5 | 910 | static int dm_table_determine_type(struct dm_table *t) |
e6ee8c0b KU |
911 | { |
912 | unsigned i; | |
169e2cc2 | 913 | unsigned bio_based = 0, request_based = 0, hybrid = 0; |
5b8c01f7 | 914 | unsigned sq_count = 0, mq_count = 0; |
e6ee8c0b KU |
915 | struct dm_target *tgt; |
916 | struct dm_dev_internal *dd; | |
e83068a5 | 917 | struct list_head *devices = dm_table_get_devices(t); |
7e0d574f | 918 | enum dm_queue_mode live_md_type = dm_get_md_type(t->md); |
e6ee8c0b | 919 | |
e83068a5 MS |
920 | if (t->type != DM_TYPE_NONE) { |
921 | /* target already set the table's type */ | |
922 | if (t->type == DM_TYPE_BIO_BASED) | |
923 | return 0; | |
545ed20e | 924 | BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED); |
e83068a5 MS |
925 | goto verify_rq_based; |
926 | } | |
927 | ||
e6ee8c0b KU |
928 | for (i = 0; i < t->num_targets; i++) { |
929 | tgt = t->targets + i; | |
169e2cc2 MS |
930 | if (dm_target_hybrid(tgt)) |
931 | hybrid = 1; | |
932 | else if (dm_target_request_based(tgt)) | |
e6ee8c0b KU |
933 | request_based = 1; |
934 | else | |
935 | bio_based = 1; | |
936 | ||
937 | if (bio_based && request_based) { | |
938 | DMWARN("Inconsistent table: different target types" | |
939 | " can't be mixed up"); | |
940 | return -EINVAL; | |
941 | } | |
942 | } | |
943 | ||
169e2cc2 MS |
944 | if (hybrid && !bio_based && !request_based) { |
945 | /* | |
946 | * The targets can work either way. | |
947 | * Determine the type from the live device. | |
948 | * Default to bio-based if device is new. | |
949 | */ | |
15b94a69 | 950 | if (__table_type_request_based(live_md_type)) |
169e2cc2 MS |
951 | request_based = 1; |
952 | else | |
953 | bio_based = 1; | |
954 | } | |
955 | ||
e6ee8c0b KU |
956 | if (bio_based) { |
957 | /* We must use this table as bio-based */ | |
958 | t->type = DM_TYPE_BIO_BASED; | |
f8df1fdf MS |
959 | if (dm_table_supports_dax(t) || |
960 | (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) | |
545ed20e | 961 | t->type = DM_TYPE_DAX_BIO_BASED; |
e6ee8c0b KU |
962 | return 0; |
963 | } | |
964 | ||
965 | BUG_ON(!request_based); /* No targets in this table */ | |
966 | ||
e83068a5 MS |
967 | /* |
968 | * The only way to establish DM_TYPE_MQ_REQUEST_BASED is by | |
969 | * having a compatible target use dm_table_set_type. | |
970 | */ | |
971 | t->type = DM_TYPE_REQUEST_BASED; | |
972 | ||
973 | verify_rq_based: | |
65803c20 MS |
974 | /* |
975 | * Request-based dm supports only tables that have a single target now. | |
976 | * To support multiple targets, request splitting support is needed, | |
977 | * and that needs lots of changes in the block-layer. | |
978 | * (e.g. request completion process for partial completion.) | |
979 | */ | |
980 | if (t->num_targets > 1) { | |
981 | DMWARN("Request-based dm doesn't support multiple targets yet"); | |
982 | return -EINVAL; | |
983 | } | |
984 | ||
6936c12c MS |
985 | if (list_empty(devices)) { |
986 | int srcu_idx; | |
987 | struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx); | |
988 | ||
989 | /* inherit live table's type and all_blk_mq */ | |
990 | if (live_table) { | |
991 | t->type = live_table->type; | |
992 | t->all_blk_mq = live_table->all_blk_mq; | |
993 | } | |
994 | dm_put_live_table(t->md, srcu_idx); | |
995 | return 0; | |
996 | } | |
997 | ||
e6ee8c0b | 998 | /* Non-request-stackable devices can't be used for request-based dm */ |
e6ee8c0b | 999 | list_for_each_entry(dd, devices, list) { |
e5863d9a MS |
1000 | struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev); |
1001 | ||
1002 | if (!blk_queue_stackable(q)) { | |
1003 | DMERR("table load rejected: including" | |
1004 | " non-request-stackable devices"); | |
e6ee8c0b KU |
1005 | return -EINVAL; |
1006 | } | |
e5863d9a MS |
1007 | |
1008 | if (q->mq_ops) | |
5b8c01f7 BVA |
1009 | mq_count++; |
1010 | else | |
1011 | sq_count++; | |
e5863d9a | 1012 | } |
5b8c01f7 BVA |
1013 | if (sq_count && mq_count) { |
1014 | DMERR("table load rejected: not all devices are blk-mq request-stackable"); | |
1015 | return -EINVAL; | |
e83068a5 | 1016 | } |
5b8c01f7 | 1017 | t->all_blk_mq = mq_count > 0; |
e6ee8c0b | 1018 | |
301fc3f5 BVA |
1019 | if (t->type == DM_TYPE_MQ_REQUEST_BASED && !t->all_blk_mq) { |
1020 | DMERR("table load rejected: all devices are not blk-mq request-stackable"); | |
1021 | return -EINVAL; | |
1022 | } | |
1023 | ||
e6ee8c0b KU |
1024 | return 0; |
1025 | } | |
1026 | ||
7e0d574f | 1027 | enum dm_queue_mode dm_table_get_type(struct dm_table *t) |
e6ee8c0b KU |
1028 | { |
1029 | return t->type; | |
1030 | } | |
1031 | ||
36a0456f AK |
1032 | struct target_type *dm_table_get_immutable_target_type(struct dm_table *t) |
1033 | { | |
1034 | return t->immutable_target_type; | |
1035 | } | |
1036 | ||
16f12266 MS |
1037 | struct dm_target *dm_table_get_immutable_target(struct dm_table *t) |
1038 | { | |
1039 | /* Immutable target is implicitly a singleton */ | |
1040 | if (t->num_targets > 1 || | |
1041 | !dm_target_is_immutable(t->targets[0].type)) | |
1042 | return NULL; | |
1043 | ||
1044 | return t->targets; | |
1045 | } | |
1046 | ||
f083b09b MS |
