blob: 3ac415675b6c778b5dd22aaaf4ee6c2dc4ca48eb [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>
Paul Gortmakerdaaa5f72011-05-27 15:50:58 -040013#include <linux/export.h>
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000014#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 Patocka4454a622009-12-10 23:52:29 +000059#define NUM_SNAPSHOT_HDR_CHUNKS 1
60
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000061struct disk_header {
Alasdair G Kergon283a8322011-08-02 12:32:01 +010062 __le32 magic;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000063
64 /*
65 * Is this snapshot valid. There is no way of recovering
66 * an invalid snapshot.
67 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +010068 __le32 valid;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000069
70 /*
71 * Simple, incrementing version. no backward
72 * compatibility.
73 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +010074 __le32 version;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000075
76 /* In sectors */
Alasdair G Kergon283a8322011-08-02 12:32:01 +010077 __le32 chunk_size;
78} __packed;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000079
80struct disk_exception {
Alasdair G Kergon283a8322011-08-02 12:32:01 +010081 __le64 old_chunk;
82 __le64 new_chunk;
83} __packed;
84
85struct core_exception {
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000086 uint64_t old_chunk;
87 uint64_t new_chunk;
88};
89
90struct 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 */
98struct pstore {
Jonathan Brassow71fab002009-04-02 19:55:33 +010099 struct dm_exception_store *store;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000100 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 Patocka61578dc2009-09-04 20:40:39 +0100117 * 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 Kergon4db6bfe2009-01-06 03:05:17 +0000124 * 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 Patocka4454a622009-12-10 23:52:29 +0000131 *
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 Kergon4db6bfe2009-01-06 03:05:17 +0000145 */
Mikulas Patocka4454a622009-12-10 23:52:29 +0000146
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000147 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 Kergon4db6bfe2009-01-06 03:05:17 +0000163static int alloc_area(struct pstore *ps)
164{
165 int r = -ENOMEM;
166 size_t len;
167
Jonathan Brassow71fab002009-04-02 19:55:33 +0100168 len = ps->store->chunk_size << SECTOR_SHIFT;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000169
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 Patocka61578dc2009-09-04 20:40:39 +0100176 goto err_area;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000177
Joe Perchese29e65a2011-08-02 12:32:02 +0100178 ps->zero_area = vzalloc(len);
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100179 if (!ps->zero_area)
180 goto err_zero_area;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000181
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
Mikulas Patockabda8efe2011-05-29 13:03:09 +0100321 ps->io_client = dm_io_client_create();
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000322 if (IS_ERR(ps->io_client))
323 return PTR_ERR(ps->io_client);
324
325 r = alloc_area(ps);
326 if (r)
327 return r;
328
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100329 r = chunk_io(ps, ps->header_area, 0, READ, 1);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000330 if (r)
331 goto bad;
332
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100333 dh = ps->header_area;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000334
335 if (le32_to_cpu(dh->magic) == 0) {
336 *new_snapshot = 1;
337 return 0;
338 }
339
340 if (le32_to_cpu(dh->magic) != SNAP_MAGIC) {
341 DMWARN("Invalid or corrupt snapshot");
342 r = -ENXIO;
343 goto bad;
344 }
345
346 *new_snapshot = 0;
347 ps->valid = le32_to_cpu(dh->valid);
348 ps->version = le32_to_cpu(dh->version);
349 chunk_size = le32_to_cpu(dh->chunk_size);
350
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100351 if (ps->store->chunk_size == chunk_size)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000352 return 0;
353
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100354 if (chunk_size_supplied)
Mikulas Patockadf96eee2009-10-16 23:18:17 +0100355 DMWARN("chunk size %u in device metadata overrides "
356 "table chunk size of %u.",
357 chunk_size, ps->store->chunk_size);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000358
359 /* We had a bogus chunk_size. Fix stuff up. */
360 free_area(ps);
361
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100362 r = dm_exception_store_set_chunk_size(ps->store, chunk_size,
363 &chunk_err);
364 if (r) {
Mikulas Patockadf96eee2009-10-16 23:18:17 +0100365 DMERR("invalid on-disk chunk size %u: %s.",
366 chunk_size, chunk_err);
