blob: d1f1d70171038cf9836832b68c63ad4b1d15297a [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 {
Alasdair G Kergon283a8322011-08-02 12:32:01 +010061 __le32 magic;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000062
63 /*
64 * Is this snapshot valid. There is no way of recovering
65 * an invalid snapshot.
66 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +010067 __le32 valid;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000068
69 /*
70 * Simple, incrementing version. no backward
71 * compatibility.
72 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +010073 __le32 version;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000074
75 /* In sectors */
Alasdair G Kergon283a8322011-08-02 12:32:01 +010076 __le32 chunk_size;
77} __packed;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000078
79struct disk_exception {
Alasdair G Kergon283a8322011-08-02 12:32:01 +010080 __le64 old_chunk;
81 __le64 new_chunk;
82} __packed;
83
84struct core_exception {
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000085 uint64_t old_chunk;
86 uint64_t new_chunk;
87};
88
89struct commit_callback {
90 void (*callback)(void *, int success);
91 void *context;
92};
93
94/*
95 * The top level structure for a persistent exception store.
96 */
97struct pstore {
Jonathan Brassow71fab002009-04-02 19:55:33 +010098 struct dm_exception_store *store;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000099 int version;
100 int valid;
101 uint32_t exceptions_per_area;
102
103 /*
104 * Now that we have an asynchronous kcopyd there is no
105 * need for large chunk sizes, so it wont hurt to have a
106 * whole chunks worth of metadata in memory at once.
107 */
108 void *area;
109
110 /*
111 * An area of zeros used to clear the next area.
112 */
113 void *zero_area;
114
115 /*
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100116 * An area used for header. The header can be written
117 * concurrently with metadata (when invalidating the snapshot),
118 * so it needs a separate buffer.
119 */
120 void *header_area;
121
122 /*
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000123 * Used to keep track of which metadata area the data in
124 * 'chunk' refers to.
125 */
126 chunk_t current_area;
127
128 /*
129 * The next free chunk for an exception.
Mikulas Patocka4454a622009-12-10 23:52:29 +0000130 *
131 * When creating exceptions, all the chunks here and above are
132 * free. It holds the next chunk to be allocated. On rare
133 * occasions (e.g. after a system crash) holes can be left in
134 * the exception store because chunks can be committed out of
135 * order.
136 *
137 * When merging exceptions, it does not necessarily mean all the
138 * chunks here and above are free. It holds the value it would
139 * have held if all chunks had been committed in order of
140 * allocation. Consequently the value may occasionally be
141 * slightly too low, but since it's only used for 'status' and
142 * it can never reach its minimum value too early this doesn't
143 * matter.
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000144 */
Mikulas Patocka4454a622009-12-10 23:52:29 +0000145
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000146 chunk_t next_free;
147
148 /*
149 * The index of next free exception in the current
150 * metadata area.
151 */
152 uint32_t current_committed;
153
154 atomic_t pending_count;
155 uint32_t callback_count;
156 struct commit_callback *callbacks;
157 struct dm_io_client *io_client;
158
159 struct workqueue_struct *metadata_wq;
160};
161
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000162static 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
Joe Perchese29e65a2011-08-02 12:32:02 +0100177 ps->zero_area = vzalloc(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
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100181 ps->header_area = vmalloc(len);
182 if (!ps->header_area)
183 goto err_header_area;
184
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000185 return 0;
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100186
187err_header_area:
188 vfree(ps->zero_area);
189
190err_zero_area:
191 vfree(ps->area);
192
193err_area:
194 return r;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000195}
196
197static void free_area(struct pstore *ps)
198{
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100199 if (ps->area)
200 vfree(ps->area);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000201 ps->area = NULL;
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100202
203 if (ps->zero_area)
204 vfree(ps->zero_area);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000205 ps->zero_area = NULL;
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100206
207 if (ps->header_area)
208 vfree(ps->header_area);
209 ps->header_area = NULL;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000210}
211
212struct mdata_req {
213 struct dm_io_region *where;
214 struct dm_io_request *io_req;
215 struct work_struct work;
216 int result;
217};
218
219static void do_metadata(struct work_struct *work)
220{
221 struct mdata_req *req = container_of(work, struct mdata_req, work);
222
223 req->result = dm_io(req->io_req, 1, req->where, NULL);
224}
225
226/*
227 * Read or write a chunk aligned and sized block of data from a device.
