blob: 169275050c0b76ccf41bdbec58812728bde9b461 [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>
Mikulas Patocka55494bf2014-01-13 19:12:36 -050016#include "dm-bufio.h"
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000017
18#define DM_MSG_PREFIX "persistent snapshot"
19#define DM_CHUNK_SIZE_DEFAULT_SECTORS 32 /* 16KB */
20
21/*-----------------------------------------------------------------
22 * Persistent snapshots, by persistent we mean that the snapshot
23 * will survive a reboot.
24 *---------------------------------------------------------------*/
25
26/*
27 * We need to store a record of which parts of the origin have
28 * been copied to the snapshot device. The snapshot code
29 * requires that we copy exception chunks to chunk aligned areas
30 * of the COW store. It makes sense therefore, to store the
31 * metadata in chunk size blocks.
32 *
33 * There is no backward or forward compatibility implemented,
34 * snapshots with different disk versions than the kernel will
35 * not be usable. It is expected that "lvcreate" will blank out
36 * the start of a fresh COW device before calling the snapshot
37 * constructor.
38 *
39 * The first chunk of the COW device just contains the header.
40 * After this there is a chunk filled with exception metadata,
41 * followed by as many exception chunks as can fit in the
42 * metadata areas.
43 *
44 * All on disk structures are in little-endian format. The end
45 * of the exceptions info is indicated by an exception with a
46 * new_chunk of 0, which is invalid since it would point to the
47 * header chunk.
48 */
49
50/*
51 * Magic for persistent snapshots: "SnAp" - Feeble isn't it.
52 */
53#define SNAP_MAGIC 0x70416e53
54
55/*
56 * The on-disk version of the metadata.
57 */
58#define SNAPSHOT_DISK_VERSION 1
59
Mikulas Patocka4454a622009-12-10 23:52:29 +000060#define NUM_SNAPSHOT_HDR_CHUNKS 1
61
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000062struct disk_header {
Alasdair G Kergon283a8322011-08-02 12:32:01 +010063 __le32 magic;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000064
65 /*
66 * Is this snapshot valid. There is no way of recovering
67 * an invalid snapshot.
68 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +010069 __le32 valid;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000070
71 /*
72 * Simple, incrementing version. no backward
73 * compatibility.
74 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +010075 __le32 version;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000076
77 /* In sectors */
Alasdair G Kergon283a8322011-08-02 12:32:01 +010078 __le32 chunk_size;
79} __packed;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000080
81struct disk_exception {
Alasdair G Kergon283a8322011-08-02 12:32:01 +010082 __le64 old_chunk;
83 __le64 new_chunk;
84} __packed;
85
86struct core_exception {
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000087 uint64_t old_chunk;
88 uint64_t new_chunk;
89};
90
91struct commit_callback {
92 void (*callback)(void *, int success);
93 void *context;
94};
95
96/*
97 * The top level structure for a persistent exception store.
98 */
99struct pstore {
Jonathan Brassow71fab002009-04-02 19:55:33 +0100100 struct dm_exception_store *store;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000101 int version;
102 int valid;
103 uint32_t exceptions_per_area;
104
105 /*
106 * Now that we have an asynchronous kcopyd there is no
107 * need for large chunk sizes, so it wont hurt to have a
108 * whole chunks worth of metadata in memory at once.
109 */
110 void *area;
111
112 /*
113 * An area of zeros used to clear the next area.
114 */
115 void *zero_area;
116
117 /*
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100118 * An area used for header. The header can be written
119 * concurrently with metadata (when invalidating the snapshot),
120 * so it needs a separate buffer.
121 */
122 void *header_area;
123
124 /*
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000125 * Used to keep track of which metadata area the data in
126 * 'chunk' refers to.
127 */
128 chunk_t current_area;
129
130 /*
131 * The next free chunk for an exception.
Mikulas Patocka4454a622009-12-10 23:52:29 +0000132 *
133 * When creating exceptions, all the chunks here and above are
134 * free. It holds the next chunk to be allocated. On rare
135 * occasions (e.g. after a system crash) holes can be left in
136 * the exception store because chunks can be committed out of
137 * order.
