Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2001-2002 Sistina Software (UK) Limited. |
| 3 | * Copyright (C) 2006-2008 Red Hat GmbH |
| 4 | * |
| 5 | * This file is released under the GPL. |
| 6 | */ |
| 7 | |
| 8 | #include "dm-exception-store.h" |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 9 | |
| 10 | #include <linux/mm.h> |
| 11 | #include <linux/pagemap.h> |
| 12 | #include <linux/vmalloc.h> |
| 13 | #include <linux/slab.h> |
| 14 | #include <linux/dm-io.h> |
| 15 | |
| 16 | #define DM_MSG_PREFIX "persistent snapshot" |
| 17 | #define DM_CHUNK_SIZE_DEFAULT_SECTORS 32 /* 16KB */ |
| 18 | |
| 19 | /*----------------------------------------------------------------- |
| 20 | * Persistent snapshots, by persistent we mean that the snapshot |
| 21 | * will survive a reboot. |
| 22 | *---------------------------------------------------------------*/ |
| 23 | |
| 24 | /* |
| 25 | * We need to store a record of which parts of the origin have |
| 26 | * been copied to the snapshot device. The snapshot code |
| 27 | * requires that we copy exception chunks to chunk aligned areas |
| 28 | * of the COW store. It makes sense therefore, to store the |
| 29 | * metadata in chunk size blocks. |
| 30 | * |
| 31 | * There is no backward or forward compatibility implemented, |
| 32 | * snapshots with different disk versions than the kernel will |
| 33 | * not be usable. It is expected that "lvcreate" will blank out |
| 34 | * the start of a fresh COW device before calling the snapshot |
| 35 | * constructor. |
| 36 | * |
| 37 | * The first chunk of the COW device just contains the header. |
| 38 | * After this there is a chunk filled with exception metadata, |
| 39 | * followed by as many exception chunks as can fit in the |
| 40 | * metadata areas. |
| 41 | * |
| 42 | * All on disk structures are in little-endian format. The end |
| 43 | * of the exceptions info is indicated by an exception with a |
| 44 | * new_chunk of 0, which is invalid since it would point to the |
| 45 | * header chunk. |
| 46 | */ |
| 47 | |
| 48 | /* |
| 49 | * Magic for persistent snapshots: "SnAp" - Feeble isn't it. |
| 50 | */ |
| 51 | #define SNAP_MAGIC 0x70416e53 |
| 52 | |
| 53 | /* |
| 54 | * The on-disk version of the metadata. |
| 55 | */ |
| 56 | #define SNAPSHOT_DISK_VERSION 1 |
| 57 | |
| 58 | struct disk_header { |
| 59 | uint32_t magic; |
| 60 | |
| 61 | /* |
| 62 | * Is this snapshot valid. There is no way of recovering |
| 63 | * an invalid snapshot. |
| 64 | */ |
| 65 | uint32_t valid; |
| 66 | |
| 67 | /* |
| 68 | * Simple, incrementing version. no backward |
| 69 | * compatibility. |
| 70 | */ |
| 71 | uint32_t version; |
| 72 | |
| 73 | /* In sectors */ |
| 74 | uint32_t chunk_size; |
| 75 | }; |
| 76 | |
| 77 | struct disk_exception { |
| 78 | uint64_t old_chunk; |
| 79 | uint64_t new_chunk; |
| 80 | }; |
| 81 | |
| 82 | struct commit_callback { |
| 83 | void (*callback)(void *, int success); |
| 84 | void *context; |
| 85 | }; |
| 86 | |
| 87 | /* |
| 88 | * The top level structure for a persistent exception store. |
| 89 | */ |
| 90 | struct pstore { |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 91 | struct dm_exception_store *store; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 92 | int version; |
| 93 | int valid; |
| 94 | uint32_t exceptions_per_area; |
| 95 | |
| 96 | /* |
| 97 | * Now that we have an asynchronous kcopyd there is no |
| 98 | * need for large chunk sizes, so it wont hurt to have a |
| 99 | * whole chunks worth of metadata in memory at once. |
| 100 | */ |
| 101 | void *area; |
