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> |
Paul Gortmaker | daaa5f7 | 2011-05-27 15:50:58 -0400 | [diff] [blame] | 13 | #include <linux/export.h> |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 14 | #include <linux/slab.h> |
| 15 | #include <linux/dm-io.h> |
| 16 | |
| 17 | #define DM_MSG_PREFIX "persistent snapshot" |
| 18 | #define DM_CHUNK_SIZE_DEFAULT_SECTORS 32 /* 16KB */ |
| 19 | |
| 20 | /*----------------------------------------------------------------- |
| 21 | * Persistent snapshots, by persistent we mean that the snapshot |
| 22 | * will survive a reboot. |
| 23 | *---------------------------------------------------------------*/ |
| 24 | |
| 25 | /* |
| 26 | * We need to store a record of which parts of the origin have |
| 27 | * been copied to the snapshot device. The snapshot code |
| 28 | * requires that we copy exception chunks to chunk aligned areas |
| 29 | * of the COW store. It makes sense therefore, to store the |
| 30 | * metadata in chunk size blocks. |
| 31 | * |
| 32 | * There is no backward or forward compatibility implemented, |
| 33 | * snapshots with different disk versions than the kernel will |
| 34 | * not be usable. It is expected that "lvcreate" will blank out |
| 35 | * the start of a fresh COW device before calling the snapshot |
| 36 | * constructor. |
| 37 | * |
| 38 | * The first chunk of the COW device just contains the header. |
| 39 | * After this there is a chunk filled with exception metadata, |
| 40 | * followed by as many exception chunks as can fit in the |
| 41 | * metadata areas. |
| 42 | * |
| 43 | * All on disk structures are in little-endian format. The end |
| 44 | * of the exceptions info is indicated by an exception with a |
| 45 | * new_chunk of 0, which is invalid since it would point to the |
| 46 | * header chunk. |
| 47 | */ |
| 48 | |
| 49 | /* |
| 50 | * Magic for persistent snapshots: "SnAp" - Feeble isn't it. |
| 51 | */ |
| 52 | #define SNAP_MAGIC 0x70416e53 |
| 53 | |
| 54 | /* |
| 55 | * The on-disk version of the metadata. |
| 56 | */ |
| 57 | #define SNAPSHOT_DISK_VERSION 1 |
| 58 | |
Mikulas Patocka | 4454a62 | 2009-12-10 23:52:29 +0000 | [diff] [blame] | 59 | #define NUM_SNAPSHOT_HDR_CHUNKS 1 |
| 60 | |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 61 | struct disk_header { |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 62 | __le32 magic; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 63 | |
| 64 | /* |
| 65 | * Is this snapshot valid. There is no way of recovering |
| 66 | * an invalid snapshot. |
| 67 | */ |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 68 | __le32 valid; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 69 | |
| 70 | /* |
| 71 | * Simple, incrementing version. no backward |
| 72 | * compatibility. |
| 73 | */ |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 74 | __le32 version; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 75 | |
| 76 | /* In sectors */ |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 77 | __le32 chunk_size; |
| 78 | } __packed; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 79 | |
| 80 | struct disk_exception { |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 81 | __le64 old_chunk; |
| 82 | __le64 new_chunk; |
| 83 | } __packed; |
| 84 | |
| 85 | struct core_exception { |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 86 | uint64_t old_chunk; |
| 87 | uint64_t new_chunk; |
| 88 | }; |
| 89 | |
| 90 | struct commit_callback { |
| 91 | void (*callback)(void *, int success); |
| 92 | void *context; |
| 93 | }; |
| 94 | |
| 95 | /* |
| 96 | * The top level structure for a persistent exception store. |
| 97 | */ |
| 98 | struct pstore { |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 99 | struct dm_exception_store *store; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 100 | int version; |
| 101 | int valid; |
| 102 | uint32_t exceptions_per_area; |
| 103 | |
| 104 | /* |
| 105 | * Now that we have an asynchronous kcopyd there is no |
| 106 | * need for large chunk sizes, so it wont hurt to have a |
| 107 | * whole chunks worth of metadata in memory at once. |
| 108 | */ |
| 109 | void *area; |
| 110 | |
| 111 | /* |
| 112 | * An area of zeros used to clear the next area. |
| 113 | */ |
| 114 | void *zero_area; |
| 115 | |
| 116 | /* |
Mikulas Patocka | 61578dc | 2009-09-04 20:40:39 +0100 | [diff] [blame] | 117 | * An area used for header. The header can be written |
| 118 | * concurrently with metadata (when invalidating the snapshot), |
| 119 | * so it needs a separate buffer. |
| 120 | */ |
| 121 | void *header_area; |
| 122 | |
| 123 | /* |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 124 | * Used to keep track of which metadata area the data in |
| 125 | * 'chunk' refers to. |
| 126 | */ |
| 127 | chunk_t current_area; |
| 128 | |
| 129 | /* |
| 130 | * The next free chunk for an exception. |
Mikulas Patocka | 4454a62 | 2009-12-10 23:52:29 +0000 | [diff] [blame] | 131 | * |
| 132 | * When creating exceptions, all the chunks here and above are |
| 133 | * free. It holds the next chunk to be allocated. On rare |
| 134 | * occasions (e.g. after a system crash) holes can be left in |