138 *
139 * When merging exceptions, it does not necessarily mean all the
140 * chunks here and above are free. It holds the value it would
141 * have held if all chunks had been committed in order of
142 * allocation. Consequently the value may occasionally be
143 * slightly too low, but since it's only used for 'status' and
144 * it can never reach its minimum value too early this doesn't
145 * matter.
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000146 */
Mikulas Patocka4454a622009-12-10 23:52:29 +0000147
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000148 chunk_t next_free;
149
150 /*
151 * The index of next free exception in the current
152 * metadata area.
153 */
154 uint32_t current_committed;
155
156 atomic_t pending_count;
157 uint32_t callback_count;
158 struct commit_callback *callbacks;
159 struct dm_io_client *io_client;
160
161 struct workqueue_struct *metadata_wq;
162};
163
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000164static int alloc_area(struct pstore *ps)
165{
166 int r = -ENOMEM;
167 size_t len;
168
Jonathan Brassow71fab002009-04-02 19:55:33 +0100169 len = ps->store->chunk_size << SECTOR_SHIFT;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000170
171 /*
172 * Allocate the chunk_size block of memory that will hold
173 * a single metadata area.
174 */
175 ps->area = vmalloc(len);
176 if (!ps->area)
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100177 goto err_area;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000178
Joe Perchese29e65a2011-08-02 12:32:02 +0100179 ps->zero_area = vzalloc(len);
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100180 if (!ps->zero_area)
181 goto err_zero_area;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000182
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100183 ps->header_area = vmalloc(len);
184 if (!ps->header_area)
185 goto err_header_area;
186
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000187 return 0;
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100188
189err_header_area:
190 vfree(ps->zero_area);
191
192err_zero_area:
193 vfree(ps->area);
194
195err_area:
196 return r;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000197}
198
199static void free_area(struct pstore *ps)
200{
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100201 if (ps->area)
202 vfree(ps->area);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000203 ps->area = NULL;
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100204
205 if (ps->zero_area)
206 vfree(ps->zero_area);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000207 ps->zero_area = NULL;
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100208
209 if (ps->header_area)
210 vfree(ps->header_area);
211 ps->header_area = NULL;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000212}
213
214struct mdata_req {
215 struct dm_io_region *where;
216 struct dm_io_request *io_req;
217 struct work_struct work;
218 int result;
219};
220
221static void do_metadata(struct work_struct *work)
222{
223 struct mdata_req *req = container_of(work, struct mdata_req, work);
224
225 req->result = dm_io(req->io_req, 1, req->where, NULL);
226}
227
228/*
229 * Read or write a chunk aligned and sized block of data from a device.
230 */
Mikulas Patocka02d2fd32009-09-04 20:40:37 +0100231static int chunk_io(struct pstore *ps, void *area, chunk_t chunk, int rw,
232 int metadata)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000233{
234 struct dm_io_region where = {
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000235 .bdev = dm_snap_cow(ps->store->snap)->bdev,
Jonathan Brassow71fab002009-04-02 19:55:33 +0100236 .sector = ps->store->chunk_size * chunk,
237 .count = ps->store->chunk_size,
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000238 };
239 struct dm_io_request io_req = {
240 .bi_rw = rw,
241 .mem.type = DM_IO_VMA,
Mikulas Patocka02d2fd32009-09-04 20:40:37 +0100242 .mem.ptr.vma = area,
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000243 .client = ps->io_client,
244 .notify.fn = NULL,
245 };
246 struct mdata_req req;
247
248 if (!metadata)
249 return dm_io(&io_req, 1, &where, NULL);
250
251 req.where = &where;
252 req.io_req = &io_req;
253
254 /*
255 * Issue the synchronous I/O from a different thread
256 * to avoid generic_make_request recursion.
257 */
Andrew Mortonca1cab32010-10-26 14:22:34 -0700258 INIT_WORK_ONSTACK(&req.work, do_metadata);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000259 queue_work(ps->metadata_wq, &req.work);
Mikulas Patocka5ea330a2013-09-18 19:14:22 -0400260 flush_workqueue(ps->metadata_wq);
Chuansheng Liuc1a64162014-01-07 16:56:18 +0800261 destroy_work_on_stack(&req.work);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000262
263 return req.result;
264}
265
266/*
267 * Convert a metadata area index to a chunk index.