| 102 | |
| 103 | /* |
| 104 | * An area of zeros used to clear the next area. |
| 105 | */ |
| 106 | void *zero_area; |
| 107 | |
| 108 | /* |
| 109 | * Used to keep track of which metadata area the data in |
| 110 | * 'chunk' refers to. |
| 111 | */ |
| 112 | chunk_t current_area; |
| 113 | |
| 114 | /* |
| 115 | * The next free chunk for an exception. |
| 116 | */ |
| 117 | chunk_t next_free; |
| 118 | |
| 119 | /* |
| 120 | * The index of next free exception in the current |
| 121 | * metadata area. |
| 122 | */ |
| 123 | uint32_t current_committed; |
| 124 | |
| 125 | atomic_t pending_count; |
| 126 | uint32_t callback_count; |
| 127 | struct commit_callback *callbacks; |
| 128 | struct dm_io_client *io_client; |
| 129 | |
| 130 | struct workqueue_struct *metadata_wq; |
| 131 | }; |
| 132 | |
| 133 | static unsigned sectors_to_pages(unsigned sectors) |
| 134 | { |
| 135 | return DIV_ROUND_UP(sectors, PAGE_SIZE >> 9); |
| 136 | } |
| 137 | |
| 138 | static int alloc_area(struct pstore *ps) |
| 139 | { |
| 140 | int r = -ENOMEM; |
| 141 | size_t len; |
| 142 | |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 143 | len = ps->store->chunk_size << SECTOR_SHIFT; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 144 | |
| 145 | /* |
| 146 | * Allocate the chunk_size block of memory that will hold |
| 147 | * a single metadata area. |
| 148 | */ |
| 149 | ps->area = vmalloc(len); |
| 150 | if (!ps->area) |
| 151 | return r; |
| 152 | |
| 153 | ps->zero_area = vmalloc(len); |
| 154 | if (!ps->zero_area) { |
| 155 | vfree(ps->area); |
| 156 | return r; |
| 157 | } |
| 158 | memset(ps->zero_area, 0, len); |
| 159 | |
| 160 | return 0; |
| 161 | } |
| 162 | |
| 163 | static void free_area(struct pstore *ps) |
| 164 | { |
Jonathan Brassow | a32079c | 2009-04-02 19:55:35 +0100 | [diff] [blame] | 165 | if (ps->area) |
| 166 | vfree(ps->area); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 167 | ps->area = NULL; |
Jonathan Brassow | a32079c | 2009-04-02 19:55:35 +0100 | [diff] [blame] | 168 | |
| 169 | if (ps->zero_area) |
| 170 | vfree(ps->zero_area); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 171 | ps->zero_area = NULL; |
| 172 | } |
| 173 | |
| 174 | struct mdata_req { |
| 175 | struct dm_io_region *where; |
| 176 | struct dm_io_request *io_req; |
| 177 | struct work_struct work; |
| 178 | int result; |
| 179 | }; |
| 180 | |
| 181 | static void do_metadata(struct work_struct *work) |
| 182 | { |
| 183 | struct mdata_req *req = container_of(work, struct mdata_req, work); |
| 184 | |
| 185 | req->result = dm_io(req->io_req, 1, req->where, NULL); |
| 186 | } |
| 187 | |
| 188 | /* |
| 189 | * Read or write a chunk aligned and sized block of data from a device. |
| 190 | */ |
| 191 | static int chunk_io(struct pstore *ps, chunk_t chunk, int rw, int metadata) |
| 192 | { |
| 193 | struct dm_io_region where = { |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 194 | .bdev = ps->store->cow->bdev, |
| 195 | .sector = ps->store->chunk_size * chunk, |
| 196 | .count = ps->store->chunk_size, |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 197 | }; |
| 198 | struct dm_io_request io_req = { |
| 199 | .bi_rw = rw, |
| 200 | .mem.type = DM_IO_VMA, |
| 201 | .mem.ptr.vma = ps->area, |
| 202 | .client = ps->io_client, |
| 203 | .notify.fn = NULL, |
| 204 | }; |
| 205 | struct mdata_req req; |
| 206 | |
| 207 | if (!metadata) |
| 208 | return dm_io(&io_req, 1, &where, NULL); |
| 209 | |
| 210 | req.where = &where; |
| 211 | req.io_req = &io_req; |
| 212 | |
| 213 | /* |
| 214 | * Issue the synchronous I/O from a different thread |