| 135 | * the exception store because chunks can be committed out of |
| 136 | * order. |
| 137 | * |
| 138 | * When merging exceptions, it does not necessarily mean all the |
| 139 | * chunks here and above are free. It holds the value it would |
| 140 | * have held if all chunks had been committed in order of |
| 141 | * allocation. Consequently the value may occasionally be |
| 142 | * slightly too low, but since it's only used for 'status' and |
| 143 | * it can never reach its minimum value too early this doesn't |
| 144 | * matter. |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 145 | */ |
Mikulas Patocka | 4454a62 | 2009-12-10 23:52:29 +0000 | [diff] [blame] | 146 | |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 147 | chunk_t next_free; |
| 148 | |
| 149 | /* |
| 150 | * The index of next free exception in the current |
| 151 | * metadata area. |
| 152 | */ |
| 153 | uint32_t current_committed; |
| 154 | |
| 155 | atomic_t pending_count; |
| 156 | uint32_t callback_count; |
| 157 | struct commit_callback *callbacks; |
| 158 | struct dm_io_client *io_client; |
| 159 | |
| 160 | struct workqueue_struct *metadata_wq; |
| 161 | }; |
| 162 | |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 163 | static int alloc_area(struct pstore *ps) |
| 164 | { |
| 165 | int r = -ENOMEM; |
| 166 | size_t len; |
| 167 | |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 168 | len = ps->store->chunk_size << SECTOR_SHIFT; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 169 | |
| 170 | /* |
| 171 | * Allocate the chunk_size block of memory that will hold |
| 172 | * a single metadata area. |
| 173 | */ |
| 174 | ps->area = vmalloc(len); |
| 175 | if (!ps->area) |
Mikulas Patocka | 61578dc | 2009-09-04 20:40:39 +0100 | [diff] [blame] | 176 | goto err_area; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 177 | |
Joe Perches | e29e65a | 2011-08-02 12:32:02 +0100 | [diff] [blame] | 178 | ps->zero_area = vzalloc(len); |
Mikulas Patocka | 61578dc | 2009-09-04 20:40:39 +0100 | [diff] [blame] | 179 | if (!ps->zero_area) |
| 180 | goto err_zero_area; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 181 | |
Mikulas Patocka | 61578dc | 2009-09-04 20:40:39 +0100 | [diff] [blame] | 182 | ps->header_area = vmalloc(len); |
| 183 | if (!ps->header_area) |
| 184 | goto err_header_area; |
| 185 | |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 186 | return 0; |
Mikulas Patocka | 61578dc | 2009-09-04 20:40:39 +0100 | [diff] [blame] | 187 | |
| 188 | err_header_area: |
| 189 | vfree(ps->zero_area); |
| 190 | |
| 191 | err_zero_area: |
| 192 | vfree(ps->area); |
| 193 | |
| 194 | err_area: |
| 195 | return r; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 196 | } |
| 197 | |
| 198 | static void free_area(struct pstore *ps) |
| 199 | { |
Jonathan Brassow | a32079c | 2009-04-02 19:55:35 +0100 | [diff] [blame] | 200 | if (ps->area) |
| 201 | vfree(ps->area); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 202 | ps->area = NULL; |
Jonathan Brassow | a32079c | 2009-04-02 19:55:35 +0100 | [diff] [blame] | 203 | |
| 204 | if (ps->zero_area) |
| 205 | vfree(ps->zero_area); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 206 | ps->zero_area = NULL; |
Mikulas Patocka | 61578dc | 2009-09-04 20:40:39 +0100 | [diff] [blame] | 207 | |
| 208 | if (ps->header_area) |
| 209 | vfree(ps->header_area); |
| 210 | ps->header_area = NULL; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 211 | } |
| 212 | |
| 213 | struct mdata_req { |
| 214 | struct dm_io_region *where; |
| 215 | struct dm_io_request *io_req; |
| 216 | struct work_struct work; |
| 217 | int result; |
| 218 | }; |
| 219 | |
| 220 | static void do_metadata(struct work_struct *work) |
| 221 | { |
| 222 | struct mdata_req *req = container_of(work, struct mdata_req, work); |
| 223 | |
| 224 | req->result = dm_io(req->io_req, 1, req->where, NULL); |
| 225 | } |
| 226 | |
| 227 | /* |
| 228 | * Read or write a chunk aligned and sized block of data from a device. |
| 229 | */ |
Mikulas Patocka | 02d2fd3 | 2009-09-04 20:40:37 +0100 | [diff] [blame] | 230 | static int chunk_io(struct pstore *ps, void *area, chunk_t chunk, int rw, |
| 231 | int metadata) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 232 | { |
| 233 | struct dm_io_region where = { |
Mike Snitzer | fc56f6f | 2009-12-10 23:52:12 +0000 | [diff] [blame] | 234 | .bdev = dm_snap_cow(ps->store->snap)->bdev, |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 235 | .sector = ps->store->chunk_size * chunk, |
| 236 | .count = ps->store->chunk_size, |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 237 | }; |
| 238 | struct dm_io_request io_req = { |
| 239 | .bi_rw = rw, |
| 240 | .mem.type = DM_IO_VMA, |
Mikulas Patocka | 02d2fd3 | 2009-09-04 20:40:37 +0100 | [diff] [blame] | 241 | .mem.ptr.vma = area, |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 242 | .client = ps->io_client, |
| 243 | .notify.fn = NULL, |
| 244 | }; |
| 245 | struct mdata_req req; |
| 246 | |
| 247 | if (!metadata) |
| 248 | return dm_io(&io_req, 1, &where, NULL); |
| 249 | |
| 250 | req.where = &where; |
| 251 | req.io_req = &io_req; |
| 252 | |
| 253 | /* |
| 254 | * Issue the synchronous I/O from a different thread |