228 */
Mikulas Patocka02d2fd32009-09-04 20:40:37 +0100229static int chunk_io(struct pstore *ps, void *area, chunk_t chunk, int rw,
230 int metadata)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000231{
232 struct dm_io_region where = {
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000233 .bdev = dm_snap_cow(ps->store->snap)->bdev,
Jonathan Brassow71fab002009-04-02 19:55:33 +0100234 .sector = ps->store->chunk_size * chunk,
235 .count = ps->store->chunk_size,
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000236 };
237 struct dm_io_request io_req = {
238 .bi_rw = rw,
239 .mem.type = DM_IO_VMA,
Mikulas Patocka02d2fd32009-09-04 20:40:37 +0100240 .mem.ptr.vma = area,
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000241 .client = ps->io_client,
242 .notify.fn = NULL,
243 };
244 struct mdata_req req;
245
246 if (!metadata)
247 return dm_io(&io_req, 1, &where, NULL);
248
249 req.where = &where;
250 req.io_req = &io_req;
251
252 /*
253 * Issue the synchronous I/O from a different thread
254 * to avoid generic_make_request recursion.
255 */
Andrew Mortonca1cab32010-10-26 14:22:34 -0700256 INIT_WORK_ONSTACK(&req.work, do_metadata);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000257 queue_work(ps->metadata_wq, &req.work);
Tejun Heo239c8dd2011-01-13 19:59:59 +0000258 flush_work(&req.work);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000259
260 return req.result;
261}
262
263/*
264 * Convert a metadata area index to a chunk index.
265 */
266static chunk_t area_location(struct pstore *ps, chunk_t area)
267{
Tomohiro Kusumi87c961c2010-08-12 04:13:59 +0100268 return NUM_SNAPSHOT_HDR_CHUNKS + ((ps->exceptions_per_area + 1) * area);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000269}
270
271/*
272 * Read or write a metadata area. Remembering to skip the first
273 * chunk which holds the header.
274 */
275static int area_io(struct pstore *ps, int rw)
276{
277 int r;
278 chunk_t chunk;
279
280 chunk = area_location(ps, ps->current_area);
281
Mikulas Patocka02d2fd32009-09-04 20:40:37 +0100282 r = chunk_io(ps, ps->area, chunk, rw, 0);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000283 if (r)
284 return r;
285
286 return 0;
287}
288
289static void zero_memory_area(struct pstore *ps)
290{
Jonathan Brassow71fab002009-04-02 19:55:33 +0100291 memset(ps->area, 0, ps->store->chunk_size << SECTOR_SHIFT);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000292}
293
294static int zero_disk_area(struct pstore *ps, chunk_t area)
295{
Mikulas Patocka02d2fd32009-09-04 20:40:37 +0100296 return chunk_io(ps, ps->zero_area, area_location(ps, area), WRITE, 0);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000297}
298
299static int read_header(struct pstore *ps, int *new_snapshot)
300{
301 int r;
302 struct disk_header *dh;
Mikulas Patockadf96eee2009-10-16 23:18:17 +0100303 unsigned chunk_size;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000304 int chunk_size_supplied = 1;
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100305 char *chunk_err;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000306
307 /*
Mikulas Patockadf96eee2009-10-16 23:18:17 +0100308 * Use default chunk size (or logical_block_size, if larger)
309 * if none supplied
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000310 */
Jonathan Brassow71fab002009-04-02 19:55:33 +0100311 if (!ps->store->chunk_size) {
312 ps->store->chunk_size = max(DM_CHUNK_SIZE_DEFAULT_SECTORS,
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000313 bdev_logical_block_size(dm_snap_cow(ps->store->snap)->
314 bdev) >> 9);
Jonathan Brassow71fab002009-04-02 19:55:33 +0100315 ps->store->chunk_mask = ps->store->chunk_size - 1;
316 ps->store->chunk_shift = ffs(ps->store->chunk_size) - 1;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000317 chunk_size_supplied = 0;
318 }
319
Mikulas Patockabda8efe2011-05-29 13:03:09 +0100320 ps->io_client = dm_io_client_create();
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000321 if (IS_ERR(ps->io_client))
322 return PTR_ERR(ps->io_client);
323
324 r = alloc_area(ps);
325 if (r)
326 return r;
327
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100328 r = chunk_io(ps, ps->header_area, 0, READ, 1);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000329 if (r)
330 goto bad;
331
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100332 dh = ps->header_area;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000333
334 if (le32_to_cpu(dh->magic) == 0) {
335 *new_snapshot = 1;
336 return 0;
