Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2001 Sistina Software (UK) Limited. |
| 3 | * Copyright (C) 2004 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/slab.h> |
| 16 | #include <linux/interrupt.h> |
Arjan van de Ven | 48c9c27 | 2006-03-27 01:18:20 -0800 | [diff] [blame] | 17 | #include <linux/mutex.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | #include <asm/atomic.h> |
| 19 | |
| 20 | #define MAX_DEPTH 16 |
| 21 | #define NODE_SIZE L1_CACHE_BYTES |
| 22 | #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t)) |
| 23 | #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1) |
| 24 | |
| 25 | struct dm_table { |
Mike Anderson | 1134e5a | 2006-03-27 01:17:54 -0800 | [diff] [blame] | 26 | struct mapped_device *md; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 | atomic_t holders; |
| 28 | |
| 29 | /* btree table */ |
| 30 | unsigned int depth; |
| 31 | unsigned int counts[MAX_DEPTH]; /* in nodes */ |
| 32 | sector_t *index[MAX_DEPTH]; |
| 33 | |
| 34 | unsigned int num_targets; |
| 35 | unsigned int num_allocated; |
| 36 | sector_t *highs; |
| 37 | struct dm_target *targets; |
| 38 | |
| 39 | /* |
| 40 | * Indicates the rw permissions for the new logical |
| 41 | * device. This should be a combination of FMODE_READ |
| 42 | * and FMODE_WRITE. |
| 43 | */ |
| 44 | int mode; |
| 45 | |
| 46 | /* a list of devices used by this table */ |
| 47 | struct list_head devices; |
| 48 | |
| 49 | /* |
| 50 | * These are optimistic limits taken from all the |
| 51 | * targets, some targets will need smaller limits. |
| 52 | */ |
| 53 | struct io_restrictions limits; |
| 54 | |
| 55 | /* events get handed up using this callback */ |
| 56 | void (*event_fn)(void *); |
| 57 | void *event_context; |
| 58 | }; |
| 59 | |
| 60 | /* |
| 61 | * Similar to ceiling(log_size(n)) |
| 62 | */ |
| 63 | static unsigned int int_log(unsigned int n, unsigned int base) |
| 64 | { |
| 65 | int result = 0; |
| 66 | |
| 67 | while (n > 1) { |
| 68 | n = dm_div_up(n, base); |
| 69 | result++; |
| 70 | } |
| 71 | |
| 72 | return result; |
| 73 | } |
| 74 | |
| 75 | /* |
| 76 | * Returns the minimum that is _not_ zero, unless both are zero. |
| 77 | */ |
| 78 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) |
| 79 | |
| 80 | /* |
| 81 | * Combine two io_restrictions, always taking the lower value. |
| 82 | */ |
| 83 | static void combine_restrictions_low(struct io_restrictions *lhs, |
| 84 | struct io_restrictions *rhs) |
| 85 | { |
| 86 | lhs->max_sectors = |
| 87 | min_not_zero(lhs->max_sectors, rhs->max_sectors); |
| 88 | |
| 89 | lhs->max_phys_segments = |
| 90 | min_not_zero(lhs->max_phys_segments, rhs->max_phys_segments); |
| 91 | |
| 92 | lhs->max_hw_segments = |
| 93 | min_not_zero(lhs->max_hw_segments, rhs->max_hw_segments); |
| 94 | |
| 95 | lhs->hardsect_size = max(lhs->hardsect_size, rhs->hardsect_size); |
| 96 | |
| 97 | lhs->max_segment_size = |
| 98 | min_not_zero(lhs->max_segment_size, rhs->max_segment_size); |
| 99 | |
| 100 | lhs->seg_boundary_mask = |
| 101 | min_not_zero(lhs->seg_boundary_mask, rhs->seg_boundary_mask); |
NeilBrown | 969429b | 2006-03-27 01:17:49 -0800 | [diff] [blame] | 102 | |
| 103 | lhs->no_cluster |= rhs->no_cluster; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 104 | } |
| 105 | |
| 106 | /* |
| 107 | * Calculate the index of the child node of the n'th node k'th key. |
| 108 | */ |
| 109 | static inline unsigned int get_child(unsigned int n, unsigned int k) |
| 110 | { |
| 111 | return (n * CHILDREN_PER_NODE) + k; |
| 112 | } |
| 113 | |
| 114 | /* |
| 115 | * Return the n'th node of level l from table t. |
| 116 | */ |
| 117 | static inline sector_t *get_node(struct dm_table *t, |
| 118 | unsigned int l, unsigned int n) |
| 119 | { |
| 120 | return t->index[l] + (n * KEYS_PER_NODE); |
| 121 | } |
| 122 | |
| 123 | /* |
| 124 | * Return the highest key that you could lookup from the n'th |
| 125 | * node on level l of the btree. |
| 126 | */ |
| 127 | static sector_t high(struct dm_table *t, unsigned int l, unsigned int n) |