| 255 | * to avoid generic_make_request recursion. |
| 256 | */ |
Andrew Morton | ca1cab3 | 2010-10-26 14:22:34 -0700 | [diff] [blame] | 257 | INIT_WORK_ONSTACK(&req.work, do_metadata); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 258 | queue_work(ps->metadata_wq, &req.work); |
Mikulas Patocka | 82e4370 | 2013-09-18 19:14:22 -0400 | [diff] [blame] | 259 | flush_workqueue(ps->metadata_wq); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 260 | |
| 261 | return req.result; |
| 262 | } |
| 263 | |
| 264 | /* |
| 265 | * Convert a metadata area index to a chunk index. |
| 266 | */ |
| 267 | static chunk_t area_location(struct pstore *ps, chunk_t area) |
| 268 | { |
Tomohiro Kusumi | 87c961c | 2010-08-12 04:13:59 +0100 | [diff] [blame] | 269 | return NUM_SNAPSHOT_HDR_CHUNKS + ((ps->exceptions_per_area + 1) * area); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 270 | } |
| 271 | |
Mikulas Patocka | cf5d2d7 | 2013-10-16 03:17:47 +0100 | [diff] [blame] | 272 | static void skip_metadata(struct pstore *ps) |
| 273 | { |
| 274 | uint32_t stride = ps->exceptions_per_area + 1; |
| 275 | chunk_t next_free = ps->next_free; |
| 276 | if (sector_div(next_free, stride) == NUM_SNAPSHOT_HDR_CHUNKS) |
| 277 | ps->next_free++; |
| 278 | } |
| 279 | |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 280 | /* |
| 281 | * Read or write a metadata area. Remembering to skip the first |
| 282 | * chunk which holds the header. |
| 283 | */ |
| 284 | static int area_io(struct pstore *ps, int rw) |
| 285 | { |
| 286 | int r; |
| 287 | chunk_t chunk; |
| 288 | |
| 289 | chunk = area_location(ps, ps->current_area); |
| 290 | |
Mikulas Patocka | 02d2fd3 | 2009-09-04 20:40:37 +0100 | [diff] [blame] | 291 | r = chunk_io(ps, ps->area, chunk, rw, 0); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 292 | if (r) |
| 293 | return r; |
| 294 | |
| 295 | return 0; |
| 296 | } |
| 297 | |
| 298 | static void zero_memory_area(struct pstore *ps) |
| 299 | { |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 300 | memset(ps->area, 0, ps->store->chunk_size << SECTOR_SHIFT); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 301 | } |
| 302 | |
| 303 | static int zero_disk_area(struct pstore *ps, chunk_t area) |
| 304 | { |
Mikulas Patocka | 02d2fd3 | 2009-09-04 20:40:37 +0100 | [diff] [blame] | 305 | return chunk_io(ps, ps->zero_area, area_location(ps, area), WRITE, 0); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 306 | } |
| 307 | |
| 308 | static int read_header(struct pstore *ps, int *new_snapshot) |
| 309 | { |
| 310 | int r; |
| 311 | struct disk_header *dh; |
Mikulas Patocka | df96eee | 2009-10-16 23:18:17 +0100 | [diff] [blame] | 312 | unsigned chunk_size; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 313 | int chunk_size_supplied = 1; |
Mikulas Patocka | ae0b744 | 2009-09-04 20:40:43 +0100 | [diff] [blame] | 314 | char *chunk_err; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 315 | |
| 316 | /* |
Mikulas Patocka | df96eee | 2009-10-16 23:18:17 +0100 | [diff] [blame] | 317 | * Use default chunk size (or logical_block_size, if larger) |
| 318 | * if none supplied |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 319 | */ |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 320 | if (!ps->store->chunk_size) { |
| 321 | ps->store->chunk_size = max(DM_CHUNK_SIZE_DEFAULT_SECTORS, |
Mike Snitzer | fc56f6f | 2009-12-10 23:52:12 +0000 | [diff] [blame] | 322 | bdev_logical_block_size(dm_snap_cow(ps->store->snap)-> |
| 323 | bdev) >> 9); |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 324 | ps->store->chunk_mask = ps->store->chunk_size - 1; |
| 325 | ps->store->chunk_shift = ffs(ps->store->chunk_size) - 1; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 326 | chunk_size_supplied = 0; |
| 327 | } |
| 328 | |
Mikulas Patocka | bda8efe | 2011-05-29 13:03:09 +0100 | [diff] [blame] | 329 | ps->io_client = dm_io_client_create(); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 330 | if (IS_ERR(ps->io_client)) |
| 331 | return PTR_ERR(ps->io_client); |
| 332 | |
| 333 | r = alloc_area(ps); |
| 334 | if (r) |
| 335 | return r; |
| 336 | |
Mikulas Patocka | 61578dc | 2009-09-04 20:40:39 +0100 | [diff] [blame] | 337 | r = chunk_io(ps, ps->header_area, 0, READ, 1); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 338 | if (r) |
| 339 | goto bad; |
| 340 | |
Mikulas Patocka | 61578dc | 2009-09-04 20:40:39 +0100 | [diff] [blame] | 341 | dh = ps->header_area; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 342 | |
| 343 | if (le32_to_cpu(dh->magic) == 0) { |
| 344 | *new_snapshot = 1; |
| 345 | return 0; |
| 346 | } |
| 347 | |
| 348 | if (le32_to_cpu(dh->magic) != SNAP_MAGIC) { |
| 349 | DMWARN("Invalid or corrupt snapshot"); |
| 350 | r = -ENXIO; |
| 351 | goto bad; |
| 352 | } |
| 353 | |
| 354 | *new_snapshot = 0; |
| 355 | ps->valid = le32_to_cpu(dh->valid); |
| 356 | ps->version = le32_to_cpu(dh->version); |
| 357 | chunk_size = le32_to_cpu(dh->chunk_size); |
| 358 | |
Mikulas Patocka | ae0b744 | 2009-09-04 20:40:43 +0100 | [diff] [blame] | 359 | if (ps->store->chunk_size == chunk_size) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 360 | return 0; |