268 */
269static chunk_t area_location(struct pstore *ps, chunk_t area)
270{
Tomohiro Kusumi87c961c2010-08-12 04:13:59 +0100271 return NUM_SNAPSHOT_HDR_CHUNKS + ((ps->exceptions_per_area + 1) * area);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000272}
273
Mikulas Patockae9c6a182013-10-16 03:17:47 +0100274static void skip_metadata(struct pstore *ps)
275{
276 uint32_t stride = ps->exceptions_per_area + 1;
277 chunk_t next_free = ps->next_free;
278 if (sector_div(next_free, stride) == NUM_SNAPSHOT_HDR_CHUNKS)
279 ps->next_free++;
280}
281
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000282/*
283 * Read or write a metadata area. Remembering to skip the first
284 * chunk which holds the header.
285 */
286static int area_io(struct pstore *ps, int rw)
287{
288 int r;
289 chunk_t chunk;
290
291 chunk = area_location(ps, ps->current_area);
292
Mikulas Patocka02d2fd32009-09-04 20:40:37 +0100293 r = chunk_io(ps, ps->area, chunk, rw, 0);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000294 if (r)
295 return r;
296
297 return 0;
298}
299
300static void zero_memory_area(struct pstore *ps)
301{
Jonathan Brassow71fab002009-04-02 19:55:33 +0100302 memset(ps->area, 0, ps->store->chunk_size << SECTOR_SHIFT);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000303}
304
305static int zero_disk_area(struct pstore *ps, chunk_t area)
306{
Mikulas Patocka02d2fd32009-09-04 20:40:37 +0100307 return chunk_io(ps, ps->zero_area, area_location(ps, area), WRITE, 0);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000308}
309
310static int read_header(struct pstore *ps, int *new_snapshot)
311{
312 int r;
313 struct disk_header *dh;
Mikulas Patockadf96eee2009-10-16 23:18:17 +0100314 unsigned chunk_size;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000315 int chunk_size_supplied = 1;
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100316 char *chunk_err;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000317
318 /*
Mikulas Patockadf96eee2009-10-16 23:18:17 +0100319 * Use default chunk size (or logical_block_size, if larger)
320 * if none supplied
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000321 */
Jonathan Brassow71fab002009-04-02 19:55:33 +0100322 if (!ps->store->chunk_size) {
323 ps->store->chunk_size = max(DM_CHUNK_SIZE_DEFAULT_SECTORS,
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000324 bdev_logical_block_size(dm_snap_cow(ps->store->snap)->
325 bdev) >> 9);
Jonathan Brassow71fab002009-04-02 19:55:33 +0100326 ps->store->chunk_mask = ps->store->chunk_size - 1;
327 ps->store->chunk_shift = ffs(ps->store->chunk_size) - 1;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000328 chunk_size_supplied = 0;
329 }
330
Mikulas Patockabda8efe2011-05-29 13:03:09 +0100331 ps->io_client = dm_io_client_create();
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000332 if (IS_ERR(ps->io_client))
333 return PTR_ERR(ps->io_client);
334
335 r = alloc_area(ps);
336 if (r)
337 return r;
338
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100339 r = chunk_io(ps, ps->header_area, 0, READ, 1);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000340 if (r)
341 goto bad;
342
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100343 dh = ps->header_area;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000344
345 if (le32_to_cpu(dh->magic) == 0) {
346 *new_snapshot = 1;
347 return 0;
348 }
349
350 if (le32_to_cpu(dh->magic) != SNAP_MAGIC) {
351 DMWARN("Invalid or corrupt snapshot");
352 r = -ENXIO;
353 goto bad;
354 }
355
356 *new_snapshot = 0;
357 ps->valid = le32_to_cpu(dh->valid);
358 ps->version = le32_to_cpu(dh->version);
359 chunk_size = le32_to_cpu(dh->chunk_size);
360
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100361 if (ps->store->chunk_size == chunk_size)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000362 return 0;
363
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100364 if (chunk_size_supplied)
Mikulas Patockadf96eee2009-10-16 23:18:17 +0100365 DMWARN("chunk size %u in device metadata overrides "
366 "table chunk size of %u.",
367 chunk_size, ps->store->chunk_size);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000368
369 /* We had a bogus chunk_size. Fix stuff up. */
370 free_area(ps);
371
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100372 r = dm_exception_store_set_chunk_size(ps->store, chunk_size,
373 &chunk_err);
374 if (r) {
Mikulas Patockadf96eee2009-10-16 23:18:17 +0100375 DMERR("invalid on-disk chunk size %u: %s.",
376 chunk_size, chunk_err);
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100377 return r;
378 }
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000379
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000380 r = alloc_area(ps);
381 return r;
382
383bad:
384 free_area(ps);
385 return r;
386}
387
388static int write_header(struct pstore *ps)
389{
390 struct disk_header *dh;
391
