| 1 | /* |
| 2 | * Copyright (C) 2001 Sistina Software (UK) Limited. |
| 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
| 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/string.h> |
| 16 | #include <linux/slab.h> |
| 17 | #include <linux/interrupt.h> |
| 18 | #include <linux/mutex.h> |
| 19 | #include <linux/delay.h> |
| 20 | #include <linux/atomic.h> |
| 21 | #include <linux/blk-mq.h> |
| 22 | #include <linux/mount.h> |
| 23 | |
| 24 | #define DM_MSG_PREFIX "table" |
| 25 | |
| 26 | #define MAX_DEPTH 16 |
| 27 | #define NODE_SIZE L1_CACHE_BYTES |
| 28 | #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t)) |
| 29 | #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1) |
| 30 | |
| 31 | struct dm_table { |
| 32 | struct mapped_device *md; |
| 33 | unsigned type; |
| 34 | |
| 35 | /* btree table */ |
| 36 | unsigned int depth; |
| 37 | unsigned int counts[MAX_DEPTH]; /* in nodes */ |
| 38 | sector_t *index[MAX_DEPTH]; |
| 39 | |
| 40 | unsigned int num_targets; |
| 41 | unsigned int num_allocated; |
| 42 | sector_t *highs; |
| 43 | struct dm_target *targets; |
| 44 | |
| 45 | struct target_type *immutable_target_type; |
| 46 | unsigned integrity_supported:1; |
| 47 | unsigned singleton:1; |
| 48 | |
| 49 | /* |
| 50 | * Indicates the rw permissions for the new logical |
| 51 | * device. This should be a combination of FMODE_READ |
| 52 | * and FMODE_WRITE. |
| 53 | */ |
| 54 | fmode_t mode; |
| 55 | |
| 56 | /* a list of devices used by this table */ |
| 57 | struct list_head devices; |
| 58 | |
| 59 | /* events get handed up using this callback */ |
| 60 | void (*event_fn)(void *); |
| 61 | void *event_context; |
| 62 | |
| 63 | struct dm_md_mempools *mempools; |
| 64 | |
| 65 | struct list_head target_callbacks; |
| 66 | }; |
| 67 | |
| 68 | /* |
| 69 | * Similar to ceiling(log_size(n)) |
| 70 | */ |
| 71 | static unsigned int int_log(unsigned int n, unsigned int base) |
| 72 | { |
| 73 | int result = 0; |
| 74 | |
| 75 | while (n > 1) { |
| 76 | n = dm_div_up(n, base); |
| 77 | result++; |
| 78 | } |
| 79 | |
| 80 | return result; |
| 81 | } |
| 82 | |
| 83 | /* |
| 84 | * Calculate the index of the child node of the n'th node k'th key. |
| 85 | */ |
| 86 | static inline unsigned int get_child(unsigned int n, unsigned int k) |
| 87 | { |
| 88 | return (n * CHILDREN_PER_NODE) + k; |
| 89 | } |
| 90 | |
| 91 | /* |
| 92 | * Return the n'th node of level l from table t. |
| 93 | */ |
| 94 | static inline sector_t *get_node(struct dm_table *t, |
| 95 | unsigned int l, unsigned int n) |
| 96 | { |
| 97 | return t->index[l] + (n * KEYS_PER_NODE); |
| 98 | } |
| 99 | |
| 100 | /* |
| 101 | * Return the highest key that you could lookup from the n'th |
| 102 | * node on level l of the btree. |
| 103 | */ |
| 104 | static sector_t high(struct dm_table *t, unsigned int l, unsigned int n) |
| 105 | { |
| 106 | for (; l < t->depth - 1; l++) |
| 107 | n = get_child(n, CHILDREN_PER_NODE - 1); |
| 108 | |
| 109 | if (n >= t->counts[l]) |
| 110 | return (sector_t) - 1; |
| 111 | |
| 112 | return get_node(t, l, n)[KEYS_PER_NODE - 1]; |
| 113 | } |
| 114 | |
| 115 | /* |
| 116 | * Fills in a level of the btree based on the highs of the level |
| 117 | * below it. |
| 118 | */ |
| 119 | static int setup_btree_index(unsigned int l, struct dm_table *t) |
| 120 | { |
| 121 | unsigned int n, k; |
| 122 | sector_t *node; |
| 123 | |
| 124 | for (n = 0U; n < t->counts[l]; n++) { |
| 125 | node = get_node(t, l, n); |
| 126 | |
| 127 | for (k = 0U; k < KEYS_PER_NODE; k++) |
| 128 | node[k] = high(t, l + 1, get_child(n, k)); |
| 129 | } |
| 130 | |
| 131 | return 0; |
| 132 | } |
| 133 | |
| 134 | void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size) |
| 135 | { |
| 136 | unsigned long size; |
| 137 | void *addr; |
| 138 | |
| 139 | /* |
| 140 | * Check that we're not going to overflow. |
| 141 | */ |
| 142 | if (nmemb > (ULONG_MAX / elem_size)) |
| 143 | return NULL; |
| 144 | |
| 145 | size = nmemb * elem_size; |
| 146 | addr = vzalloc(size); |
| 147 | |
| 148 | return addr; |
| 149 | } |
| 150 | EXPORT_SYMBOL(dm_vcalloc); |
| 151 | |
| 152 | /* |
| 153 | * highs, and targets are managed as dynamic arrays during a |
| 154 | * table load. |
| 155 | */ |
| 156 | static int alloc_targets(struct dm_table *t, unsigned int num) |
| 157 | { |
| 158 | sector_t *n_highs; |
| 159 | struct dm_target *n_targets; |
| 160 | |
| 161 | /* |
| 162 | * Allocate both the target array and offset array at once. |
| 163 | * Append an empty entry to catch sectors beyond the end of |
| 164 | * the device. |
| 165 | */ |
| 166 | n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) + |
| 167 | sizeof(sector_t)); |
| 168 | if (!n_highs) |
| 169 | return -ENOMEM; |
| 170 | |
| 171 | n_targets = (struct dm_target *) (n_highs + num); |
| 172 | |
| 173 | memset(n_highs, -1, sizeof(*n_highs) * num); |
| 174 | vfree(t->highs); |
| 175 | |
| 176 | t->num_allocated = num; |
| 177 | t->highs = n_highs; |
| 178 | t->targets = n_targets; |
| 179 | |
| 180 | return 0; |
| 181 | } |
| 182 | |
| 183 | int dm_table_create(struct dm_table **result, fmode_t mode, |
| 184 | unsigned num_targets, struct mapped_device *md) |
| 185 | { |
| 186 | struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL); |
| 187 | |
| 188 | if (!t) |
| 189 | return -ENOMEM; |
| 190 | |
| 191 | INIT_LIST_HEAD(&t->devices); |
| 192 | INIT_LIST_HEAD(&t->target_callbacks); |
| 193 | |
| 194 | if (!num_targets) |
| 195 | num_targets = KEYS_PER_NODE; |
| 196 | |
| 197 | num_targets = dm_round_up(num_targets, KEYS_PER_NODE); |
| 198 | |
| 199 | if (!num_targets) { |
| 200 | kfree(t); |
| 201 | return -ENOMEM; |
| 202 | } |
| 203 | |
| 204 | if (alloc_targets(t, num_targets)) { |
| 205 | kfree(t); |
| 206 | return -ENOMEM; |
| 207 | } |
| 208 | |
| 209 | t->mode = mode; |
| 210 | t->md = md; |
| 211 | *result = t; |
| 212 | return 0; |
| 213 | } |
| 214 | |
| 215 | static void free_devices(struct list_head *devices, struct mapped_device *md) |
| 216 | { |
| 217 | struct list_head *tmp, *next; |
| 218 | |
| 219 | list_for_each_safe(tmp, next, devices) { |
| 220 | struct dm_dev_internal *dd = |
| 221 | list_entry(tmp, struct dm_dev_internal, list); |
| 222 | DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s", |
| 223 | dm_device_name(md), dd->dm_dev->name); |
| 224 | dm_put_table_device(md, dd->dm_dev); |
| 225 | kfree(dd); |
| 226 | } |
| 227 | } |
| 228 | |
| 229 | void dm_table_destroy(struct dm_table *t) |
| 230 | { |
| 231 | unsigned int i; |
| 232 | |
| 233 | if (!t) |
| 234 | return; |
| 235 | |
| 236 | /* free the indexes */ |
| 237 | if (t->depth >= 2) |
| 238 | vfree(t->index[t->depth - 2]); |
| 239 | |
| 240 | /* free the targets */ |
| 241 | for (i = 0; i < t->num_targets; i++) { |
| 242 | struct dm_target *tgt = t->targets + i; |
| 243 | |
| 244 | if (tgt->type->dtr) |
| 245 | tgt->type->dtr(tgt); |
| 246 | |
| 247 | dm_put_target_type(tgt->type); |