| 361 | |
Mikulas Patocka | ae0b744 | 2009-09-04 20:40:43 +0100 | [diff] [blame] | 362 | if (chunk_size_supplied) |
Mikulas Patocka | df96eee | 2009-10-16 23:18:17 +0100 | [diff] [blame] | 363 | DMWARN("chunk size %u in device metadata overrides " |
| 364 | "table chunk size of %u.", |
| 365 | chunk_size, ps->store->chunk_size); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 366 | |
| 367 | /* We had a bogus chunk_size. Fix stuff up. */ |
| 368 | free_area(ps); |
| 369 | |
Mikulas Patocka | ae0b744 | 2009-09-04 20:40:43 +0100 | [diff] [blame] | 370 | r = dm_exception_store_set_chunk_size(ps->store, chunk_size, |
| 371 | &chunk_err); |
| 372 | if (r) { |
Mikulas Patocka | df96eee | 2009-10-16 23:18:17 +0100 | [diff] [blame] | 373 | DMERR("invalid on-disk chunk size %u: %s.", |
| 374 | chunk_size, chunk_err); |
Mikulas Patocka | ae0b744 | 2009-09-04 20:40:43 +0100 | [diff] [blame] | 375 | return r; |
| 376 | } |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 377 | |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 378 | r = alloc_area(ps); |
| 379 | return r; |
| 380 | |
| 381 | bad: |
| 382 | free_area(ps); |
| 383 | return r; |
| 384 | } |
| 385 | |
| 386 | static int write_header(struct pstore *ps) |
| 387 | { |
| 388 | struct disk_header *dh; |
| 389 | |
Mikulas Patocka | 61578dc | 2009-09-04 20:40:39 +0100 | [diff] [blame] | 390 | memset(ps->header_area, 0, ps->store->chunk_size << SECTOR_SHIFT); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 391 | |
Mikulas Patocka | 61578dc | 2009-09-04 20:40:39 +0100 | [diff] [blame] | 392 | dh = ps->header_area; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 393 | dh->magic = cpu_to_le32(SNAP_MAGIC); |
| 394 | dh->valid = cpu_to_le32(ps->valid); |
| 395 | dh->version = cpu_to_le32(ps->version); |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 396 | dh->chunk_size = cpu_to_le32(ps->store->chunk_size); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 397 | |
Mikulas Patocka | 61578dc | 2009-09-04 20:40:39 +0100 | [diff] [blame] | 398 | return chunk_io(ps, ps->header_area, 0, WRITE, 1); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 399 | } |
| 400 | |
| 401 | /* |
| 402 | * Access functions for the disk exceptions, these do the endian conversions. |
| 403 | */ |
| 404 | static struct disk_exception *get_exception(struct pstore *ps, uint32_t index) |
| 405 | { |
| 406 | BUG_ON(index >= ps->exceptions_per_area); |
| 407 | |
| 408 | return ((struct disk_exception *) ps->area) + index; |
| 409 | } |
| 410 | |
| 411 | static void read_exception(struct pstore *ps, |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 412 | uint32_t index, struct core_exception *result) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 413 | { |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 414 | struct disk_exception *de = get_exception(ps, index); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 415 | |
| 416 | /* copy it */ |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 417 | result->old_chunk = le64_to_cpu(de->old_chunk); |
| 418 | result->new_chunk = le64_to_cpu(de->new_chunk); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 419 | } |
| 420 | |
| 421 | static void write_exception(struct pstore *ps, |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 422 | uint32_t index, struct core_exception *e) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 423 | { |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 424 | struct disk_exception *de = get_exception(ps, index); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 425 | |
| 426 | /* copy it */ |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 427 | de->old_chunk = cpu_to_le64(e->old_chunk); |
| 428 | de->new_chunk = cpu_to_le64(e->new_chunk); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 429 | } |
| 430 | |
Mikulas Patocka | 4454a62 | 2009-12-10 23:52:29 +0000 | [diff] [blame] | 431 | static void clear_exception(struct pstore *ps, uint32_t index) |
| 432 | { |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 433 | struct disk_exception *de = get_exception(ps, index); |
Mikulas Patocka | 4454a62 | 2009-12-10 23:52:29 +0000 | [diff] [blame] | 434 | |
| 435 | /* clear it */ |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 436 | de->old_chunk = 0; |
| 437 | de->new_chunk = 0; |
Mikulas Patocka | 4454a62 | 2009-12-10 23:52:29 +0000 | [diff] [blame] | 438 | } |
| 439 | |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 440 | /* |
| 441 | * Registers the exceptions that are present in the current area. |
| 442 | * 'full' is filled in to indicate if the area has been |
| 443 | * filled. |
| 444 | */ |
Jonathan Brassow | a159c1a | 2009-01-06 03:05:19 +0000 | [diff] [blame] | 445 | static int insert_exceptions(struct pstore *ps, |
| 446 | int (*callback)(void *callback_context, |
| 447 | chunk_t old, chunk_t new), |
| 448 | void *callback_context, |
| 449 | int *full) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 450 | { |
| 451 | int r; |
| 452 | unsigned int i; |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 453 | struct core_exception e; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 454 | |
