blob: 95891dfcbca021563a465751dacf064aee3a46df [file] [log] [blame]
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00001/*
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 Kergon4db6bfe2009-01-06 03:05:17 +00009
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
Mikulas Patocka4454a622009-12-10 23:52:29 +000058#define NUM_SNAPSHOT_HDR_CHUNKS 1
59
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000060struct disk_header {
61 uint32_t magic;
62
63 /*
64 * Is this snapshot valid. There is no way of recovering
65 * an invalid snapshot.
66 */
67 uint32_t valid;
68
69 /*
70 * Simple, incrementing version. no backward
71 * compatibility.
72 */
73 uint32_t version;
74
75 /* In sectors */
76 uint32_t chunk_size;
77};
78
79struct disk_exception {
80 uint64_t old_chunk;
81 uint64_t new_chunk;
82};
83
84struct commit_callback {
85 void (*callback)(void *, int success);
86 void *context;
87};
88
89/*
90 * The top level structure for a persistent exception store.
91 */
92struct pstore {
Jonathan Brassow71fab002009-04-02 19:55:33 +010093 struct dm_exception_store *store;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000094 int version;
95 int valid;
96 uint32_t exceptions_per_area;
97
98 /*
99 * Now that we have an asynchronous kcopyd there is no
100 * need for large chunk sizes, so it wont hurt to have a
101 * whole chunks worth of metadata in memory at once.
102 */
103 void *area;
104
105 /*
106 * An area of zeros used to clear the next area.
107 */
108 void *zero_area;
109
110 /*
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100111 * An area used for header. The header can be written
112 * concurrently with metadata (when invalidating the snapshot),
113 * so it needs a separate buffer.
114 */
115 void *header_area;
116
117 /*
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000118 * Used to keep track of which metadata area the data in
119 * 'chunk' refers to.
120 */
121 chunk_t current_area;
122
123 /*
124 * The next free chunk for an exception.
Mikulas Patocka4454a622009-12-10 23:52:29 +0000125 *
126 * When creating exceptions, all the chunks here and above are
127 * free. It holds the next chunk to be allocated. On rare
128 * occasions (e.g. after a system crash) holes can be left in
129 * the exception store because chunks can be committed out of
130 * order.
131 *
132 * When merging exceptions, it does not necessarily mean all the
133 * chunks here and above are free. It holds the value it would
134 * have held if all chunks had been committed in order of
135 * allocation. Consequently the value may occasionally be
136 * slightly too low, but since it's only used for 'status' and
137 * it can never reach its minimum value too early this doesn't
138 * matter.
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000139 */
Mikulas Patocka4454a622009-12-10 23:52:29 +0000140
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000141 chunk_t next_free;
142
143 /*
144 * The index of next free exception in the current
145 * metadata area.
146 */
147 uint32_t current_committed;
148
149 atomic_t pending_count;
150 uint32_t callback_count;
151 struct commit_callback *callbacks;
152 struct dm_io_client *io_client;
153
154 struct workqueue_struct *metadata_wq;
155};
156
157static unsigned sectors_to_pages(unsigned sectors)
158{
159 return DIV_ROUND_UP(sectors, PAGE_SIZE >> 9);
160}
161
162static int alloc_area(struct pstore *ps)
163{
164 int r = -ENOMEM;
165 size_t len;
166
Jonathan Brassow71fab002009-04-02 19:55:33 +0100167 len = ps->store->chunk_size << SECTOR_SHIFT;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000168
169 /*
170 * Allocate the chunk_size block of memory that will hold
171 * a single metadata area.