| 248 | } |
| 249 | |
| 250 | vfree(t->highs); |
| 251 | |
| 252 | /* free the device list */ |
| 253 | free_devices(&t->devices, t->md); |
| 254 | |
| 255 | dm_free_md_mempools(t->mempools); |
| 256 | |
| 257 | kfree(t); |
| 258 | } |
| 259 | |
| 260 | /* |
| 261 | * See if we've already got a device in the list. |
| 262 | */ |
| 263 | static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev) |
| 264 | { |
| 265 | struct dm_dev_internal *dd; |
| 266 | |
| 267 | list_for_each_entry (dd, l, list) |
| 268 | if (dd->dm_dev->bdev->bd_dev == dev) |
| 269 | return dd; |
| 270 | |
| 271 | return NULL; |
| 272 | } |
| 273 | |
| 274 | /* |
| 275 | * If possible, this checks an area of a destination device is invalid. |
| 276 | */ |
| 277 | static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev, |
| 278 | sector_t start, sector_t len, void *data) |
| 279 | { |
| 280 | struct request_queue *q; |
| 281 | struct queue_limits *limits = data; |
| 282 | struct block_device *bdev = dev->bdev; |
| 283 | sector_t dev_size = |
| 284 | i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; |
| 285 | unsigned short logical_block_size_sectors = |
| 286 | limits->logical_block_size >> SECTOR_SHIFT; |
| 287 | char b[BDEVNAME_SIZE]; |
| 288 | |
| 289 | /* |
| 290 | * Some devices exist without request functions, |
| 291 | * such as loop devices not yet bound to backing files. |
| 292 | * Forbid the use of such devices. |
| 293 | */ |
| 294 | q = bdev_get_queue(bdev); |
| 295 | if (!q || !q->make_request_fn) { |
| 296 | DMWARN("%s: %s is not yet initialised: " |
| 297 | "start=%llu, len=%llu, dev_size=%llu", |
| 298 | dm_device_name(ti->table->md), bdevname(bdev, b), |
| 299 | (unsigned long long)start, |
| 300 | (unsigned long long)len, |
| 301 | (unsigned long long)dev_size); |
| 302 | return 1; |
| 303 | } |
| 304 | |
| 305 | if (!dev_size) |
| 306 | return 0; |
| 307 | |
| 308 | if ((start >= dev_size) || (start + len > dev_size)) { |
| 309 | DMWARN("%s: %s too small for target: " |
| 310 | "start=%llu, len=%llu, dev_size=%llu", |
| 311 | dm_device_name(ti->table->md), bdevname(bdev, b), |
| 312 | (unsigned long long)start, |
| 313 | (unsigned long long)len, |
| 314 | (unsigned long long)dev_size); |
| 315 | return 1; |
| 316 | } |
| 317 | |
| 318 | if (logical_block_size_sectors <= 1) |
| 319 | return 0; |
| 320 | |
| 321 | if (start & (logical_block_size_sectors - 1)) { |
| 322 | DMWARN("%s: start=%llu not aligned to h/w " |
| 323 | "logical block size %u of %s", |
| 324 | dm_device_name(ti->table->md), |
| 325 | (unsigned long long)start, |
| 326 | limits->logical_block_size, bdevname(bdev, b)); |
| 327 | return 1; |
| 328 | } |
| 329 | |
| 330 | if (len & (logical_block_size_sectors - 1)) { |
| 331 | DMWARN("%s: len=%llu not aligned to h/w " |
| 332 | "logical block size %u of %s", |
| 333 | dm_device_name(ti->table->md), |
| 334 | (unsigned long long)len, |
| 335 | limits->logical_block_size, bdevname(bdev, b)); |
| 336 | return 1; |
| 337 | } |
| 338 | |
| 339 | return 0; |
| 340 | } |
| 341 | |
| 342 | /* |
| 343 | * This upgrades the mode on an already open dm_dev, being |
| 344 | * careful to leave things as they were if we fail to reopen the |
| 345 | * device and not to touch the existing bdev field in case |
| 346 | * it is accessed concurrently inside dm_table_any_congested(). |
| 347 | */ |
| 348 | static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode, |
| 349 | struct mapped_device *md) |
| 350 | { |
| 351 | int r; |
| 352 | struct dm_dev *old_dev, *new_dev; |
| 353 | |
| 354 | old_dev = dd->dm_dev; |
| 355 | |
| 356 | r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev, |
| 357 | dd->dm_dev->mode | new_mode, &new_dev); |
| 358 | if (r) |
| 359 | return r; |
| 360 | |
| 361 | dd->dm_dev = new_dev; |
| 362 | dm_put_table_device(md, old_dev); |
| 363 | |
| 364 | return 0; |
| 365 | } |
| 366 | |
| 367 | /* |
| 368 | * Add a device to the list, or just increment the usage count if |
| 369 | * it's already present. |
| 370 | */ |
| 371 | int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode, |
| 372 | struct dm_dev **result) |
| 373 | { |
| 374 | int r; |
| 375 | dev_t uninitialized_var(dev); |
| 376 | struct dm_dev_internal *dd; |
| 377 | struct dm_table *t = ti->table; |
| 378 | struct block_device *bdev; |
| 379 | |
| 380 | BUG_ON(!t); |
| 381 | |
| 382 | /* convert the path to a device */ |
| 383 | bdev = lookup_bdev(path); |
| 384 | if (IS_ERR(bdev)) { |
| 385 | dev = name_to_dev_t(path); |
| 386 | if (!dev) |
| 387 | return -ENODEV; |
| 388 | } else { |
| 389 | dev = bdev->bd_dev; |
| 390 | bdput(bdev); |
| 391 | } |
| 392 | |
| 393 | dd = find_device(&t->devices, dev); |
| 394 | if (!dd) { |
| 395 | dd = kmalloc(sizeof(*dd), GFP_KERNEL); |
| 396 | if (!dd) |
| 397 | return -ENOMEM; |
| 398 | |
| 399 | if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) { |
| 400 | kfree(dd); |
| 401 | return r; |
| 402 | } |
| 403 | |
| 404 | atomic_set(&dd->count, 0); |
| 405 | list_add(&dd->list, &t->devices); |
| 406 | |
| 407 | } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) { |
| 408 | r = upgrade_mode(dd, mode, t->md); |
| 409 | if (r) |
| 410 | return r; |
| 411 | } |
| 412 | atomic_inc(&dd->count); |
| 413 | |
| 414 | *result = dd->dm_dev; |
| 415 | return 0; |
| 416 | } |
| 417 | EXPORT_SYMBOL(dm_get_device); |
| 418 | |
| 419 | static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev, |
| 420 | sector_t start, sector_t len, void *data) |
| 421 | { |
| 422 | struct queue_limits *limits = data; |
| 423 | struct block_device *bdev = dev->bdev; |
| 424 | struct request_queue *q = bdev_get_queue(bdev); |
| 425 | char b[BDEVNAME_SIZE]; |
| 426 | |
| 427 | if (unlikely(!q)) { |
| 428 | DMWARN("%s: Cannot set limits for nonexistent device %s", |
| 429 | dm_device_name(ti->table->md), bdevname(bdev, b)); |
| 430 | return 0; |
| 431 | } |
| 432 | |
| 433 | if (bdev_stack_limits(limits, bdev, start) < 0) |
| 434 | DMWARN("%s: adding target device %s caused an alignment inconsistency: " |
| 435 | "physical_block_size=%u, logical_block_size=%u, " |
| 436 | "alignment_offset=%u, start=%llu", |
| 437 | dm_device_name(ti->table->md), bdevname(bdev, b), |
| 438 | q->limits.physical_block_size, |
| 439 | q->limits.logical_block_size, |
| 440 | q->limits.alignment_offset, |
| 441 | (unsigned long long) start << SECTOR_SHIFT); |
| 442 | |
| 443 | /* |
| 444 | * Check if merge fn is supported. |
| 445 | * If not we'll force DM to use PAGE_SIZE or |
| 446 | * smaller I/O, just to be safe. |
| 447 | */ |
| 448 | if (dm_queue_merge_is_compulsory(q) && !ti->type->merge) |
| 449 | blk_limits_max_hw_sectors(limits, |
| 450 | (unsigned int) (PAGE_SIZE >> 9)); |
| 451 | return 0; |
| 452 | } |
| 453 | |
| 454 | /* |
| 455 | * Decrement a device's use count and remove it if necessary. |
| 456 | */ |
| 457 | void dm_put_device(struct dm_target *ti, struct dm_dev *d) |
| 458 | { |