| 455 | /* presume the area is full */ |
| 456 | *full = 1; |
| 457 | |
| 458 | for (i = 0; i < ps->exceptions_per_area; i++) { |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 459 | read_exception(ps, i, &e); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 460 | |
| 461 | /* |
| 462 | * If the new_chunk is pointing at the start of |
| 463 | * the COW device, where the first metadata area |
| 464 | * is we know that we've hit the end of the |
| 465 | * exceptions. Therefore the area is not full. |
| 466 | */ |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 467 | if (e.new_chunk == 0LL) { |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 468 | ps->current_committed = i; |
| 469 | *full = 0; |
| 470 | break; |
| 471 | } |
| 472 | |
| 473 | /* |
| 474 | * Keep track of the start of the free chunks. |
| 475 | */ |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 476 | if (ps->next_free <= e.new_chunk) |
| 477 | ps->next_free = e.new_chunk + 1; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 478 | |
| 479 | /* |
| 480 | * Otherwise we add the exception to the snapshot. |
| 481 | */ |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 482 | r = callback(callback_context, e.old_chunk, e.new_chunk); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 483 | if (r) |
| 484 | return r; |
| 485 | } |
| 486 | |
| 487 | return 0; |
| 488 | } |
| 489 | |
Jonathan Brassow | a159c1a | 2009-01-06 03:05:19 +0000 | [diff] [blame] | 490 | static int read_exceptions(struct pstore *ps, |
| 491 | int (*callback)(void *callback_context, chunk_t old, |
| 492 | chunk_t new), |
| 493 | void *callback_context) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 494 | { |
| 495 | int r, full = 1; |
| 496 | |
| 497 | /* |
| 498 | * Keeping reading chunks and inserting exceptions until |
| 499 | * we find a partially full area. |
| 500 | */ |
| 501 | for (ps->current_area = 0; full; ps->current_area++) { |
| 502 | r = area_io(ps, READ); |
| 503 | if (r) |
| 504 | return r; |
| 505 | |
Jonathan Brassow | a159c1a | 2009-01-06 03:05:19 +0000 | [diff] [blame] | 506 | r = insert_exceptions(ps, callback, callback_context, &full); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 507 | if (r) |
| 508 | return r; |
| 509 | } |
| 510 | |
| 511 | ps->current_area--; |
| 512 | |
Mikulas Patocka | cf5d2d7 | 2013-10-16 03:17:47 +0100 | [diff] [blame] | 513 | skip_metadata(ps); |
| 514 | |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 515 | return 0; |
| 516 | } |
| 517 | |
| 518 | static struct pstore *get_info(struct dm_exception_store *store) |
| 519 | { |
| 520 | return (struct pstore *) store->context; |
| 521 | } |
| 522 | |
Mike Snitzer | 985903b | 2009-12-10 23:52:11 +0000 | [diff] [blame] | 523 | static void persistent_usage(struct dm_exception_store *store, |
| 524 | sector_t *total_sectors, |
| 525 | sector_t *sectors_allocated, |
| 526 | sector_t *metadata_sectors) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 527 | { |
Mike Snitzer | 985903b | 2009-12-10 23:52:11 +0000 | [diff] [blame] | 528 | struct pstore *ps = get_info(store); |
| 529 | |
| 530 | *sectors_allocated = ps->next_free * store->chunk_size; |
Mike Snitzer | fc56f6f | 2009-12-10 23:52:12 +0000 | [diff] [blame] | 531 | *total_sectors = get_dev_size(dm_snap_cow(store->snap)->bdev); |
Mike Snitzer | 985903b | 2009-12-10 23:52:11 +0000 | [diff] [blame] | 532 | |
| 533 | /* |
| 534 | * First chunk is the fixed header. |
| 535 | * Then there are (ps->current_area + 1) metadata chunks, each one |
| 536 | * separated from the next by ps->exceptions_per_area data chunks. |
| 537 | */ |
Mikulas Patocka | 4454a62 | 2009-12-10 23:52:29 +0000 | [diff] [blame] | 538 | *metadata_sectors = (ps->current_area + 1 + NUM_SNAPSHOT_HDR_CHUNKS) * |
| 539 | store->chunk_size; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 540 | } |
| 541 | |
Jonathan Brassow | 493df71 | 2009-04-02 19:55:31 +0100 | [diff] [blame] | 542 | static void persistent_dtr(struct dm_exception_store *store) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 543 | { |
| 544 | struct pstore *ps = get_info(store); |
| 545 | |
| 546 | destroy_workqueue(ps->metadata_wq); |
Jonathan Brassow | a32079c | 2009-04-02 19:55:35 +0100 | [diff] [blame] | 547 | |
| 548 | /* Created in read_header */ |
| 549 | if (ps->io_client) |
| 550 | dm_io_client_destroy(ps->io_client); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 551 | free_area(ps); |
Jonathan Brassow | a32079c | 2009-04-02 19:55:35 +0100 | [diff] [blame] | 552 | |
| 553 | /* Allocated in persistent_read_metadata */ |
| 554 | if (ps->callbacks) |
| 555 | vfree(ps->callbacks); |
| 556 | |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 557 | kfree(ps); |
| 558 | } |
| 559 | |
Jonathan Brassow | a159c1a | 2009-01-06 03:05:19 +0000 | [diff] [blame] | 560 | static int persistent_read_metadata(struct dm_exception_store *store, |
| 561 | int (*callback)(void *callback_context, |
| 562 | chunk_t old, chunk_t new), |
| 563 | void *callback_context) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 564 | { |
| 565 | int r, uninitialized_var(new_snapshot); |
| 566 | struct pstore *ps = get_info(store); |
| 567 | |