| 215 | * to avoid generic_make_request recursion. |
| 216 | */ |
| 217 | INIT_WORK(&req.work, do_metadata); |
| 218 | queue_work(ps->metadata_wq, &req.work); |
| 219 | flush_workqueue(ps->metadata_wq); |
| 220 | |
| 221 | return req.result; |
| 222 | } |
| 223 | |
| 224 | /* |
| 225 | * Convert a metadata area index to a chunk index. |
| 226 | */ |
| 227 | static chunk_t area_location(struct pstore *ps, chunk_t area) |
| 228 | { |
| 229 | return 1 + ((ps->exceptions_per_area + 1) * area); |
| 230 | } |
| 231 | |
| 232 | /* |
| 233 | * Read or write a metadata area. Remembering to skip the first |
| 234 | * chunk which holds the header. |
| 235 | */ |
| 236 | static int area_io(struct pstore *ps, int rw) |
| 237 | { |
| 238 | int r; |
| 239 | chunk_t chunk; |
| 240 | |
| 241 | chunk = area_location(ps, ps->current_area); |
| 242 | |
| 243 | r = chunk_io(ps, chunk, rw, 0); |
| 244 | if (r) |
| 245 | return r; |
| 246 | |
| 247 | return 0; |
| 248 | } |
| 249 | |
| 250 | static void zero_memory_area(struct pstore *ps) |
| 251 | { |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 252 | memset(ps->area, 0, ps->store->chunk_size << SECTOR_SHIFT); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 253 | } |
| 254 | |
| 255 | static int zero_disk_area(struct pstore *ps, chunk_t area) |
| 256 | { |
| 257 | struct dm_io_region where = { |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 258 | .bdev = ps->store->cow->bdev, |
| 259 | .sector = ps->store->chunk_size * area_location(ps, area), |
| 260 | .count = ps->store->chunk_size, |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 261 | }; |
| 262 | struct dm_io_request io_req = { |
| 263 | .bi_rw = WRITE, |
| 264 | .mem.type = DM_IO_VMA, |
| 265 | .mem.ptr.vma = ps->zero_area, |
| 266 | .client = ps->io_client, |
| 267 | .notify.fn = NULL, |
| 268 | }; |
| 269 | |
| 270 | return dm_io(&io_req, 1, &where, NULL); |
| 271 | } |
| 272 | |
| 273 | static int read_header(struct pstore *ps, int *new_snapshot) |
| 274 | { |
| 275 | int r; |
| 276 | struct disk_header *dh; |
| 277 | chunk_t chunk_size; |
| 278 | int chunk_size_supplied = 1; |
| 279 | |
| 280 | /* |
| 281 | * Use default chunk size (or hardsect_size, if larger) if none supplied |
| 282 | */ |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 283 | if (!ps->store->chunk_size) { |
| 284 | ps->store->chunk_size = max(DM_CHUNK_SIZE_DEFAULT_SECTORS, |
| 285 | bdev_hardsect_size(ps->store->cow->bdev) >> 9); |
| 286 | ps->store->chunk_mask = ps->store->chunk_size - 1; |
| 287 | ps->store->chunk_shift = ffs(ps->store->chunk_size) - 1; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 288 | chunk_size_supplied = 0; |
| 289 | } |
| 290 | |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 291 | ps->io_client = dm_io_client_create(sectors_to_pages(ps->store-> |
| 292 | chunk_size)); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 293 | if (IS_ERR(ps->io_client)) |
| 294 | return PTR_ERR(ps->io_client); |
| 295 | |
| 296 | r = alloc_area(ps); |
| 297 | if (r) |
| 298 | return r; |
| 299 | |
| 300 | r = chunk_io(ps, 0, READ, 1); |
| 301 | if (r) |
| 302 | goto bad; |
| 303 | |
| 304 | dh = (struct disk_header *) ps->area; |
| 305 | |
| 306 | if (le32_to_cpu(dh->magic) == 0) { |
| 307 | *new_snapshot = 1; |
| 308 | return 0; |
| 309 | } |
| 310 | |
| 311 | if (le32_to_cpu(dh->magic) != SNAP_MAGIC) { |
| 312 | DMWARN("Invalid or corrupt snapshot"); |
| 313 | r = -ENXIO; |
| 314 | goto bad; |
| 315 | } |
| 316 | |
| 317 | *new_snapshot = 0; |
| 318 | ps->valid = le32_to_cpu(dh->valid); |