1047 | struct dm_target *dm_table_get_wildcard_target(struct dm_table *t) |
1048 | { | |
3c120169 MP |
1049 | struct dm_target *ti; |
1050 | unsigned i; | |
f083b09b | 1051 | |
3c120169 MP |
1052 | for (i = 0; i < dm_table_get_num_targets(t); i++) { |
1053 | ti = dm_table_get_target(t, i); | |
f083b09b MS |
1054 | if (dm_target_is_wildcard(ti->type)) |
1055 | return ti; | |
1056 | } | |
1057 | ||
1058 | return NULL; | |
1059 | } | |
1060 | ||
545ed20e TK |
1061 | bool dm_table_bio_based(struct dm_table *t) |
1062 | { | |
1063 | return __table_type_bio_based(dm_table_get_type(t)); | |
1064 | } | |
1065 | ||
e6ee8c0b KU |
1066 | bool dm_table_request_based(struct dm_table *t) |
1067 | { | |
15b94a69 | 1068 | return __table_type_request_based(dm_table_get_type(t)); |
e5863d9a MS |
1069 | } |
1070 | ||
e83068a5 | 1071 | bool dm_table_all_blk_mq_devices(struct dm_table *t) |
e5863d9a | 1072 | { |
e83068a5 | 1073 | return t->all_blk_mq; |
e6ee8c0b KU |
1074 | } |
1075 | ||
17e149b8 | 1076 | static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md) |
e6ee8c0b | 1077 | { |
7e0d574f | 1078 | enum dm_queue_mode type = dm_table_get_type(t); |
30187e1d | 1079 | unsigned per_io_data_size = 0; |
78d8e58a | 1080 | struct dm_target *tgt; |
c0820cf5 | 1081 | unsigned i; |
e6ee8c0b | 1082 | |
78d8e58a | 1083 | if (unlikely(type == DM_TYPE_NONE)) { |
e6ee8c0b KU |
1084 | DMWARN("no table type is set, can't allocate mempools"); |
1085 | return -EINVAL; | |
1086 | } | |
1087 | ||
545ed20e | 1088 | if (__table_type_bio_based(type)) |
78d8e58a MS |
1089 | for (i = 0; i < t->num_targets; i++) { |
1090 | tgt = t->targets + i; | |
30187e1d | 1091 | per_io_data_size = max(per_io_data_size, tgt->per_io_data_size); |
78d8e58a MS |
1092 | } |
1093 | ||
30187e1d | 1094 | t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported, per_io_data_size); |
4e6e36c3 MS |
1095 | if (!t->mempools) |
1096 | return -ENOMEM; | |
e6ee8c0b KU |
1097 | |
1098 | return 0; | |
1099 | } | |
1100 | ||
1101 | void dm_table_free_md_mempools(struct dm_table *t) | |
1102 | { | |
1103 | dm_free_md_mempools(t->mempools); | |
1104 | t->mempools = NULL; | |
1105 | } | |
1106 | ||
1107 | struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t) | |
1108 | { | |
1109 | return t->mempools; | |
1110 | } | |
1111 | ||
1da177e4 LT |
1112 | static int setup_indexes(struct dm_table *t) |
1113 | { | |
1114 | int i; | |
1115 | unsigned int total = 0; | |
1116 | sector_t *indexes; | |
1117 | ||
1118 | /* allocate the space for *all* the indexes */ | |
1119 | for (i = t->depth - 2; i >= 0; i--) { | |
1120 | t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE); | |
1121 | total += t->counts[i]; | |
1122 | } | |
1123 | ||
1124 | indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE); | |
1125 | if (!indexes) | |
1126 | return -ENOMEM; | |
1127 | ||
1128 | /* set up internal nodes, bottom-up */ | |
82d601dc | 1129 | for (i = t->depth - 2; i >= 0; i--) { |
1da177e4 LT |
1130 | t->index[i] = indexes; |
1131 | indexes += (KEYS_PER_NODE * t->counts[i]); | |
1132 | setup_btree_index(i, t); | |
1133 | } | |
1134 | ||
1135 | return 0; | |
1136 | } | |
1137 | ||
1138 | /* | |
1139 | * Builds the btree to index the map. | |
1140 | */ | |
26803b9f | 1141 | static int dm_table_build_index(struct dm_table *t) |
1da177e4 LT |
1142 | { |
1143 | int r = 0; | |
1144 | unsigned int leaf_nodes; | |
1145 | ||
1da177e4 LT |
1146 | /* how many indexes will the btree have ? */ |
1147 | leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE); | |
1148 | t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE); | |
1149 | ||
1150 | /* leaf layer has already been set up */ | |
1151 | t->counts[t->depth - 1] = leaf_nodes; | |
1152 | t->index[t->depth - 1] = t->highs; | |
1153 | ||
1154 | if (t->depth >= 2) | |
1155 | r = setup_indexes(t); | |
1156 | ||
1157 | return r; | |
1158 | } | |
1159 | ||
25520d55 MP |
1160 | static bool integrity_profile_exists(struct gendisk *disk) |
1161 | { | |
1162 | return !!blk_get_integrity(disk); | |
1163 | } | |
1164 | ||
a63a5cf8 MS |
1165 | /* |
1166 | * Get a disk whose integrity profile reflects the table's profile. | |
a63a5cf8 MS |
1167 | * Returns NULL if integrity support was inconsistent or unavailable. |
1168 | */ | |
25520d55 | 1169 | static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t) |
a63a5cf8 MS |
1170 | { |
1171 | struct list_head *devices = dm_table_get_devices(t); | |
1172 | struct dm_dev_internal *dd = NULL; | |
1173 | struct gendisk *prev_disk = NULL, *template_disk = NULL; | |
e2460f2a MP |
1174 | unsigned i; |
1175 | ||
1176 | for (i = 0; i < dm_table_get_num_targets(t); i++) { | |
1177 | struct dm_target *ti = dm_table_get_target(t, i); | |
1178 | if (!dm_target_passes_integrity(ti->type)) | |
1179 | goto no_integrity; | |
1180 | } | |
a63a5cf8 MS |
1181 | |
1182 | list_for_each_entry(dd, devices, list) { | |
86f1152b | 1183 | template_disk = dd->dm_dev->bdev->bd_disk; |
25520d55 | 1184 | if (!integrity_profile_exists(template_disk)) |
a63a5cf8 | 1185 | goto no_integrity; |
a63a5cf8 MS |
1186 | else if (prev_disk && |
1187 | blk_integrity_compare(prev_disk, template_disk) < 0) | |