172 */
173 ps->area = vmalloc(len);
174 if (!ps->area)
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100175 goto err_area;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000176
177 ps->zero_area = vmalloc(len);
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100178 if (!ps->zero_area)
179 goto err_zero_area;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000180 memset(ps->zero_area, 0, len);
181
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100182 ps->header_area = vmalloc(len);
183 if (!ps->header_area)
184 goto err_header_area;
185
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000186 return 0;
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100187
188err_header_area:
189 vfree(ps->zero_area);
190
191err_zero_area:
192 vfree(ps->area);
193
194err_area:
195 return r;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000196}
197
198static void free_area(struct pstore *ps)
199{
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100200 if (ps->area)
201 vfree(ps->area);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000202 ps->area = NULL;
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100203
204 if (ps->zero_area)
205 vfree(ps->zero_area);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000206 ps->zero_area = NULL;
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100207
208 if (ps->header_area)
209 vfree(ps->header_area);
210 ps->header_area = NULL;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000211}
212
213struct 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
220static 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 Patocka02d2fd32009-09-04 20:40:37 +0100230static int chunk_io(struct pstore *ps, void *area, chunk_t chunk, int rw,
231 int metadata)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000232{
233 struct dm_io_region where = {
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000234 .bdev = dm_snap_cow(ps->store->snap)->bdev,
Jonathan Brassow71fab002009-04-02 19:55:33 +0100235 .sector = ps->store->chunk_size * chunk,
236 .count = ps->store->chunk_size,
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000237 };
238 struct dm_io_request io_req = {
239 .bi_rw = rw,
240 .mem.type = DM_IO_VMA,
Mikulas Patocka02d2fd32009-09-04 20:40:37 +0100241 .mem.ptr.vma = area,
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000242 .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 Mortonca1cab32010-10-26 14:22:34 -0700257 INIT_WORK_ONSTACK(&req.work, do_metadata);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000258 queue_work(ps->metadata_wq, &req.work);
Tejun Heo239c8dd2011-01-13 19:59:59 +0000259 flush_work(&req.work);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000260
261 return req.result;
262}
263
264/*
265 * Convert a metadata area index to a chunk index.
266 */
267static chunk_t area_location(struct pstore *ps, chunk_t area)
268{
Tomohiro Kusumi87c961c2010-08-12 04:13:59 +0100269 return NUM_SNAPSHOT_HDR_CHUNKS + ((ps->exceptions_per_area + 1) * area);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000270}
271
272/*
273 * Read or write a metadata area. Remembering to skip the first
274 * chunk which holds the header.
275 */
276static int area_io(struct pstore *ps, int rw)
277{
278 int r;
279 chunk_t chunk;
280
281 chunk = area_location(ps, ps->current_area);
282
Mikulas Patocka02d2fd32009-09-04 20:40:37 +0100283 r = chunk_io(ps, ps->area, chunk, rw, 0);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000284 if (r)
285 return r;
286
287 return 0;
288}
289
290static void zero_memory_area(struct pstore *ps)
291{
Jonathan Brassow71fab002009-04-02 19:55:33 +0100292 memset(ps->area, 0, ps->store->chunk_size << SECTOR_SHIFT);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000293}
294
295static int zero_disk_area(struct pstore *ps, chunk_t area)
296{
Mikulas Patocka02d2fd32009-09-04 20:40:37 +0100297 return chunk_io(ps, ps->zero_area, area_location(ps, area), WRITE, 0);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000298}
299
300static int read_header(struct pstore *ps, int *new_snapshot)
301{
302 int r;
303 struct disk_header *dh;
Mikulas Patockadf96eee2009-10-16 23:18:17 +0100304 unsigned chunk_size;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000305 int chunk_size_supplied = 1;
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100306 char *chunk_err;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000307
308 /*
Mikulas Patockadf96eee2009-10-16 23:18:17 +0100309 * Use default chunk size (or logical_block_size, if larger)
310 * if none supplied
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000311 */
Jonathan Brassow71fab002009-04-02 19:55:33 +0100312 if (!ps->store->chunk_size) {
313 ps->store->chunk_size = max(DM_CHUNK_SIZE_DEFAULT_SECTORS,