| 319 | ps->version = le32_to_cpu(dh->version); |
| 320 | chunk_size = le32_to_cpu(dh->chunk_size); |
| 321 | |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 322 | if (!chunk_size_supplied || ps->store->chunk_size == chunk_size) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 323 | return 0; |
| 324 | |
| 325 | DMWARN("chunk size %llu in device metadata overrides " |
| 326 | "table chunk size of %llu.", |
| 327 | (unsigned long long)chunk_size, |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 328 | (unsigned long long)ps->store->chunk_size); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 329 | |
| 330 | /* We had a bogus chunk_size. Fix stuff up. */ |
| 331 | free_area(ps); |
| 332 | |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 333 | ps->store->chunk_size = chunk_size; |
| 334 | ps->store->chunk_mask = chunk_size - 1; |
| 335 | ps->store->chunk_shift = ffs(chunk_size) - 1; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 336 | |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 337 | r = dm_io_client_resize(sectors_to_pages(ps->store->chunk_size), |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 338 | ps->io_client); |
| 339 | if (r) |
| 340 | return r; |
| 341 | |
| 342 | r = alloc_area(ps); |
| 343 | return r; |
| 344 | |
| 345 | bad: |
| 346 | free_area(ps); |
| 347 | return r; |
| 348 | } |
| 349 | |
| 350 | static int write_header(struct pstore *ps) |
| 351 | { |
| 352 | struct disk_header *dh; |
| 353 | |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 354 | memset(ps->area, 0, ps->store->chunk_size << SECTOR_SHIFT); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 355 | |
| 356 | dh = (struct disk_header *) ps->area; |
| 357 | dh->magic = cpu_to_le32(SNAP_MAGIC); |
| 358 | dh->valid = cpu_to_le32(ps->valid); |
| 359 | dh->version = cpu_to_le32(ps->version); |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 360 | dh->chunk_size = cpu_to_le32(ps->store->chunk_size); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 361 | |
| 362 | return chunk_io(ps, 0, WRITE, 1); |
| 363 | } |
| 364 | |
| 365 | /* |
| 366 | * Access functions for the disk exceptions, these do the endian conversions. |
| 367 | */ |
| 368 | static struct disk_exception *get_exception(struct pstore *ps, uint32_t index) |
| 369 | { |
| 370 | BUG_ON(index >= ps->exceptions_per_area); |
| 371 | |
| 372 | return ((struct disk_exception *) ps->area) + index; |
| 373 | } |
| 374 | |
| 375 | static void read_exception(struct pstore *ps, |
| 376 | uint32_t index, struct disk_exception *result) |
| 377 | { |
| 378 | struct disk_exception *e = get_exception(ps, index); |
| 379 | |
| 380 | /* copy it */ |
| 381 | result->old_chunk = le64_to_cpu(e->old_chunk); |
| 382 | result->new_chunk = le64_to_cpu(e->new_chunk); |
| 383 | } |
| 384 | |
| 385 | static void write_exception(struct pstore *ps, |
| 386 | uint32_t index, struct disk_exception *de) |
| 387 | { |
| 388 | struct disk_exception *e = get_exception(ps, index); |
| 389 | |
| 390 | /* copy it */ |
| 391 | e->old_chunk = cpu_to_le64(de->old_chunk); |
| 392 | e->new_chunk = cpu_to_le64(de->new_chunk); |
| 393 | } |
| 394 | |
| 395 | /* |
| 396 | * Registers the exceptions that are present in the current area. |
| 397 | * 'full' is filled in to indicate if the area has been |
| 398 | * filled. |
| 399 | */ |
Jonathan Brassow | a159c1a | 2009-01-06 03:05:19 +0000 | [diff] [blame] | 400 | static int insert_exceptions(struct pstore *ps, |
| 401 | int (*callback)(void *callback_context, |
| 402 | chunk_t old, chunk_t new), |
