blob: 4d3909393f2cce5488ced8843ccab1375d40d2d9 [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
Mike Snitzerb0d3cc02015-10-08 18:05:41 -040010#include <linux/ctype.h>
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000011#include <linux/mm.h>
12#include <linux/pagemap.h>
13#include <linux/vmalloc.h>
Paul Gortmakerdaaa5f72011-05-27 15:50:58 -040014#include <linux/export.h>
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000015#include <linux/slab.h>
16#include <linux/dm-io.h>
Mikulas Patocka55494bf2014-01-13 19:12:36 -050017#include "dm-bufio.h"
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000018
19#define DM_MSG_PREFIX "persistent snapshot"
20#define DM_CHUNK_SIZE_DEFAULT_SECTORS 32 /* 16KB */
21
Mikulas Patocka55b082e2014-01-13 19:13:05 -050022#define DM_PREFETCH_CHUNKS 12
23
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000024/*-----------------------------------------------------------------
25 * Persistent snapshots, by persistent we mean that the snapshot
26 * will survive a reboot.
27 *---------------------------------------------------------------*/
28
29/*
30 * We need to store a record of which parts of the origin have
31 * been copied to the snapshot device. The snapshot code
32 * requires that we copy exception chunks to chunk aligned areas
33 * of the COW store. It makes sense therefore, to store the
34 * metadata in chunk size blocks.
35 *
36 * There is no backward or forward compatibility implemented,
37 * snapshots with different disk versions than the kernel will
38 * not be usable. It is expected that "lvcreate" will blank out
39 * the start of a fresh COW device before calling the snapshot
40 * constructor.
41 *
42 * The first chunk of the COW device just contains the header.
43 * After this there is a chunk filled with exception metadata,
44 * followed by as many exception chunks as can fit in the
45 * metadata areas.
46 *
47 * All on disk structures are in little-endian format. The end
48 * of the exceptions info is indicated by an exception with a
49 * new_chunk of 0, which is invalid since it would point to the
50 * header chunk.
51 */
52
53/*
54 * Magic for persistent snapshots: "SnAp" - Feeble isn't it.
55 */
56#define SNAP_MAGIC 0x70416e53
57
58/*
59 * The on-disk version of the metadata.
60 */
61#define SNAPSHOT_DISK_VERSION 1
62
Mikulas Patocka4454a622009-12-10 23:52:29 +000063#define NUM_SNAPSHOT_HDR_CHUNKS 1
64
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000065struct disk_header {
Alasdair G Kergon283a8322011-08-02 12:32:01 +010066 __le32 magic;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000067
68 /*
69 * Is this snapshot valid. There is no way of recovering
70 * an invalid snapshot.
71 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +010072 __le32 valid;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000073
74 /*
75 * Simple, incrementing version. no backward
76 * compatibility.
77 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +010078 __le32 version;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000079
80 /* In sectors */
Alasdair G Kergon283a8322011-08-02 12:32:01 +010081 __le32 chunk_size;
82} __packed;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000083
84struct disk_exception {
Alasdair G Kergon283a8322011-08-02 12:32:01 +010085 __le64 old_chunk;
86 __le64 new_chunk;
87} __packed;
88
89struct core_exception {
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +000090 uint64_t old_chunk;
91 uint64_t new_chunk;
92};
93
94struct commit_callback {
95 void (*callback)(void *, int success);
96 void *context;
97};
98
99/*
100 * The top level structure for a persistent exception store.
101 */
102struct pstore {
Jonathan Brassow71fab002009-04-02 19:55:33 +0100103 struct dm_exception_store *store;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000104 int version;
105 int valid;
106 uint32_t exceptions_per_area;
107
108 /*
109 * Now that we have an asynchronous kcopyd there is no
110 * need for large chunk sizes, so it wont hurt to have a
111 * whole chunks worth of metadata in memory at once.
112 */
113 void *area;
114
115 /*
116 * An area of zeros used to clear the next area.
117 */
118 void *zero_area;
119
120 /*
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100121 * An area used for header. The header can be written
122 * concurrently with metadata (when invalidating the snapshot),
123 * so it needs a separate buffer.
124 */
125 void *header_area;
126
127 /*
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000128 * Used to keep track of which metadata area the data in
129 * 'chunk' refers to.
130 */
131 chunk_t current_area;
132
133 /*
134 * The next free chunk for an exception.