| 459 | int found = 0; |
| 460 | struct list_head *devices = &ti->table->devices; |
| 461 | struct dm_dev_internal *dd; |
| 462 | |
| 463 | list_for_each_entry(dd, devices, list) { |
| 464 | if (dd->dm_dev == d) { |
| 465 | found = 1; |
| 466 | break; |
| 467 | } |
| 468 | } |
| 469 | if (!found) { |
| 470 | DMWARN("%s: device %s not in table devices list", |
| 471 | dm_device_name(ti->table->md), d->name); |
| 472 | return; |
| 473 | } |
| 474 | if (atomic_dec_and_test(&dd->count)) { |
| 475 | dm_put_table_device(ti->table->md, d); |
| 476 | list_del(&dd->list); |
| 477 | kfree(dd); |
| 478 | } |
| 479 | } |
| 480 | EXPORT_SYMBOL(dm_put_device); |
| 481 | |
| 482 | /* |
| 483 | * Checks to see if the target joins onto the end of the table. |
| 484 | */ |
| 485 | static int adjoin(struct dm_table *table, struct dm_target *ti) |
| 486 | { |
| 487 | struct dm_target *prev; |
| 488 | |
| 489 | if (!table->num_targets) |
| 490 | return !ti->begin; |
| 491 | |
| 492 | prev = &table->targets[table->num_targets - 1]; |
| 493 | return (ti->begin == (prev->begin + prev->len)); |
| 494 | } |
| 495 | |
| 496 | /* |
| 497 | * Used to dynamically allocate the arg array. |
| 498 | * |
| 499 | * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must |
| 500 | * process messages even if some device is suspended. These messages have a |
| 501 | * small fixed number of arguments. |
| 502 | * |
| 503 | * On the other hand, dm-switch needs to process bulk data using messages and |
| 504 | * excessive use of GFP_NOIO could cause trouble. |
| 505 | */ |
| 506 | static char **realloc_argv(unsigned *array_size, char **old_argv) |
| 507 | { |
| 508 | char **argv; |
| 509 | unsigned new_size; |
| 510 | gfp_t gfp; |
| 511 | |
| 512 | if (*array_size) { |
| 513 | new_size = *array_size * 2; |
| 514 | gfp = GFP_KERNEL; |
| 515 | } else { |
| 516 | new_size = 8; |
| 517 | gfp = GFP_NOIO; |
| 518 | } |
| 519 | argv = kmalloc(new_size * sizeof(*argv), gfp); |
| 520 | if (argv) { |
| 521 | memcpy(argv, old_argv, *array_size * sizeof(*argv)); |
| 522 | *array_size = new_size; |
| 523 | } |
| 524 | |
| 525 | kfree(old_argv); |
| 526 | return argv; |
| 527 | } |
| 528 | |
| 529 | /* |
| 530 | * Destructively splits up the argument list to pass to ctr. |
| 531 | */ |
| 532 | int dm_split_args(int *argc, char ***argvp, char *input) |
| 533 | { |
| 534 | char *start, *end = input, *out, **argv = NULL; |
| 535 | unsigned array_size = 0; |
| 536 | |
| 537 | *argc = 0; |
| 538 | |
| 539 | if (!input) { |
| 540 | *argvp = NULL; |
| 541 | return 0; |
| 542 | } |
| 543 | |
| 544 | argv = realloc_argv(&array_size, argv); |
| 545 | if (!argv) |
| 546 | return -ENOMEM; |
| 547 | |
| 548 | while (1) { |
| 549 | /* Skip whitespace */ |
| 550 | start = skip_spaces(end); |
| 551 | |
| 552 | if (!*start) |
| 553 | break; /* success, we hit the end */ |
| 554 | |
| 555 | /* 'out' is used to remove any back-quotes */ |
| 556 | end = out = start; |
| 557 | while (*end) { |
| 558 | /* Everything apart from '\0' can be quoted */ |
| 559 | if (*end == '\\' && *(end + 1)) { |
| 560 | *out++ = *(end + 1); |
| 561 | end += 2; |
| 562 | continue; |
| 563 | } |
| 564 | |
| 565 | if (isspace(*end)) |
| 566 | break; /* end of token */ |
| 567 | |
| 568 | *out++ = *end++; |
| 569 | } |
| 570 | |
| 571 | /* have we already filled the array ? */ |
| 572 | if ((*argc + 1) > array_size) { |
| 573 | argv = realloc_argv(&array_size, argv); |
| 574 | if (!argv) |
| 575 | return -ENOMEM; |
| 576 | } |
| 577 | |
| 578 | /* we know this is whitespace */ |
| 579 | if (*end) |
| 580 | end++; |
| 581 | |
| 582 | /* terminate the string and put it in the array */ |
| 583 | *out = '\0'; |
| 584 | argv[*argc] = start; |
| 585 | (*argc)++; |
| 586 | } |
| 587 | |
| 588 | *argvp = argv; |
| 589 | return 0; |
| 590 | } |
| 591 | |
| 592 | /* |
| 593 | * Impose necessary and sufficient conditions on a devices's table such |
| 594 | * that any incoming bio which respects its logical_block_size can be |
| 595 | * processed successfully. If it falls across the boundary between |
| 596 | * two or more targets, the size of each piece it gets split into must |
| 597 | * be compatible with the logical_block_size of the target processing it. |
| 598 | */ |
| 599 | static int validate_hardware_logical_block_alignment(struct dm_table *table, |
| 600 | struct queue_limits *limits) |
| 601 | { |
| 602 | /* |
| 603 | * This function uses arithmetic modulo the logical_block_size |
| 604 | * (in units of 512-byte sectors). |
| 605 | */ |
| 606 | unsigned short device_logical_block_size_sects = |
| 607 | limits->logical_block_size >> SECTOR_SHIFT; |
| 608 | |
| 609 | /* |
| 610 | * Offset of the start of the next table entry, mod logical_block_size. |
| 611 | */ |
| 612 | unsigned short next_target_start = 0; |
| 613 | |
| 614 | /* |
| 615 | * Given an aligned bio that extends beyond the end of a |
| 616 | * target, how many sectors must the next target handle? |
| 617 | */ |
| 618 | unsigned short remaining = 0; |
| 619 | |
| 620 | struct dm_target *uninitialized_var(ti); |
| 621 | struct queue_limits ti_limits; |
| 622 | unsigned i = 0; |
| 623 | |
| 624 | /* |
| 625 | * Check each entry in the table in turn. |
| 626 | */ |
| 627 | while (i < dm_table_get_num_targets(table)) { |
| 628 | ti = dm_table_get_target(table, i++); |
| 629 | |
| 630 | blk_set_stacking_limits(&ti_limits); |
| 631 | |
| 632 | /* combine all target devices' limits */ |
| 633 | if (ti->type->iterate_devices) |
| 634 | ti->type->iterate_devices(ti, dm_set_device_limits, |
| 635 | &ti_limits); |
| 636 | |
| 637 | /* |
| 638 | * If the remaining sectors fall entirely within this |
| 639 | * table entry are they compatible with its logical_block_size? |
| 640 | */ |
| 641 | if (remaining < ti->len && |
| 642 | remaining & ((ti_limits.logical_block_size >> |
| 643 | SECTOR_SHIFT) - 1)) |
| 644 | break; /* Error */ |
| 645 | |
| 646 | next_target_start = |
| 647 | (unsigned short) ((next_target_start + ti->len) & |
| 648 | (device_logical_block_size_sects - 1)); |
| 649 | remaining = next_target_start ? |
| 650 | device_logical_block_size_sects - next_target_start : 0; |
| 651 | } |
| 652 | |
| 653 | if (remaining) { |
| 654 | DMWARN("%s: table line %u (start sect %llu len %llu) " |
| 655 | "not aligned to h/w logical block size %u", |
| 656 | dm_device_name(table->md), i, |
| 657 | (unsigned long long) ti->begin, |
| 658 | (unsigned long long) ti->len, |
| 659 | limits->logical_block_size); |
| 660 | return -EINVAL; |
| 661 | } |
| 662 | |
| 663 | return 0; |
| 664 | } |
| 665 | |
| 666 | int dm_table_add_target(struct dm_table *t, const char *type, |
| 667 | sector_t start, sector_t len, char *params) |
| 668 | { |
| 669 | int r = -EINVAL, argc; |
| 670 | char **argv; |
| 671 | struct dm_target *tgt; |
| 672 | |
| 673 | if (t->singleton) { |
| 674 | DMERR("%s: target type %s must appear alone in table", |