| 403 | void *callback_context, |
| 404 | int *full) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 405 | { |
| 406 | int r; |
| 407 | unsigned int i; |
| 408 | struct disk_exception de; |
| 409 | |
| 410 | /* presume the area is full */ |
| 411 | *full = 1; |
| 412 | |
| 413 | for (i = 0; i < ps->exceptions_per_area; i++) { |
| 414 | read_exception(ps, i, &de); |
| 415 | |
| 416 | /* |
| 417 | * If the new_chunk is pointing at the start of |
| 418 | * the COW device, where the first metadata area |
| 419 | * is we know that we've hit the end of the |
| 420 | * exceptions. Therefore the area is not full. |
| 421 | */ |
| 422 | if (de.new_chunk == 0LL) { |
| 423 | ps->current_committed = i; |
| 424 | *full = 0; |
| 425 | break; |
| 426 | } |
| 427 | |
| 428 | /* |
| 429 | * Keep track of the start of the free chunks. |
| 430 | */ |
| 431 | if (ps->next_free <= de.new_chunk) |
| 432 | ps->next_free = de.new_chunk + 1; |
| 433 | |
| 434 | /* |
| 435 | * Otherwise we add the exception to the snapshot. |
| 436 | */ |
Jonathan Brassow | a159c1a | 2009-01-06 03:05:19 +0000 | [diff] [blame] | 437 | r = callback(callback_context, de.old_chunk, de.new_chunk); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 438 | if (r) |
| 439 | return r; |
| 440 | } |
| 441 | |
| 442 | return 0; |
| 443 | } |
| 444 | |
Jonathan Brassow | a159c1a | 2009-01-06 03:05:19 +0000 | [diff] [blame] | 445 | static int read_exceptions(struct pstore *ps, |
| 446 | int (*callback)(void *callback_context, chunk_t old, |
| 447 | chunk_t new), |
| 448 | void *callback_context) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 449 | { |
| 450 | int r, full = 1; |
| 451 | |
| 452 | /* |
| 453 | * Keeping reading chunks and inserting exceptions until |
| 454 | * we find a partially full area. |
| 455 | */ |
| 456 | for (ps->current_area = 0; full; ps->current_area++) { |
| 457 | r = area_io(ps, READ); |
| 458 | if (r) |
| 459 | return r; |
| 460 | |
Jonathan Brassow | a159c1a | 2009-01-06 03:05:19 +0000 | [diff] [blame] | 461 | r = insert_exceptions(ps, callback, callback_context, &full); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 462 | if (r) |
| 463 | return r; |
| 464 | } |
| 465 | |
| 466 | ps->current_area--; |
| 467 | |
| 468 | return 0; |
| 469 | } |
| 470 | |
| 471 | static struct pstore *get_info(struct dm_exception_store *store) |
| 472 | { |
| 473 | return (struct pstore *) store->context; |
| 474 | } |
| 475 | |
| 476 | static void persistent_fraction_full(struct dm_exception_store *store, |
| 477 | sector_t *numerator, sector_t *denominator) |
| 478 | { |
Jonathan Brassow | d021684 | 2009-04-02 19:55:32 +0100 | [diff] [blame] | 479 | *numerator = get_info(store)->next_free * store->chunk_size; |
Jonathan Brassow | 49beb2b | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 480 | *denominator = get_dev_size(store->cow->bdev); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 481 | } |
| 482 | |
Jonathan Brassow | 493df71 | 2009-04-02 19:55:31 +0100 | [diff] [blame] | 483 | static void persistent_dtr(struct dm_exception_store *store) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 484 | { |
| 485 | struct pstore *ps = get_info(store); |
| 486 | |
| 487 | destroy_workqueue(ps->metadata_wq); |
Jonathan Brassow | a32079c | 2009-04-02 19:55:35 +0100 | [diff] [blame] | 488 | |
| 489 | /* Created in read_header */ |
| 490 | if (ps->io_client) |
| 491 | dm_io_client_destroy(ps->io_client); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 492 | free_area(ps); |
Jonathan Brassow | a32079c | 2009-04-02 19:55:35 +0100 | [diff] [blame] | 493 | |