Mikulas Patocka4454a622009-12-10 23:52:29 +0000135 *
136 * When creating exceptions, all the chunks here and above are
137 * free. It holds the next chunk to be allocated. On rare
138 * occasions (e.g. after a system crash) holes can be left in
139 * the exception store because chunks can be committed out of
140 * order.
141 *
142 * When merging exceptions, it does not necessarily mean all the
143 * chunks here and above are free. It holds the value it would
144 * have held if all chunks had been committed in order of
145 * allocation. Consequently the value may occasionally be
146 * slightly too low, but since it's only used for 'status' and
147 * it can never reach its minimum value too early this doesn't
148 * matter.
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000149 */
Mikulas Patocka4454a622009-12-10 23:52:29 +0000150
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000151 chunk_t next_free;
152
153 /*
154 * The index of next free exception in the current
155 * metadata area.
156 */
157 uint32_t current_committed;
158
159 atomic_t pending_count;
160 uint32_t callback_count;
161 struct commit_callback *callbacks;
162 struct dm_io_client *io_client;
163
164 struct workqueue_struct *metadata_wq;
165};
166
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000167static int alloc_area(struct pstore *ps)
168{
169 int r = -ENOMEM;
170 size_t len;
171
Jonathan Brassow71fab002009-04-02 19:55:33 +0100172 len = ps->store->chunk_size << SECTOR_SHIFT;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000173
174 /*
175 * Allocate the chunk_size block of memory that will hold
176 * a single metadata area.
177 */
178 ps->area = vmalloc(len);
179 if (!ps->area)
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100180 goto err_area;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000181
Joe Perchese29e65a2011-08-02 12:32:02 +0100182 ps->zero_area = vzalloc(len);
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100183 if (!ps->zero_area)
184 goto err_zero_area;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000185
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100186 ps->header_area = vmalloc(len);
187 if (!ps->header_area)
188 goto err_header_area;
189
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000190 return 0;
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100191
192err_header_area:
193 vfree(ps->zero_area);
194
195err_zero_area:
196 vfree(ps->area);
197
198err_area:
199 return r;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000200}
201
202static void free_area(struct pstore *ps)
203{
Markus Elfring0c8f8632015-02-02 14:38:29 +0100204 vfree(ps->area);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000205 ps->area = NULL;
Markus Elfring0c8f8632015-02-02 14:38:29 +0100206 vfree(ps->zero_area);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000207 ps->zero_area = NULL;
Markus Elfring0c8f8632015-02-02 14:38:29 +0100208 vfree(ps->header_area);
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100209 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);
Mikulas Patocka5ea330a2013-09-18 19:14:22 -0400258 flush_workqueue(ps->metadata_wq);
Chuansheng Liuc1a64162014-01-07 16:56:18 +0800259 destroy_work_on_stack(&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
Mikulas Patockae9c6a182013-10-16 03:17:47 +0100272static void skip_metadata(struct pstore *ps)
273{
274 uint32_t stride = ps->exceptions_per_area + 1;
275 chunk_t next_free = ps->next_free;
276 if (sector_div(next_free, stride) == NUM_SNAPSHOT_HDR_CHUNKS)
277 ps->next_free++;
278}
279
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000280/*
281 * Read or write a metadata area. Remembering to skip the first
282 * chunk which holds the header.