| 128 | { |
| 129 | for (; l < t->depth - 1; l++) |
| 130 | n = get_child(n, CHILDREN_PER_NODE - 1); |
| 131 | |
| 132 | if (n >= t->counts[l]) |
| 133 | return (sector_t) - 1; |
| 134 | |
| 135 | return get_node(t, l, n)[KEYS_PER_NODE - 1]; |
| 136 | } |
| 137 | |
| 138 | /* |
| 139 | * Fills in a level of the btree based on the highs of the level |
| 140 | * below it. |
| 141 | */ |
| 142 | static int setup_btree_index(unsigned int l, struct dm_table *t) |
| 143 | { |
| 144 | unsigned int n, k; |
| 145 | sector_t *node; |
| 146 | |
| 147 | for (n = 0U; n < t->counts[l]; n++) { |
| 148 | node = get_node(t, l, n); |
| 149 | |
| 150 | for (k = 0U; k < KEYS_PER_NODE; k++) |
| 151 | node[k] = high(t, l + 1, get_child(n, k)); |
| 152 | } |
| 153 | |
| 154 | return 0; |
| 155 | } |
| 156 | |
| 157 | void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size) |
| 158 | { |
| 159 | unsigned long size; |
| 160 | void *addr; |
| 161 | |
| 162 | /* |
| 163 | * Check that we're not going to overflow. |
| 164 | */ |
| 165 | if (nmemb > (ULONG_MAX / elem_size)) |
| 166 | return NULL; |
| 167 | |
| 168 | size = nmemb * elem_size; |
| 169 | addr = vmalloc(size); |
| 170 | if (addr) |
| 171 | memset(addr, 0, size); |
| 172 | |
| 173 | return addr; |
| 174 | } |
| 175 | |
| 176 | /* |
| 177 | * highs, and targets are managed as dynamic arrays during a |
| 178 | * table load. |
| 179 | */ |
| 180 | static int alloc_targets(struct dm_table *t, unsigned int num) |
| 181 | { |
| 182 | sector_t *n_highs; |
| 183 | struct dm_target *n_targets; |
| 184 | int n = t->num_targets; |
| 185 | |
| 186 | /* |
| 187 | * Allocate both the target array and offset array at once. |
| 188 | */ |
| 189 | n_highs = (sector_t *) dm_vcalloc(num, sizeof(struct dm_target) + |
| 190 | sizeof(sector_t)); |
| 191 | if (!n_highs) |
| 192 | return -ENOMEM; |
| 193 | |
| 194 | n_targets = (struct dm_target *) (n_highs + num); |
| 195 | |
| 196 | if (n) { |
| 197 | memcpy(n_highs, t->highs, sizeof(*n_highs) * n); |
| 198 | memcpy(n_targets, t->targets, sizeof(*n_targets) * n); |
| 199 | } |
| 200 | |
| 201 | memset(n_highs + n, -1, sizeof(*n_highs) * (num - n)); |
| 202 | vfree(t->highs); |
| 203 | |
| 204 | t->num_allocated = num; |
| 205 | t->highs = n_highs; |
| 206 | t->targets = n_targets; |
| 207 | |
| 208 | return 0; |
| 209 | } |
| 210 | |
Mike Anderson | 1134e5a | 2006-03-27 01:17:54 -0800 | [diff] [blame] | 211 | int dm_table_create(struct dm_table **result, int mode, |
| 212 | unsigned num_targets, struct mapped_device *md) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 213 | { |
| 214 | struct dm_table *t = kmalloc(sizeof(*t), GFP_KERNEL); |
| 215 | |
| 216 | if (!t) |
| 217 | return -ENOMEM; |
| 218 | |
| 219 | memset(t, 0, sizeof(*t)); |
| 220 | INIT_LIST_HEAD(&t->devices); |
| 221 | atomic_set(&t->holders, 1); |
| 222 | |
| 223 | if (!num_targets) |
| 224 | num_targets = KEYS_PER_NODE; |
| 225 | |
| 226 | num_targets = dm_round_up(num_targets, KEYS_PER_NODE); |
| 227 | |
| 228 | if (alloc_targets(t, num_targets)) { |
| 229 | kfree(t); |
| 230 | t = NULL; |
| 231 | return -ENOMEM; |
| 232 | } |
| 233 | |
| 234 | t->mode = mode; |
Mike Anderson | 1134e5a | 2006-03-27 01:17:54 -0800 | [diff] [blame] | 235 | t->md = md; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 236 | *result = t; |
| 237 | return 0; |
| 238 | } |
| 239 | |
| 240 | static void free_devices(struct list_head *devices) |
| 241 | { |
| 242 | struct list_head *tmp, *next; |
| 243 | |
| 244 | for (tmp = devices->next; tmp != devices; tmp = next) { |
| 245 | struct dm_dev *dd = list_entry(tmp, struct dm_dev, list); |
| 246 | next = tmp->next; |
| 247 | kfree(dd); |
| 248 | } |
| 249 | } |
| 250 | |
Alasdair G Kergon | 5e198d9 | 2005-05-05 16:16:09 -0700 | [diff] [blame] | 251 | static void table_destroy(struct dm_table *t) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 252 | { |
| 253 | unsigned int i; |
| 254 | |
| 255 | /* free the indexes (see dm_table_complete) */ |
| 256 | if (t->depth >= 2) |
| 257 | vfree(t->index[t->depth - 2]); |