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000314 bdev_logical_block_size(dm_snap_cow(ps->store->snap)->
315 bdev) >> 9);
Jonathan Brassow71fab002009-04-02 19:55:33 +0100316 ps->store->chunk_mask = ps->store->chunk_size - 1;
317 ps->store->chunk_shift = ffs(ps->store->chunk_size) - 1;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000318 chunk_size_supplied = 0;
319 }
320
Jonathan Brassow71fab002009-04-02 19:55:33 +0100321 ps->io_client = dm_io_client_create(sectors_to_pages(ps->store->
322 chunk_size));
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000323 if (IS_ERR(ps->io_client))
324 return PTR_ERR(ps->io_client);
325
326 r = alloc_area(ps);
327 if (r)
328 return r;
329
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100330 r = chunk_io(ps, ps->header_area, 0, READ, 1);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000331 if (r)
332 goto bad;
333
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100334 dh = ps->header_area;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000335
336 if (le32_to_cpu(dh->magic) == 0) {
337 *new_snapshot = 1;
338 return 0;
339 }
340
341 if (le32_to_cpu(dh->magic) != SNAP_MAGIC) {
342 DMWARN("Invalid or corrupt snapshot");
343 r = -ENXIO;
344 goto bad;
345 }
346
347 *new_snapshot = 0;
348 ps->valid = le32_to_cpu(dh->valid);
349 ps->version = le32_to_cpu(dh->version);
350 chunk_size = le32_to_cpu(dh->chunk_size);
351
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100352 if (ps->store->chunk_size == chunk_size)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000353 return 0;
354
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100355 if (chunk_size_supplied)
Mikulas Patockadf96eee2009-10-16 23:18:17 +0100356 DMWARN("chunk size %u in device metadata overrides "
357 "table chunk size of %u.",
358 chunk_size, ps->store->chunk_size);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000359
360 /* We had a bogus chunk_size. Fix stuff up. */
361 free_area(ps);
362
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100363 r = dm_exception_store_set_chunk_size(ps->store, chunk_size,
364 &chunk_err);
365 if (r) {
Mikulas Patockadf96eee2009-10-16 23:18:17 +0100366 DMERR("invalid on-disk chunk size %u: %s.",
367 chunk_size, chunk_err);
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100368 return r;
369 }
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000370
Jonathan Brassow71fab002009-04-02 19:55:33 +0100371 r = dm_io_client_resize(sectors_to_pages(ps->store->chunk_size),
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000372 ps->io_client);
373 if (r)
374 return r;
375
376 r = alloc_area(ps);
377 return r;
378
379bad:
380 free_area(ps);
381 return r;
382}
383
384static int write_header(struct pstore *ps)
385{
386 struct disk_header *dh;
387
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100388 memset(ps->header_area, 0, ps->store->chunk_size << SECTOR_SHIFT);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000389
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100390 dh = ps->header_area;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000391 dh->magic = cpu_to_le32(SNAP_MAGIC);
392 dh->valid = cpu_to_le32(ps->valid);
393 dh->version = cpu_to_le32(ps->version);
Jonathan Brassow71fab002009-04-02 19:55:33 +0100394 dh->chunk_size = cpu_to_le32(ps->store->chunk_size);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000395
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100396 return chunk_io(ps, ps->header_area, 0, WRITE, 1);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000397}
398
399/*
400 * Access functions for the disk exceptions, these do the endian conversions.
401 */
402static struct disk_exception *get_exception(struct pstore *ps, uint32_t index)
403{
404 BUG_ON(index >= ps->exceptions_per_area);
405
406 return ((struct disk_exception *) ps->area) + index;
407}
408
409static void read_exception(struct pstore *ps,
410 uint32_t index, struct disk_exception *result)
411{
412 struct disk_exception *e = get_exception(ps, index);
413
414 /* copy it */
415 result->old_chunk = le64_to_cpu(e->old_chunk);
416 result->new_chunk = le64_to_cpu(e->new_chunk);
417}
418
419static void write_exception(struct pstore *ps,
420 uint32_t index, struct disk_exception *de)
421{
422 struct disk_exception *e = get_exception(ps, index);
423
424 /* copy it */
425 e->old_chunk = cpu_to_le64(de->old_chunk);
426 e->new_chunk = cpu_to_le64(de->new_chunk);
427}
428
Mikulas Patocka4454a622009-12-10 23:52:29 +0000429static void clear_exception(struct pstore *ps, uint32_t index)
430{
431 struct disk_exception *e = get_exception(ps, index);
432
433 /* clear it */
434 e->old_chunk = 0;
435 e->new_chunk = 0;
436}
437
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000438/*
439 * Registers the exceptions that are present in the current area.