283 */
284static int area_io(struct pstore *ps, int rw)
285{
286 int r;
287 chunk_t chunk;
288
289 chunk = area_location(ps, ps->current_area);
290
Mikulas Patocka02d2fd32009-09-04 20:40:37 +0100291 r = chunk_io(ps, ps->area, chunk, rw, 0);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000292 if (r)
293 return r;
294
295 return 0;
296}
297
298static void zero_memory_area(struct pstore *ps)
299{
Jonathan Brassow71fab002009-04-02 19:55:33 +0100300 memset(ps->area, 0, ps->store->chunk_size << SECTOR_SHIFT);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000301}
302
303static int zero_disk_area(struct pstore *ps, chunk_t area)
304{
Mikulas Patocka02d2fd32009-09-04 20:40:37 +0100305 return chunk_io(ps, ps->zero_area, area_location(ps, area), WRITE, 0);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000306}
307
308static int read_header(struct pstore *ps, int *new_snapshot)
309{
310 int r;
311 struct disk_header *dh;
Mikulas Patockadf96eee2009-10-16 23:18:17 +0100312 unsigned chunk_size;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000313 int chunk_size_supplied = 1;
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100314 char *chunk_err;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000315
316 /*
Mikulas Patockadf96eee2009-10-16 23:18:17 +0100317 * Use default chunk size (or logical_block_size, if larger)
318 * if none supplied
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000319 */
Jonathan Brassow71fab002009-04-02 19:55:33 +0100320 if (!ps->store->chunk_size) {
321 ps->store->chunk_size = max(DM_CHUNK_SIZE_DEFAULT_SECTORS,
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000322 bdev_logical_block_size(dm_snap_cow(ps->store->snap)->
323 bdev) >> 9);
Jonathan Brassow71fab002009-04-02 19:55:33 +0100324 ps->store->chunk_mask = ps->store->chunk_size - 1;
Mikulas Patockaa3d939a2015-10-02 11:21:24 -0400325 ps->store->chunk_shift = __ffs(ps->store->chunk_size);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000326 chunk_size_supplied = 0;
327 }
328
Mikulas Patockabda8efe2011-05-29 13:03:09 +0100329 ps->io_client = dm_io_client_create();
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000330 if (IS_ERR(ps->io_client))
331 return PTR_ERR(ps->io_client);
332
333 r = alloc_area(ps);
334 if (r)
335 return r;
336
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100337 r = chunk_io(ps, ps->header_area, 0, READ, 1);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000338 if (r)
339 goto bad;
340
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100341 dh = ps->header_area;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000342
343 if (le32_to_cpu(dh->magic) == 0) {
344 *new_snapshot = 1;
345 return 0;
346 }
347
348 if (le32_to_cpu(dh->magic) != SNAP_MAGIC) {
349 DMWARN("Invalid or corrupt snapshot");
350 r = -ENXIO;
351 goto bad;
352 }
353
354 *new_snapshot = 0;
355 ps->valid = le32_to_cpu(dh->valid);
356 ps->version = le32_to_cpu(dh->version);
357 chunk_size = le32_to_cpu(dh->chunk_size);
358
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100359 if (ps->store->chunk_size == chunk_size)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000360 return 0;
361
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100362 if (chunk_size_supplied)
Mikulas Patockadf96eee2009-10-16 23:18:17 +0100363 DMWARN("chunk size %u in device metadata overrides "
364 "table chunk size of %u.",
365 chunk_size, ps->store->chunk_size);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000366
367 /* We had a bogus chunk_size. Fix stuff up. */
368 free_area(ps);
369
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100370 r = dm_exception_store_set_chunk_size(ps->store, chunk_size,
371 &chunk_err);
372 if (r) {
Mikulas Patockadf96eee2009-10-16 23:18:17 +0100373 DMERR("invalid on-disk chunk size %u: %s.",
374 chunk_size, chunk_err);
Mikulas Patockaae0b7442009-09-04 20:40:43 +0100375 return r;
376 }
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000377
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000378 r = alloc_area(ps);
379 return r;
380
381bad:
382 free_area(ps);
383 return r;
384}
385
386static int write_header(struct pstore *ps)
387{
388 struct disk_header *dh;
389
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100390 memset(ps->header_area, 0, ps->store->chunk_size << SECTOR_SHIFT);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000391
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100392 dh = ps->header_area;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000393 dh->magic = cpu_to_le32(SNAP_MAGIC);
394 dh->valid = cpu_to_le32(ps->valid);
395 dh->version = cpu_to_le32(ps->version);
Jonathan Brassow71fab002009-04-02 19:55:33 +0100396 dh->chunk_size = cpu_to_le32(ps->store->chunk_size);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000397
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100398 return chunk_io(ps, ps->header_area, 0, WRITE, 1);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000399}
400
401/*
402 * Access functions for the disk exceptions, these do the endian conversions.
403 */
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500404static struct disk_exception *get_exception(struct pstore *ps, void *ps_area,
405 uint32_t index)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000406{
407 BUG_ON(index >= ps->exceptions_per_area);
408
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500409 return ((struct disk_exception *) ps_area) + index;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000410}
411
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500412static void read_exception(struct pstore *ps, void *ps_area,
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100413 uint32_t index, struct core_exception *result)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000414{
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500415 struct disk_exception *de = get_exception(ps, ps_area, index);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000416
417 /* copy it */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100418 result->old_chunk = le64_to_cpu(de->old_chunk);
419 result->new_chunk = le64_to_cpu(de->new_chunk);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000420}
421
422static void write_exception(struct pstore *ps,
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100423 uint32_t index, struct core_exception *e)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000424{
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500425 struct disk_exception *de = get_exception(ps, ps->area, index);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000426
427 /* copy it */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100428 de->old_chunk = cpu_to_le64(e->old_chunk);
429 de->new_chunk = cpu_to_le64(e->new_chunk);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000430}
431
Mikulas Patocka4454a622009-12-10 23:52:29 +0000432static void clear_exception(struct pstore *ps, uint32_t index)
433{
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500434 struct disk_exception *de = get_exception(ps, ps->area, index);
Mikulas Patocka4454a622009-12-10 23:52:29 +0000435
436 /* clear it */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100437 de->old_chunk = 0;
438 de->new_chunk = 0;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000439}
440
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000441/*
442 * Registers the exceptions that are present in the current area.