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100367 return r;
368 }
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000369
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000370 r = alloc_area(ps);
371 return r;
372
373bad:
374 free_area(ps);
375 return r;
376}
377
378static int write_header(struct pstore *ps)
379{
380 struct disk_header *dh;
381
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100382 memset(ps->header_area, 0, ps->store->chunk_size << SECTOR_SHIFT);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000383
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100384 dh = ps->header_area;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000385 dh->magic = cpu_to_le32(SNAP_MAGIC);
386 dh->valid = cpu_to_le32(ps->valid);
387 dh->version = cpu_to_le32(ps->version);
Jonathan Brassow71fab002009-04-02 19:55:33 +0100388 dh->chunk_size = cpu_to_le32(ps->store->chunk_size);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000389
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100390 return chunk_io(ps, ps->header_area, 0, WRITE, 1);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000391}
392
393/*
394 * Access functions for the disk exceptions, these do the endian conversions.
395 */
396static struct disk_exception *get_exception(struct pstore *ps, uint32_t index)
397{
398 BUG_ON(index >= ps->exceptions_per_area);
399
400 return ((struct disk_exception *) ps->area) + index;
401}
402
403static void read_exception(struct pstore *ps,
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100404 uint32_t index, struct core_exception *result)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000405{
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100406 struct disk_exception *de = get_exception(ps, index);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000407
408 /* copy it */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100409 result->old_chunk = le64_to_cpu(de->old_chunk);
410 result->new_chunk = le64_to_cpu(de->new_chunk);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000411}
412
413static void write_exception(struct pstore *ps,
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100414 uint32_t index, struct core_exception *e)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000415{
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100416 struct disk_exception *de = get_exception(ps, index);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000417
418 /* copy it */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100419 de->old_chunk = cpu_to_le64(e->old_chunk);
420 de->new_chunk = cpu_to_le64(e->new_chunk);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000421}
422
Mikulas Patocka4454a622009-12-10 23:52:29 +0000423static void clear_exception(struct pstore *ps, uint32_t index)
424{
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100425 struct disk_exception *de = get_exception(ps, index);
Mikulas Patocka4454a622009-12-10 23:52:29 +0000426
427 /* clear it */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100428 de->old_chunk = 0;
429 de->new_chunk = 0;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000430}
431
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000432/*
433 * Registers the exceptions that are present in the current area.
434 * 'full' is filled in to indicate if the area has been
435 * filled.
436 */
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000437static int insert_exceptions(struct pstore *ps,
438 int (*callback)(void *callback_context,
439 chunk_t old, chunk_t new),
440 void *callback_context,
441 int *full)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000442{
443 int r;
444 unsigned int i;
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100445 struct core_exception e;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000446
447 /* presume the area is full */
448 *full = 1;
449
450 for (i = 0; i < ps->exceptions_per_area; i++) {
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100451 read_exception(ps, i, &e);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000452
453 /*
454 * If the new_chunk is pointing at the start of
455 * the COW device, where the first metadata area
456 * is we know that we've hit the end of the
457 * exceptions. Therefore the area is not full.
458 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100459 if (e.new_chunk == 0LL) {
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000460 ps->current_committed = i;
461 *full = 0;
462 break;
463 }
464
465 /*
466 * Keep track of the start of the free chunks.
467 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100468 if (ps->next_free <= e.new_chunk)
469 ps->next_free = e.new_chunk + 1;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000470
471 /*
472 * Otherwise we add the exception to the snapshot.