| 258 | |
| 259 | /* free the targets */ |
| 260 | for (i = 0; i < t->num_targets; i++) { |
| 261 | struct dm_target *tgt = t->targets + i; |
| 262 | |
| 263 | if (tgt->type->dtr) |
| 264 | tgt->type->dtr(tgt); |
| 265 | |
| 266 | dm_put_target_type(tgt->type); |
| 267 | } |
| 268 | |
| 269 | vfree(t->highs); |
| 270 | |
| 271 | /* free the device list */ |
| 272 | if (t->devices.next != &t->devices) { |
| 273 | DMWARN("devices still present during destroy: " |
| 274 | "dm_table_remove_device calls missing"); |
| 275 | |
| 276 | free_devices(&t->devices); |
| 277 | } |
| 278 | |
| 279 | kfree(t); |
| 280 | } |
| 281 | |
| 282 | void dm_table_get(struct dm_table *t) |
| 283 | { |
| 284 | atomic_inc(&t->holders); |
| 285 | } |
| 286 | |
| 287 | void dm_table_put(struct dm_table *t) |
| 288 | { |
| 289 | if (!t) |
| 290 | return; |
| 291 | |
| 292 | if (atomic_dec_and_test(&t->holders)) |
| 293 | table_destroy(t); |
| 294 | } |
| 295 | |
| 296 | /* |
| 297 | * Checks to see if we need to extend highs or targets. |
| 298 | */ |
| 299 | static inline int check_space(struct dm_table *t) |
| 300 | { |
| 301 | if (t->num_targets >= t->num_allocated) |
| 302 | return alloc_targets(t, t->num_allocated * 2); |
| 303 | |
| 304 | return 0; |
| 305 | } |
| 306 | |
| 307 | /* |
| 308 | * Convert a device path to a dev_t. |
| 309 | */ |
| 310 | static int lookup_device(const char *path, dev_t *dev) |
| 311 | { |
| 312 | int r; |
| 313 | struct nameidata nd; |
| 314 | struct inode *inode; |
| 315 | |
| 316 | if ((r = path_lookup(path, LOOKUP_FOLLOW, &nd))) |
| 317 | return r; |
| 318 | |
| 319 | inode = nd.dentry->d_inode; |
| 320 | if (!inode) { |
| 321 | r = -ENOENT; |
| 322 | goto out; |
| 323 | } |
| 324 | |
| 325 | if (!S_ISBLK(inode->i_mode)) { |
| 326 | r = -ENOTBLK; |
| 327 | goto out; |
| 328 | } |
| 329 | |
| 330 | *dev = inode->i_rdev; |
| 331 | |
| 332 | out: |
| 333 | path_release(&nd); |
| 334 | return r; |
| 335 | } |
| 336 | |
| 337 | /* |
| 338 | * See if we've already got a device in the list. |
| 339 | */ |
| 340 | static struct dm_dev *find_device(struct list_head *l, dev_t dev) |
| 341 | { |
| 342 | struct dm_dev *dd; |
| 343 | |
| 344 | list_for_each_entry (dd, l, list) |
| 345 | if (dd->bdev->bd_dev == dev) |
| 346 | return dd; |
| 347 | |
| 348 | return NULL; |
| 349 | } |
| 350 | |
| 351 | /* |
| 352 | * Open a device so we can use it as a map destination. |
| 353 | */ |
Jun'ichi Nomura | f165921 | 2006-03-27 01:17:59 -0800 | [diff] [blame] | 354 | static int open_dev(struct dm_dev *d, dev_t dev, struct mapped_device *md) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 355 | { |
| 356 | static char *_claim_ptr = "I belong to device-mapper"; |
| 357 | struct block_device *bdev; |
| 358 | |
| 359 | int r; |
| 360 | |
Eric Sesterhenn | 547bc92 | 2006-03-26 18:22:50 +0200 | [diff] [blame] | 361 | BUG_ON(d->bdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 362 | |
| 363 | bdev = open_by_devnum(dev, d->mode); |
| 364 | if (IS_ERR(bdev)) |
| 365 | return PTR_ERR(bdev); |
Jun'ichi Nomura | f165921 | 2006-03-27 01:17:59 -0800 | [diff] [blame] | 366 | r = bd_claim_by_disk(bdev, _claim_ptr, dm_disk(md)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 367 | if (r) |
| 368 | blkdev_put(bdev); |
| 369 | else |
| 370 | d->bdev = bdev; |
| 371 | return r; |
| 372 | } |
| 373 | |
| 374 | /* |
| 375 | * Close a device that we've been using. |
| 376 | */ |
Jun'ichi Nomura | f165921 | 2006-03-27 01:17:59 -0800 | [diff] [blame] | 377 | static void close_dev(struct dm_dev *d, struct mapped_device *md) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 378 | { |
| 379 | if (!d->bdev) |
| 380 | return; |
| 381 | |
Jun'ichi Nomura | f165921 | 2006-03-27 01:17:59 -0800 | [diff] [blame] | 382 | bd_release_from_disk(d->bdev, dm_disk(md)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 383 | blkdev_put(d->bdev); |
| 384 | d->bdev = NULL; |
| 385 | } |
| 386 | |
| 387 | /* |
| 388 | * If possible (ie. blk_size[major] is set), this checks an area |