443 * 'full' is filled in to indicate if the area has been
444 * filled.
445 */
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500446static int insert_exceptions(struct pstore *ps, void *ps_area,
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000447 int (*callback)(void *callback_context,
448 chunk_t old, chunk_t new),
449 void *callback_context,
450 int *full)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000451{
452 int r;
453 unsigned int i;
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100454 struct core_exception e;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000455
456 /* presume the area is full */
457 *full = 1;
458
459 for (i = 0; i < ps->exceptions_per_area; i++) {
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500460 read_exception(ps, ps_area, i, &e);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000461
462 /*
463 * If the new_chunk is pointing at the start of
464 * the COW device, where the first metadata area
465 * is we know that we've hit the end of the
466 * exceptions. Therefore the area is not full.
467 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100468 if (e.new_chunk == 0LL) {
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000469 ps->current_committed = i;
470 *full = 0;
471 break;
472 }
473
474 /*
475 * Keep track of the start of the free chunks.
476 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100477 if (ps->next_free <= e.new_chunk)
478 ps->next_free = e.new_chunk + 1;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000479
480 /*
481 * Otherwise we add the exception to the snapshot.
482 */
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100483 r = callback(callback_context, e.old_chunk, e.new_chunk);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000484 if (r)
485 return r;
486 }
487
488 return 0;
489}
490
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000491static int read_exceptions(struct pstore *ps,
492 int (*callback)(void *callback_context, chunk_t old,
493 chunk_t new),
494 void *callback_context)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000495{
496 int r, full = 1;
Mikulas Patocka55494bf2014-01-13 19:12:36 -0500497 struct dm_bufio_client *client;
Mikulas Patocka55b082e2014-01-13 19:13:05 -0500498 chunk_t prefetch_area = 0;
Mikulas Patocka55494bf2014-01-13 19:12:36 -0500499
500 client = dm_bufio_client_create(dm_snap_cow(ps->store->snap)->bdev,
501 ps->store->chunk_size << SECTOR_SHIFT,
502 1, 0, NULL, NULL);
503
504 if (IS_ERR(client))
505 return PTR_ERR(client);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000506
507 /*
Mikulas Patocka55b082e2014-01-13 19:13:05 -0500508 * Setup for one current buffer + desired readahead buffers.
509 */
510 dm_bufio_set_minimum_buffers(client, 1 + DM_PREFETCH_CHUNKS);
511
512 /*
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000513 * Keeping reading chunks and inserting exceptions until
514 * we find a partially full area.
515 */
516 for (ps->current_area = 0; full; ps->current_area++) {
Mikulas Patocka55494bf2014-01-13 19:12:36 -0500517 struct dm_buffer *bp;
518 void *area;
Mikulas Patocka55b082e2014-01-13 19:13:05 -0500519 chunk_t chunk;
520
521 if (unlikely(prefetch_area < ps->current_area))
522 prefetch_area = ps->current_area;
523
524 if (DM_PREFETCH_CHUNKS) do {
525 chunk_t pf_chunk = area_location(ps, prefetch_area);
526 if (unlikely(pf_chunk >= dm_bufio_get_device_size(client)))
527 break;
528 dm_bufio_prefetch(client, pf_chunk, 1);
529 prefetch_area++;
530 if (unlikely(!prefetch_area))
531 break;
532 } while (prefetch_area <= ps->current_area + DM_PREFETCH_CHUNKS);
533
534 chunk = area_location(ps, ps->current_area);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000535
Mikulas Patocka55494bf2014-01-13 19:12:36 -0500536 area = dm_bufio_read(client, chunk, &bp);
viresh kumarfc0a4462015-08-10 11:42:26 +0530537 if (IS_ERR(area)) {
Mikulas Patocka55494bf2014-01-13 19:12:36 -0500538 r = PTR_ERR(area);
539 goto ret_destroy_bufio;
540 }
541
542 r = insert_exceptions(ps, area, callback, callback_context,
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500543 &full);
Mikulas Patocka55494bf2014-01-13 19:12:36 -0500544
Mikulas Patocka2c945822014-03-03 17:19:22 -0500545 if (!full)
546 memcpy(ps->area, area, ps->store->chunk_size << SECTOR_SHIFT);
547