473 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100474 r = callback(callback_context, e.old_chunk, e.new_chunk);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000475 if (r)
476 return r;
477 }
478
479 return 0;
480}
481
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000482static int read_exceptions(struct pstore *ps,
483 int (*callback)(void *callback_context, chunk_t old,
484 chunk_t new),
485 void *callback_context)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000486{
487 int r, full = 1;
488
489 /*
490 * Keeping reading chunks and inserting exceptions until
491 * we find a partially full area.
492 */
493 for (ps->current_area = 0; full; ps->current_area++) {
494 r = area_io(ps, READ);
495 if (r)
496 return r;
497
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000498 r = insert_exceptions(ps, callback, callback_context, &full);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000499 if (r)
500 return r;
501 }
502
503 ps->current_area--;
504
505 return 0;
506}
507
508static struct pstore *get_info(struct dm_exception_store *store)
509{
510 return (struct pstore *) store->context;
511}
512
Mike Snitzer985903b2009-12-10 23:52:11 +0000513static void persistent_usage(struct dm_exception_store *store,
514 sector_t *total_sectors,
515 sector_t *sectors_allocated,
516 sector_t *metadata_sectors)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000517{
Mike Snitzer985903b2009-12-10 23:52:11 +0000518 struct pstore *ps = get_info(store);
519
520 *sectors_allocated = ps->next_free * store->chunk_size;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000521 *total_sectors = get_dev_size(dm_snap_cow(store->snap)->bdev);
Mike Snitzer985903b2009-12-10 23:52:11 +0000522
523 /*
524 * First chunk is the fixed header.
525 * Then there are (ps->current_area + 1) metadata chunks, each one
526 * separated from the next by ps->exceptions_per_area data chunks.
527 */
Mikulas Patocka4454a622009-12-10 23:52:29 +0000528 *metadata_sectors = (ps->current_area + 1 + NUM_SNAPSHOT_HDR_CHUNKS) *
529 store->chunk_size;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000530}
531
Jonathan Brassow493df712009-04-02 19:55:31 +0100532static void persistent_dtr(struct dm_exception_store *store)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000533{
534 struct pstore *ps = get_info(store);
535
536 destroy_workqueue(ps->metadata_wq);
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100537
538 /* Created in read_header */
539 if (ps->io_client)
540 dm_io_client_destroy(ps->io_client);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000541 free_area(ps);
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100542
543 /* Allocated in persistent_read_metadata */
544 if (ps->callbacks)
545 vfree(ps->callbacks);
546
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000547 kfree(ps);
548}
549
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000550static int persistent_read_metadata(struct dm_exception_store *store,
551 int (*callback)(void *callback_context,
552 chunk_t old, chunk_t new),
553 void *callback_context)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000554{
555 int r, uninitialized_var(new_snapshot);
556 struct pstore *ps = get_info(store);
557
558 /*
559 * Read the snapshot header.
560 */
561 r = read_header(ps, &new_snapshot);
562 if (r)
563 return r;
564
565 /*
566 * Now we know correct chunk_size, complete the initialisation.
567 */
Jonathan Brassow71fab002009-04-02 19:55:33 +0100568 ps->exceptions_per_area = (ps->store->chunk_size << SECTOR_SHIFT) /
569 sizeof(struct disk_exception);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000570 ps->callbacks = dm_vcalloc(ps->exceptions_per_area,
Jonathan Brassowa2d2b032011-08-02 12:32:03 +0100571 sizeof(*ps->callbacks));
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000572 if (!ps->callbacks)
573 return -ENOMEM;
574
575 /*
576 * Do we need to setup a new snapshot ?
577 */
578 if (new_snapshot) {
579 r = write_header(ps);
580 if (r) {
581 DMWARN("write_header failed");
582 return r;
583 }
584
585 ps->current_area = 0;
586 zero_memory_area(ps);
587 r = zero_disk_area(ps, 0);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000588 if (r)
Jon Brassowf5acc832009-12-10 23:52:07 +0000589 DMWARN("zero_disk_area(0) failed");
590 return r;
591 }
592 /*
593 * Sanity checks.