1188 | goto no_integrity; | |
1189 | prev_disk = template_disk; | |
1190 | } | |
1191 | ||
1192 | return template_disk; | |
1193 | ||
1194 | no_integrity: | |
1195 | if (prev_disk) | |
1196 | DMWARN("%s: integrity not set: %s and %s profile mismatch", | |
1197 | dm_device_name(t->md), | |
1198 | prev_disk->disk_name, | |
1199 | template_disk->disk_name); | |
1200 | return NULL; | |
1201 | } | |
1202 | ||
26803b9f | 1203 | /* |
25520d55 MP |
1204 | * Register the mapped device for blk_integrity support if the |
1205 | * underlying devices have an integrity profile. But all devices may | |
1206 | * not have matching profiles (checking all devices isn't reliable | |
a63a5cf8 | 1207 | * during table load because this table may use other DM device(s) which |
25520d55 MP |
1208 | * must be resumed before they will have an initialized integity |
1209 | * profile). Consequently, stacked DM devices force a 2 stage integrity | |
1210 | * profile validation: First pass during table load, final pass during | |
1211 | * resume. | |
26803b9f | 1212 | */ |
25520d55 | 1213 | static int dm_table_register_integrity(struct dm_table *t) |
26803b9f | 1214 | { |
25520d55 | 1215 | struct mapped_device *md = t->md; |
a63a5cf8 | 1216 | struct gendisk *template_disk = NULL; |
26803b9f | 1217 | |
9b4b5a79 MB |
1218 | /* If target handles integrity itself do not register it here. */ |
1219 | if (t->integrity_added) | |
1220 | return 0; | |
1221 | ||
25520d55 | 1222 | template_disk = dm_table_get_integrity_disk(t); |
a63a5cf8 MS |
1223 | if (!template_disk) |
1224 | return 0; | |
26803b9f | 1225 | |
25520d55 | 1226 | if (!integrity_profile_exists(dm_disk(md))) { |
e83068a5 | 1227 | t->integrity_supported = true; |
25520d55 MP |
1228 | /* |
1229 | * Register integrity profile during table load; we can do | |
1230 | * this because the final profile must match during resume. | |
1231 | */ | |
1232 | blk_integrity_register(dm_disk(md), | |
1233 | blk_get_integrity(template_disk)); | |
1234 | return 0; | |
a63a5cf8 MS |
1235 | } |
1236 | ||
1237 | /* | |
25520d55 | 1238 | * If DM device already has an initialized integrity |
a63a5cf8 MS |
1239 | * profile the new profile should not conflict. |
1240 | */ | |
25520d55 | 1241 | if (blk_integrity_compare(dm_disk(md), template_disk) < 0) { |
a63a5cf8 MS |
1242 | DMWARN("%s: conflict with existing integrity profile: " |
1243 | "%s profile mismatch", | |
1244 | dm_device_name(t->md), | |
1245 | template_disk->disk_name); | |
1246 | return 1; | |
1247 | } | |
1248 | ||
25520d55 | 1249 | /* Preserve existing integrity profile */ |
e83068a5 | 1250 | t->integrity_supported = true; |
26803b9f WD |
1251 | return 0; |
1252 | } | |
1253 | ||
1254 | /* | |
1255 | * Prepares the table for use by building the indices, | |
1256 | * setting the type, and allocating mempools. | |
1257 | */ | |
1258 | int dm_table_complete(struct dm_table *t) | |
1259 | { | |
1260 | int r; | |
1261 | ||
e83068a5 | 1262 | r = dm_table_determine_type(t); |
26803b9f | 1263 | if (r) { |
e83068a5 | 1264 | DMERR("unable to determine table type"); |
26803b9f WD |
1265 | return r; |
1266 | } | |
1267 | ||
1268 | r = dm_table_build_index(t); | |
1269 | if (r) { | |
1270 | DMERR("unable to build btrees"); | |
1271 | return r; | |
1272 | } | |
1273 | ||
25520d55 | 1274 | r = dm_table_register_integrity(t); |
26803b9f WD |
1275 | if (r) { |
1276 | DMERR("could not register integrity profile."); | |
1277 | return r; | |
1278 | } | |
1279 | ||
17e149b8 | 1280 | r = dm_table_alloc_md_mempools(t, t->md); |
26803b9f WD |
1281 | if (r) |
1282 | DMERR("unable to allocate mempools"); | |
1283 | ||
1284 | return r; | |
1285 | } | |
1286 | ||
48c9c27b | 1287 | static DEFINE_MUTEX(_event_lock); |
1da177e4 LT |
1288 | void dm_table_event_callback(struct dm_table *t, |
1289 | void (*fn)(void *), void *context) | |
1290 | { | |
48c9c27b | 1291 | mutex_lock(&_event_lock); |
1da177e4 LT |
1292 | t->event_fn = fn; |
1293 | t->event_context = context; | |
48c9c27b | 1294 | mutex_unlock(&_event_lock); |
1da177e4 LT |
1295 | } |
1296 | ||
1297 | void dm_table_event(struct dm_table *t) | |
1298 | { | |
1299 | /* | |
1300 | * You can no longer call dm_table_event() from interrupt | |
1301 | * context, use a bottom half instead. | |
1302 | */ | |
1303 | BUG_ON(in_interrupt()); | |
1304 | ||
48c9c27b | 1305 | mutex_lock(&_event_lock); |
1da177e4 LT |
1306 | if (t->event_fn) |
1307 | t->event_fn(t->event_context); | |
48c9c27b | 1308 | mutex_unlock(&_event_lock); |
1da177e4 | 1309 | } |
08649012 | 1310 | EXPORT_SYMBOL(dm_table_event); |
1da177e4 LT |
1311 | |
1312 | sector_t dm_table_get_size(struct dm_table *t) | |
1313 | { | |
1314 | return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0; | |
1315 | } | |
08649012 | 1316 | EXPORT_SYMBOL(dm_table_get_size); |
1da177e4 LT |
1317 | |
1318 | struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index) | |
1319 | { | |
14353539 | 1320 | if (index >= t->num_targets) |
1da177e4 LT |
1321 | return NULL; |
1322 | ||
1323 | return t->targets + index; | |
1324 | } | |
1325 | ||
1326 | /* | |
1327 | * Search the btree for the correct target. | |