| 675 | dm_device_name(t->md), t->targets->type->name); |
| 676 | return -EINVAL; |
| 677 | } |
| 678 | |
| 679 | BUG_ON(t->num_targets >= t->num_allocated); |
| 680 | |
| 681 | tgt = t->targets + t->num_targets; |
| 682 | memset(tgt, 0, sizeof(*tgt)); |
| 683 | |
| 684 | if (!len) { |
| 685 | DMERR("%s: zero-length target", dm_device_name(t->md)); |
| 686 | return -EINVAL; |
| 687 | } |
| 688 | |
| 689 | tgt->type = dm_get_target_type(type); |
| 690 | if (!tgt->type) { |
| 691 | DMERR("%s: %s: unknown target type", dm_device_name(t->md), |
| 692 | type); |
| 693 | return -EINVAL; |
| 694 | } |
| 695 | |
| 696 | if (dm_target_needs_singleton(tgt->type)) { |
| 697 | if (t->num_targets) { |
| 698 | DMERR("%s: target type %s must appear alone in table", |
| 699 | dm_device_name(t->md), type); |
| 700 | return -EINVAL; |
| 701 | } |
| 702 | t->singleton = 1; |
| 703 | } |
| 704 | |
| 705 | if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) { |
| 706 | DMERR("%s: target type %s may not be included in read-only tables", |
| 707 | dm_device_name(t->md), type); |
| 708 | return -EINVAL; |
| 709 | } |
| 710 | |
| 711 | if (t->immutable_target_type) { |
| 712 | if (t->immutable_target_type != tgt->type) { |
| 713 | DMERR("%s: immutable target type %s cannot be mixed with other target types", |
| 714 | dm_device_name(t->md), t->immutable_target_type->name); |
| 715 | return -EINVAL; |
| 716 | } |
| 717 | } else if (dm_target_is_immutable(tgt->type)) { |
| 718 | if (t->num_targets) { |
| 719 | DMERR("%s: immutable target type %s cannot be mixed with other target types", |
| 720 | dm_device_name(t->md), tgt->type->name); |
| 721 | return -EINVAL; |
| 722 | } |
| 723 | t->immutable_target_type = tgt->type; |
| 724 | } |
| 725 | |
| 726 | tgt->table = t; |
| 727 | tgt->begin = start; |
| 728 | tgt->len = len; |
| 729 | tgt->error = "Unknown error"; |
| 730 | |
| 731 | /* |
| 732 | * Does this target adjoin the previous one ? |
| 733 | */ |
| 734 | if (!adjoin(t, tgt)) { |
| 735 | tgt->error = "Gap in table"; |
| 736 | r = -EINVAL; |
| 737 | goto bad; |
| 738 | } |
| 739 | |
| 740 | r = dm_split_args(&argc, &argv, params); |
| 741 | if (r) { |
| 742 | tgt->error = "couldn't split parameters (insufficient memory)"; |
| 743 | goto bad; |
| 744 | } |
| 745 | |
| 746 | r = tgt->type->ctr(tgt, argc, argv); |
| 747 | kfree(argv); |
| 748 | if (r) |
| 749 | goto bad; |
| 750 | |
| 751 | t->highs[t->num_targets++] = tgt->begin + tgt->len - 1; |
| 752 | |
| 753 | if (!tgt->num_discard_bios && tgt->discards_supported) |
| 754 | DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.", |
| 755 | dm_device_name(t->md), type); |
| 756 | |
| 757 | return 0; |
| 758 | |
| 759 | bad: |
| 760 | DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error); |
| 761 | dm_put_target_type(tgt->type); |
| 762 | return r; |
| 763 | } |
| 764 | |
| 765 | /* |
| 766 | * Target argument parsing helpers. |
| 767 | */ |
| 768 | static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set, |
| 769 | unsigned *value, char **error, unsigned grouped) |
| 770 | { |
| 771 | const char *arg_str = dm_shift_arg(arg_set); |
| 772 | char dummy; |
| 773 | |
| 774 | if (!arg_str || |
| 775 | (sscanf(arg_str, "%u%c", value, &dummy) != 1) || |
| 776 | (*value < arg->min) || |
| 777 | (*value > arg->max) || |
| 778 | (grouped && arg_set->argc < *value)) { |
| 779 | *error = arg->error; |
| 780 | return -EINVAL; |
| 781 | } |
| 782 | |
| 783 | return 0; |
| 784 | } |
| 785 | |
| 786 | int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set, |
| 787 | unsigned *value, char **error) |
| 788 | { |
| 789 | return validate_next_arg(arg, arg_set, value, error, 0); |
| 790 | } |
| 791 | EXPORT_SYMBOL(dm_read_arg); |
| 792 | |
| 793 | int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set, |
| 794 | unsigned *value, char **error) |
| 795 | { |
| 796 | return validate_next_arg(arg, arg_set, value, error, 1); |
| 797 | } |
| 798 | EXPORT_SYMBOL(dm_read_arg_group); |
| 799 | |
| 800 | const char *dm_shift_arg(struct dm_arg_set *as) |
| 801 | { |
| 802 | char *r; |
| 803 | |
| 804 | if (as->argc) { |
| 805 | as->argc--; |
| 806 | r = *as->argv; |
| 807 | as->argv++; |
| 808 | return r; |
| 809 | } |
| 810 | |
| 811 | return NULL; |
| 812 | } |
| 813 | EXPORT_SYMBOL(dm_shift_arg); |
| 814 | |
| 815 | void dm_consume_args(struct dm_arg_set *as, unsigned num_args) |
| 816 | { |
| 817 | BUG_ON(as->argc < num_args); |
| 818 | as->argc -= num_args; |
| 819 | as->argv += num_args; |
| 820 | } |
| 821 | EXPORT_SYMBOL(dm_consume_args); |
| 822 | |
| 823 | static bool __table_type_request_based(unsigned table_type) |
| 824 | { |
| 825 | return (table_type == DM_TYPE_REQUEST_BASED || |
| 826 | table_type == DM_TYPE_MQ_REQUEST_BASED); |
| 827 | } |
| 828 | |
| 829 | static int dm_table_set_type(struct dm_table *t) |
| 830 | { |
| 831 | unsigned i; |
| 832 | unsigned bio_based = 0, request_based = 0, hybrid = 0; |
| 833 | bool use_blk_mq = false; |
| 834 | struct dm_target *tgt; |
| 835 | struct dm_dev_internal *dd; |
| 836 | struct list_head *devices; |
| 837 | unsigned live_md_type = dm_get_md_type(t->md); |
| 838 | |
| 839 | for (i = 0; i < t->num_targets; i++) { |
| 840 | tgt = t->targets + i; |
| 841 | if (dm_target_hybrid(tgt)) |
| 842 | hybrid = 1; |
| 843 | else if (dm_target_request_based(tgt)) |
| 844 | request_based = 1; |
| 845 | else |
| 846 | bio_based = 1; |
| 847 | |
| 848 | if (bio_based && request_based) { |
| 849 | DMWARN("Inconsistent table: different target types" |
| 850 | " can't be mixed up"); |
| 851 | return -EINVAL; |
| 852 | } |
| 853 | } |
| 854 | |
| 855 | if (hybrid && !bio_based && !request_based) { |
| 856 | /* |
| 857 | * The targets can work either way. |
| 858 | * Determine the type from the live device. |
| 859 | * Default to bio-based if device is new. |
| 860 | */ |
| 861 | if (__table_type_request_based(live_md_type)) |
| 862 | request_based = 1; |
| 863 | else |
| 864 | bio_based = 1; |
| 865 | } |
| 866 | |
| 867 | if (bio_based) { |
| 868 | /* We must use this table as bio-based */ |
| 869 | t->type = DM_TYPE_BIO_BASED; |
| 870 | return 0; |
| 871 | } |
| 872 | |
| 873 | BUG_ON(!request_based); /* No targets in this table */ |
| 874 | |
| 875 | /* |
| 876 | * Request-based dm supports only tables that have a single target now. |
| 877 | * To support multiple targets, request splitting support is needed, |
| 878 | * and that needs lots of changes in the block-layer. |
| 879 | * (e.g. request completion process for partial completion.) |
| 880 | */ |
| 881 | if (t->num_targets > 1) { |
| 882 | DMWARN("Request-based dm doesn't support multiple targets yet"); |
| 883 | return -EINVAL; |
| 884 | } |
| 885 | |
| 886 | /* Non-request-stackable devices can't be used for request-based dm */ |
| 887 | devices = dm_table_get_devices(t); |
| 888 | list_for_each_entry(dd, devices, list) { |
| 889 | struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev); |