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100392 memset(ps->header_area, 0, ps->store->chunk_size << SECTOR_SHIFT);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000393
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100394 dh = ps->header_area;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000395 dh->magic = cpu_to_le32(SNAP_MAGIC);
396 dh->valid = cpu_to_le32(ps->valid);
397 dh->version = cpu_to_le32(ps->version);
Jonathan Brassow71fab002009-04-02 19:55:33 +0100398 dh->chunk_size = cpu_to_le32(ps->store->chunk_size);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000399
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100400 return chunk_io(ps, ps->header_area, 0, WRITE, 1);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000401}
402
403/*
404 * Access functions for the disk exceptions, these do the endian conversions.
405 */
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500406static struct disk_exception *get_exception(struct pstore *ps, void *ps_area,
407 uint32_t index)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000408{
409 BUG_ON(index >= ps->exceptions_per_area);
410
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500411 return ((struct disk_exception *) ps_area) + index;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000412}
413
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500414static void read_exception(struct pstore *ps, void *ps_area,
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100415 uint32_t index, struct core_exception *result)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000416{
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500417 struct disk_exception *de = get_exception(ps, ps_area, index);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000418
419 /* copy it */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100420 result->old_chunk = le64_to_cpu(de->old_chunk);
421 result->new_chunk = le64_to_cpu(de->new_chunk);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000422}
423
424static void write_exception(struct pstore *ps,
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100425 uint32_t index, struct core_exception *e)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000426{
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500427 struct disk_exception *de = get_exception(ps, ps->area, index);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000428
429 /* copy it */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100430 de->old_chunk = cpu_to_le64(e->old_chunk);
431 de->new_chunk = cpu_to_le64(e->new_chunk);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000432}
433
Mikulas Patocka4454a622009-12-10 23:52:29 +0000434static void clear_exception(struct pstore *ps, uint32_t index)
435{
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500436 struct disk_exception *de = get_exception(ps, ps->area, index);
Mikulas Patocka4454a622009-12-10 23:52:29 +0000437
438 /* clear it */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100439 de->old_chunk = 0;
440 de->new_chunk = 0;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000441}
442
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000443/*
444 * Registers the exceptions that are present in the current area.
445 * 'full' is filled in to indicate if the area has been
446 * filled.
447 */
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500448static int insert_exceptions(struct pstore *ps, void *ps_area,
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000449 int (*callback)(void *callback_context,
450 chunk_t old, chunk_t new),
451 void *callback_context,
452 int *full)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000453{
454 int r;
455 unsigned int i;
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100456 struct core_exception e;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000457
458 /* presume the area is full */
459 *full = 1;
460
461 for (i = 0; i < ps->exceptions_per_area; i++) {
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500462 read_exception(ps, ps_area, i, &e);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000463
464 /*
465 * If the new_chunk is pointing at the start of
466 * the COW device, where the first metadata area
467 * is we know that we've hit the end of the
468 * exceptions. Therefore the area is not full.
469 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100470 if (e.new_chunk == 0LL) {
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000471 ps->current_committed = i;
472 *full = 0;
473 break;
474 }
475
476 /*
477 * Keep track of the start of the free chunks.
478 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100479 if (ps->next_free <= e.new_chunk)
480 ps->next_free = e.new_chunk + 1;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000481
482 /*
483 * Otherwise we add the exception to the snapshot.