512875bd JN |
1328 | * |
1329 | * Caller should check returned pointer with dm_target_is_valid() | |
1330 | * to trap I/O beyond end of device. | |
1da177e4 LT |
1331 | */ |
1332 | struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector) | |
1333 | { | |
1334 | unsigned int l, n = 0, k = 0; | |
1335 | sector_t *node; | |
1336 | ||
1337 | for (l = 0; l < t->depth; l++) { | |
1338 | n = get_child(n, k); | |
1339 | node = get_node(t, l, n); | |
1340 | ||
1341 | for (k = 0; k < KEYS_PER_NODE; k++) | |
1342 | if (node[k] >= sector) | |
1343 | break; | |
1344 | } | |
1345 | ||
1346 | return &t->targets[(KEYS_PER_NODE * n) + k]; | |
1347 | } | |
1348 | ||
3ae70656 MS |
1349 | static int count_device(struct dm_target *ti, struct dm_dev *dev, |
1350 | sector_t start, sector_t len, void *data) | |
1351 | { | |
1352 | unsigned *num_devices = data; | |
1353 | ||
1354 | (*num_devices)++; | |
1355 | ||
1356 | return 0; | |
1357 | } | |
1358 | ||
1359 | /* | |
1360 | * Check whether a table has no data devices attached using each | |
1361 | * target's iterate_devices method. | |
1362 | * Returns false if the result is unknown because a target doesn't | |
1363 | * support iterate_devices. | |
1364 | */ | |
1365 | bool dm_table_has_no_data_devices(struct dm_table *table) | |
1366 | { | |
3c120169 MP |
1367 | struct dm_target *ti; |
1368 | unsigned i, num_devices; | |
3ae70656 | 1369 | |
3c120169 MP |
1370 | for (i = 0; i < dm_table_get_num_targets(table); i++) { |
1371 | ti = dm_table_get_target(table, i); | |
3ae70656 MS |
1372 | |
1373 | if (!ti->type->iterate_devices) | |
1374 | return false; | |
1375 | ||
3c120169 | 1376 | num_devices = 0; |
3ae70656 MS |
1377 | ti->type->iterate_devices(ti, count_device, &num_devices); |
1378 | if (num_devices) | |
1379 | return false; | |
1380 | } | |
1381 | ||
1382 | return true; | |
1383 | } | |
1384 | ||
dd88d313 DLM |
1385 | static int device_is_zoned_model(struct dm_target *ti, struct dm_dev *dev, |
1386 | sector_t start, sector_t len, void *data) | |
1387 | { | |
1388 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
1389 | enum blk_zoned_model *zoned_model = data; | |
1390 | ||
1391 | return q && blk_queue_zoned_model(q) == *zoned_model; | |
1392 | } | |
1393 | ||
1394 | static bool dm_table_supports_zoned_model(struct dm_table *t, | |
1395 | enum blk_zoned_model zoned_model) | |
1396 | { | |
1397 | struct dm_target *ti; | |
1398 | unsigned i; | |
1399 | ||
1400 | for (i = 0; i < dm_table_get_num_targets(t); i++) { | |
1401 | ti = dm_table_get_target(t, i); | |
1402 | ||
1403 | if (zoned_model == BLK_ZONED_HM && | |
1404 | !dm_target_supports_zoned_hm(ti->type)) | |
1405 | return false; | |
1406 | ||
1407 | if (!ti->type->iterate_devices || | |
1408 | !ti->type->iterate_devices(ti, device_is_zoned_model, &zoned_model)) | |
1409 | return false; | |
1410 | } | |
1411 | ||
1412 | return true; | |
1413 | } | |
1414 | ||
1415 | static int device_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev, | |
1416 | sector_t start, sector_t len, void *data) | |
1417 | { | |
1418 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
1419 | unsigned int *zone_sectors = data; | |
1420 | ||
1421 | return q && blk_queue_zone_sectors(q) == *zone_sectors; | |
1422 | } | |
1423 | ||
1424 | static bool dm_table_matches_zone_sectors(struct dm_table *t, | |
1425 | unsigned int zone_sectors) | |
1426 | { | |
1427 | struct dm_target *ti; | |
1428 | unsigned i; | |
1429 | ||
1430 | for (i = 0; i < dm_table_get_num_targets(t); i++) { | |
1431 | ti = dm_table_get_target(t, i); | |
1432 | ||
1433 | if (!ti->type->iterate_devices || | |
1434 | !ti->type->iterate_devices(ti, device_matches_zone_sectors, &zone_sectors)) | |
1435 | return false; | |
1436 | } | |
1437 | ||
1438 | return true; | |
1439 | } | |
1440 | ||
1441 | static int validate_hardware_zoned_model(struct dm_table *table, | |
1442 | enum blk_zoned_model zoned_model, | |
1443 | unsigned int zone_sectors) | |
1444 | { | |
1445 | if (zoned_model == BLK_ZONED_NONE) | |
1446 | return 0; | |
1447 | ||
1448 | if (!dm_table_supports_zoned_model(table, zoned_model)) { | |
1449 | DMERR("%s: zoned model is not consistent across all devices", | |
1450 | dm_device_name(table->md)); | |
1451 | return -EINVAL; | |
1452 | } | |
1453 | ||
1454 | /* Check zone size validity and compatibility */ | |
1455 | if (!zone_sectors || !is_power_of_2(zone_sectors)) | |
1456 | return -EINVAL; | |
1457 | ||
1458 | if (!dm_table_matches_zone_sectors(table, zone_sectors)) { | |
1459 | DMERR("%s: zone sectors is not consistent across all devices", | |
1460 | dm_device_name(table->md)); | |
1461 | return -EINVAL; | |
1462 | } | |
1463 | ||
1464 | return 0; | |
1465 | } | |
1466 | ||
754c5fc7 MS |
1467 | /* |
1468 | * Establish the new table's queue_limits and validate them. | |
1469 | */ | |
1470 | int dm_calculate_queue_limits(struct dm_table *table, | |
1471 | struct queue_limits *limits) | |
1472 | { | |
3c120169 | 1473 | struct dm_target *ti; |
754c5fc7 | 1474 | struct queue_limits ti_limits; |
3c120169 | 1475 | unsigned i; |
dd88d313 DLM |