| 890 | |
| 891 | if (!blk_queue_stackable(q)) { |
| 892 | DMERR("table load rejected: including" |
| 893 | " non-request-stackable devices"); |
| 894 | return -EINVAL; |
| 895 | } |
| 896 | |
| 897 | if (q->mq_ops) |
| 898 | use_blk_mq = true; |
| 899 | } |
| 900 | |
| 901 | if (use_blk_mq) { |
| 902 | /* verify _all_ devices in the table are blk-mq devices */ |
| 903 | list_for_each_entry(dd, devices, list) |
| 904 | if (!bdev_get_queue(dd->dm_dev->bdev)->mq_ops) { |
| 905 | DMERR("table load rejected: not all devices" |
| 906 | " are blk-mq request-stackable"); |
| 907 | return -EINVAL; |
| 908 | } |
| 909 | t->type = DM_TYPE_MQ_REQUEST_BASED; |
| 910 | |
| 911 | } else if (list_empty(devices) && __table_type_request_based(live_md_type)) { |
| 912 | /* inherit live MD type */ |
| 913 | t->type = live_md_type; |
| 914 | |
| 915 | } else |
| 916 | t->type = DM_TYPE_REQUEST_BASED; |
| 917 | |
| 918 | return 0; |
| 919 | } |
| 920 | |
| 921 | unsigned dm_table_get_type(struct dm_table *t) |
| 922 | { |
| 923 | return t->type; |
| 924 | } |
| 925 | |
| 926 | struct target_type *dm_table_get_immutable_target_type(struct dm_table *t) |
| 927 | { |
| 928 | return t->immutable_target_type; |
| 929 | } |
| 930 | |
| 931 | bool dm_table_request_based(struct dm_table *t) |
| 932 | { |
| 933 | return __table_type_request_based(dm_table_get_type(t)); |
| 934 | } |
| 935 | |
| 936 | bool dm_table_mq_request_based(struct dm_table *t) |
| 937 | { |
| 938 | return dm_table_get_type(t) == DM_TYPE_MQ_REQUEST_BASED; |
| 939 | } |
| 940 | |
| 941 | static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md) |
| 942 | { |
| 943 | unsigned type = dm_table_get_type(t); |
| 944 | unsigned per_bio_data_size = 0; |
| 945 | unsigned i; |
| 946 | |
| 947 | switch (type) { |
| 948 | case DM_TYPE_BIO_BASED: |
| 949 | for (i = 0; i < t->num_targets; i++) { |
| 950 | struct dm_target *tgt = t->targets + i; |
| 951 | |
| 952 | per_bio_data_size = max(per_bio_data_size, |
| 953 | tgt->per_bio_data_size); |
| 954 | } |
| 955 | t->mempools = dm_alloc_bio_mempools(t->integrity_supported, |
| 956 | per_bio_data_size); |
| 957 | break; |
| 958 | case DM_TYPE_REQUEST_BASED: |
| 959 | case DM_TYPE_MQ_REQUEST_BASED: |
| 960 | t->mempools = dm_alloc_rq_mempools(md, type); |
| 961 | break; |
| 962 | default: |
| 963 | DMWARN("no table type is set, can't allocate mempools"); |
| 964 | return -EINVAL; |
| 965 | } |
| 966 | |
| 967 | if (!t->mempools) |
| 968 | return -ENOMEM; |
| 969 | |
| 970 | return 0; |
| 971 | } |
| 972 | |
| 973 | void dm_table_free_md_mempools(struct dm_table *t) |
| 974 | { |
| 975 | dm_free_md_mempools(t->mempools); |
| 976 | t->mempools = NULL; |
| 977 | } |
| 978 | |
| 979 | struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t) |
| 980 | { |
| 981 | return t->mempools; |
| 982 | } |
| 983 | |
| 984 | static int setup_indexes(struct dm_table *t) |
| 985 | { |
| 986 | int i; |
| 987 | unsigned int total = 0; |
| 988 | sector_t *indexes; |
| 989 | |
| 990 | /* allocate the space for *all* the indexes */ |
| 991 | for (i = t->depth - 2; i >= 0; i--) { |
| 992 | t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE); |
| 993 | total += t->counts[i]; |
| 994 | } |
| 995 | |
| 996 | indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE); |
| 997 | if (!indexes) |
| 998 | return -ENOMEM; |
| 999 | |
| 1000 | /* set up internal nodes, bottom-up */ |
| 1001 | for (i = t->depth - 2; i >= 0; i--) { |
| 1002 | t->index[i] = indexes; |
| 1003 | indexes += (KEYS_PER_NODE * t->counts[i]); |
| 1004 | setup_btree_index(i, t); |
| 1005 | } |
| 1006 | |
| 1007 | return 0; |
| 1008 | } |
| 1009 | |
| 1010 | /* |
| 1011 | * Builds the btree to index the map. |
| 1012 | */ |
| 1013 | static int dm_table_build_index(struct dm_table *t) |
| 1014 | { |
| 1015 | int r = 0; |
| 1016 | unsigned int leaf_nodes; |
| 1017 | |
| 1018 | /* how many indexes will the btree have ? */ |
| 1019 | leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE); |
| 1020 | t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE); |
| 1021 | |
| 1022 | /* leaf layer has already been set up */ |
| 1023 | t->counts[t->depth - 1] = leaf_nodes; |
| 1024 | t->index[t->depth - 1] = t->highs; |
| 1025 | |
| 1026 | if (t->depth >= 2) |
| 1027 | r = setup_indexes(t); |
| 1028 | |
| 1029 | return r; |
| 1030 | } |
| 1031 | |
| 1032 | /* |
| 1033 | * Get a disk whose integrity profile reflects the table's profile. |
| 1034 | * If %match_all is true, all devices' profiles must match. |
| 1035 | * If %match_all is false, all devices must at least have an |
| 1036 | * allocated integrity profile; but uninitialized is ok. |
| 1037 | * Returns NULL if integrity support was inconsistent or unavailable. |
| 1038 | */ |
| 1039 | static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t, |
| 1040 | bool match_all) |
| 1041 | { |
| 1042 | struct list_head *devices = dm_table_get_devices(t); |
| 1043 | struct dm_dev_internal *dd = NULL; |
| 1044 | struct gendisk *prev_disk = NULL, *template_disk = NULL; |
| 1045 | |
| 1046 | list_for_each_entry(dd, devices, list) { |
| 1047 | template_disk = dd->dm_dev->bdev->bd_disk; |
| 1048 | if (!blk_get_integrity(template_disk)) |
| 1049 | goto no_integrity; |
| 1050 | if (!match_all && !blk_integrity_is_initialized(template_disk)) |
| 1051 | continue; /* skip uninitialized profiles */ |
| 1052 | else if (prev_disk && |
| 1053 | blk_integrity_compare(prev_disk, template_disk) < 0) |
| 1054 | goto no_integrity; |
| 1055 | prev_disk = template_disk; |
| 1056 | } |
| 1057 | |
| 1058 | return template_disk; |
| 1059 | |
| 1060 | no_integrity: |
| 1061 | if (prev_disk) |
| 1062 | DMWARN("%s: integrity not set: %s and %s profile mismatch", |
| 1063 | dm_device_name(t->md), |
| 1064 | prev_disk->disk_name, |
| 1065 | template_disk->disk_name); |
| 1066 | return NULL; |
| 1067 | } |
| 1068 | |
| 1069 | /* |
| 1070 | * Register the mapped device for blk_integrity support if |
| 1071 | * the underlying devices have an integrity profile. But all devices |
| 1072 | * may not have matching profiles (checking all devices isn't reliable |
| 1073 | * during table load because this table may use other DM device(s) which |
| 1074 | * must be resumed before they will have an initialized integity profile). |
| 1075 | * Stacked DM devices force a 2 stage integrity profile validation: |
| 1076 | * 1 - during load, validate all initialized integrity profiles match |
| 1077 | * 2 - during resume, validate all integrity profiles match |
| 1078 | */ |
| 1079 | static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md) |
| 1080 | { |
| 1081 | struct gendisk *template_disk = NULL; |
| 1082 | |
| 1083 | template_disk = dm_table_get_integrity_disk(t, false); |
| 1084 | if (!template_disk) |
| 1085 | return 0; |
| 1086 | |
| 1087 | if (!blk_integrity_is_initialized(dm_disk(md))) { |
| 1088 | t->integrity_supported = 1; |
| 1089 | return blk_integrity_register(dm_disk(md), NULL); |