| 568 | /* |
| 569 | * Read the snapshot header. |
| 570 | */ |
| 571 | r = read_header(ps, &new_snapshot); |
| 572 | if (r) |
| 573 | return r; |
| 574 | |
| 575 | /* |
| 576 | * Now we know correct chunk_size, complete the initialisation. |
| 577 | */ |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 578 | ps->exceptions_per_area = (ps->store->chunk_size << SECTOR_SHIFT) / |
| 579 | sizeof(struct disk_exception); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 580 | ps->callbacks = dm_vcalloc(ps->exceptions_per_area, |
Jonathan Brassow | a2d2b03 | 2011-08-02 12:32:03 +0100 | [diff] [blame] | 581 | sizeof(*ps->callbacks)); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 582 | if (!ps->callbacks) |
| 583 | return -ENOMEM; |
| 584 | |
| 585 | /* |
| 586 | * Do we need to setup a new snapshot ? |
| 587 | */ |
| 588 | if (new_snapshot) { |
| 589 | r = write_header(ps); |
| 590 | if (r) { |
| 591 | DMWARN("write_header failed"); |
| 592 | return r; |
| 593 | } |
| 594 | |
| 595 | ps->current_area = 0; |
| 596 | zero_memory_area(ps); |
| 597 | r = zero_disk_area(ps, 0); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 598 | if (r) |
Jon Brassow | f5acc83 | 2009-12-10 23:52:07 +0000 | [diff] [blame] | 599 | DMWARN("zero_disk_area(0) failed"); |
| 600 | return r; |
| 601 | } |
| 602 | /* |
| 603 | * Sanity checks. |
| 604 | */ |
| 605 | if (ps->version != SNAPSHOT_DISK_VERSION) { |
| 606 | DMWARN("unable to handle snapshot disk version %d", |
| 607 | ps->version); |
| 608 | return -EINVAL; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 609 | } |
| 610 | |
Jon Brassow | f5acc83 | 2009-12-10 23:52:07 +0000 | [diff] [blame] | 611 | /* |
| 612 | * Metadata are valid, but snapshot is invalidated |
| 613 | */ |
| 614 | if (!ps->valid) |
| 615 | return 1; |
| 616 | |
| 617 | /* |
| 618 | * Read the metadata. |
| 619 | */ |
| 620 | r = read_exceptions(ps, callback, callback_context); |
| 621 | |
| 622 | return r; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 623 | } |
| 624 | |
Jonathan Brassow | a159c1a | 2009-01-06 03:05:19 +0000 | [diff] [blame] | 625 | static int persistent_prepare_exception(struct dm_exception_store *store, |
Jon Brassow | 1d4989c | 2009-12-10 23:52:10 +0000 | [diff] [blame] | 626 | struct dm_exception *e) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 627 | { |
| 628 | struct pstore *ps = get_info(store); |
Mike Snitzer | fc56f6f | 2009-12-10 23:52:12 +0000 | [diff] [blame] | 629 | sector_t size = get_dev_size(dm_snap_cow(store->snap)->bdev); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 630 | |
| 631 | /* Is there enough room ? */ |
Jonathan Brassow | d021684 | 2009-04-02 19:55:32 +0100 | [diff] [blame] | 632 | if (size < ((ps->next_free + 1) * store->chunk_size)) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 633 | return -ENOSPC; |
| 634 | |
| 635 | e->new_chunk = ps->next_free; |
| 636 | |
| 637 | /* |
| 638 | * Move onto the next free pending, making sure to take |
| 639 | * into account the location of the metadata chunks. |
| 640 | */ |
Mikulas Patocka | cf5d2d7 | 2013-10-16 03:17:47 +0100 | [diff] [blame] | 641 | ps->next_free++; |
| 642 | skip_metadata(ps); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 643 | |
| 644 | atomic_inc(&ps->pending_count); |
| 645 | return 0; |
| 646 | } |
| 647 | |
Jonathan Brassow | a159c1a | 2009-01-06 03:05:19 +0000 | [diff] [blame] | 648 | static void persistent_commit_exception(struct dm_exception_store *store, |
Jon Brassow | 1d4989c | 2009-12-10 23:52:10 +0000 | [diff] [blame] | 649 | struct dm_exception *e, |
Jonathan Brassow | a159c1a | 2009-01-06 03:05:19 +0000 | [diff] [blame] | 650 | void (*callback) (void *, int success), |
| 651 | void *callback_context) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 652 | { |
| 653 | unsigned int i; |
| 654 | struct pstore *ps = get_info(store); |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 655 | struct core_exception ce; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 656 | struct commit_callback *cb; |
| 657 | |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 658 | ce.old_chunk = e->old_chunk; |
| 659 | ce.new_chunk = e->new_chunk; |
| 660 | write_exception(ps, ps->current_committed++, &ce); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 661 | |
| 662 | /* |
| 663 | * Add the callback to the back of the array. This code |
| 664 | * is the only place where the callback array is |
| 665 | * manipulated, and we know that it will never be called |
| 666 | * multiple times concurrently. |
| 667 | */ |
| 668 | cb = ps->callbacks + ps->callback_count++; |
| 669 | cb->callback = callback; |
| 670 | cb->context = callback_context; |
| 671 | |
| 672 | /* |
| 673 | * If there are exceptions in flight and we have not yet |
| 674 | * filled this metadata area there's nothing more to do. |
| 675 | */ |
| 676 | if (!atomic_dec_and_test(&ps->pending_count) && |
| 677 | (ps->current_committed != ps->exceptions_per_area)) |
| 678 | return; |
| 679 | |
| 680 | /* |
| 681 | * If we completely filled the current area, then wipe the next one. |
| 682 | */ |
| 683 | if ((ps->current_committed == ps->exceptions_per_area) && |