Mikulas Patocka55494bf2014-01-13 19:12:36 -0500548 dm_bufio_release(bp);
549
550 dm_bufio_forget(client, chunk);
551
552 if (unlikely(r))
553 goto ret_destroy_bufio;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000554 }
555
556 ps->current_area--;
557
Mikulas Patockae9c6a182013-10-16 03:17:47 +0100558 skip_metadata(ps);
559
Mikulas Patocka55494bf2014-01-13 19:12:36 -0500560 r = 0;
561
562ret_destroy_bufio:
563 dm_bufio_client_destroy(client);
564
565 return r;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000566}
567
568static struct pstore *get_info(struct dm_exception_store *store)
569{
570 return (struct pstore *) store->context;
571}
572
Mike Snitzer985903b2009-12-10 23:52:11 +0000573static void persistent_usage(struct dm_exception_store *store,
574 sector_t *total_sectors,
575 sector_t *sectors_allocated,
576 sector_t *metadata_sectors)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000577{
Mike Snitzer985903b2009-12-10 23:52:11 +0000578 struct pstore *ps = get_info(store);
579
580 *sectors_allocated = ps->next_free * store->chunk_size;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000581 *total_sectors = get_dev_size(dm_snap_cow(store->snap)->bdev);
Mike Snitzer985903b2009-12-10 23:52:11 +0000582
583 /*
584 * First chunk is the fixed header.
585 * Then there are (ps->current_area + 1) metadata chunks, each one
586 * separated from the next by ps->exceptions_per_area data chunks.
587 */
Mikulas Patocka4454a622009-12-10 23:52:29 +0000588 *metadata_sectors = (ps->current_area + 1 + NUM_SNAPSHOT_HDR_CHUNKS) *
589 store->chunk_size;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000590}
591
Jonathan Brassow493df712009-04-02 19:55:31 +0100592static void persistent_dtr(struct dm_exception_store *store)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000593{
594 struct pstore *ps = get_info(store);
595
596 destroy_workqueue(ps->metadata_wq);
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100597
598 /* Created in read_header */
599 if (ps->io_client)
600 dm_io_client_destroy(ps->io_client);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000601 free_area(ps);
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100602
603 /* Allocated in persistent_read_metadata */
Markus Elfring0c8f8632015-02-02 14:38:29 +0100604 vfree(ps->callbacks);
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100605
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000606 kfree(ps);
607}
608
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000609static int persistent_read_metadata(struct dm_exception_store *store,
610 int (*callback)(void *callback_context,
611 chunk_t old, chunk_t new),
612 void *callback_context)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000613{
614 int r, uninitialized_var(new_snapshot);
615 struct pstore *ps = get_info(store);
616
617 /*
618 * Read the snapshot header.
619 */
620 r = read_header(ps, &new_snapshot);
621 if (r)
622 return r;
623
624 /*
625 * Now we know correct chunk_size, complete the initialisation.
626 */
Jonathan Brassow71fab002009-04-02 19:55:33 +0100627 ps->exceptions_per_area = (ps->store->chunk_size << SECTOR_SHIFT) /
628 sizeof(struct disk_exception);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000629 ps->callbacks = dm_vcalloc(ps->exceptions_per_area,
Jonathan Brassowa2d2b032011-08-02 12:32:03 +0100630 sizeof(*ps->callbacks));
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000631 if (!ps->callbacks)
632 return -ENOMEM;
633
634 /*
635 * Do we need to setup a new snapshot ?
636 */
637 if (new_snapshot) {
638 r = write_header(ps);
639 if (r) {
640 DMWARN("write_header failed");
641 return r;
642 }
643
644 ps->current_area = 0;
645 zero_memory_area(ps);
646 r = zero_disk_area(ps, 0);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000647 if (r)
Jon Brassowf5acc832009-12-10 23:52:07 +0000648 DMWARN("zero_disk_area(0) failed");
649 return r;
650 }
651 /*
652 * Sanity checks.
653 */
654 if (ps->version != SNAPSHOT_DISK_VERSION) {
655 DMWARN("unable to handle snapshot disk version %d",
656 ps->version);
657 return -EINVAL;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000658 }
659
Jon Brassowf5acc832009-12-10 23:52:07 +0000660 /*
661 * Metadata are valid, but snapshot is invalidated
662 */
663 if (!ps->valid)
664 return 1;
665
666 /*
667 * Read the metadata.