| 1090 | } |
| 1091 | |
| 1092 | /* |
| 1093 | * If DM device already has an initalized integrity |
| 1094 | * profile the new profile should not conflict. |
| 1095 | */ |
| 1096 | if (blk_integrity_is_initialized(template_disk) && |
| 1097 | blk_integrity_compare(dm_disk(md), template_disk) < 0) { |
| 1098 | DMWARN("%s: conflict with existing integrity profile: " |
| 1099 | "%s profile mismatch", |
| 1100 | dm_device_name(t->md), |
| 1101 | template_disk->disk_name); |
| 1102 | return 1; |
| 1103 | } |
| 1104 | |
| 1105 | /* Preserve existing initialized integrity profile */ |
| 1106 | t->integrity_supported = 1; |
| 1107 | return 0; |
| 1108 | } |
| 1109 | |
| 1110 | /* |
| 1111 | * Prepares the table for use by building the indices, |
| 1112 | * setting the type, and allocating mempools. |
| 1113 | */ |
| 1114 | int dm_table_complete(struct dm_table *t) |
| 1115 | { |
| 1116 | int r; |
| 1117 | |
| 1118 | r = dm_table_set_type(t); |
| 1119 | if (r) { |
| 1120 | DMERR("unable to set table type"); |
| 1121 | return r; |
| 1122 | } |
| 1123 | |
| 1124 | r = dm_table_build_index(t); |
| 1125 | if (r) { |
| 1126 | DMERR("unable to build btrees"); |
| 1127 | return r; |
| 1128 | } |
| 1129 | |
| 1130 | r = dm_table_prealloc_integrity(t, t->md); |
| 1131 | if (r) { |
| 1132 | DMERR("could not register integrity profile."); |
| 1133 | return r; |
| 1134 | } |
| 1135 | |
| 1136 | r = dm_table_alloc_md_mempools(t, t->md); |
| 1137 | if (r) |
| 1138 | DMERR("unable to allocate mempools"); |
| 1139 | |
| 1140 | return r; |
| 1141 | } |
| 1142 | |
| 1143 | static DEFINE_MUTEX(_event_lock); |
| 1144 | void dm_table_event_callback(struct dm_table *t, |
| 1145 | void (*fn)(void *), void *context) |
| 1146 | { |
| 1147 | mutex_lock(&_event_lock); |
| 1148 | t->event_fn = fn; |
| 1149 | t->event_context = context; |
| 1150 | mutex_unlock(&_event_lock); |
| 1151 | } |
| 1152 | |
| 1153 | void dm_table_event(struct dm_table *t) |
| 1154 | { |
| 1155 | /* |
| 1156 | * You can no longer call dm_table_event() from interrupt |
| 1157 | * context, use a bottom half instead. |
| 1158 | */ |
| 1159 | BUG_ON(in_interrupt()); |
| 1160 | |
| 1161 | mutex_lock(&_event_lock); |
| 1162 | if (t->event_fn) |
| 1163 | t->event_fn(t->event_context); |
| 1164 | mutex_unlock(&_event_lock); |
| 1165 | } |
| 1166 | EXPORT_SYMBOL(dm_table_event); |
| 1167 | |
| 1168 | sector_t dm_table_get_size(struct dm_table *t) |
| 1169 | { |
| 1170 | return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0; |
| 1171 | } |
| 1172 | EXPORT_SYMBOL(dm_table_get_size); |
| 1173 | |
| 1174 | struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index) |
| 1175 | { |
| 1176 | if (index >= t->num_targets) |
| 1177 | return NULL; |
| 1178 | |
| 1179 | return t->targets + index; |
| 1180 | } |
| 1181 | |
| 1182 | /* |
| 1183 | * Search the btree for the correct target. |
| 1184 | * |
| 1185 | * Caller should check returned pointer with dm_target_is_valid() |
| 1186 | * to trap I/O beyond end of device. |
| 1187 | */ |
| 1188 | struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector) |
| 1189 | { |
| 1190 | unsigned int l, n = 0, k = 0; |
| 1191 | sector_t *node; |
| 1192 | |
| 1193 | for (l = 0; l < t->depth; l++) { |
| 1194 | n = get_child(n, k); |
| 1195 | node = get_node(t, l, n); |
| 1196 | |
| 1197 | for (k = 0; k < KEYS_PER_NODE; k++) |
| 1198 | if (node[k] >= sector) |
| 1199 | break; |
| 1200 | } |
| 1201 | |
| 1202 | return &t->targets[(KEYS_PER_NODE * n) + k]; |
| 1203 | } |
| 1204 | |
| 1205 | static int count_device(struct dm_target *ti, struct dm_dev *dev, |
| 1206 | sector_t start, sector_t len, void *data) |
| 1207 | { |
| 1208 | unsigned *num_devices = data; |
| 1209 | |
| 1210 | (*num_devices)++; |
| 1211 | |
| 1212 | return 0; |
| 1213 | } |
| 1214 | |
| 1215 | /* |
| 1216 | * Check whether a table has no data devices attached using each |
| 1217 | * target's iterate_devices method. |
| 1218 | * Returns false if the result is unknown because a target doesn't |
| 1219 | * support iterate_devices. |
| 1220 | */ |
| 1221 | bool dm_table_has_no_data_devices(struct dm_table *table) |
| 1222 | { |
| 1223 | struct dm_target *uninitialized_var(ti); |
| 1224 | unsigned i = 0, num_devices = 0; |
| 1225 | |
| 1226 | while (i < dm_table_get_num_targets(table)) { |
| 1227 | ti = dm_table_get_target(table, i++); |
| 1228 | |
| 1229 | if (!ti->type->iterate_devices) |
| 1230 | return false; |
| 1231 | |
| 1232 | ti->type->iterate_devices(ti, count_device, &num_devices); |
| 1233 | if (num_devices) |
| 1234 | return false; |
| 1235 | } |
| 1236 | |
| 1237 | return true; |
| 1238 | } |
| 1239 | |
| 1240 | /* |
| 1241 | * Establish the new table's queue_limits and validate them. |
| 1242 | */ |
| 1243 | int dm_calculate_queue_limits(struct dm_table *table, |
| 1244 | struct queue_limits *limits) |
| 1245 | { |
| 1246 | struct dm_target *uninitialized_var(ti); |
| 1247 | struct queue_limits ti_limits; |
| 1248 | unsigned i = 0; |
| 1249 | |
| 1250 | blk_set_stacking_limits(limits); |
| 1251 | |
| 1252 | while (i < dm_table_get_num_targets(table)) { |
| 1253 | blk_set_stacking_limits(&ti_limits); |
| 1254 | |
| 1255 | ti = dm_table_get_target(table, i++); |
| 1256 | |
| 1257 | if (!ti->type->iterate_devices) |
| 1258 | goto combine_limits; |
| 1259 | |
| 1260 | /* |
| 1261 | * Combine queue limits of all the devices this target uses. |
| 1262 | */ |
| 1263 | ti->type->iterate_devices(ti, dm_set_device_limits, |
| 1264 | &ti_limits); |
| 1265 | |
| 1266 | /* Set I/O hints portion of queue limits */ |
| 1267 | if (ti->type->io_hints) |
| 1268 | ti->type->io_hints(ti, &ti_limits); |
| 1269 | |
| 1270 | /* |
| 1271 | * Check each device area is consistent with the target's |
| 1272 | * overall queue limits. |
| 1273 | */ |
| 1274 | if (ti->type->iterate_devices(ti, device_area_is_invalid, |
| 1275 | &ti_limits)) |
| 1276 | return -EINVAL; |
| 1277 | |
| 1278 | combine_limits: |
| 1279 | /* |
| 1280 | * Merge this target's queue limits into the overall limits |
| 1281 | * for the table. |
| 1282 | */ |
| 1283 | if (blk_stack_limits(limits, &ti_limits, 0) < 0) |
| 1284 | DMWARN("%s: adding target device " |
| 1285 | "(start sect %llu len %llu) " |
| 1286 | "caused an alignment inconsistency", |
| 1287 | dm_device_name(table->md), |
| 1288 | (unsigned long long) ti->begin, |
| 1289 | (unsigned long long) ti->len); |
| 1290 | } |
| 1291 | |
| 1292 | return validate_hardware_logical_block_alignment(table, limits); |
| 1293 | } |
| 1294 | |
| 1295 | /* |
| 1296 | * Set the integrity profile for this device if all devices used have |
| 1297 | * matching profiles. We're quite deep in the resume path but still |
| 1298 | * don't know if all devices (particularly DM devices this device |
| 1299 | * may be stacked on) have matching profiles. Even if the profiles |
| 1300 | * don't match we have no way to fail (to resume) at this point. |