484 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100485 r = callback(callback_context, e.old_chunk, e.new_chunk);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000486 if (r)
487 return r;
488 }
489
490 return 0;
491}
492
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000493static int read_exceptions(struct pstore *ps,
494 int (*callback)(void *callback_context, chunk_t old,
495 chunk_t new),
496 void *callback_context)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000497{
498 int r, full = 1;
Mikulas Patocka55494bf2014-01-13 19:12:36 -0500499 struct dm_bufio_client *client;
500
501 client = dm_bufio_client_create(dm_snap_cow(ps->store->snap)->bdev,
502 ps->store->chunk_size << SECTOR_SHIFT,
503 1, 0, NULL, NULL);
504
505 if (IS_ERR(client))
506 return PTR_ERR(client);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000507
508 /*
509 * Keeping reading chunks and inserting exceptions until
510 * we find a partially full area.
511 */
512 for (ps->current_area = 0; full; ps->current_area++) {
Mikulas Patocka55494bf2014-01-13 19:12:36 -0500513 struct dm_buffer *bp;
514 void *area;
515 chunk_t chunk = area_location(ps, ps->current_area);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000516
Mikulas Patocka55494bf2014-01-13 19:12:36 -0500517 area = dm_bufio_read(client, chunk, &bp);
518 if (unlikely(IS_ERR(area))) {
519 r = PTR_ERR(area);
520 goto ret_destroy_bufio;
521 }
522
523 r = insert_exceptions(ps, area, callback, callback_context,
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500524 &full);
Mikulas Patocka55494bf2014-01-13 19:12:36 -0500525
526 dm_bufio_release(bp);
527
528 dm_bufio_forget(client, chunk);
529
530 if (unlikely(r))
531 goto ret_destroy_bufio;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000532 }
533
534 ps->current_area--;
535
Mikulas Patockae9c6a182013-10-16 03:17:47 +0100536 skip_metadata(ps);
537
Mikulas Patocka55494bf2014-01-13 19:12:36 -0500538 r = 0;
539
540ret_destroy_bufio:
541 dm_bufio_client_destroy(client);
542
543 return r;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000544}
545
546static struct pstore *get_info(struct dm_exception_store *store)
547{
548 return (struct pstore *) store->context;
549}
550
Mike Snitzer985903b2009-12-10 23:52:11 +0000551static void persistent_usage(struct dm_exception_store *store,
552 sector_t *total_sectors,
553 sector_t *sectors_allocated,
554 sector_t *metadata_sectors)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000555{
Mike Snitzer985903b2009-12-10 23:52:11 +0000556 struct pstore *ps = get_info(store);
557
558 *sectors_allocated = ps->next_free * store->chunk_size;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000559 *total_sectors = get_dev_size(dm_snap_cow(store->snap)->bdev);
Mike Snitzer985903b2009-12-10 23:52:11 +0000560
561 /*
562 * First chunk is the fixed header.
563 * Then there are (ps->current_area + 1) metadata chunks, each one
564 * separated from the next by ps->exceptions_per_area data chunks.
565 */
Mikulas Patocka4454a622009-12-10 23:52:29 +0000566 *metadata_sectors = (ps->current_area + 1 + NUM_SNAPSHOT_HDR_CHUNKS) *
567 store->chunk_size;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000568}
569
Jonathan Brassow493df712009-04-02 19:55:31 +0100570static void persistent_dtr(struct dm_exception_store *store)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000571{
572 struct pstore *ps = get_info(store);
573
574 destroy_workqueue(ps->metadata_wq);
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100575
576 /* Created in read_header */
577 if (ps->io_client)
578 dm_io_client_destroy(ps->io_client);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000579 free_area(ps);
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100580
581 /* Allocated in persistent_read_metadata */
582 if (ps->callbacks)
583 vfree(ps->callbacks);
584
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000585 kfree(ps);
586}
587
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000588static int persistent_read_metadata(struct dm_exception_store *store,
589 int (*callback)(void *callback_context,
590 chunk_t old, chunk_t new),
591 void *callback_context)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000592{
593 int r, uninitialized_var(new_snapshot);
594 struct pstore *ps = get_info(store);
595
596 /*
597 * Read the snapshot header.
598 */
599 r = read_header(ps, &new_snapshot);
600 if (r)
601 return r;
602
603 /*
604 * Now we know correct chunk_size, complete the initialisation.