| 494 | /* Allocated in persistent_read_metadata */ |
| 495 | if (ps->callbacks) |
| 496 | vfree(ps->callbacks); |
| 497 | |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 498 | kfree(ps); |
| 499 | } |
| 500 | |
Jonathan Brassow | a159c1a | 2009-01-06 03:05:19 +0000 | [diff] [blame] | 501 | static int persistent_read_metadata(struct dm_exception_store *store, |
| 502 | int (*callback)(void *callback_context, |
| 503 | chunk_t old, chunk_t new), |
| 504 | void *callback_context) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 505 | { |
| 506 | int r, uninitialized_var(new_snapshot); |
| 507 | struct pstore *ps = get_info(store); |
| 508 | |
| 509 | /* |
| 510 | * Read the snapshot header. |
| 511 | */ |
| 512 | r = read_header(ps, &new_snapshot); |
| 513 | if (r) |
| 514 | return r; |
| 515 | |
| 516 | /* |
| 517 | * Now we know correct chunk_size, complete the initialisation. |
| 518 | */ |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 519 | ps->exceptions_per_area = (ps->store->chunk_size << SECTOR_SHIFT) / |
| 520 | sizeof(struct disk_exception); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 521 | ps->callbacks = dm_vcalloc(ps->exceptions_per_area, |
| 522 | sizeof(*ps->callbacks)); |
| 523 | if (!ps->callbacks) |
| 524 | return -ENOMEM; |
| 525 | |
| 526 | /* |
| 527 | * Do we need to setup a new snapshot ? |
| 528 | */ |
| 529 | if (new_snapshot) { |
| 530 | r = write_header(ps); |
| 531 | if (r) { |
| 532 | DMWARN("write_header failed"); |
| 533 | return r; |
| 534 | } |
| 535 | |
| 536 | ps->current_area = 0; |
| 537 | zero_memory_area(ps); |
| 538 | r = zero_disk_area(ps, 0); |
| 539 | if (r) { |
| 540 | DMWARN("zero_disk_area(0) failed"); |
| 541 | return r; |
| 542 | } |
| 543 | } else { |
| 544 | /* |
| 545 | * Sanity checks. |
| 546 | */ |
| 547 | if (ps->version != SNAPSHOT_DISK_VERSION) { |
| 548 | DMWARN("unable to handle snapshot disk version %d", |
| 549 | ps->version); |
| 550 | return -EINVAL; |
| 551 | } |
| 552 | |
| 553 | /* |
| 554 | * Metadata are valid, but snapshot is invalidated |
| 555 | */ |
| 556 | if (!ps->valid) |
| 557 | return 1; |
| 558 | |
| 559 | /* |
| 560 | * Read the metadata. |
| 561 | */ |
Jonathan Brassow | a159c1a | 2009-01-06 03:05:19 +0000 | [diff] [blame] | 562 | r = read_exceptions(ps, callback, callback_context); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 563 | if (r) |
| 564 | return r; |
| 565 | } |
| 566 | |
| 567 | return 0; |
| 568 | } |
| 569 | |
Jonathan Brassow | a159c1a | 2009-01-06 03:05:19 +0000 | [diff] [blame] | 570 | static int persistent_prepare_exception(struct dm_exception_store *store, |
| 571 | struct dm_snap_exception *e) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 572 | { |
| 573 | struct pstore *ps = get_info(store); |
| 574 | uint32_t stride; |
| 575 | chunk_t next_free; |
Jonathan Brassow | 49beb2b | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 576 | sector_t size = get_dev_size(store->cow->bdev); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 577 | |
| 578 | /* Is there enough room ? */ |
Jonathan Brassow | d021684 | 2009-04-02 19:55:32 +0100 | [diff] [blame] | 579 | if (size < ((ps->next_free + 1) * store->chunk_size)) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 580 | return -ENOSPC; |
| 581 | |
| 582 | e->new_chunk = ps->next_free; |
| 583 | |
| 584 | /* |
| 585 | * Move onto the next free pending, making sure to take |
| 586 | * into account the location of the metadata chunks. |
| 587 | */ |
| 588 | stride = (ps->exceptions_per_area + 1); |
| 589 | next_free = ++ps->next_free; |