337 }
338
339 if (le32_to_cpu(dh->magic) != SNAP_MAGIC) {
340 DMWARN("Invalid or corrupt snapshot");
341 r = -ENXIO;
342 goto bad;
343 }
344
345 *new_snapshot = 0;
346 ps->valid = le32_to_cpu(dh->valid);
347 ps->version = le32_to_cpu(dh->version);
348 chunk_size = le32_to_cpu(dh->chunk_size);
349
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100350 if (ps->store->chunk_size == chunk_size)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000351 return 0;
352
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100353 if (chunk_size_supplied)
Mikulas Patockadf96eee2009-10-16 23:18:17 +0100354 DMWARN("chunk size %u in device metadata overrides "
355 "table chunk size of %u.",
356 chunk_size, ps->store->chunk_size);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000357
358 /* We had a bogus chunk_size. Fix stuff up. */
359 free_area(ps);
360
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100361 r = dm_exception_store_set_chunk_size(ps->store, chunk_size,
362 &chunk_err);
363 if (r) {
Mikulas Patockadf96eee2009-10-16 23:18:17 +0100364 DMERR("invalid on-disk chunk size %u: %s.",
365 chunk_size, chunk_err);
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100366 return r;
367 }
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000368
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000369 r = alloc_area(ps);
370 return r;
371
372bad:
373 free_area(ps);
374 return r;
375}
376
377static int write_header(struct pstore *ps)
378{
379 struct disk_header *dh;
380
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100381 memset(ps->header_area, 0, ps->store->chunk_size << SECTOR_SHIFT);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000382
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100383 dh = ps->header_area;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000384 dh->magic = cpu_to_le32(SNAP_MAGIC);
385 dh->valid = cpu_to_le32(ps->valid);
386 dh->version = cpu_to_le32(ps->version);
Jonathan Brassow71fab002009-04-02 19:55:33 +0100387 dh->chunk_size = cpu_to_le32(ps->store->chunk_size);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000388
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100389 return chunk_io(ps, ps->header_area, 0, WRITE, 1);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000390}
391
392/*
393 * Access functions for the disk exceptions, these do the endian conversions.
394 */
395static struct disk_exception *get_exception(struct pstore *ps, uint32_t index)
396{
397 BUG_ON(index >= ps->exceptions_per_area);
398
399 return ((struct disk_exception *) ps->area) + index;
400}
401
402static void read_exception(struct pstore *ps,
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100403 uint32_t index, struct core_exception *result)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000404{
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100405 struct disk_exception *de = get_exception(ps, index);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000406
407 /* copy it */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100408 result->old_chunk = le64_to_cpu(de->old_chunk);
409 result->new_chunk = le64_to_cpu(de->new_chunk);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000410}
411
412static void write_exception(struct pstore *ps,
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100413 uint32_t index, struct core_exception *e)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000414{
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100415 struct disk_exception *de = get_exception(ps, index);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000416
417 /* copy it */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100418 de->old_chunk = cpu_to_le64(e->old_chunk);
419 de->new_chunk = cpu_to_le64(e->new_chunk);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000420}
421
Mikulas Patocka4454a622009-12-10 23:52:29 +0000422static void clear_exception(struct pstore *ps, uint32_t index)
423{
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100424 struct disk_exception *de = get_exception(ps, index);
Mikulas Patocka4454a622009-12-10 23:52:29 +0000425
426 /* clear it */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100427 de->old_chunk = 0;
428 de->new_chunk = 0;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000429}
430
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000431/*
432 * Registers the exceptions that are present in the current area.
433 * 'full' is filled in to indicate if the area has been
434 * filled.