Jonathan Brassow | a2d2b03 | 2011-08-02 12:32:03 +0100 | [diff] [blame] | 684 | zero_disk_area(ps, ps->current_area + 1)) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 685 | ps->valid = 0; |
| 686 | |
| 687 | /* |
| 688 | * Commit exceptions to disk. |
| 689 | */ |
Tejun Heo | d87f4c1 | 2010-09-03 11:56:19 +0200 | [diff] [blame] | 690 | if (ps->valid && area_io(ps, WRITE_FLUSH_FUA)) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 691 | ps->valid = 0; |
| 692 | |
| 693 | /* |
| 694 | * Advance to the next area if this one is full. |
| 695 | */ |
| 696 | if (ps->current_committed == ps->exceptions_per_area) { |
| 697 | ps->current_committed = 0; |
| 698 | ps->current_area++; |
| 699 | zero_memory_area(ps); |
| 700 | } |
| 701 | |
| 702 | for (i = 0; i < ps->callback_count; i++) { |
| 703 | cb = ps->callbacks + i; |
| 704 | cb->callback(cb->context, ps->valid); |
| 705 | } |
| 706 | |
| 707 | ps->callback_count = 0; |
| 708 | } |
| 709 | |
Mikulas Patocka | 4454a62 | 2009-12-10 23:52:29 +0000 | [diff] [blame] | 710 | static int persistent_prepare_merge(struct dm_exception_store *store, |
| 711 | chunk_t *last_old_chunk, |
| 712 | chunk_t *last_new_chunk) |
| 713 | { |
| 714 | struct pstore *ps = get_info(store); |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 715 | struct core_exception ce; |
Mikulas Patocka | 4454a62 | 2009-12-10 23:52:29 +0000 | [diff] [blame] | 716 | int nr_consecutive; |
| 717 | int r; |
| 718 | |
| 719 | /* |
| 720 | * When current area is empty, move back to preceding area. |
| 721 | */ |
| 722 | if (!ps->current_committed) { |
| 723 | /* |
| 724 | * Have we finished? |
| 725 | */ |
| 726 | if (!ps->current_area) |
| 727 | return 0; |
| 728 | |
| 729 | ps->current_area--; |
| 730 | r = area_io(ps, READ); |
| 731 | if (r < 0) |
| 732 | return r; |
| 733 | ps->current_committed = ps->exceptions_per_area; |
| 734 | } |
| 735 | |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 736 | read_exception(ps, ps->current_committed - 1, &ce); |
| 737 | *last_old_chunk = ce.old_chunk; |
| 738 | *last_new_chunk = ce.new_chunk; |
Mikulas Patocka | 4454a62 | 2009-12-10 23:52:29 +0000 | [diff] [blame] | 739 | |
| 740 | /* |
| 741 | * Find number of consecutive chunks within the current area, |
| 742 | * working backwards. |
| 743 | */ |
| 744 | for (nr_consecutive = 1; nr_consecutive < ps->current_committed; |
| 745 | nr_consecutive++) { |
| 746 | read_exception(ps, ps->current_committed - 1 - nr_consecutive, |
Alasdair G Kergon | 283a832 | 2011-08-02 12:32:01 +0100 | [diff] [blame] | 747 | &ce); |
| 748 | if (ce.old_chunk != *last_old_chunk - nr_consecutive || |
| 749 | ce.new_chunk != *last_new_chunk - nr_consecutive) |
Mikulas Patocka | 4454a62 | 2009-12-10 23:52:29 +0000 | [diff] [blame] | 750 | break; |
| 751 | } |
| 752 | |
| 753 | return nr_consecutive; |
| 754 | } |
| 755 | |
| 756 | static int persistent_commit_merge(struct dm_exception_store *store, |
| 757 | int nr_merged) |
| 758 | { |
| 759 | int r, i; |
| 760 | struct pstore *ps = get_info(store); |
| 761 | |
| 762 | BUG_ON(nr_merged > ps->current_committed); |
| 763 | |
| 764 | for (i = 0; i < nr_merged; i++) |
| 765 | clear_exception(ps, ps->current_committed - 1 - i); |
| 766 | |
Mikulas Patocka | 762a80d | 2011-08-02 12:32:00 +0100 | [diff] [blame] | 767 | r = area_io(ps, WRITE_FLUSH_FUA); |
Mikulas Patocka | 4454a62 | 2009-12-10 23:52:29 +0000 | [diff] [blame] | 768 | if (r < 0) |
| 769 | return r; |
| 770 | |
| 771 | ps->current_committed -= nr_merged; |
| 772 | |
| 773 | /* |
| 774 | * At this stage, only persistent_usage() uses ps->next_free, so |
| 775 | * we make no attempt to keep ps->next_free strictly accurate |
| 776 | * as exceptions may have been committed out-of-order originally. |
| 777 | * Once a snapshot has become merging, we set it to the value it |
| 778 | * would have held had all the exceptions been committed in order. |
| 779 | * |
| 780 | * ps->current_area does not get reduced by prepare_merge() until |
| 781 | * after commit_merge() has removed the nr_merged previous exceptions. |
| 782 | */ |
Tomohiro Kusumi | 87c961c | 2010-08-12 04:13:59 +0100 | [diff] [blame] | 783 | ps->next_free = area_location(ps, ps->current_area) + |
| 784 | ps->current_committed + 1; |
Mikulas Patocka | 4454a62 | 2009-12-10 23:52:29 +0000 | [diff] [blame] | 785 | |
| 786 | return 0; |
| 787 | } |
| 788 | |
Jonathan Brassow | a159c1a | 2009-01-06 03:05:19 +0000 | [diff] [blame] | 789 | static void persistent_drop_snapshot(struct dm_exception_store *store) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 790 | { |
| 791 | struct pstore *ps = get_info(store); |
| 792 | |
| 793 | ps->valid = 0; |
| 794 | if (write_header(ps)) |
| 795 | DMWARN("write header failed"); |
| 796 | } |
| 797 | |
Jonathan Brassow | 493df71 | 2009-04-02 19:55:31 +0100 | [diff] [blame] | 798 | static int persistent_ctr(struct dm_exception_store *store, |
| 799 | unsigned argc, char **argv) |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 800 | { |
| 801 | struct pstore *ps; |
| 802 | |
| 803 | /* allocate the pstore */ |
Jonathan Brassow | a32079c | 2009-04-02 19:55:35 +0100 | [diff] [blame] | 804 | ps = kzalloc(sizeof(*ps), GFP_KERNEL); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 805 | if (!ps) |