605 */
Jonathan Brassow71fab002009-04-02 19:55:33 +0100606 ps->exceptions_per_area = (ps->store->chunk_size << SECTOR_SHIFT) /
607 sizeof(struct disk_exception);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000608 ps->callbacks = dm_vcalloc(ps->exceptions_per_area,
Jonathan Brassowa2d2b032011-08-02 12:32:03 +0100609 sizeof(*ps->callbacks));
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000610 if (!ps->callbacks)
611 return -ENOMEM;
612
613 /*
614 * Do we need to setup a new snapshot ?
615 */
616 if (new_snapshot) {
617 r = write_header(ps);
618 if (r) {
619 DMWARN("write_header failed");
620 return r;
621 }
622
623 ps->current_area = 0;
624 zero_memory_area(ps);
625 r = zero_disk_area(ps, 0);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000626 if (r)
Jon Brassowf5acc832009-12-10 23:52:07 +0000627 DMWARN("zero_disk_area(0) failed");
628 return r;
629 }
630 /*
631 * Sanity checks.
632 */
633 if (ps->version != SNAPSHOT_DISK_VERSION) {
634 DMWARN("unable to handle snapshot disk version %d",
635 ps->version);
636 return -EINVAL;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000637 }
638
Jon Brassowf5acc832009-12-10 23:52:07 +0000639 /*
640 * Metadata are valid, but snapshot is invalidated
641 */
642 if (!ps->valid)
643 return 1;
644
645 /*
646 * Read the metadata.
647 */
648 r = read_exceptions(ps, callback, callback_context);
649
650 return r;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000651}
652
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000653static int persistent_prepare_exception(struct dm_exception_store *store,
Jon Brassow1d4989c2009-12-10 23:52:10 +0000654 struct dm_exception *e)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000655{
656 struct pstore *ps = get_info(store);
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000657 sector_t size = get_dev_size(dm_snap_cow(store->snap)->bdev);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000658
659 /* Is there enough room ? */
Jonathan Brassowd0216842009-04-02 19:55:32 +0100660 if (size < ((ps->next_free + 1) * store->chunk_size))
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000661 return -ENOSPC;
662
663 e->new_chunk = ps->next_free;
664
665 /*
666 * Move onto the next free pending, making sure to take
667 * into account the location of the metadata chunks.
668 */
Mikulas Patockae9c6a182013-10-16 03:17:47 +0100669 ps->next_free++;
670 skip_metadata(ps);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000671
672 atomic_inc(&ps->pending_count);
673 return 0;
674}
675
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000676static void persistent_commit_exception(struct dm_exception_store *store,
Jon Brassow1d4989c2009-12-10 23:52:10 +0000677 struct dm_exception *e,
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000678 void (*callback) (void *, int success),
679 void *callback_context)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000680{
681 unsigned int i;
682 struct pstore *ps = get_info(store);
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100683 struct core_exception ce;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000684 struct commit_callback *cb;
685
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100686 ce.old_chunk = e->old_chunk;
687 ce.new_chunk = e->new_chunk;
688 write_exception(ps, ps->current_committed++, &ce);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000689
690 /*
691 * Add the callback to the back of the array. This code
692 * is the only place where the callback array is
693 * manipulated, and we know that it will never be called
694 * multiple times concurrently.
695 */
696 cb = ps->callbacks + ps->callback_count++;
697 cb->callback = callback;
698 cb->context = callback_context;
699
700 /*
701 * If there are exceptions in flight and we have not yet
702 * filled this metadata area there's nothing more to do.
703 */
704 if (!atomic_dec_and_test(&ps->pending_count) &&
705 (ps->current_committed != ps->exceptions_per_area))
706 return;
707
708 /*
709 * If we completely filled the current area, then wipe the next one.
710 */
711 if ((ps->current_committed == ps->exceptions_per_area) &&
Jonathan Brassowa2d2b032011-08-02 12:32:03 +0100712 zero_disk_area(ps, ps->current_area + 1))
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000713 ps->valid = 0;
714
715 /*
716 * Commit exceptions to disk.
717 */
Tejun Heod87f4c12010-09-03 11:56:19 +0200718 if (ps->valid && area_io(ps, WRITE_FLUSH_FUA))
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000719 ps->valid = 0;
720
721 /*
722 * Advance to the next area if this one is full.