435 */
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000436static int insert_exceptions(struct pstore *ps,
437 int (*callback)(void *callback_context,
438 chunk_t old, chunk_t new),
439 void *callback_context,
440 int *full)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000441{
442 int r;
443 unsigned int i;
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100444 struct core_exception e;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000445
446 /* presume the area is full */
447 *full = 1;
448
449 for (i = 0; i < ps->exceptions_per_area; i++) {
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100450 read_exception(ps, i, &e);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000451
452 /*
453 * If the new_chunk is pointing at the start of
454 * the COW device, where the first metadata area
455 * is we know that we've hit the end of the
456 * exceptions. Therefore the area is not full.
457 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100458 if (e.new_chunk == 0LL) {
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000459 ps->current_committed = i;
460 *full = 0;
461 break;
462 }
463
464 /*
465 * Keep track of the start of the free chunks.
466 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100467 if (ps->next_free <= e.new_chunk)
468 ps->next_free = e.new_chunk + 1;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000469
470 /*
471 * Otherwise we add the exception to the snapshot.
472 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100473 r = callback(callback_context, e.old_chunk, e.new_chunk);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000474 if (r)
475 return r;
476 }
477
478 return 0;
479}
480
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000481static int read_exceptions(struct pstore *ps,
482 int (*callback)(void *callback_context, chunk_t old,
483 chunk_t new),
484 void *callback_context)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000485{
486 int r, full = 1;
487
488 /*
489 * Keeping reading chunks and inserting exceptions until
490 * we find a partially full area.
491 */
492 for (ps->current_area = 0; full; ps->current_area++) {
493 r = area_io(ps, READ);
494 if (r)
495 return r;
496
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000497 r = insert_exceptions(ps, callback, callback_context, &full);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000498 if (r)
499 return r;
500 }
501
502 ps->current_area--;
503
504 return 0;
505}
506
507static struct pstore *get_info(struct dm_exception_store *store)
508{
509 return (struct pstore *) store->context;
510}
511
Mike Snitzer985903b2009-12-10 23:52:11 +0000512static void persistent_usage(struct dm_exception_store *store,
513 sector_t *total_sectors,
514 sector_t *sectors_allocated,
515 sector_t *metadata_sectors)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000516{
Mike Snitzer985903b2009-12-10 23:52:11 +0000517 struct pstore *ps = get_info(store);
518
519 *sectors_allocated = ps->next_free * store->chunk_size;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000520 *total_sectors = get_dev_size(dm_snap_cow(store->snap)->bdev);
Mike Snitzer985903b2009-12-10 23:52:11 +0000521
522 /*
523 * First chunk is the fixed header.
524 * Then there are (ps->current_area + 1) metadata chunks, each one
525 * separated from the next by ps->exceptions_per_area data chunks.
526 */
Mikulas Patocka4454a622009-12-10 23:52:29 +0000527 *metadata_sectors = (ps->current_area + 1 + NUM_SNAPSHOT_HDR_CHUNKS) *
528 store->chunk_size;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000529}
530
Jonathan Brassow493df712009-04-02 19:55:31 +0100531static void persistent_dtr(struct dm_exception_store *store)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000532{
533 struct pstore *ps = get_info(store);
534
535 destroy_workqueue(ps->metadata_wq);
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100536
537 /* Created in read_header */
538 if (ps->io_client)
539 dm_io_client_destroy(ps->io_client);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000540 free_area(ps);
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100541
542 /* Allocated in persistent_read_metadata */
543 if (ps->callbacks)
544 vfree(ps->callbacks);
545
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000546 kfree(ps);
547}
548
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000549static int persistent_read_metadata(struct dm_exception_store *store,
550 int (*callback)(void *callback_context,
551 chunk_t old, chunk_t new),
552 void *callback_context)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000553{
554 int r, uninitialized_var(new_snapshot);
555 struct pstore *ps = get_info(store);
556
557 /*
558 * Read the snapshot header.
559 */
560 r = read_header(ps, &new_snapshot);
561 if (r)
562 return r;
563
564 /*
565 * Now we know correct chunk_size, complete the initialisation.