440 * 'full' is filled in to indicate if the area has been
441 * filled.
442 */
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000443static int insert_exceptions(struct pstore *ps,
444 int (*callback)(void *callback_context,
445 chunk_t old, chunk_t new),
446 void *callback_context,
447 int *full)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000448{
449 int r;
450 unsigned int i;
451 struct disk_exception de;
452
453 /* presume the area is full */
454 *full = 1;
455
456 for (i = 0; i < ps->exceptions_per_area; i++) {
457 read_exception(ps, i, &de);
458
459 /*
460 * If the new_chunk is pointing at the start of
461 * the COW device, where the first metadata area
462 * is we know that we've hit the end of the
463 * exceptions. Therefore the area is not full.
464 */
465 if (de.new_chunk == 0LL) {
466 ps->current_committed = i;
467 *full = 0;
468 break;
469 }
470
471 /*
472 * Keep track of the start of the free chunks.
473 */
474 if (ps->next_free <= de.new_chunk)
475 ps->next_free = de.new_chunk + 1;
476
477 /*
478 * Otherwise we add the exception to the snapshot.
479 */
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000480 r = callback(callback_context, de.old_chunk, de.new_chunk);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000481 if (r)
482 return r;
483 }
484
485 return 0;
486}
487
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000488static int read_exceptions(struct pstore *ps,
489 int (*callback)(void *callback_context, chunk_t old,
490 chunk_t new),
491 void *callback_context)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000492{
493 int r, full = 1;
494
495 /*
496 * Keeping reading chunks and inserting exceptions until
497 * we find a partially full area.
498 */
499 for (ps->current_area = 0; full; ps->current_area++) {
500 r = area_io(ps, READ);
501 if (r)
502 return r;
503
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000504 r = insert_exceptions(ps, callback, callback_context, &full);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000505 if (r)
506 return r;
507 }
508
509 ps->current_area--;
510
511 return 0;
512}
513
514static struct pstore *get_info(struct dm_exception_store *store)
515{
516 return (struct pstore *) store->context;
517}
518
Mike Snitzer985903b2009-12-10 23:52:11 +0000519static void persistent_usage(struct dm_exception_store *store,
520 sector_t *total_sectors,
521 sector_t *sectors_allocated,
522 sector_t *metadata_sectors)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000523{
Mike Snitzer985903b2009-12-10 23:52:11 +0000524 struct pstore *ps = get_info(store);
525
526 *sectors_allocated = ps->next_free * store->chunk_size;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000527 *total_sectors = get_dev_size(dm_snap_cow(store->snap)->bdev);
Mike Snitzer985903b2009-12-10 23:52:11 +0000528
529 /*
530 * First chunk is the fixed header.
531 * Then there are (ps->current_area + 1) metadata chunks, each one
532 * separated from the next by ps->exceptions_per_area data chunks.
533 */
Mikulas Patocka4454a622009-12-10 23:52:29 +0000534 *metadata_sectors = (ps->current_area + 1 + NUM_SNAPSHOT_HDR_CHUNKS) *
535 store->chunk_size;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000536}
537
Jonathan Brassow493df712009-04-02 19:55:31 +0100538static void persistent_dtr(struct dm_exception_store *store)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000539{
540 struct pstore *ps = get_info(store);
541
542 destroy_workqueue(ps->metadata_wq);
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100543
544 /* Created in read_header */
545 if (ps->io_client)
546 dm_io_client_destroy(ps->io_client);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000547 free_area(ps);
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100548
549 /* Allocated in persistent_read_metadata */
550 if (ps->callbacks)
551 vfree(ps->callbacks);
552
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000553 kfree(ps);
554}
555
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000556static int persistent_read_metadata(struct dm_exception_store *store,
557 int (*callback)(void *callback_context,
558 chunk_t old, chunk_t new),
559 void *callback_context)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000560{
561 int r, uninitialized_var(new_snapshot);
562 struct pstore *ps = get_info(store);
563
564 /*
565 * Read the snapshot header.