723 */
724 if (ps->current_committed == ps->exceptions_per_area) {
725 ps->current_committed = 0;
726 ps->current_area++;
727 zero_memory_area(ps);
728 }
729
730 for (i = 0; i < ps->callback_count; i++) {
731 cb = ps->callbacks + i;
732 cb->callback(cb->context, ps->valid);
733 }
734
735 ps->callback_count = 0;
736}
737
Mikulas Patocka4454a622009-12-10 23:52:29 +0000738static int persistent_prepare_merge(struct dm_exception_store *store,
739 chunk_t *last_old_chunk,
740 chunk_t *last_new_chunk)
741{
742 struct pstore *ps = get_info(store);
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100743 struct core_exception ce;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000744 int nr_consecutive;
745 int r;
746
747 /*
748 * When current area is empty, move back to preceding area.
749 */
750 if (!ps->current_committed) {
751 /*
752 * Have we finished?
753 */
754 if (!ps->current_area)
755 return 0;
756
757 ps->current_area--;
758 r = area_io(ps, READ);
759 if (r < 0)
760 return r;
761 ps->current_committed = ps->exceptions_per_area;
762 }
763
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500764 read_exception(ps, ps->area, ps->current_committed - 1, &ce);
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100765 *last_old_chunk = ce.old_chunk;
766 *last_new_chunk = ce.new_chunk;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000767
768 /*
769 * Find number of consecutive chunks within the current area,
770 * working backwards.
771 */
772 for (nr_consecutive = 1; nr_consecutive < ps->current_committed;
773 nr_consecutive++) {
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500774 read_exception(ps, ps->area,
775 ps->current_committed - 1 - nr_consecutive, &ce);
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100776 if (ce.old_chunk != *last_old_chunk - nr_consecutive ||
777 ce.new_chunk != *last_new_chunk - nr_consecutive)
Mikulas Patocka4454a622009-12-10 23:52:29 +0000778 break;
779 }
780
781 return nr_consecutive;
782}
783
784static int persistent_commit_merge(struct dm_exception_store *store,
785 int nr_merged)
786{
787 int r, i;
788 struct pstore *ps = get_info(store);
789
790 BUG_ON(nr_merged > ps->current_committed);
791
792 for (i = 0; i < nr_merged; i++)
793 clear_exception(ps, ps->current_committed - 1 - i);
794
Mikulas Patocka762a80d2011-08-02 12:32:00 +0100795 r = area_io(ps, WRITE_FLUSH_FUA);
Mikulas Patocka4454a622009-12-10 23:52:29 +0000796 if (r < 0)
797 return r;
798
799 ps->current_committed -= nr_merged;
800
801 /*
802 * At this stage, only persistent_usage() uses ps->next_free, so
803 * we make no attempt to keep ps->next_free strictly accurate
804 * as exceptions may have been committed out-of-order originally.
805 * Once a snapshot has become merging, we set it to the value it
806 * would have held had all the exceptions been committed in order.
807 *
808 * ps->current_area does not get reduced by prepare_merge() until
809 * after commit_merge() has removed the nr_merged previous exceptions.