566 */
Jonathan Brassow71fab002009-04-02 19:55:33 +0100567 ps->exceptions_per_area = (ps->store->chunk_size << SECTOR_SHIFT) /
568 sizeof(struct disk_exception);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000569 ps->callbacks = dm_vcalloc(ps->exceptions_per_area,
Jonathan Brassowa2d2b032011-08-02 12:32:03 +0100570 sizeof(*ps->callbacks));
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000571 if (!ps->callbacks)
572 return -ENOMEM;
573
574 /*
575 * Do we need to setup a new snapshot ?
576 */
577 if (new_snapshot) {
578 r = write_header(ps);
579 if (r) {
580 DMWARN("write_header failed");
581 return r;
582 }
583
584 ps->current_area = 0;
585 zero_memory_area(ps);
586 r = zero_disk_area(ps, 0);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000587 if (r)
Jon Brassowf5acc832009-12-10 23:52:07 +0000588 DMWARN("zero_disk_area(0) failed");
589 return r;
590 }
591 /*
592 * Sanity checks.
593 */
594 if (ps->version != SNAPSHOT_DISK_VERSION) {
595 DMWARN("unable to handle snapshot disk version %d",
596 ps->version);
597 return -EINVAL;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000598 }
599
Jon Brassowf5acc832009-12-10 23:52:07 +0000600 /*
601 * Metadata are valid, but snapshot is invalidated
602 */
603 if (!ps->valid)
604 return 1;
605
606 /*
607 * Read the metadata.
608 */
609 r = read_exceptions(ps, callback, callback_context);
610
611 return r;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000612}
613
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000614static int persistent_prepare_exception(struct dm_exception_store *store,
Jon Brassow1d4989c2009-12-10 23:52:10 +0000615 struct dm_exception *e)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000616{
617 struct pstore *ps = get_info(store);
618 uint32_t stride;
619 chunk_t next_free;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000620 sector_t size = get_dev_size(dm_snap_cow(store->snap)->bdev);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000621
622 /* Is there enough room ? */
Jonathan Brassowd0216842009-04-02 19:55:32 +0100623 if (size < ((ps->next_free + 1) * store->chunk_size))
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000624 return -ENOSPC;
625
626 e->new_chunk = ps->next_free;
627
628 /*
629 * Move onto the next free pending, making sure to take
630 * into account the location of the metadata chunks.
631 */
632 stride = (ps->exceptions_per_area + 1);
633 next_free = ++ps->next_free;
634 if (sector_div(next_free, stride) == 1)
635 ps->next_free++;
636
637 atomic_inc(&ps->pending_count);
638 return 0;
639}
640
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000641static void persistent_commit_exception(struct dm_exception_store *store,
Jon Brassow1d4989c2009-12-10 23:52:10 +0000642 struct dm_exception *e,
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000643 void (*callback) (void *, int success),
644 void *callback_context)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000645{
646 unsigned int i;
647 struct pstore *ps = get_info(store);
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100648 struct core_exception ce;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000649 struct commit_callback *cb;
650
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100651 ce.old_chunk = e->old_chunk;
652 ce.new_chunk = e->new_chunk;
653 write_exception(ps, ps->current_committed++, &ce);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000654
655 /*
656 * Add the callback to the back of the array. This code
657 * is the only place where the callback array is
658 * manipulated, and we know that it will never be called
659 * multiple times concurrently.
660 */
661 cb = ps->callbacks + ps->callback_count++;
662 cb->callback = callback;
663 cb->context = callback_context;
664
665 /*
666 * If there are exceptions in flight and we have not yet
667 * filled this metadata area there's nothing more to do.
668 */
669 if (!atomic_dec_and_test(&ps->pending_count) &&
670 (ps->current_committed != ps->exceptions_per_area))
671 return;
672
673 /*
674 * If we completely filled the current area, then wipe the next one.
675 */
676 if ((ps->current_committed == ps->exceptions_per_area) &&
Jonathan Brassowa2d2b032011-08-02 12:32:03 +0100677 zero_disk_area(ps, ps->current_area + 1))
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000678 ps->valid = 0;
679
680 /*
681 * Commit exceptions to disk.
682 */
Tejun Heod87f4c12010-09-03 11:56:19 +0200683 if (ps->valid && area_io(ps, WRITE_FLUSH_FUA))
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000684 ps->valid = 0;
685
686 /*
687 * Advance to the next area if this one is full.