| 1301 | */ |
| 1302 | static void dm_table_set_integrity(struct dm_table *t) |
| 1303 | { |
| 1304 | struct gendisk *template_disk = NULL; |
| 1305 | |
| 1306 | if (!blk_get_integrity(dm_disk(t->md))) |
| 1307 | return; |
| 1308 | |
| 1309 | template_disk = dm_table_get_integrity_disk(t, true); |
| 1310 | if (template_disk) |
| 1311 | blk_integrity_register(dm_disk(t->md), |
| 1312 | blk_get_integrity(template_disk)); |
| 1313 | else if (blk_integrity_is_initialized(dm_disk(t->md))) |
| 1314 | DMWARN("%s: device no longer has a valid integrity profile", |
| 1315 | dm_device_name(t->md)); |
| 1316 | else |
| 1317 | DMWARN("%s: unable to establish an integrity profile", |
| 1318 | dm_device_name(t->md)); |
| 1319 | } |
| 1320 | |
| 1321 | static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev, |
| 1322 | sector_t start, sector_t len, void *data) |
| 1323 | { |
| 1324 | unsigned flush = (*(unsigned *)data); |
| 1325 | struct request_queue *q = bdev_get_queue(dev->bdev); |
| 1326 | |
| 1327 | return q && (q->flush_flags & flush); |
| 1328 | } |
| 1329 | |
| 1330 | static bool dm_table_supports_flush(struct dm_table *t, unsigned flush) |
| 1331 | { |
| 1332 | struct dm_target *ti; |
| 1333 | unsigned i = 0; |
| 1334 | |
| 1335 | /* |
| 1336 | * Require at least one underlying device to support flushes. |
| 1337 | * t->devices includes internal dm devices such as mirror logs |
| 1338 | * so we need to use iterate_devices here, which targets |
| 1339 | * supporting flushes must provide. |
| 1340 | */ |
| 1341 | while (i < dm_table_get_num_targets(t)) { |
| 1342 | ti = dm_table_get_target(t, i++); |
| 1343 | |
| 1344 | if (!ti->num_flush_bios) |
| 1345 | continue; |
| 1346 | |
| 1347 | if (ti->flush_supported) |
| 1348 | return true; |
| 1349 | |
| 1350 | if (ti->type->iterate_devices && |
| 1351 | ti->type->iterate_devices(ti, device_flush_capable, &flush)) |
| 1352 | return true; |
| 1353 | } |
| 1354 | |
| 1355 | return false; |
| 1356 | } |
| 1357 | |
| 1358 | static bool dm_table_discard_zeroes_data(struct dm_table *t) |
| 1359 | { |
| 1360 | struct dm_target *ti; |
| 1361 | unsigned i = 0; |
| 1362 | |
| 1363 | /* Ensure that all targets supports discard_zeroes_data. */ |
| 1364 | while (i < dm_table_get_num_targets(t)) { |
| 1365 | ti = dm_table_get_target(t, i++); |
| 1366 | |
| 1367 | if (ti->discard_zeroes_data_unsupported) |
| 1368 | return false; |
| 1369 | } |
| 1370 | |
| 1371 | return true; |
| 1372 | } |
| 1373 | |
| 1374 | static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev, |
| 1375 | sector_t start, sector_t len, void *data) |
| 1376 | { |
| 1377 | struct request_queue *q = bdev_get_queue(dev->bdev); |
| 1378 | |
| 1379 | return q && blk_queue_nonrot(q); |
| 1380 | } |
| 1381 | |
| 1382 | static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev, |
| 1383 | sector_t start, sector_t len, void *data) |
| 1384 | { |
| 1385 | struct request_queue *q = bdev_get_queue(dev->bdev); |
| 1386 | |
| 1387 | return q && !blk_queue_add_random(q); |
| 1388 | } |
| 1389 | |
| 1390 | static int queue_supports_sg_merge(struct dm_target *ti, struct dm_dev *dev, |
| 1391 | sector_t start, sector_t len, void *data) |
| 1392 | { |
| 1393 | struct request_queue *q = bdev_get_queue(dev->bdev); |
| 1394 | |
| 1395 | return q && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags); |
| 1396 | } |
| 1397 | |
| 1398 | static int queue_supports_sg_gaps(struct dm_target *ti, struct dm_dev *dev, |
| 1399 | sector_t start, sector_t len, void *data) |
| 1400 | { |
| 1401 | struct request_queue *q = bdev_get_queue(dev->bdev); |
| 1402 | |
| 1403 | return q && !test_bit(QUEUE_FLAG_SG_GAPS, &q->queue_flags); |
| 1404 | } |
| 1405 | |
| 1406 | static bool dm_table_all_devices_attribute(struct dm_table *t, |
| 1407 | iterate_devices_callout_fn func) |
| 1408 | { |
| 1409 | struct dm_target *ti; |
| 1410 | unsigned i = 0; |
| 1411 | |
| 1412 | while (i < dm_table_get_num_targets(t)) { |
| 1413 | ti = dm_table_get_target(t, i++); |
| 1414 | |
| 1415 | if (!ti->type->iterate_devices || |
| 1416 | !ti->type->iterate_devices(ti, func, NULL)) |
| 1417 | return false; |
| 1418 | } |
| 1419 | |
| 1420 | return true; |
| 1421 | } |
| 1422 | |
| 1423 | static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev, |
| 1424 | sector_t start, sector_t len, void *data) |
| 1425 | { |
| 1426 | struct request_queue *q = bdev_get_queue(dev->bdev); |
| 1427 | |
| 1428 | return q && !q->limits.max_write_same_sectors; |
| 1429 | } |
| 1430 | |
| 1431 | static bool dm_table_supports_write_same(struct dm_table *t) |
| 1432 | { |
| 1433 | struct dm_target *ti; |
| 1434 | unsigned i = 0; |
| 1435 | |
| 1436 | while (i < dm_table_get_num_targets(t)) { |
| 1437 | ti = dm_table_get_target(t, i++); |
| 1438 | |
| 1439 | if (!ti->num_write_same_bios) |
| 1440 | return false; |
| 1441 | |
| 1442 | if (!ti->type->iterate_devices || |
| 1443 | ti->type->iterate_devices(ti, device_not_write_same_capable, NULL)) |
| 1444 | return false; |
| 1445 | } |
| 1446 | |
| 1447 | return true; |
| 1448 | } |
| 1449 | |
| 1450 | static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev, |
| 1451 | sector_t start, sector_t len, void *data) |
| 1452 | { |
| 1453 | struct request_queue *q = bdev_get_queue(dev->bdev); |
| 1454 | |
| 1455 | return q && blk_queue_discard(q); |
| 1456 | } |
| 1457 | |
| 1458 | static bool dm_table_supports_discards(struct dm_table *t) |
| 1459 | { |
| 1460 | struct dm_target *ti; |
| 1461 | unsigned i = 0; |
| 1462 | |
| 1463 | /* |
| 1464 | * Unless any target used by the table set discards_supported, |
| 1465 | * require at least one underlying device to support discards. |
| 1466 | * t->devices includes internal dm devices such as mirror logs |
| 1467 | * so we need to use iterate_devices here, which targets |
| 1468 | * supporting discard selectively must provide. |
| 1469 | */ |
| 1470 | while (i < dm_table_get_num_targets(t)) { |
| 1471 | ti = dm_table_get_target(t, i++); |
| 1472 | |
| 1473 | if (!ti->num_discard_bios) |
| 1474 | continue; |
| 1475 | |
| 1476 | if (ti->discards_supported) |
| 1477 | return true; |
| 1478 | |
| 1479 | if (ti->type->iterate_devices && |
| 1480 | ti->type->iterate_devices(ti, device_discard_capable, NULL)) |
| 1481 | return true; |
| 1482 | } |
| 1483 | |
| 1484 | return false; |
| 1485 | } |
| 1486 | |
| 1487 | void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q, |
| 1488 | struct queue_limits *limits) |
| 1489 | { |
| 1490 | unsigned flush = 0; |
| 1491 | |
| 1492 | /* |
| 1493 | * Copy table's limits to the DM device's request_queue |
| 1494 | */ |
| 1495 | q->limits = *limits; |
| 1496 | |
| 1497 | if (!dm_table_supports_discards(t)) |
| 1498 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); |
| 1499 | else |
| 1500 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); |
| 1501 | |
| 1502 | if (dm_table_supports_flush(t, REQ_FLUSH)) { |