1476 | enum blk_zoned_model zoned_model = BLK_ZONED_NONE; |
1477 | unsigned int zone_sectors = 0; | |
754c5fc7 | 1478 | |
b1bd055d | 1479 | blk_set_stacking_limits(limits); |
754c5fc7 | 1480 | |
3c120169 | 1481 | for (i = 0; i < dm_table_get_num_targets(table); i++) { |
b1bd055d | 1482 | blk_set_stacking_limits(&ti_limits); |
754c5fc7 | 1483 | |
3c120169 | 1484 | ti = dm_table_get_target(table, i); |
754c5fc7 MS |
1485 | |
1486 | if (!ti->type->iterate_devices) | |
1487 | goto combine_limits; | |
1488 | ||
1489 | /* | |
1490 | * Combine queue limits of all the devices this target uses. | |
1491 | */ | |
1492 | ti->type->iterate_devices(ti, dm_set_device_limits, | |
1493 | &ti_limits); | |
1494 | ||
dd88d313 DLM |
1495 | if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) { |
1496 | /* | |
1497 | * After stacking all limits, validate all devices | |
1498 | * in table support this zoned model and zone sectors. | |
1499 | */ | |
1500 | zoned_model = ti_limits.zoned; | |
1501 | zone_sectors = ti_limits.chunk_sectors; | |
1502 | } | |
1503 | ||
40bea431 MS |
1504 | /* Set I/O hints portion of queue limits */ |
1505 | if (ti->type->io_hints) | |
1506 | ti->type->io_hints(ti, &ti_limits); | |
1507 | ||
754c5fc7 MS |
1508 | /* |
1509 | * Check each device area is consistent with the target's | |
1510 | * overall queue limits. | |
1511 | */ | |
f6a1ed10 MP |
1512 | if (ti->type->iterate_devices(ti, device_area_is_invalid, |
1513 | &ti_limits)) | |
754c5fc7 MS |
1514 | return -EINVAL; |
1515 | ||
1516 | combine_limits: | |
1517 | /* | |
1518 | * Merge this target's queue limits into the overall limits | |
1519 | * for the table. | |
1520 | */ | |
1521 | if (blk_stack_limits(limits, &ti_limits, 0) < 0) | |
b27d7f16 | 1522 | DMWARN("%s: adding target device " |
754c5fc7 | 1523 | "(start sect %llu len %llu) " |
b27d7f16 | 1524 | "caused an alignment inconsistency", |
754c5fc7 MS |
1525 | dm_device_name(table->md), |
1526 | (unsigned long long) ti->begin, | |
1527 | (unsigned long long) ti->len); | |
dd88d313 DLM |
1528 | |
1529 | /* | |
1530 | * FIXME: this should likely be moved to blk_stack_limits(), would | |
1531 | * also eliminate limits->zoned stacking hack in dm_set_device_limits() | |
1532 | */ | |
1533 | if (limits->zoned == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) { | |
1534 | /* | |
1535 | * By default, the stacked limits zoned model is set to | |
1536 | * BLK_ZONED_NONE in blk_set_stacking_limits(). Update | |
1537 | * this model using the first target model reported | |
1538 | * that is not BLK_ZONED_NONE. This will be either the | |
1539 | * first target device zoned model or the model reported | |
1540 | * by the target .io_hints. | |
1541 | */ | |
1542 | limits->zoned = ti_limits.zoned; | |
1543 | } | |
754c5fc7 MS |
1544 | } |
1545 | ||
dd88d313 DLM |
1546 | /* |
1547 | * Verify that the zoned model and zone sectors, as determined before | |
1548 | * any .io_hints override, are the same across all devices in the table. | |
1549 | * - this is especially relevant if .io_hints is emulating a disk-managed | |
1550 | * zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices. | |
1551 | * BUT... | |
1552 | */ | |
1553 | if (limits->zoned != BLK_ZONED_NONE) { | |
1554 | /* | |
1555 | * ...IF the above limits stacking determined a zoned model | |
1556 | * validate that all of the table's devices conform to it. | |
1557 | */ | |
1558 | zoned_model = limits->zoned; | |
1559 | zone_sectors = limits->chunk_sectors; | |
1560 | } | |
1561 | if (validate_hardware_zoned_model(table, zoned_model, zone_sectors)) | |
1562 | return -EINVAL; | |
1563 | ||
754c5fc7 MS |
1564 | return validate_hardware_logical_block_alignment(table, limits); |
1565 | } | |
1566 | ||
9c47008d | 1567 | /* |
25520d55 MP |
1568 | * Verify that all devices have an integrity profile that matches the |
1569 | * DM device's registered integrity profile. If the profiles don't | |
1570 | * match then unregister the DM device's integrity profile. | |
9c47008d | 1571 | */ |
25520d55 | 1572 | static void dm_table_verify_integrity(struct dm_table *t) |
9c47008d | 1573 | { |
a63a5cf8 | 1574 | struct gendisk *template_disk = NULL; |
9c47008d | 1575 | |
9b4b5a79 MB |
1576 | if (t->integrity_added) |
1577 | return; | |
1578 | ||
25520d55 MP |
1579 | if (t->integrity_supported) { |
1580 | /* | |
1581 | * Verify that the original integrity profile | |
1582 | * matches all the devices in this table. | |
1583 | */ | |
1584 | template_disk = dm_table_get_integrity_disk(t); | |
1585 | if (template_disk && | |
1586 | blk_integrity_compare(dm_disk(t->md), template_disk) >= 0) | |
1587 | return; | |
1588 | } | |
9c47008d | 1589 | |
25520d55 | 1590 | if (integrity_profile_exists(dm_disk(t->md))) { |
876fbba1 MS |
1591 | DMWARN("%s: unable to establish an integrity profile", |
1592 | dm_device_name(t->md)); | |
25520d55 MP |
1593 | blk_integrity_unregister(dm_disk(t->md)); |
1594 | } | |
9c47008d MP |
1595 | } |
1596 | ||
ed8b752b MS |
1597 | static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev, |
1598 | sector_t start, sector_t len, void *data) | |
1599 | { | |