688 */
689 if (ps->current_committed == ps->exceptions_per_area) {
690 ps->current_committed = 0;
691 ps->current_area++;
692 zero_memory_area(ps);
693 }
694
695 for (i = 0; i < ps->callback_count; i++) {
696 cb = ps->callbacks + i;
697 cb->callback(cb->context, ps->valid);
698 }
699
700 ps->callback_count = 0;
701}
702
Mikulas Patocka4454a622009-12-10 23:52:29 +0000703static int persistent_prepare_merge(struct dm_exception_store *store,
704 chunk_t *last_old_chunk,
705 chunk_t *last_new_chunk)
706{
707 struct pstore *ps = get_info(store);
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100708 struct core_exception ce;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000709 int nr_consecutive;
710 int r;
711
712 /*
713 * When current area is empty, move back to preceding area.
714 */
715 if (!ps->current_committed) {
716 /*
717 * Have we finished?
718 */
719 if (!ps->current_area)
720 return 0;
721
722 ps->current_area--;
723 r = area_io(ps, READ);
724 if (r < 0)
725 return r;
726 ps->current_committed = ps->exceptions_per_area;
727 }
728
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100729 read_exception(ps, ps->current_committed - 1, &ce);
730 *last_old_chunk = ce.old_chunk;
731 *last_new_chunk = ce.new_chunk;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000732
733 /*
734 * Find number of consecutive chunks within the current area,
735 * working backwards.
736 */
737 for (nr_consecutive = 1; nr_consecutive < ps->current_committed;
738 nr_consecutive++) {
739 read_exception(ps, ps->current_committed - 1 - nr_consecutive,
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100740 &ce);
741 if (ce.old_chunk != *last_old_chunk - nr_consecutive ||
742 ce.new_chunk != *last_new_chunk - nr_consecutive)
Mikulas Patocka4454a622009-12-10 23:52:29 +0000743 break;
744 }
745
746 return nr_consecutive;
747}
748
749static int persistent_commit_merge(struct dm_exception_store *store,
750 int nr_merged)
751{
752 int r, i;
753 struct pstore *ps = get_info(store);
754
755 BUG_ON(nr_merged > ps->current_committed);
756
757 for (i = 0; i < nr_merged; i++)
758 clear_exception(ps, ps->current_committed - 1 - i);
759
Mikulas Patocka762a80d2011-08-02 12:32:00 +0100760 r = area_io(ps, WRITE_FLUSH_FUA);
Mikulas Patocka4454a622009-12-10 23:52:29 +0000761 if (r < 0)
762 return r;
763
764 ps->current_committed -= nr_merged;
765
766 /*
767 * At this stage, only persistent_usage() uses ps->next_free, so
768 * we make no attempt to keep ps->next_free strictly accurate
769 * as exceptions may have been committed out-of-order originally.
770 * Once a snapshot has become merging, we set it to the value it
771 * would have held had all the exceptions been committed in order.
772 *
773 * ps->current_area does not get reduced by prepare_merge() until
774 * after commit_merge() has removed the nr_merged previous exceptions.