| 806 | return -ENOMEM; |
| 807 | |
Jonathan Brassow | 71fab00 | 2009-04-02 19:55:33 +0100 | [diff] [blame] | 808 | ps->store = store; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 809 | ps->valid = 1; |
| 810 | ps->version = SNAPSHOT_DISK_VERSION; |
| 811 | ps->area = NULL; |
Mikulas Patocka | 61578dc | 2009-09-04 20:40:39 +0100 | [diff] [blame] | 812 | ps->zero_area = NULL; |
| 813 | ps->header_area = NULL; |
Mikulas Patocka | 4454a62 | 2009-12-10 23:52:29 +0000 | [diff] [blame] | 814 | ps->next_free = NUM_SNAPSHOT_HDR_CHUNKS + 1; /* header and 1st area */ |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 815 | ps->current_committed = 0; |
| 816 | |
| 817 | ps->callback_count = 0; |
| 818 | atomic_set(&ps->pending_count, 0); |
| 819 | ps->callbacks = NULL; |
| 820 | |
Tejun Heo | 239c8dd | 2011-01-13 19:59:59 +0000 | [diff] [blame] | 821 | ps->metadata_wq = alloc_workqueue("ksnaphd", WQ_MEM_RECLAIM, 0); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 822 | if (!ps->metadata_wq) { |
| 823 | kfree(ps); |
| 824 | DMERR("couldn't start header metadata update thread"); |
| 825 | return -ENOMEM; |
| 826 | } |
| 827 | |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 828 | store->context = ps; |
| 829 | |
| 830 | return 0; |
| 831 | } |
| 832 | |
Jonathan Brassow | 1e302a9 | 2009-04-02 19:55:35 +0100 | [diff] [blame] | 833 | static unsigned persistent_status(struct dm_exception_store *store, |
| 834 | status_type_t status, char *result, |
| 835 | unsigned maxlen) |
Jonathan Brassow | 493df71 | 2009-04-02 19:55:31 +0100 | [diff] [blame] | 836 | { |
Jonathan Brassow | 1e302a9 | 2009-04-02 19:55:35 +0100 | [diff] [blame] | 837 | unsigned sz = 0; |
| 838 | |
| 839 | switch (status) { |
| 840 | case STATUSTYPE_INFO: |
| 841 | break; |
| 842 | case STATUSTYPE_TABLE: |
Mike Snitzer | fc56f6f | 2009-12-10 23:52:12 +0000 | [diff] [blame] | 843 | DMEMIT(" P %llu", (unsigned long long)store->chunk_size); |
Jonathan Brassow | 1e302a9 | 2009-04-02 19:55:35 +0100 | [diff] [blame] | 844 | } |
Jonathan Brassow | 493df71 | 2009-04-02 19:55:31 +0100 | [diff] [blame] | 845 | |
| 846 | return sz; |
| 847 | } |
| 848 | |
| 849 | static struct dm_exception_store_type _persistent_type = { |
| 850 | .name = "persistent", |
| 851 | .module = THIS_MODULE, |
| 852 | .ctr = persistent_ctr, |
| 853 | .dtr = persistent_dtr, |
| 854 | .read_metadata = persistent_read_metadata, |
| 855 | .prepare_exception = persistent_prepare_exception, |
| 856 | .commit_exception = persistent_commit_exception, |
Mikulas Patocka | 4454a62 | 2009-12-10 23:52:29 +0000 | [diff] [blame] | 857 | .prepare_merge = persistent_prepare_merge, |
| 858 | .commit_merge = persistent_commit_merge, |
Jonathan Brassow | 493df71 | 2009-04-02 19:55:31 +0100 | [diff] [blame] | 859 | .drop_snapshot = persistent_drop_snapshot, |
Mike Snitzer | 985903b | 2009-12-10 23:52:11 +0000 | [diff] [blame] | 860 | .usage = persistent_usage, |
Jonathan Brassow | 493df71 | 2009-04-02 19:55:31 +0100 | [diff] [blame] | 861 | .status = persistent_status, |
| 862 | }; |
| 863 | |
| 864 | static struct dm_exception_store_type _persistent_compat_type = { |
| 865 | .name = "P", |
| 866 | .module = THIS_MODULE, |
| 867 | .ctr = persistent_ctr, |
| 868 | .dtr = persistent_dtr, |
| 869 | .read_metadata = persistent_read_metadata, |
| 870 | .prepare_exception = persistent_prepare_exception, |
| 871 | .commit_exception = persistent_commit_exception, |
Mikulas Patocka | 4454a62 | 2009-12-10 23:52:29 +0000 | [diff] [blame] | 872 | .prepare_merge = persistent_prepare_merge, |
| 873 | .commit_merge = persistent_commit_merge, |
Jonathan Brassow | 493df71 | 2009-04-02 19:55:31 +0100 | [diff] [blame] | 874 | .drop_snapshot = persistent_drop_snapshot, |
Mike Snitzer | 985903b | 2009-12-10 23:52:11 +0000 | [diff] [blame] | 875 | .usage = persistent_usage, |
Jonathan Brassow | 493df71 | 2009-04-02 19:55:31 +0100 | [diff] [blame] | 876 | .status = persistent_status, |
| 877 | }; |
| 878 | |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 879 | int dm_persistent_snapshot_init(void) |
| 880 | { |
Jonathan Brassow | 493df71 | 2009-04-02 19:55:31 +0100 | [diff] [blame] | 881 | int r; |
| 882 | |
| 883 | r = dm_exception_store_type_register(&_persistent_type); |
| 884 | if (r) { |
| 885 | DMERR("Unable to register persistent exception store type"); |
| 886 | return r; |
| 887 | } |
| 888 | |
| 889 | r = dm_exception_store_type_register(&_persistent_compat_type); |
| 890 | if (r) { |
| 891 | DMERR("Unable to register old-style persistent exception " |
| 892 | "store type"); |
| 893 | dm_exception_store_type_unregister(&_persistent_type); |
| 894 | return r; |
| 895 | } |
| 896 | |
| 897 | return r; |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 898 | } |
| 899 | |
| 900 | void dm_persistent_snapshot_exit(void) |
| 901 | { |
Jonathan Brassow | 493df71 | 2009-04-02 19:55:31 +0100 | [diff] [blame] | 902 | dm_exception_store_type_unregister(&_persistent_type); |
| 903 | dm_exception_store_type_unregister(&_persistent_compat_type); |
Alasdair G Kergon | 4db6bfe | 2009-01-06 03:05:17 +0000 | [diff] [blame] | 904 | } |