c888a8f9 | 1600 | unsigned long flush = (unsigned long) data; |
ed8b752b MS |
1601 | struct request_queue *q = bdev_get_queue(dev->bdev); |
1602 | ||
c888a8f9 | 1603 | return q && (q->queue_flags & flush); |
ed8b752b MS |
1604 | } |
1605 | ||
c888a8f9 | 1606 | static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush) |
ed8b752b MS |
1607 | { |
1608 | struct dm_target *ti; | |
3c120169 | 1609 | unsigned i; |
ed8b752b MS |
1610 | |
1611 | /* | |
1612 | * Require at least one underlying device to support flushes. | |
1613 | * t->devices includes internal dm devices such as mirror logs | |
1614 | * so we need to use iterate_devices here, which targets | |
1615 | * supporting flushes must provide. | |
1616 | */ | |
3c120169 MP |
1617 | for (i = 0; i < dm_table_get_num_targets(t); i++) { |
1618 | ti = dm_table_get_target(t, i); | |
ed8b752b | 1619 | |
55a62eef | 1620 | if (!ti->num_flush_bios) |
ed8b752b MS |
1621 | continue; |
1622 | ||
0e9c24ed | 1623 | if (ti->flush_supported) |
7f61f5a0 | 1624 | return true; |
0e9c24ed | 1625 | |
ed8b752b | 1626 | if (ti->type->iterate_devices && |
c888a8f9 | 1627 | ti->type->iterate_devices(ti, device_flush_capable, (void *) flush)) |
7f61f5a0 | 1628 | return true; |
ed8b752b MS |
1629 | } |
1630 | ||
7f61f5a0 | 1631 | return false; |
ed8b752b MS |
1632 | } |
1633 | ||
273752c9 VG |
1634 | static int device_dax_write_cache_enabled(struct dm_target *ti, |
1635 | struct dm_dev *dev, sector_t start, | |
1636 | sector_t len, void *data) | |
1637 | { | |
1638 | struct dax_device *dax_dev = dev->dax_dev; | |
1639 | ||
1640 | if (!dax_dev) | |
1641 | return false; | |
1642 | ||
1643 | if (dax_write_cache_enabled(dax_dev)) | |
1644 | return true; | |
1645 | return false; | |
1646 | } | |
1647 | ||
1648 | static int dm_table_supports_dax_write_cache(struct dm_table *t) | |
1649 | { | |
1650 | struct dm_target *ti; | |
1651 | unsigned i; | |
1652 | ||
1653 | for (i = 0; i < dm_table_get_num_targets(t); i++) { | |
1654 | ti = dm_table_get_target(t, i); | |
1655 | ||
1656 | if (ti->type->iterate_devices && | |
1657 | ti->type->iterate_devices(ti, | |
1658 | device_dax_write_cache_enabled, NULL)) | |
1659 | return true; | |
1660 | } | |
1661 | ||
1662 | return false; | |
1663 | } | |
1664 | ||
4693c966 MSB |
1665 | static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev, |
1666 | sector_t start, sector_t len, void *data) | |
1667 | { | |
1668 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
1669 | ||
1670 | return q && blk_queue_nonrot(q); | |
1671 | } | |
1672 | ||
c3c4555e MB |
1673 | static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev, |
1674 | sector_t start, sector_t len, void *data) | |
1675 | { | |
1676 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
1677 | ||
1678 | return q && !blk_queue_add_random(q); | |
1679 | } | |
1680 | ||
200612ec JM |
1681 | static int queue_supports_sg_merge(struct dm_target *ti, struct dm_dev *dev, |
1682 | sector_t start, sector_t len, void *data) | |
1683 | { | |
1684 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
1685 | ||
1686 | return q && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags); | |
1687 | } | |
1688 | ||
c3c4555e MB |
1689 | static bool dm_table_all_devices_attribute(struct dm_table *t, |
1690 | iterate_devices_callout_fn func) | |
4693c966 MSB |
1691 | { |
1692 | struct dm_target *ti; | |
3c120169 | 1693 | unsigned i; |
4693c966 | 1694 | |
3c120169 MP |
1695 | for (i = 0; i < dm_table_get_num_targets(t); i++) { |
1696 | ti = dm_table_get_target(t, i); | |
4693c966 MSB |
1697 | |
1698 | if (!ti->type->iterate_devices || | |
c3c4555e | 1699 | !ti->type->iterate_devices(ti, func, NULL)) |
7f61f5a0 | 1700 | return false; |
4693c966 MSB |
1701 | } |
1702 | ||
7f61f5a0 | 1703 | return true; |
4693c966 MSB |
1704 | } |
1705 | ||
d54eaa5a MS |
1706 | static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev, |
1707 | sector_t start, sector_t len, void *data) | |
1708 | { | |
1709 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
1710 | ||
1711 | return q && !q->limits.max_write_same_sectors; | |
1712 | } | |
1713 | ||
1714 | static bool dm_table_supports_write_same(struct dm_table *t) | |
1715 | { | |
1716 | struct dm_target *ti; | |
3c120169 | 1717 | unsigned i; |
d54eaa5a | 1718 | |
3c120169 MP |
1719 | for (i = 0; i < dm_table_get_num_targets(t); i++) { |
1720 | ti = dm_table_get_target(t, i); | |
d54eaa5a | 1721 | |
55a62eef | 1722 | if (!ti->num_write_same_bios) |
d54eaa5a MS |
1723 | return false; |
1724 | ||
1725 | if (!ti->type->iterate_devices || | |
dc019b21 | 1726 | ti->type->iterate_devices(ti, device_not_write_same_capable, NULL)) |
d54eaa5a MS |
1727 | return false; |
1728 | } | |
1729 | ||
1730 | return true; | |
1731 | } | |
1732 | ||
ac62d620 CH |
1733 | static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev, |
1734 | sector_t start, sector_t len, void *data) | |
1735 | { | |
1736 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
1737 | ||
1738 | return q && !q->limits.max_write_zeroes_sectors; | |
1739 | } | |
1740 | ||
1741 | static bool dm_table_supports_write_zeroes(struct dm_table *t) | |
1742 | { | |