| 590 | if (sector_div(next_free, stride) == 1) |
| 591 | ps->next_free++; |
| 592 | |
| 593 | atomic_inc(&ps->pending_count); |
| 594 | return 0; |
| 595 | } |
| 596 | |
Jonathan Brassow | a159c1a | 2009-01-06 03:05:19 +0000 | [diff] [blame] | 597 | static void persistent_commit_exception(struct dm_exception_store *store, |
| 598 | struct dm_snap_exception *e, |
| 599 | void (*callback) (void *, int success), |
| 600 | void *callback_context) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 601 | { |
| 602 | unsigned int i; |
| 603 | struct pstore *ps = get_info(store); |
| 604 | struct disk_exception de; |
| 605 | struct commit_callback *cb; |
| 606 | |
| 607 | de.old_chunk = e->old_chunk; |
| 608 | de.new_chunk = e->new_chunk; |
| 609 | write_exception(ps, ps->current_committed++, &de); |
| 610 | |
| 611 | /* |
| 612 | * Add the callback to the back of the array. This code |
| 613 | * is the only place where the callback array is |
| 614 | * manipulated, and we know that it will never be called |
| 615 | * multiple times concurrently. |
| 616 | */ |
| 617 | cb = ps->callbacks + ps->callback_count++; |
| 618 | cb->callback = callback; |
| 619 | cb->context = callback_context; |
| 620 | |
| 621 | /* |
| 622 | * If there are exceptions in flight and we have not yet |
| 623 | * filled this metadata area there's nothing more to do. |
| 624 | */ |
| 625 | if (!atomic_dec_and_test(&ps->pending_count) && |
| 626 | (ps->current_committed != ps->exceptions_per_area)) |
| 627 | return; |
| 628 | |
| 629 | /* |
| 630 | * If we completely filled the current area, then wipe the next one. |
| 631 | */ |
| 632 | if ((ps->current_committed == ps->exceptions_per_area) && |
| 633 | zero_disk_area(ps, ps->current_area + 1)) |
| 634 | ps->valid = 0; |
| 635 | |
| 636 | /* |
| 637 | * Commit exceptions to disk. |
| 638 | */ |
| 639 | if (ps->valid && area_io(ps, WRITE)) |
| 640 | ps->valid = 0; |
| 641 | |
| 642 | /* |
| 643 | * Advance to the next area if this one is full. |
| 644 | */ |
| 645 | if (ps->current_committed == ps->exceptions_per_area) { |
| 646 | ps->current_committed = 0; |
| 647 | ps->current_area++; |
| 648 | zero_memory_area(ps); |
| 649 | } |
| 650 | |
| 651 | for (i = 0; i < ps->callback_count; i++) { |
| 652 | cb = ps->callbacks + i; |
| 653 | cb->callback(cb->context, ps->valid); |
| 654 | } |
| 655 | |
| 656 | ps->callback_count = 0; |
| 657 | } |
| 658 | |
Jonathan Brassow | a159c1a | 2009-01-06 03:05:19 +0000 | [diff] [blame] | 659 | static void persistent_drop_snapshot(struct dm_exception_store *store) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 660 | { |
| 661 | struct pstore *ps = get_info(store); |
| 662 | |
| 663 | ps->valid = 0; |
| 664 | if (write_header(ps)) |
| 665 | DMWARN("write header failed"); |
| 666 | } |
| 667 | |
Jonathan Brassow | 493df71 | 2009-04-02 19:55:31 +0100 | [diff] [blame] | 668 | static int persistent_ctr(struct dm_exception_store *store, |
| 669 | unsigned argc, char **argv) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 670 | { |
| 671 | struct pstore *ps; |
| 672 | |
| 673 | /* allocate the pstore */ |
Jonathan Brassow | a32079c | 2009-04-02 19:55:35 +0100 | [diff] [blame] | 674 | ps = kzalloc(sizeof(*ps), GFP_KERNEL); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 675 | if (!ps) |
| 676 | return -ENOMEM; |
| 677 | |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 678 | ps->store = store; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 679 | ps->valid = 1; |
| 680 | ps->version = SNAPSHOT_DISK_VERSION; |
| 681 | ps->area = NULL; |