775 */
Tomohiro Kusumi87c961c2010-08-12 04:13:59 +0100776 ps->next_free = area_location(ps, ps->current_area) +
777 ps->current_committed + 1;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000778
779 return 0;
780}
781
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000782static void persistent_drop_snapshot(struct dm_exception_store *store)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000783{
784 struct pstore *ps = get_info(store);
785
786 ps->valid = 0;
787 if (write_header(ps))
788 DMWARN("write header failed");
789}
790
Jonathan Brassow493df712009-04-02 19:55:31 +0100791static int persistent_ctr(struct dm_exception_store *store,
792 unsigned argc, char **argv)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000793{
794 struct pstore *ps;
795
796 /* allocate the pstore */
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100797 ps = kzalloc(sizeof(*ps), GFP_KERNEL);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000798 if (!ps)
799 return -ENOMEM;
800
Jonathan Brassow71fab002009-04-02 19:55:33 +0100801 ps->store = store;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000802 ps->valid = 1;
803 ps->version = SNAPSHOT_DISK_VERSION;
804 ps->area = NULL;
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100805 ps->zero_area = NULL;
806 ps->header_area = NULL;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000807 ps->next_free = NUM_SNAPSHOT_HDR_CHUNKS + 1; /* header and 1st area */
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000808 ps->current_committed = 0;
809
810 ps->callback_count = 0;
811 atomic_set(&ps->pending_count, 0);
812 ps->callbacks = NULL;
813
Tejun Heo239c8dd2011-01-13 19:59:59 +0000814 ps->metadata_wq = alloc_workqueue("ksnaphd", WQ_MEM_RECLAIM, 0);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000815 if (!ps->metadata_wq) {
816 kfree(ps);
817 DMERR("couldn't start header metadata update thread");
818 return -ENOMEM;
819 }
820
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000821 store->context = ps;
822
823 return 0;
824}
825
Jonathan Brassow1e302a92009-04-02 19:55:35 +0100826static unsigned persistent_status(struct dm_exception_store *store,
827 status_type_t status, char *result,
828 unsigned maxlen)
Jonathan Brassow493df712009-04-02 19:55:31 +0100829{
Jonathan Brassow1e302a92009-04-02 19:55:35 +0100830 unsigned sz = 0;
831
832 switch (status) {
833 case STATUSTYPE_INFO:
834 break;
835 case STATUSTYPE_TABLE:
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000836 DMEMIT(" P %llu", (unsigned long long)store->chunk_size);
Jonathan Brassow1e302a92009-04-02 19:55:35 +0100837 }
Jonathan Brassow493df712009-04-02 19:55:31 +0100838
839 return sz;
840}
841
842static struct dm_exception_store_type _persistent_type = {
843 .name = "persistent",
844 .module = THIS_MODULE,
845 .ctr = persistent_ctr,
846 .dtr = persistent_dtr,
847 .read_metadata = persistent_read_metadata,
848 .prepare_exception = persistent_prepare_exception,
849 .commit_exception = persistent_commit_exception,
Mikulas Patocka4454a622009-12-10 23:52:29 +0000850 .prepare_merge = persistent_prepare_merge,
851 .commit_merge = persistent_commit_merge,
Jonathan Brassow493df712009-04-02 19:55:31 +0100852 .drop_snapshot = persistent_drop_snapshot,
Mike Snitzer985903b2009-12-10 23:52:11 +0000853 .usage = persistent_usage,
Jonathan Brassow493df712009-04-02 19:55:31 +0100854 .status = persistent_status,
855};
856
857static struct dm_exception_store_type _persistent_compat_type = {
858 .name = "P",
859 .module = THIS_MODULE,
860 .ctr = persistent_ctr,
861 .dtr = persistent_dtr,
862 .read_metadata = persistent_read_metadata,
863 .prepare_exception = persistent_prepare_exception,
864 .commit_exception = persistent_commit_exception,
Mikulas Patocka4454a622009-12-10 23:52:29 +0000865 .prepare_merge = persistent_prepare_merge,
866 .commit_merge = persistent_commit_merge,
Jonathan Brassow493df712009-04-02 19:55:31 +0100867 .drop_snapshot = persistent_drop_snapshot,
Mike Snitzer985903b2009-12-10 23:52:11 +0000868 .usage = persistent_usage,
Jonathan Brassow493df712009-04-02 19:55:31 +0100869 .status = persistent_status,
870};
871
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000872int dm_persistent_snapshot_init(void)
873{
Jonathan Brassow493df712009-04-02 19:55:31 +0100874 int r;
875
876 r = dm_exception_store_type_register(&_persistent_type);
877 if (r) {
878 DMERR("Unable to register persistent exception store type");
879 return r;
880 }
881
882 r = dm_exception_store_type_register(&_persistent_compat_type);
883 if (r) {
884 DMERR("Unable to register old-style persistent exception "
885 "store type");
886 dm_exception_store_type_unregister(&_persistent_type);
887 return r;
888 }
889
890 return r;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000891}
892
893void dm_persistent_snapshot_exit(void)
894{
Jonathan Brassow493df712009-04-02 19:55:31 +0100895 dm_exception_store_type_unregister(&_persistent_type);
896 dm_exception_store_type_unregister(&_persistent_compat_type);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000897}