| 389 | * of a destination device is valid. |
| 390 | */ |
| 391 | static int check_device_area(struct dm_dev *dd, sector_t start, sector_t len) |
| 392 | { |
| 393 | sector_t dev_size; |
| 394 | dev_size = dd->bdev->bd_inode->i_size >> SECTOR_SHIFT; |
| 395 | return ((start < dev_size) && (len <= (dev_size - start))); |
| 396 | } |
| 397 | |
| 398 | /* |
| 399 | * This upgrades the mode on an already open dm_dev. Being |
| 400 | * careful to leave things as they were if we fail to reopen the |
| 401 | * device. |
| 402 | */ |
Jun'ichi Nomura | f165921 | 2006-03-27 01:17:59 -0800 | [diff] [blame] | 403 | static int upgrade_mode(struct dm_dev *dd, int new_mode, struct mapped_device *md) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 404 | { |
| 405 | int r; |
| 406 | struct dm_dev dd_copy; |
| 407 | dev_t dev = dd->bdev->bd_dev; |
| 408 | |
| 409 | dd_copy = *dd; |
| 410 | |
| 411 | dd->mode |= new_mode; |
| 412 | dd->bdev = NULL; |
Jun'ichi Nomura | f165921 | 2006-03-27 01:17:59 -0800 | [diff] [blame] | 413 | r = open_dev(dd, dev, md); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 414 | if (!r) |
Jun'ichi Nomura | f165921 | 2006-03-27 01:17:59 -0800 | [diff] [blame] | 415 | close_dev(&dd_copy, md); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 416 | else |
| 417 | *dd = dd_copy; |
| 418 | |
| 419 | return r; |
| 420 | } |
| 421 | |
| 422 | /* |
| 423 | * Add a device to the list, or just increment the usage count if |
| 424 | * it's already present. |
| 425 | */ |
| 426 | static int __table_get_device(struct dm_table *t, struct dm_target *ti, |
| 427 | const char *path, sector_t start, sector_t len, |
| 428 | int mode, struct dm_dev **result) |
| 429 | { |
| 430 | int r; |
| 431 | dev_t dev; |
| 432 | struct dm_dev *dd; |
| 433 | unsigned int major, minor; |
| 434 | |
Eric Sesterhenn | 547bc92 | 2006-03-26 18:22:50 +0200 | [diff] [blame] | 435 | BUG_ON(!t); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 436 | |
| 437 | if (sscanf(path, "%u:%u", &major, &minor) == 2) { |
| 438 | /* Extract the major/minor numbers */ |
| 439 | dev = MKDEV(major, minor); |
| 440 | if (MAJOR(dev) != major || MINOR(dev) != minor) |
| 441 | return -EOVERFLOW; |
| 442 | } else { |
| 443 | /* convert the path to a device */ |
| 444 | if ((r = lookup_device(path, &dev))) |
| 445 | return r; |
| 446 | } |
| 447 | |
| 448 | dd = find_device(&t->devices, dev); |
| 449 | if (!dd) { |
| 450 | dd = kmalloc(sizeof(*dd), GFP_KERNEL); |
| 451 | if (!dd) |
| 452 | return -ENOMEM; |
| 453 | |
| 454 | dd->mode = mode; |
| 455 | dd->bdev = NULL; |
| 456 | |
Jun'ichi Nomura | f165921 | 2006-03-27 01:17:59 -0800 | [diff] [blame] | 457 | if ((r = open_dev(dd, dev, t->md))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 458 | kfree(dd); |
| 459 | return r; |
| 460 | } |
| 461 | |
| 462 | format_dev_t(dd->name, dev); |
| 463 | |
| 464 | atomic_set(&dd->count, 0); |
| 465 | list_add(&dd->list, &t->devices); |
| 466 | |
| 467 | } else if (dd->mode != (mode | dd->mode)) { |
Jun'ichi Nomura | f165921 | 2006-03-27 01:17:59 -0800 | [diff] [blame] | 468 | r = upgrade_mode(dd, mode, t->md); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 469 | if (r) |
| 470 | return r; |
| 471 | } |
| 472 | atomic_inc(&dd->count); |
| 473 | |
| 474 | if (!check_device_area(dd, start, len)) { |
| 475 | DMWARN("device %s too small for target", path); |
| 476 | dm_put_device(ti, dd); |
| 477 | return -EINVAL; |
| 478 | } |
| 479 | |
| 480 | *result = dd; |
| 481 | |
| 482 | return 0; |
| 483 | } |
| 484 | |
| 485 | |
| 486 | int dm_get_device(struct dm_target *ti, const char *path, sector_t start, |
| 487 | sector_t len, int mode, struct dm_dev **result) |
| 488 | { |
| 489 | int r = __table_get_device(ti->table, ti, path, |
| 490 | start, len, mode, result); |
| 491 | if (!r) { |
| 492 | request_queue_t *q = bdev_get_queue((*result)->bdev); |
| 493 | struct io_restrictions *rs = &ti->limits; |
| 494 | |
| 495 | /* |
| 496 | * Combine the device limits low. |
| 497 | * |
| 498 | * FIXME: if we move an io_restriction struct |