1743 | struct dm_target *ti; | |
1744 | unsigned i = 0; | |
1745 | ||
1746 | while (i < dm_table_get_num_targets(t)) { | |
1747 | ti = dm_table_get_target(t, i++); | |
1748 | ||
1749 | if (!ti->num_write_zeroes_bios) | |
1750 | return false; | |
1751 | ||
1752 | if (!ti->type->iterate_devices || | |
1753 | ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL)) | |
1754 | return false; | |
1755 | } | |
1756 | ||
1757 | return true; | |
1758 | } | |
1759 | ||
1760 | ||
a7ffb6a5 MP |
1761 | static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev, |
1762 | sector_t start, sector_t len, void *data) | |
1763 | { | |
1764 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
1765 | ||
1766 | return q && blk_queue_discard(q); | |
1767 | } | |
1768 | ||
1769 | static bool dm_table_supports_discards(struct dm_table *t) | |
1770 | { | |
1771 | struct dm_target *ti; | |
3c120169 | 1772 | unsigned i; |
a7ffb6a5 MP |
1773 | |
1774 | /* | |
1775 | * Unless any target used by the table set discards_supported, | |
1776 | * require at least one underlying device to support discards. | |
1777 | * t->devices includes internal dm devices such as mirror logs | |
1778 | * so we need to use iterate_devices here, which targets | |
1779 | * supporting discard selectively must provide. | |
1780 | */ | |
3c120169 MP |
1781 | for (i = 0; i < dm_table_get_num_targets(t); i++) { |
1782 | ti = dm_table_get_target(t, i); | |
a7ffb6a5 MP |
1783 | |
1784 | if (!ti->num_discard_bios) | |
1785 | continue; | |
1786 | ||
1787 | if (ti->discards_supported) | |
7f61f5a0 | 1788 | return true; |
a7ffb6a5 MP |
1789 | |
1790 | if (ti->type->iterate_devices && | |
1791 | ti->type->iterate_devices(ti, device_discard_capable, NULL)) | |
7f61f5a0 | 1792 | return true; |
a7ffb6a5 MP |
1793 | } |
1794 | ||
7f61f5a0 | 1795 | return false; |
a7ffb6a5 MP |
1796 | } |
1797 | ||
754c5fc7 MS |
1798 | void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q, |
1799 | struct queue_limits *limits) | |
1da177e4 | 1800 | { |
519a7e16 | 1801 | bool wc = false, fua = false; |
ed8b752b | 1802 | |
1da177e4 | 1803 | /* |
1197764e | 1804 | * Copy table's limits to the DM device's request_queue |
1da177e4 | 1805 | */ |
754c5fc7 | 1806 | q->limits = *limits; |
c9a3f6d6 | 1807 | |
5ae89a87 MS |
1808 | if (!dm_table_supports_discards(t)) |
1809 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); | |
1810 | else | |
1811 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); | |
1812 | ||
c888a8f9 | 1813 | if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) { |
519a7e16 | 1814 | wc = true; |
c888a8f9 | 1815 | if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA))) |
519a7e16 | 1816 | fua = true; |
ed8b752b | 1817 | } |
519a7e16 | 1818 | blk_queue_write_cache(q, wc, fua); |
ed8b752b | 1819 | |
273752c9 VG |
1820 | if (dm_table_supports_dax_write_cache(t)) |
1821 | dax_write_cache(t->md->dax_dev, true); | |
1822 | ||
c3c4555e MB |
1823 | /* Ensure that all underlying devices are non-rotational. */ |
1824 | if (dm_table_all_devices_attribute(t, device_is_nonrot)) | |
4693c966 MSB |
1825 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q); |
1826 | else | |
1827 | queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q); | |
1828 | ||
d54eaa5a MS |
1829 | if (!dm_table_supports_write_same(t)) |
1830 | q->limits.max_write_same_sectors = 0; | |
ac62d620 CH |
1831 | if (!dm_table_supports_write_zeroes(t)) |
1832 | q->limits.max_write_zeroes_sectors = 0; | |
c1a94672 | 1833 | |
200612ec JM |
1834 | if (dm_table_all_devices_attribute(t, queue_supports_sg_merge)) |
1835 | queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q); | |
1836 | else | |
1837 | queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q); | |
1838 | ||
25520d55 | 1839 | dm_table_verify_integrity(t); |
e6ee8c0b | 1840 | |
c3c4555e MB |
1841 | /* |
1842 | * Determine whether or not this queue's I/O timings contribute | |
1843 | * to the entropy pool, Only request-based targets use this. | |
1844 | * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not | |
1845 | * have it set. | |
1846 | */ | |
1847 | if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random)) | |
1848 | queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q); | |
1849 | ||
e6ee8c0b KU |
1850 | /* |
1851 | * QUEUE_FLAG_STACKABLE must be set after all queue settings are | |
1852 | * visible to other CPUs because, once the flag is set, incoming bios | |
1853 | * are processed by request-based dm, which refers to the queue | |
1854 | * settings. | |
1855 | * Until the flag set, bios are passed to bio-based dm and queued to | |
1856 | * md->deferred where queue settings are not needed yet. | |
1857 | * Those bios are passed to request-based dm at the resume time. | |
1858 | */ | |
1859 | smp_mb(); | |
1860 | if (dm_table_request_based(t)) | |
1861 | queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q); | |
1da177e4 LT |
1862 | } |
1863 | ||
1864 | unsigned int dm_table_get_num_targets(struct dm_table *t) | |
1865 | { | |
1866 | return t->num_targets; | |
1867 | } | |
1868 | ||
1869 | struct list_head *dm_table_get_devices(struct dm_table *t) | |