566 */
567 r = read_header(ps, &new_snapshot);
568 if (r)
569 return r;
570
571 /*
572 * Now we know correct chunk_size, complete the initialisation.
573 */
Jonathan Brassow71fab002009-04-02 19:55:33 +0100574 ps->exceptions_per_area = (ps->store->chunk_size << SECTOR_SHIFT) /
575 sizeof(struct disk_exception);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000576 ps->callbacks = dm_vcalloc(ps->exceptions_per_area,
577 sizeof(*ps->callbacks));
578 if (!ps->callbacks)
579 return -ENOMEM;
580
581 /*
582 * Do we need to setup a new snapshot ?
583 */
584 if (new_snapshot) {
585 r = write_header(ps);
586 if (r) {
587 DMWARN("write_header failed");
588 return r;
589 }
590
591 ps->current_area = 0;
592 zero_memory_area(ps);
593 r = zero_disk_area(ps, 0);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000594 if (r)
Jon Brassowf5acc832009-12-10 23:52:07 +0000595 DMWARN("zero_disk_area(0) failed");
596 return r;
597 }
598 /*
599 * Sanity checks.
600 */
601 if (ps->version != SNAPSHOT_DISK_VERSION) {
602 DMWARN("unable to handle snapshot disk version %d",
603 ps->version);
604 return -EINVAL;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000605 }
606
Jon Brassowf5acc832009-12-10 23:52:07 +0000607 /*
608 * Metadata are valid, but snapshot is invalidated
609 */
610 if (!ps->valid)
611 return 1;
612
613 /*
614 * Read the metadata.
615 */
616 r = read_exceptions(ps, callback, callback_context);
617
618 return r;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000619}
620
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000621static int persistent_prepare_exception(struct dm_exception_store *store,
Jon Brassow1d4989c2009-12-10 23:52:10 +0000622 struct dm_exception *e)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000623{
624 struct pstore *ps = get_info(store);
625 uint32_t stride;
626 chunk_t next_free;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000627 sector_t size = get_dev_size(dm_snap_cow(store->snap)->bdev);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000628
629 /* Is there enough room ? */
Jonathan Brassowd0216842009-04-02 19:55:32 +0100630 if (size < ((ps->next_free + 1) * store->chunk_size))
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000631 return -ENOSPC;
632
633 e->new_chunk = ps->next_free;
634
635 /*
636 * Move onto the next free pending, making sure to take
637 * into account the location of the metadata chunks.
638 */
639 stride = (ps->exceptions_per_area + 1);
640 next_free = ++ps->next_free;
641 if (sector_div(next_free, stride) == 1)
642 ps->next_free++;
643
644 atomic_inc(&ps->pending_count);
645 return 0;
646}
647
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000648static void persistent_commit_exception(struct dm_exception_store *store,
Jon Brassow1d4989c2009-12-10 23:52:10 +0000649 struct dm_exception *e,
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000650 void (*callback) (void *, int success),
651 void *callback_context)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000652{
653 unsigned int i;
654 struct pstore *ps = get_info(store);
655 struct disk_exception de;
656 struct commit_callback *cb;
657
658 de.old_chunk = e->old_chunk;
659 de.new_chunk = e->new_chunk;
660 write_exception(ps, ps->current_committed++, &de);
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) &&
684 zero_disk_area(ps, ps->current_area + 1))
685 ps->valid = 0;
686
687 /*
688 * Commit exceptions to disk.