594 */
595 if (ps->version != SNAPSHOT_DISK_VERSION) {
596 DMWARN("unable to handle snapshot disk version %d",
597 ps->version);
598 return -EINVAL;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000599 }
600
Jon Brassowf5acc832009-12-10 23:52:07 +0000601 /*
602 * Metadata are valid, but snapshot is invalidated
603 */
604 if (!ps->valid)
605 return 1;
606
607 /*
608 * Read the metadata.
609 */
610 r = read_exceptions(ps, callback, callback_context);
611
612 return r;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000613}
614
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000615static int persistent_prepare_exception(struct dm_exception_store *store,
Jon Brassow1d4989c2009-12-10 23:52:10 +0000616 struct dm_exception *e)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000617{
618 struct pstore *ps = get_info(store);
619 uint32_t stride;
620 chunk_t next_free;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000621 sector_t size = get_dev_size(dm_snap_cow(store->snap)->bdev);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000622
623 /* Is there enough room ? */
Jonathan Brassowd0216842009-04-02 19:55:32 +0100624 if (size < ((ps->next_free + 1) * store->chunk_size))
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000625 return -ENOSPC;
626
627 e->new_chunk = ps->next_free;
628
629 /*
630 * Move onto the next free pending, making sure to take
631 * into account the location of the metadata chunks.
632 */
633 stride = (ps->exceptions_per_area + 1);
634 next_free = ++ps->next_free;
635 if (sector_div(next_free, stride) == 1)
636 ps->next_free++;
637
638 atomic_inc(&ps->pending_count);
639 return 0;
640}
641
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000642static void persistent_commit_exception(struct dm_exception_store *store,
Jon Brassow1d4989c2009-12-10 23:52:10 +0000643 struct dm_exception *e,
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000644 void (*callback) (void *, int success),
645 void *callback_context)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000646{
647 unsigned int i;
648 struct pstore *ps = get_info(store);
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100649 struct core_exception ce;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000650 struct commit_callback *cb;
651
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100652 ce.old_chunk = e->old_chunk;
653 ce.new_chunk = e->new_chunk;
654 write_exception(ps, ps->current_committed++, &ce);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000655
656 /*
657 * Add the callback to the back of the array. This code
658 * is the only place where the callback array is
659 * manipulated, and we know that it will never be called
660 * multiple times concurrently.
661 */
662 cb = ps->callbacks + ps->callback_count++;
663 cb->callback = callback;
664 cb->context = callback_context;
665
666 /*
667 * If there are exceptions in flight and we have not yet
668 * filled this metadata area there's nothing more to do.
669 */
670 if (!atomic_dec_and_test(&ps->pending_count) &&
671 (ps->current_committed != ps->exceptions_per_area))
672 return;
673
674 /*
675 * If we completely filled the current area, then wipe the next one.
676 */
677 if ((ps->current_committed == ps->exceptions_per_area) &&
Jonathan Brassowa2d2b032011-08-02 12:32:03 +0100678 zero_disk_area(ps, ps->current_area + 1))
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000679 ps->valid = 0;
680
681 /*
682 * Commit exceptions to disk.
683 */
Tejun Heod87f4c12010-09-03 11:56:19 +0200684 if (ps->valid && area_io(ps, WRITE_FLUSH_FUA))
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000685 ps->valid = 0;
686
687 /*
688 * Advance to the next area if this one is full.
689 */
690 if (ps->current_committed == ps->exceptions_per_area) {
691 ps->current_committed = 0;
692 ps->current_area++;
693 zero_memory_area(ps);
694 }
695
696 for (i = 0; i < ps->callback_count; i++) {
697 cb = ps->callbacks + i;
698 cb->callback(cb->context, ps->valid);
699 }
700
701 ps->callback_count = 0;
702}
703
Mikulas Patocka4454a622009-12-10 23:52:29 +0000704static int persistent_prepare_merge(struct dm_exception_store *store,
705 chunk_t *last_old_chunk,
706 chunk_t *last_new_chunk)
707{
708 struct pstore *ps = get_info(store);
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100709 struct core_exception ce;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000710 int nr_consecutive;
711 int r;
712
713 /*
714 * When current area is empty, move back to preceding area.