| 499 | * into q this would just be a call to |
| 500 | * combine_restrictions_low() |
| 501 | */ |
| 502 | rs->max_sectors = |
| 503 | min_not_zero(rs->max_sectors, q->max_sectors); |
| 504 | |
| 505 | /* FIXME: Device-Mapper on top of RAID-0 breaks because DM |
| 506 | * currently doesn't honor MD's merge_bvec_fn routine. |
| 507 | * In this case, we'll force DM to use PAGE_SIZE or |
| 508 | * smaller I/O, just to be safe. A better fix is in the |
| 509 | * works, but add this for the time being so it will at |
| 510 | * least operate correctly. |
| 511 | */ |
| 512 | if (q->merge_bvec_fn) |
| 513 | rs->max_sectors = |
| 514 | min_not_zero(rs->max_sectors, |
Alasdair G Kergon | 3ee247e | 2006-02-01 03:04:55 -0800 | [diff] [blame] | 515 | (unsigned int) (PAGE_SIZE >> 9)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 516 | |
| 517 | rs->max_phys_segments = |
| 518 | min_not_zero(rs->max_phys_segments, |
| 519 | q->max_phys_segments); |
| 520 | |
| 521 | rs->max_hw_segments = |
| 522 | min_not_zero(rs->max_hw_segments, q->max_hw_segments); |
| 523 | |
| 524 | rs->hardsect_size = max(rs->hardsect_size, q->hardsect_size); |
| 525 | |
| 526 | rs->max_segment_size = |
| 527 | min_not_zero(rs->max_segment_size, q->max_segment_size); |
| 528 | |
| 529 | rs->seg_boundary_mask = |
| 530 | min_not_zero(rs->seg_boundary_mask, |
| 531 | q->seg_boundary_mask); |
NeilBrown | 969429b | 2006-03-27 01:17:49 -0800 | [diff] [blame] | 532 | |
| 533 | rs->no_cluster |= !test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 534 | } |
| 535 | |
| 536 | return r; |
| 537 | } |
| 538 | |
| 539 | /* |
| 540 | * Decrement a devices use count and remove it if necessary. |
| 541 | */ |
| 542 | void dm_put_device(struct dm_target *ti, struct dm_dev *dd) |
| 543 | { |
| 544 | if (atomic_dec_and_test(&dd->count)) { |
Jun'ichi Nomura | f165921 | 2006-03-27 01:17:59 -0800 | [diff] [blame] | 545 | close_dev(dd, ti->table->md); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 546 | list_del(&dd->list); |
| 547 | kfree(dd); |
| 548 | } |
| 549 | } |
| 550 | |
| 551 | /* |
| 552 | * Checks to see if the target joins onto the end of the table. |
| 553 | */ |
| 554 | static int adjoin(struct dm_table *table, struct dm_target *ti) |
| 555 | { |
| 556 | struct dm_target *prev; |
| 557 | |
| 558 | if (!table->num_targets) |
| 559 | return !ti->begin; |
| 560 | |
| 561 | prev = &table->targets[table->num_targets - 1]; |
| 562 | return (ti->begin == (prev->begin + prev->len)); |
| 563 | } |
| 564 | |
| 565 | /* |
| 566 | * Used to dynamically allocate the arg array. |
| 567 | */ |
| 568 | static char **realloc_argv(unsigned *array_size, char **old_argv) |
| 569 | { |
| 570 | char **argv; |
| 571 | unsigned new_size; |
| 572 | |
| 573 | new_size = *array_size ? *array_size * 2 : 64; |
| 574 | argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL); |
| 575 | if (argv) { |
| 576 | memcpy(argv, old_argv, *array_size * sizeof(*argv)); |
| 577 | *array_size = new_size; |
| 578 | } |
| 579 | |
| 580 | kfree(old_argv); |
| 581 | return argv; |
| 582 | } |
| 583 | |
| 584 | /* |
| 585 | * Destructively splits up the argument list to pass to ctr. |
| 586 | */ |
| 587 | int dm_split_args(int *argc, char ***argvp, char *input) |
| 588 | { |
| 589 | char *start, *end = input, *out, **argv = NULL; |
| 590 | unsigned array_size = 0; |
| 591 | |
| 592 | *argc = 0; |
| 593 | argv = realloc_argv(&array_size, argv); |
| 594 | if (!argv) |
| 595 | return -ENOMEM; |
| 596 | |
| 597 | while (1) { |
| 598 | start = end; |
| 599 | |
| 600 | /* Skip whitespace */ |
| 601 | while (*start && isspace(*start)) |
| 602 | start++; |
| 603 | |
| 604 | if (!*start) |
| 605 | break; /* success, we hit the end */ |
| 606 | |
| 607 | /* 'out' is used to remove any back-quotes */ |
| 608 | end = out = start; |
| 609 | while (*end) { |
| 610 | /* Everything apart from '\0' can be quoted */ |
| 611 | if (*end == '\\' && *(end + 1)) { |
| 612 | *out++ = *(end + 1); |
| 613 | end += 2; |
| 614 | continue; |
| 615 | } |
| 616 | |