689 */
Tejun Heod87f4c12010-09-03 11:56:19 +0200690 if (ps->valid && area_io(ps, WRITE_FLUSH_FUA))
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000691 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 Patocka4454a622009-12-10 23:52:29 +0000710static 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);
715 struct disk_exception de;
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
736 read_exception(ps, ps->current_committed - 1, &de);
737 *last_old_chunk = de.old_chunk;
738 *last_new_chunk = de.new_chunk;
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,
747 &de);
748 if (de.old_chunk != *last_old_chunk - nr_consecutive ||
749 de.new_chunk != *last_new_chunk - nr_consecutive)
750 break;
751 }
752
753 return nr_consecutive;
754}
755
756static 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
767 r = area_io(ps, WRITE);
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 Kusumi87c961c2010-08-12 04:13:59 +0100783 ps->next_free = area_location(ps, ps->current_area) +
784 ps->current_committed + 1;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000785
786 return 0;
787}
788
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000789static void persistent_drop_snapshot(struct dm_exception_store *store)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000790{
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 Brassow493df712009-04-02 19:55:31 +0100798static int persistent_ctr(struct dm_exception_store *store,
799 unsigned argc, char **argv)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000800{
801 struct pstore *ps;
802
803 /* allocate the pstore */
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100804 ps = kzalloc(sizeof(*ps), GFP_KERNEL);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000805 if (!ps)
806 return -ENOMEM;
807
Jonathan Brassow71fab002009-04-02 19:55:33 +0100808 ps->store = store;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000809 ps->valid = 1;
810 ps->version = SNAPSHOT_DISK_VERSION;
811 ps->area = NULL;
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100812 ps->zero_area = NULL;
813 ps->header_area = NULL;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000814 ps->next_free = NUM_SNAPSHOT_HDR_CHUNKS + 1; /* header and 1st area */
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000815 ps->current_committed = 0;
816
817 ps->callback_count = 0;
818 atomic_set(&ps->pending_count, 0);
819 ps->callbacks = NULL;
820
Tejun Heo239c8dd2011-01-13 19:59:59 +0000821 ps->metadata_wq = alloc_workqueue("ksnaphd", WQ_MEM_RECLAIM, 0);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000822 if (!ps->metadata_wq) {
823 kfree(ps);
824 DMERR("couldn't start header metadata update thread");
825 return -ENOMEM;
826 }
827
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000828 store->context = ps;
829
830 return 0;
831}
832
Jonathan Brassow1e302a92009-04-02 19:55:35 +0100833static unsigned persistent_status(struct dm_exception_store *store,
834 status_type_t status, char *result,
835 unsigned maxlen)
Jonathan Brassow493df712009-04-02 19:55:31 +0100836{
Jonathan Brassow1e302a92009-04-02 19:55:35 +0100837 unsigned sz = 0;
838
839 switch (status) {
840 case STATUSTYPE_INFO:
841 break;
842 case STATUSTYPE_TABLE:
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000843 DMEMIT(" P %llu", (unsigned long long)store->chunk_size);
Jonathan Brassow1e302a92009-04-02 19:55:35 +0100844 }
Jonathan Brassow493df712009-04-02 19:55:31 +0100845
846 return sz;
847}
848
849static 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 Patocka4454a622009-12-10 23:52:29 +0000857 .prepare_merge = persistent_prepare_merge,
858 .commit_merge = persistent_commit_merge,
Jonathan Brassow493df712009-04-02 19:55:31 +0100859 .drop_snapshot = persistent_drop_snapshot,
Mike Snitzer985903b2009-12-10 23:52:11 +0000860 .usage = persistent_usage,
Jonathan Brassow493df712009-04-02 19:55:31 +0100861 .status = persistent_status,
862};
863
864static 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 Patocka4454a622009-12-10 23:52:29 +0000872 .prepare_merge = persistent_prepare_merge,
873 .commit_merge = persistent_commit_merge,
Jonathan Brassow493df712009-04-02 19:55:31 +0100874 .drop_snapshot = persistent_drop_snapshot,
Mike Snitzer985903b2009-12-10 23:52:11 +0000875 .usage = persistent_usage,
Jonathan Brassow493df712009-04-02 19:55:31 +0100876 .status = persistent_status,
877};
878
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000879int dm_persistent_snapshot_init(void)
880{
Jonathan Brassow493df712009-04-02 19:55:31 +0100881 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 Kergon4db6bfe2009-01-06 03:05:17 +0000898}
899
900void dm_persistent_snapshot_exit(void)
901{
Jonathan Brassow493df712009-04-02 19:55:31 +0100902 dm_exception_store_type_unregister(&_persistent_type);
903 dm_exception_store_type_unregister(&_persistent_compat_type);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000904}