1870 | { | |
1871 | return &t->devices; | |
1872 | } | |
1873 | ||
aeb5d727 | 1874 | fmode_t dm_table_get_mode(struct dm_table *t) |
1da177e4 LT |
1875 | { |
1876 | return t->mode; | |
1877 | } | |
08649012 | 1878 | EXPORT_SYMBOL(dm_table_get_mode); |
1da177e4 | 1879 | |
d67ee213 MS |
1880 | enum suspend_mode { |
1881 | PRESUSPEND, | |
1882 | PRESUSPEND_UNDO, | |
1883 | POSTSUSPEND, | |
1884 | }; | |
1885 | ||
1886 | static void suspend_targets(struct dm_table *t, enum suspend_mode mode) | |
1da177e4 LT |
1887 | { |
1888 | int i = t->num_targets; | |
1889 | struct dm_target *ti = t->targets; | |
1890 | ||
1ea0654e BVA |
1891 | lockdep_assert_held(&t->md->suspend_lock); |
1892 | ||
1da177e4 | 1893 | while (i--) { |
d67ee213 MS |
1894 | switch (mode) { |
1895 | case PRESUSPEND: | |
1896 | if (ti->type->presuspend) | |
1897 | ti->type->presuspend(ti); | |
1898 | break; | |
1899 | case PRESUSPEND_UNDO: | |
1900 | if (ti->type->presuspend_undo) | |
1901 | ti->type->presuspend_undo(ti); | |
1902 | break; | |
1903 | case POSTSUSPEND: | |
1da177e4 LT |
1904 | if (ti->type->postsuspend) |
1905 | ti->type->postsuspend(ti); | |
d67ee213 MS |
1906 | break; |
1907 | } | |
1da177e4 LT |
1908 | ti++; |
1909 | } | |
1910 | } | |
1911 | ||
1912 | void dm_table_presuspend_targets(struct dm_table *t) | |
1913 | { | |
cf222b37 AK |
1914 | if (!t) |
1915 | return; | |
1916 | ||
d67ee213 MS |
1917 | suspend_targets(t, PRESUSPEND); |
1918 | } | |
1919 | ||
1920 | void dm_table_presuspend_undo_targets(struct dm_table *t) | |
1921 | { | |
1922 | if (!t) | |
1923 | return; | |
1924 | ||
1925 | suspend_targets(t, PRESUSPEND_UNDO); | |
1da177e4 LT |
1926 | } |
1927 | ||
1928 | void dm_table_postsuspend_targets(struct dm_table *t) | |
1929 | { | |
cf222b37 AK |
1930 | if (!t) |
1931 | return; | |
1932 | ||
d67ee213 | 1933 | suspend_targets(t, POSTSUSPEND); |
1da177e4 LT |
1934 | } |
1935 | ||
8757b776 | 1936 | int dm_table_resume_targets(struct dm_table *t) |
1da177e4 | 1937 | { |
8757b776 MB |
1938 | int i, r = 0; |
1939 | ||
1ea0654e BVA |
1940 | lockdep_assert_held(&t->md->suspend_lock); |
1941 | ||
8757b776 MB |
1942 | for (i = 0; i < t->num_targets; i++) { |
1943 | struct dm_target *ti = t->targets + i; | |
1944 | ||
1945 | if (!ti->type->preresume) | |
1946 | continue; | |
1947 | ||
1948 | r = ti->type->preresume(ti); | |
7833b08e MS |
1949 | if (r) { |
1950 | DMERR("%s: %s: preresume failed, error = %d", | |
1951 | dm_device_name(t->md), ti->type->name, r); | |
8757b776 | 1952 | return r; |
7833b08e | 1953 | } |
8757b776 | 1954 | } |
1da177e4 LT |
1955 | |
1956 | for (i = 0; i < t->num_targets; i++) { | |
1957 | struct dm_target *ti = t->targets + i; | |
1958 | ||
1959 | if (ti->type->resume) | |
1960 | ti->type->resume(ti); | |
1961 | } | |
8757b776 MB |
1962 | |
1963 | return 0; | |
1da177e4 LT |
1964 | } |
1965 | ||
9d357b07 N |
1966 | void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb) |
1967 | { | |
1968 | list_add(&cb->list, &t->target_callbacks); | |
1969 | } | |
1970 | EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks); | |
1971 | ||
1da177e4 LT |
1972 | int dm_table_any_congested(struct dm_table *t, int bdi_bits) |
1973 | { | |
82b1519b | 1974 | struct dm_dev_internal *dd; |
afb24528 | 1975 | struct list_head *devices = dm_table_get_devices(t); |
9d357b07 | 1976 | struct dm_target_callbacks *cb; |
1da177e4 LT |
1977 | int r = 0; |
1978 | ||
afb24528 | 1979 | list_for_each_entry(dd, devices, list) { |
86f1152b | 1980 | struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev); |
0c2322e4 AK |
1981 | char b[BDEVNAME_SIZE]; |
1982 | ||
1983 | if (likely(q)) | |
dc3b17cc | 1984 | r |= bdi_congested(q->backing_dev_info, bdi_bits); |
0c2322e4 AK |
1985 | else |
1986 | DMWARN_LIMIT("%s: any_congested: nonexistent device %s", | |
1987 | dm_device_name(t->md), | |
86f1152b | 1988 | bdevname(dd->dm_dev->bdev, b)); |
1da177e4 LT |
1989 | } |
1990 | ||
9d357b07 N |
1991 | list_for_each_entry(cb, &t->target_callbacks, list) |
1992 | if (cb->congested_fn) | |
1993 | r |= cb->congested_fn(cb, bdi_bits); | |
1994 | ||
1da177e4 LT |
1995 | return r; |
1996 | } | |
1997 | ||
1134e5ae MA |
1998 | struct mapped_device *dm_table_get_md(struct dm_table *t) |
1999 | { | |
1134e5ae MA |
2000 | return t->md; |
2001 | } | |
08649012 | 2002 | EXPORT_SYMBOL(dm_table_get_md); |
1134e5ae | 2003 | |
9974fa2c MS |
2004 | void dm_table_run_md_queue_async(struct dm_table *t) |
2005 | { | |
2006 | struct mapped_device *md; | |
2007 | struct request_queue *queue; | |
2008 | unsigned long flags; | |
2009 | ||
2010 | if (!dm_table_request_based(t)) | |
2011 | return; | |
2012 | ||
2013 | md = dm_table_get_md(t); | |
2014 | queue = dm_get_md_queue(md); | |
2015 | if (queue) { | |
bfebd1cd MS |
2016 | if (queue->mq_ops) |
2017 | blk_mq_run_hw_queues(queue, true); | |
2018 | else { | |
2019 | spin_lock_irqsave(queue->queue_lock, flags); | |
2020 | blk_run_queue_async(queue); | |
2021 | spin_unlock_irqrestore(queue->queue_lock, flags); | |
2022 | } | |
9974fa2c MS |
2023 | } |
2024 | } | |
2025 | EXPORT_SYMBOL(dm_table_run_md_queue_async); | |
2026 |