| 617 | if (isspace(*end)) |
| 618 | break; /* end of token */ |
| 619 | |
| 620 | *out++ = *end++; |
| 621 | } |
| 622 | |
| 623 | /* have we already filled the array ? */ |
| 624 | if ((*argc + 1) > array_size) { |
| 625 | argv = realloc_argv(&array_size, argv); |
| 626 | if (!argv) |
| 627 | return -ENOMEM; |
| 628 | } |
| 629 | |
| 630 | /* we know this is whitespace */ |
| 631 | if (*end) |
| 632 | end++; |
| 633 | |
| 634 | /* terminate the string and put it in the array */ |
| 635 | *out = '\0'; |
| 636 | argv[*argc] = start; |
| 637 | (*argc)++; |
| 638 | } |
| 639 | |
| 640 | *argvp = argv; |
| 641 | return 0; |
| 642 | } |
| 643 | |
| 644 | static void check_for_valid_limits(struct io_restrictions *rs) |
| 645 | { |
| 646 | if (!rs->max_sectors) |
Mike Christie | defd94b | 2005-12-05 02:37:06 -0600 | [diff] [blame] | 647 | rs->max_sectors = SAFE_MAX_SECTORS; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 648 | if (!rs->max_phys_segments) |
| 649 | rs->max_phys_segments = MAX_PHYS_SEGMENTS; |
| 650 | if (!rs->max_hw_segments) |
| 651 | rs->max_hw_segments = MAX_HW_SEGMENTS; |
| 652 | if (!rs->hardsect_size) |
| 653 | rs->hardsect_size = 1 << SECTOR_SHIFT; |
| 654 | if (!rs->max_segment_size) |
| 655 | rs->max_segment_size = MAX_SEGMENT_SIZE; |
| 656 | if (!rs->seg_boundary_mask) |
| 657 | rs->seg_boundary_mask = -1; |
| 658 | } |
| 659 | |
| 660 | int dm_table_add_target(struct dm_table *t, const char *type, |
| 661 | sector_t start, sector_t len, char *params) |
| 662 | { |
| 663 | int r = -EINVAL, argc; |
| 664 | char **argv; |
| 665 | struct dm_target *tgt; |
| 666 | |
| 667 | if ((r = check_space(t))) |
| 668 | return r; |
| 669 | |
| 670 | tgt = t->targets + t->num_targets; |
| 671 | memset(tgt, 0, sizeof(*tgt)); |
| 672 | |
| 673 | if (!len) { |
| 674 | tgt->error = "zero-length target"; |
| 675 | DMERR("%s", tgt->error); |
| 676 | return -EINVAL; |
| 677 | } |
| 678 | |
| 679 | tgt->type = dm_get_target_type(type); |
| 680 | if (!tgt->type) { |
| 681 | tgt->error = "unknown target type"; |
| 682 | DMERR("%s", tgt->error); |
| 683 | return -EINVAL; |
| 684 | } |
| 685 | |
| 686 | tgt->table = t; |
| 687 | tgt->begin = start; |
| 688 | tgt->len = len; |
| 689 | tgt->error = "Unknown error"; |
| 690 | |
| 691 | /* |
| 692 | * Does this target adjoin the previous one ? |
| 693 | */ |
| 694 | if (!adjoin(t, tgt)) { |
| 695 | tgt->error = "Gap in table"; |
| 696 | r = -EINVAL; |
| 697 | goto bad; |
| 698 | } |
| 699 | |
| 700 | r = dm_split_args(&argc, &argv, params); |
| 701 | if (r) { |
| 702 | tgt->error = "couldn't split parameters (insufficient memory)"; |
| 703 | goto bad; |
| 704 | } |
| 705 | |
| 706 | r = tgt->type->ctr(tgt, argc, argv); |
| 707 | kfree(argv); |
| 708 | if (r) |
| 709 | goto bad; |
| 710 | |
| 711 | t->highs[t->num_targets++] = tgt->begin + tgt->len - 1; |
| 712 | |
| 713 | /* FIXME: the plan is to combine high here and then have |
| 714 | * the merge fn apply the target level restrictions. */ |
| 715 | combine_restrictions_low(&t->limits, &tgt->limits); |
| 716 | return 0; |
| 717 | |
| 718 | bad: |
| 719 | DMERR("%s", tgt->error); |
| 720 | dm_put_target_type(tgt->type); |
| 721 | return r; |
| 722 | } |
| 723 | |
| 724 | static int setup_indexes(struct dm_table *t) |
| 725 | { |
| 726 | int i; |
| 727 | unsigned int total = 0; |
| 728 | sector_t *indexes; |
| 729 | |
| 730 | /* allocate the space for *all* the indexes */ |
| 731 | for (i = t->depth - 2; i >= 0; i--) { |
| 732 | t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE); |
| 733 | total += t->counts[i]; |
| 734 | } |
| 735 | |
| 736 | indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE); |
| 737 | if (!indexes) |
| 738 | return -ENOMEM; |
| 739 | |
| 740 | /* set up internal nodes, bottom-up */ |
| 741 | for (i = t->depth - 2, total = 0; i >= 0; i--) { |
| 742 | t->index[i] = indexes; |
| 743 | indexes += (KEYS_PER_NODE * t->counts[i]); |
| 744 | setup_btree_index(i, t); |
| 745 | } |
| 746 | |
| 747 | return 0; |
| 748 | } |
| 749 | |
| 750 | /* |