| 1503 | flush |= REQ_FLUSH; |
| 1504 | if (dm_table_supports_flush(t, REQ_FUA)) |
| 1505 | flush |= REQ_FUA; |
| 1506 | } |
| 1507 | blk_queue_flush(q, flush); |
| 1508 | |
| 1509 | if (!dm_table_discard_zeroes_data(t)) |
| 1510 | q->limits.discard_zeroes_data = 0; |
| 1511 | |
| 1512 | /* Ensure that all underlying devices are non-rotational. */ |
| 1513 | if (dm_table_all_devices_attribute(t, device_is_nonrot)) |
| 1514 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q); |
| 1515 | else |
| 1516 | queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q); |
| 1517 | |
| 1518 | if (!dm_table_supports_write_same(t)) |
| 1519 | q->limits.max_write_same_sectors = 0; |
| 1520 | |
| 1521 | if (dm_table_all_devices_attribute(t, queue_supports_sg_merge)) |
| 1522 | queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q); |
| 1523 | else |
| 1524 | queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q); |
| 1525 | |
| 1526 | if (dm_table_all_devices_attribute(t, queue_supports_sg_gaps)) |
| 1527 | queue_flag_clear_unlocked(QUEUE_FLAG_SG_GAPS, q); |
| 1528 | else |
| 1529 | queue_flag_set_unlocked(QUEUE_FLAG_SG_GAPS, q); |
| 1530 | |
| 1531 | dm_table_set_integrity(t); |
| 1532 | |
| 1533 | /* |
| 1534 | * Determine whether or not this queue's I/O timings contribute |
| 1535 | * to the entropy pool, Only request-based targets use this. |
| 1536 | * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not |
| 1537 | * have it set. |
| 1538 | */ |
| 1539 | if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random)) |
| 1540 | queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q); |
| 1541 | |
| 1542 | /* |
| 1543 | * QUEUE_FLAG_STACKABLE must be set after all queue settings are |
| 1544 | * visible to other CPUs because, once the flag is set, incoming bios |
| 1545 | * are processed by request-based dm, which refers to the queue |
| 1546 | * settings. |
| 1547 | * Until the flag set, bios are passed to bio-based dm and queued to |
| 1548 | * md->deferred where queue settings are not needed yet. |
| 1549 | * Those bios are passed to request-based dm at the resume time. |
| 1550 | */ |
| 1551 | smp_mb(); |
| 1552 | if (dm_table_request_based(t)) |
| 1553 | queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q); |
| 1554 | } |
| 1555 | |
| 1556 | unsigned int dm_table_get_num_targets(struct dm_table *t) |
| 1557 | { |
| 1558 | return t->num_targets; |
| 1559 | } |
| 1560 | |
| 1561 | struct list_head *dm_table_get_devices(struct dm_table *t) |
| 1562 | { |
| 1563 | return &t->devices; |
| 1564 | } |
| 1565 | |
| 1566 | fmode_t dm_table_get_mode(struct dm_table *t) |
| 1567 | { |
| 1568 | return t->mode; |
| 1569 | } |
| 1570 | EXPORT_SYMBOL(dm_table_get_mode); |
| 1571 | |
| 1572 | enum suspend_mode { |
| 1573 | PRESUSPEND, |
| 1574 | PRESUSPEND_UNDO, |
| 1575 | POSTSUSPEND, |
| 1576 | }; |
| 1577 | |
| 1578 | static void suspend_targets(struct dm_table *t, enum suspend_mode mode) |
| 1579 | { |
| 1580 | int i = t->num_targets; |
| 1581 | struct dm_target *ti = t->targets; |
| 1582 | |
| 1583 | while (i--) { |
| 1584 | switch (mode) { |
| 1585 | case PRESUSPEND: |
| 1586 | if (ti->type->presuspend) |
| 1587 | ti->type->presuspend(ti); |
| 1588 | break; |
| 1589 | case PRESUSPEND_UNDO: |
| 1590 | if (ti->type->presuspend_undo) |
| 1591 | ti->type->presuspend_undo(ti); |
| 1592 | break; |
| 1593 | case POSTSUSPEND: |
| 1594 | if (ti->type->postsuspend) |
| 1595 | ti->type->postsuspend(ti); |
| 1596 | break; |
| 1597 | } |
| 1598 | ti++; |
| 1599 | } |
| 1600 | } |
| 1601 | |
| 1602 | void dm_table_presuspend_targets(struct dm_table *t) |
| 1603 | { |
| 1604 | if (!t) |
| 1605 | return; |
| 1606 | |
| 1607 | suspend_targets(t, PRESUSPEND); |
| 1608 | } |
| 1609 | |
| 1610 | void dm_table_presuspend_undo_targets(struct dm_table *t) |
| 1611 | { |
| 1612 | if (!t) |
| 1613 | return; |
| 1614 | |
| 1615 | suspend_targets(t, PRESUSPEND_UNDO); |
| 1616 | } |
| 1617 | |
| 1618 | void dm_table_postsuspend_targets(struct dm_table *t) |
| 1619 | { |
| 1620 | if (!t) |
| 1621 | return; |
| 1622 | |
| 1623 | suspend_targets(t, POSTSUSPEND); |
| 1624 | } |
| 1625 | |
| 1626 | int dm_table_resume_targets(struct dm_table *t) |
| 1627 | { |
| 1628 | int i, r = 0; |
| 1629 | |
| 1630 | for (i = 0; i < t->num_targets; i++) { |
| 1631 | struct dm_target *ti = t->targets + i; |
| 1632 | |
| 1633 | if (!ti->type->preresume) |
| 1634 | continue; |
| 1635 | |
| 1636 | r = ti->type->preresume(ti); |
| 1637 | if (r) { |
| 1638 | DMERR("%s: %s: preresume failed, error = %d", |
| 1639 | dm_device_name(t->md), ti->type->name, r); |
| 1640 | return r; |
| 1641 | } |
| 1642 | } |
| 1643 | |
| 1644 | for (i = 0; i < t->num_targets; i++) { |
| 1645 | struct dm_target *ti = t->targets + i; |
| 1646 | |
| 1647 | if (ti->type->resume) |
| 1648 | ti->type->resume(ti); |
| 1649 | } |
| 1650 | |
| 1651 | return 0; |
| 1652 | } |
| 1653 | |
| 1654 | void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb) |
| 1655 | { |
| 1656 | list_add(&cb->list, &t->target_callbacks); |
| 1657 | } |
| 1658 | EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks); |
| 1659 | |
| 1660 | int dm_table_any_congested(struct dm_table *t, int bdi_bits) |
| 1661 | { |
| 1662 | struct dm_dev_internal *dd; |
| 1663 | struct list_head *devices = dm_table_get_devices(t); |
| 1664 | struct dm_target_callbacks *cb; |
| 1665 | int r = 0; |
| 1666 | |
| 1667 | list_for_each_entry(dd, devices, list) { |
| 1668 | struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev); |
| 1669 | char b[BDEVNAME_SIZE]; |
| 1670 | |
| 1671 | if (likely(q)) |
| 1672 | r |= bdi_congested(&q->backing_dev_info, bdi_bits); |
| 1673 | else |
| 1674 | DMWARN_LIMIT("%s: any_congested: nonexistent device %s", |
| 1675 | dm_device_name(t->md), |
| 1676 | bdevname(dd->dm_dev->bdev, b)); |
| 1677 | } |
| 1678 | |
| 1679 | list_for_each_entry(cb, &t->target_callbacks, list) |
| 1680 | if (cb->congested_fn) |
| 1681 | r |= cb->congested_fn(cb, bdi_bits); |
| 1682 | |
| 1683 | return r; |
| 1684 | } |
| 1685 | |
| 1686 | struct mapped_device *dm_table_get_md(struct dm_table *t) |
| 1687 | { |
| 1688 | return t->md; |
| 1689 | } |
| 1690 | EXPORT_SYMBOL(dm_table_get_md); |
| 1691 | |
| 1692 | void dm_table_run_md_queue_async(struct dm_table *t) |
| 1693 | { |
| 1694 | struct mapped_device *md; |
| 1695 | struct request_queue *queue; |
| 1696 | unsigned long flags; |
| 1697 | |
| 1698 | if (!dm_table_request_based(t)) |
| 1699 | return; |
| 1700 | |
| 1701 | md = dm_table_get_md(t); |
| 1702 | queue = dm_get_md_queue(md); |
| 1703 | if (queue) { |
| 1704 | if (queue->mq_ops) |
| 1705 | blk_mq_run_hw_queues(queue, true); |
| 1706 | else { |
| 1707 | spin_lock_irqsave(queue->queue_lock, flags); |
| 1708 | blk_run_queue_async(queue); |
| 1709 | spin_unlock_irqrestore(queue->queue_lock, flags); |
| 1710 | } |
| 1711 | } |
| 1712 | } |
| 1713 | EXPORT_SYMBOL(dm_table_run_md_queue_async); |
| 1714 | |