810 */
Tomohiro Kusumi87c961c2010-08-12 04:13:59 +0100811 ps->next_free = area_location(ps, ps->current_area) +
812 ps->current_committed + 1;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000813
814 return 0;
815}
816
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000817static void persistent_drop_snapshot(struct dm_exception_store *store)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000818{
819 struct pstore *ps = get_info(store);
820
821 ps->valid = 0;
822 if (write_header(ps))
823 DMWARN("write header failed");
824}
825
Jonathan Brassow493df712009-04-02 19:55:31 +0100826static int persistent_ctr(struct dm_exception_store *store,
827 unsigned argc, char **argv)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000828{
829 struct pstore *ps;
830
831 /* allocate the pstore */
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100832 ps = kzalloc(sizeof(*ps), GFP_KERNEL);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000833 if (!ps)
834 return -ENOMEM;
835
Jonathan Brassow71fab002009-04-02 19:55:33 +0100836 ps->store = store;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000837 ps->valid = 1;
838 ps->version = SNAPSHOT_DISK_VERSION;
839 ps->area = NULL;
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100840 ps->zero_area = NULL;
841 ps->header_area = NULL;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000842 ps->next_free = NUM_SNAPSHOT_HDR_CHUNKS + 1; /* header and 1st area */
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000843 ps->current_committed = 0;
844
845 ps->callback_count = 0;
846 atomic_set(&ps->pending_count, 0);
847 ps->callbacks = NULL;
848
Tejun Heo239c8dd2011-01-13 19:59:59 +0000849 ps->metadata_wq = alloc_workqueue("ksnaphd", WQ_MEM_RECLAIM, 0);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000850 if (!ps->metadata_wq) {
851 kfree(ps);
852 DMERR("couldn't start header metadata update thread");
853 return -ENOMEM;
854 }
855
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000856 store->context = ps;
857
858 return 0;
859}
860
Jonathan Brassow1e302a92009-04-02 19:55:35 +0100861static unsigned persistent_status(struct dm_exception_store *store,
862 status_type_t status, char *result,
863 unsigned maxlen)
Jonathan Brassow493df712009-04-02 19:55:31 +0100864{
Jonathan Brassow1e302a92009-04-02 19:55:35 +0100865 unsigned sz = 0;
866
867 switch (status) {
868 case STATUSTYPE_INFO:
869 break;
870 case STATUSTYPE_TABLE:
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000871 DMEMIT(" P %llu", (unsigned long long)store->chunk_size);
Jonathan Brassow1e302a92009-04-02 19:55:35 +0100872 }
Jonathan Brassow493df712009-04-02 19:55:31 +0100873
874 return sz;
875}
876
877static struct dm_exception_store_type _persistent_type = {
878 .name = "persistent",
879 .module = THIS_MODULE,
880 .ctr = persistent_ctr,
881 .dtr = persistent_dtr,
882 .read_metadata = persistent_read_metadata,
883 .prepare_exception = persistent_prepare_exception,
884 .commit_exception = persistent_commit_exception,
Mikulas Patocka4454a622009-12-10 23:52:29 +0000885 .prepare_merge = persistent_prepare_merge,
886 .commit_merge = persistent_commit_merge,
Jonathan Brassow493df712009-04-02 19:55:31 +0100887 .drop_snapshot = persistent_drop_snapshot,
Mike Snitzer985903b2009-12-10 23:52:11 +0000888 .usage = persistent_usage,
Jonathan Brassow493df712009-04-02 19:55:31 +0100889 .status = persistent_status,
890};
891
892static struct dm_exception_store_type _persistent_compat_type = {
893 .name = "P",
894 .module = THIS_MODULE,
895 .ctr = persistent_ctr,
896 .dtr = persistent_dtr,
897 .read_metadata = persistent_read_metadata,
898 .prepare_exception = persistent_prepare_exception,
899 .commit_exception = persistent_commit_exception,
Mikulas Patocka4454a622009-12-10 23:52:29 +0000900 .prepare_merge = persistent_prepare_merge,
901 .commit_merge = persistent_commit_merge,
Jonathan Brassow493df712009-04-02 19:55:31 +0100902 .drop_snapshot = persistent_drop_snapshot,
Mike Snitzer985903b2009-12-10 23:52:11 +0000903 .usage = persistent_usage,
Jonathan Brassow493df712009-04-02 19:55:31 +0100904 .status = persistent_status,
905};
906
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000907int dm_persistent_snapshot_init(void)
908{
Jonathan Brassow493df712009-04-02 19:55:31 +0100909 int r;
910
911 r = dm_exception_store_type_register(&_persistent_type);
912 if (r) {
913 DMERR("Unable to register persistent exception store type");
914 return r;
915 }
916
917 r = dm_exception_store_type_register(&_persistent_compat_type);
918 if (r) {
919 DMERR("Unable to register old-style persistent exception "
920 "store type");
921 dm_exception_store_type_unregister(&_persistent_type);
922 return r;
923 }
924
925 return r;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000926}
927
928void dm_persistent_snapshot_exit(void)
929{
Jonathan Brassow493df712009-04-02 19:55:31 +0100930 dm_exception_store_type_unregister(&_persistent_type);
931 dm_exception_store_type_unregister(&_persistent_compat_type);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000932}