715 */
716 if (!ps->current_committed) {
717 /*
718 * Have we finished?
719 */
720 if (!ps->current_area)
721 return 0;
722
723 ps->current_area--;
724 r = area_io(ps, READ);
725 if (r < 0)
726 return r;
727 ps->current_committed = ps->exceptions_per_area;
728 }
729
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100730 read_exception(ps, ps->current_committed - 1, &ce);
731 *last_old_chunk = ce.old_chunk;
732 *last_new_chunk = ce.new_chunk;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000733
734 /*
735 * Find number of consecutive chunks within the current area,
736 * working backwards.
737 */
738 for (nr_consecutive = 1; nr_consecutive < ps->current_committed;
739 nr_consecutive++) {
740 read_exception(ps, ps->current_committed - 1 - nr_consecutive,
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100741 &ce);
742 if (ce.old_chunk != *last_old_chunk - nr_consecutive ||
743 ce.new_chunk != *last_new_chunk - nr_consecutive)
Mikulas Patocka4454a622009-12-10 23:52:29 +0000744 break;
745 }
746
747 return nr_consecutive;
748}
749
750static int persistent_commit_merge(struct dm_exception_store *store,
751 int nr_merged)
752{
753 int r, i;
754 struct pstore *ps = get_info(store);
755
756 BUG_ON(nr_merged > ps->current_committed);
757
758 for (i = 0; i < nr_merged; i++)
759 clear_exception(ps, ps->current_committed - 1 - i);
760
Mikulas Patocka762a80d2011-08-02 12:32:00 +0100761 r = area_io(ps, WRITE_FLUSH_FUA);
Mikulas Patocka4454a622009-12-10 23:52:29 +0000762 if (r < 0)
763 return r;
764
765 ps->current_committed -= nr_merged;
766
767 /*
768 * At this stage, only persistent_usage() uses ps->next_free, so
769 * we make no attempt to keep ps->next_free strictly accurate
770 * as exceptions may have been committed out-of-order originally.
771 * Once a snapshot has become merging, we set it to the value it
772 * would have held had all the exceptions been committed in order.
773 *
774 * ps->current_area does not get reduced by prepare_merge() until
775 * after commit_merge() has removed the nr_merged previous exceptions.
776 */
Tomohiro Kusumi87c961c2010-08-12 04:13:59 +0100777 ps->next_free = area_location(ps, ps->current_area) +
778 ps->current_committed + 1;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000779
780 return 0;
781}
782
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000783static void persistent_drop_snapshot(struct dm_exception_store *store)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000784{
785 struct pstore *ps = get_info(store);
786
787 ps->valid = 0;
788 if (write_header(ps))
789 DMWARN("write header failed");
790}
791
Jonathan Brassow493df712009-04-02 19:55:31 +0100792static int persistent_ctr(struct dm_exception_store *store,
793 unsigned argc, char **argv)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000794{
795 struct pstore *ps;
796
797 /* allocate the pstore */
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100798 ps = kzalloc(sizeof(*ps), GFP_KERNEL);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000799 if (!ps)
800 return -ENOMEM;
801
Jonathan Brassow71fab002009-04-02 19:55:33 +0100802 ps->store = store;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000803 ps->valid = 1;
804 ps->version = SNAPSHOT_DISK_VERSION;
805 ps->area = NULL;
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100806 ps->zero_area = NULL;
807 ps->header_area = NULL;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000808 ps->next_free = NUM_SNAPSHOT_HDR_CHUNKS + 1; /* header and 1st area */
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000809 ps->current_committed = 0;