| 682 | ps->next_free = 2; /* skipping the header and first area */ |
| 683 | ps->current_committed = 0; |
| 684 | |
| 685 | ps->callback_count = 0; |
| 686 | atomic_set(&ps->pending_count, 0); |
| 687 | ps->callbacks = NULL; |
| 688 | |
| 689 | ps->metadata_wq = create_singlethread_workqueue("ksnaphd"); |
| 690 | if (!ps->metadata_wq) { |
| 691 | kfree(ps); |
| 692 | DMERR("couldn't start header metadata update thread"); |
| 693 | return -ENOMEM; |
| 694 | } |
| 695 | |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 696 | store->context = ps; |
| 697 | |
| 698 | return 0; |
| 699 | } |
| 700 | |
Jonathan Brassow | 1e302a9 | 2009-04-02 19:55:35 +0100 | [diff] [blame] | 701 | static unsigned persistent_status(struct dm_exception_store *store, |
| 702 | status_type_t status, char *result, |
| 703 | unsigned maxlen) |
Jonathan Brassow | 493df71 | 2009-04-02 19:55:31 +0100 | [diff] [blame] | 704 | { |
Jonathan Brassow | 1e302a9 | 2009-04-02 19:55:35 +0100 | [diff] [blame] | 705 | unsigned sz = 0; |
| 706 | |
| 707 | switch (status) { |
| 708 | case STATUSTYPE_INFO: |
| 709 | break; |
| 710 | case STATUSTYPE_TABLE: |
| 711 | DMEMIT(" %s P %llu", store->cow->name, |
| 712 | (unsigned long long)store->chunk_size); |
| 713 | } |
Jonathan Brassow | 493df71 | 2009-04-02 19:55:31 +0100 | [diff] [blame] | 714 | |
| 715 | return sz; |
| 716 | } |
| 717 | |
| 718 | static struct dm_exception_store_type _persistent_type = { |
| 719 | .name = "persistent", |
| 720 | .module = THIS_MODULE, |
| 721 | .ctr = persistent_ctr, |
| 722 | .dtr = persistent_dtr, |
| 723 | .read_metadata = persistent_read_metadata, |
| 724 | .prepare_exception = persistent_prepare_exception, |
| 725 | .commit_exception = persistent_commit_exception, |
| 726 | .drop_snapshot = persistent_drop_snapshot, |
| 727 | .fraction_full = persistent_fraction_full, |
| 728 | .status = persistent_status, |
| 729 | }; |
| 730 | |
| 731 | static struct dm_exception_store_type _persistent_compat_type = { |
| 732 | .name = "P", |
| 733 | .module = THIS_MODULE, |
| 734 | .ctr = persistent_ctr, |
| 735 | .dtr = persistent_dtr, |
| 736 | .read_metadata = persistent_read_metadata, |
| 737 | .prepare_exception = persistent_prepare_exception, |
| 738 | .commit_exception = persistent_commit_exception, |
| 739 | .drop_snapshot = persistent_drop_snapshot, |
| 740 | .fraction_full = persistent_fraction_full, |
| 741 | .status = persistent_status, |
| 742 | }; |
| 743 | |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 744 | int dm_persistent_snapshot_init(void) |
| 745 | { |
Jonathan Brassow | 493df71 | 2009-04-02 19:55:31 +0100 | [diff] [blame] | 746 | int r; |
| 747 | |
| 748 | r = dm_exception_store_type_register(&_persistent_type); |
| 749 | if (r) { |
| 750 | DMERR("Unable to register persistent exception store type"); |
| 751 | return r; |
| 752 | } |
| 753 | |
| 754 | r = dm_exception_store_type_register(&_persistent_compat_type); |
| 755 | if (r) { |
| 756 | DMERR("Unable to register old-style persistent exception " |
| 757 | "store type"); |
| 758 | dm_exception_store_type_unregister(&_persistent_type); |
| 759 | return r; |
| 760 | } |
| 761 | |
| 762 | return r; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 763 | } |
| 764 | |
| 765 | void dm_persistent_snapshot_exit(void) |
| 766 | { |
Jonathan Brassow | 493df71 | 2009-04-02 19:55:31 +0100 | [diff] [blame] | 767 | dm_exception_store_type_unregister(&_persistent_type); |
| 768 | dm_exception_store_type_unregister(&_persistent_compat_type); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 769 | } |