| 751 | * Builds the btree to index the map. |
| 752 | */ |
| 753 | int dm_table_complete(struct dm_table *t) |
| 754 | { |
| 755 | int r = 0; |
| 756 | unsigned int leaf_nodes; |
| 757 | |
| 758 | check_for_valid_limits(&t->limits); |
| 759 | |
| 760 | /* how many indexes will the btree have ? */ |
| 761 | leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE); |
| 762 | t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE); |
| 763 | |
| 764 | /* leaf layer has already been set up */ |
| 765 | t->counts[t->depth - 1] = leaf_nodes; |
| 766 | t->index[t->depth - 1] = t->highs; |
| 767 | |
| 768 | if (t->depth >= 2) |
| 769 | r = setup_indexes(t); |
| 770 | |
| 771 | return r; |
| 772 | } |
| 773 | |
Arjan van de Ven | 48c9c27 | 2006-03-27 01:18:20 -0800 | [diff] [blame] | 774 | static DEFINE_MUTEX(_event_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 775 | void dm_table_event_callback(struct dm_table *t, |
| 776 | void (*fn)(void *), void *context) |
| 777 | { |
Arjan van de Ven | 48c9c27 | 2006-03-27 01:18:20 -0800 | [diff] [blame] | 778 | mutex_lock(&_event_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 779 | t->event_fn = fn; |
| 780 | t->event_context = context; |
Arjan van de Ven | 48c9c27 | 2006-03-27 01:18:20 -0800 | [diff] [blame] | 781 | mutex_unlock(&_event_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 782 | } |
| 783 | |
| 784 | void dm_table_event(struct dm_table *t) |
| 785 | { |
| 786 | /* |
| 787 | * You can no longer call dm_table_event() from interrupt |
| 788 | * context, use a bottom half instead. |
| 789 | */ |
| 790 | BUG_ON(in_interrupt()); |
| 791 | |
Arjan van de Ven | 48c9c27 | 2006-03-27 01:18:20 -0800 | [diff] [blame] | 792 | mutex_lock(&_event_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 793 | if (t->event_fn) |
| 794 | t->event_fn(t->event_context); |
Arjan van de Ven | 48c9c27 | 2006-03-27 01:18:20 -0800 | [diff] [blame] | 795 | mutex_unlock(&_event_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 796 | } |
| 797 | |
| 798 | sector_t dm_table_get_size(struct dm_table *t) |
| 799 | { |
| 800 | return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0; |
| 801 | } |
| 802 | |
| 803 | struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index) |
| 804 | { |
| 805 | if (index > t->num_targets) |
| 806 | return NULL; |
| 807 | |
| 808 | return t->targets + index; |
| 809 | } |
| 810 | |
| 811 | /* |
| 812 | * Search the btree for the correct target. |
| 813 | */ |
| 814 | struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector) |
| 815 | { |
| 816 | unsigned int l, n = 0, k = 0; |
| 817 | sector_t *node; |
| 818 | |
| 819 | for (l = 0; l < t->depth; l++) { |
| 820 | n = get_child(n, k); |
| 821 | node = get_node(t, l, n); |
| 822 | |
| 823 | for (k = 0; k < KEYS_PER_NODE; k++) |
| 824 | if (node[k] >= sector) |
| 825 | break; |
| 826 | } |
| 827 | |
| 828 | return &t->targets[(KEYS_PER_NODE * n) + k]; |
| 829 | } |
| 830 | |
| 831 | void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q) |
| 832 | { |
| 833 | /* |
| 834 | * Make sure we obey the optimistic sub devices |
| 835 | * restrictions. |
| 836 | */ |
| 837 | blk_queue_max_sectors(q, t->limits.max_sectors); |
| 838 | q->max_phys_segments = t->limits.max_phys_segments; |
| 839 | q->max_hw_segments = t->limits.max_hw_segments; |
| 840 | q->hardsect_size = t->limits.hardsect_size; |
| 841 | q->max_segment_size = t->limits.max_segment_size; |
| 842 | q->seg_boundary_mask = t->limits.seg_boundary_mask; |
NeilBrown | 969429b | 2006-03-27 01:17:49 -0800 | [diff] [blame] | 843 | if (t->limits.no_cluster) |
| 844 | q->queue_flags &= ~(1 << QUEUE_FLAG_CLUSTER); |
| 845 | else |
| 846 | q->queue_flags |= (1 << QUEUE_FLAG_CLUSTER); |
| 847 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 848 | } |
| 849 | |
| 850 | unsigned int dm_table_get_num_targets(struct dm_table *t) |
| 851 | { |
| 852 | return t->num_targets; |
| 853 | } |
| 854 | |
| 855 | struct list_head *dm_table_get_devices(struct dm_table *t) |
| 856 | { |
| 857 | return &t->devices; |