810
811 ps->callback_count = 0;
812 atomic_set(&ps->pending_count, 0);
813 ps->callbacks = NULL;
814
Tejun Heo239c8dd2011-01-13 19:59:59 +0000815 ps->metadata_wq = alloc_workqueue("ksnaphd", WQ_MEM_RECLAIM, 0);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000816 if (!ps->metadata_wq) {
817 kfree(ps);
818 DMERR("couldn't start header metadata update thread");
819 return -ENOMEM;
820 }
821
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000822 store->context = ps;
823
824 return 0;
825}
826
Jonathan Brassow1e302a92009-04-02 19:55:35 +0100827static unsigned persistent_status(struct dm_exception_store *store,
828 status_type_t status, char *result,
829 unsigned maxlen)
Jonathan Brassow493df712009-04-02 19:55:31 +0100830{
Jonathan Brassow1e302a92009-04-02 19:55:35 +0100831 unsigned sz = 0;
832
833 switch (status) {
834 case STATUSTYPE_INFO:
835 break;
836 case STATUSTYPE_TABLE:
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000837 DMEMIT(" P %llu", (unsigned long long)store->chunk_size);
Jonathan Brassow1e302a92009-04-02 19:55:35 +0100838 }
Jonathan Brassow493df712009-04-02 19:55:31 +0100839
840 return sz;
841}
842
843static struct dm_exception_store_type _persistent_type = {
844 .name = "persistent",
845 .module = THIS_MODULE,
846 .ctr = persistent_ctr,
847 .dtr = persistent_dtr,
848 .read_metadata = persistent_read_metadata,
849 .prepare_exception = persistent_prepare_exception,
850 .commit_exception = persistent_commit_exception,
Mikulas Patocka4454a622009-12-10 23:52:29 +0000851 .prepare_merge = persistent_prepare_merge,
852 .commit_merge = persistent_commit_merge,
Jonathan Brassow493df712009-04-02 19:55:31 +0100853 .drop_snapshot = persistent_drop_snapshot,
Mike Snitzer985903b2009-12-10 23:52:11 +0000854 .usage = persistent_usage,
Jonathan Brassow493df712009-04-02 19:55:31 +0100855 .status = persistent_status,
856};
857
858static struct dm_exception_store_type _persistent_compat_type = {
859 .name = "P",
860 .module = THIS_MODULE,
861 .ctr = persistent_ctr,
862 .dtr = persistent_dtr,
863 .read_metadata = persistent_read_metadata,
864 .prepare_exception = persistent_prepare_exception,
865 .commit_exception = persistent_commit_exception,
Mikulas Patocka4454a622009-12-10 23:52:29 +0000866 .prepare_merge = persistent_prepare_merge,
867 .commit_merge = persistent_commit_merge,
Jonathan Brassow493df712009-04-02 19:55:31 +0100868 .drop_snapshot = persistent_drop_snapshot,
Mike Snitzer985903b2009-12-10 23:52:11 +0000869 .usage = persistent_usage,
Jonathan Brassow493df712009-04-02 19:55:31 +0100870 .status = persistent_status,
871};
872
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000873int dm_persistent_snapshot_init(void)
874{
Jonathan Brassow493df712009-04-02 19:55:31 +0100875 int r;
876
877 r = dm_exception_store_type_register(&_persistent_type);
878 if (r) {
879 DMERR("Unable to register persistent exception store type");
880 return r;
881 }
882
883 r = dm_exception_store_type_register(&_persistent_compat_type);
884 if (r) {
885 DMERR("Unable to register old-style persistent exception "
886 "store type");
887 dm_exception_store_type_unregister(&_persistent_type);
888 return r;
889 }
890
891 return r;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000892}
893
894void dm_persistent_snapshot_exit(void)
895{
Jonathan Brassow493df712009-04-02 19:55:31 +0100896 dm_exception_store_type_unregister(&_persistent_type);
897 dm_exception_store_type_unregister(&_persistent_compat_type);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000898}