| 858 | } |
| 859 | |
| 860 | int dm_table_get_mode(struct dm_table *t) |
| 861 | { |
| 862 | return t->mode; |
| 863 | } |
| 864 | |
| 865 | static void suspend_targets(struct dm_table *t, unsigned postsuspend) |
| 866 | { |
| 867 | int i = t->num_targets; |
| 868 | struct dm_target *ti = t->targets; |
| 869 | |
| 870 | while (i--) { |
| 871 | if (postsuspend) { |
| 872 | if (ti->type->postsuspend) |
| 873 | ti->type->postsuspend(ti); |
| 874 | } else if (ti->type->presuspend) |
| 875 | ti->type->presuspend(ti); |
| 876 | |
| 877 | ti++; |
| 878 | } |
| 879 | } |
| 880 | |
| 881 | void dm_table_presuspend_targets(struct dm_table *t) |
| 882 | { |
Alasdair G Kergon | cf222b3 | 2005-07-28 21:15:57 -0700 | [diff] [blame] | 883 | if (!t) |
| 884 | return; |
| 885 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 886 | return suspend_targets(t, 0); |
| 887 | } |
| 888 | |
| 889 | void dm_table_postsuspend_targets(struct dm_table *t) |
| 890 | { |
Alasdair G Kergon | cf222b3 | 2005-07-28 21:15:57 -0700 | [diff] [blame] | 891 | if (!t) |
| 892 | return; |
| 893 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 894 | return suspend_targets(t, 1); |
| 895 | } |
| 896 | |
| 897 | void dm_table_resume_targets(struct dm_table *t) |
| 898 | { |
| 899 | int i; |
| 900 | |
| 901 | for (i = 0; i < t->num_targets; i++) { |
| 902 | struct dm_target *ti = t->targets + i; |
| 903 | |
| 904 | if (ti->type->resume) |
| 905 | ti->type->resume(ti); |
| 906 | } |
| 907 | } |
| 908 | |
| 909 | int dm_table_any_congested(struct dm_table *t, int bdi_bits) |
| 910 | { |
| 911 | struct list_head *d, *devices; |
| 912 | int r = 0; |
| 913 | |
| 914 | devices = dm_table_get_devices(t); |
| 915 | for (d = devices->next; d != devices; d = d->next) { |
| 916 | struct dm_dev *dd = list_entry(d, struct dm_dev, list); |
| 917 | request_queue_t *q = bdev_get_queue(dd->bdev); |
| 918 | r |= bdi_congested(&q->backing_dev_info, bdi_bits); |
| 919 | } |
| 920 | |
| 921 | return r; |
| 922 | } |
| 923 | |
| 924 | void dm_table_unplug_all(struct dm_table *t) |
| 925 | { |
| 926 | struct list_head *d, *devices = dm_table_get_devices(t); |
| 927 | |
| 928 | for (d = devices->next; d != devices; d = d->next) { |
| 929 | struct dm_dev *dd = list_entry(d, struct dm_dev, list); |
| 930 | request_queue_t *q = bdev_get_queue(dd->bdev); |
| 931 | |
| 932 | if (q->unplug_fn) |
| 933 | q->unplug_fn(q); |
| 934 | } |
| 935 | } |
| 936 | |
| 937 | int dm_table_flush_all(struct dm_table *t) |
| 938 | { |
| 939 | struct list_head *d, *devices = dm_table_get_devices(t); |
| 940 | int ret = 0; |
| 941 | |
| 942 | for (d = devices->next; d != devices; d = d->next) { |
| 943 | struct dm_dev *dd = list_entry(d, struct dm_dev, list); |
| 944 | request_queue_t *q = bdev_get_queue(dd->bdev); |
| 945 | int err; |
| 946 | |
| 947 | if (!q->issue_flush_fn) |
| 948 | err = -EOPNOTSUPP; |
| 949 | else |
| 950 | err = q->issue_flush_fn(q, dd->bdev->bd_disk, NULL); |
| 951 | |
| 952 | if (!ret) |
| 953 | ret = err; |
| 954 | } |
| 955 | |
| 956 | return ret; |
| 957 | } |
| 958 | |
Mike Anderson | 1134e5a | 2006-03-27 01:17:54 -0800 | [diff] [blame] | 959 | struct mapped_device *dm_table_get_md(struct dm_table *t) |
| 960 | { |
| 961 | dm_get(t->md); |
| 962 | |
| 963 | return t->md; |
| 964 | } |
| 965 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 966 | EXPORT_SYMBOL(dm_vcalloc); |
| 967 | EXPORT_SYMBOL(dm_get_device); |
| 968 | EXPORT_SYMBOL(dm_put_device); |
| 969 | EXPORT_SYMBOL(dm_table_event); |
Alasdair G Kergon | d5e404c | 2005-07-12 15:53:05 -0700 | [diff] [blame] | 970 | EXPORT_SYMBOL(dm_table_get_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 971 | EXPORT_SYMBOL(dm_table_get_mode); |
Mike Anderson | 1134e5a | 2006-03-27 01:17:54 -0800 | [diff] [blame] | 972 | EXPORT_SYMBOL(dm_table_get_md); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 973 | EXPORT_SYMBOL(dm_table_put); |
| 974 | EXPORT_SYMBOL(dm_table_get); |
| 975 | EXPORT_SYMBOL(dm_table_unplug_all); |
| 976 | EXPORT_SYMBOL(dm_table_flush_all); |