668 */
669 r = read_exceptions(ps, callback, callback_context);
670
671 return r;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000672}
673
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000674static int persistent_prepare_exception(struct dm_exception_store *store,
Jon Brassow1d4989c2009-12-10 23:52:10 +0000675 struct dm_exception *e)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000676{
677 struct pstore *ps = get_info(store);
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000678 sector_t size = get_dev_size(dm_snap_cow(store->snap)->bdev);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000679
680 /* Is there enough room ? */
Jonathan Brassowd0216842009-04-02 19:55:32 +0100681 if (size < ((ps->next_free + 1) * store->chunk_size))
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000682 return -ENOSPC;
683
684 e->new_chunk = ps->next_free;
685
686 /*
687 * Move onto the next free pending, making sure to take
688 * into account the location of the metadata chunks.
689 */
Mikulas Patockae9c6a182013-10-16 03:17:47 +0100690 ps->next_free++;
691 skip_metadata(ps);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000692
693 atomic_inc(&ps->pending_count);
694 return 0;
695}
696
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000697static void persistent_commit_exception(struct dm_exception_store *store,
Mikulas Patocka385277b2016-01-08 19:07:55 -0500698 struct dm_exception *e, int valid,
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000699 void (*callback) (void *, int success),
700 void *callback_context)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000701{
702 unsigned int i;
703 struct pstore *ps = get_info(store);
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100704 struct core_exception ce;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000705 struct commit_callback *cb;
706
Mikulas Patocka385277b2016-01-08 19:07:55 -0500707 if (!valid)
708 ps->valid = 0;
709
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100710 ce.old_chunk = e->old_chunk;
711 ce.new_chunk = e->new_chunk;
712 write_exception(ps, ps->current_committed++, &ce);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000713
714 /*
715 * Add the callback to the back of the array. This code
716 * is the only place where the callback array is
717 * manipulated, and we know that it will never be called
718 * multiple times concurrently.
719 */
720 cb = ps->callbacks + ps->callback_count++;
721 cb->callback = callback;
722 cb->context = callback_context;
723
724 /*
725 * If there are exceptions in flight and we have not yet
726 * filled this metadata area there's nothing more to do.
727 */
728 if (!atomic_dec_and_test(&ps->pending_count) &&
729 (ps->current_committed != ps->exceptions_per_area))
730 return;
731
732 /*
733 * If we completely filled the current area, then wipe the next one.
734 */
735 if ((ps->current_committed == ps->exceptions_per_area) &&
Jonathan Brassowa2d2b032011-08-02 12:32:03 +0100736 zero_disk_area(ps, ps->current_area + 1))
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000737 ps->valid = 0;
738
739 /*
740 * Commit exceptions to disk.
741 */
Tejun Heod87f4c12010-09-03 11:56:19 +0200742 if (ps->valid && area_io(ps, WRITE_FLUSH_FUA))
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000743 ps->valid = 0;
744
745 /*
746 * Advance to the next area if this one is full.
747 */
748 if (ps->current_committed == ps->exceptions_per_area) {
749 ps->current_committed = 0;
750 ps->current_area++;
751 zero_memory_area(ps);
752 }
753
754 for (i = 0; i < ps->callback_count; i++) {
755 cb = ps->callbacks + i;
756 cb->callback(cb->context, ps->valid);
757 }
758
759 ps->callback_count = 0;
760}
761
Mikulas Patocka4454a622009-12-10 23:52:29 +0000762static int persistent_prepare_merge(struct dm_exception_store *store,
763 chunk_t *last_old_chunk,
764 chunk_t *last_new_chunk)
765{
766 struct pstore *ps = get_info(store);
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100767 struct core_exception ce;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000768 int nr_consecutive;
769 int r;
770
771 /*
772 * When current area is empty, move back to preceding area.
773 */
774 if (!ps->current_committed) {
775 /*
776 * Have we finished?
777 */
778 if (!ps->current_area)
779 return 0;
780
781 ps->current_area--;
782 r = area_io(ps, READ);
783 if (r < 0)
784 return r;
785 ps->current_committed = ps->exceptions_per_area;
786 }
787
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500788 read_exception(ps, ps->area, ps->current_committed - 1, &ce);
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100789 *last_old_chunk = ce.old_chunk;
790 *last_new_chunk = ce.new_chunk;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000791
792 /*
793 * Find number of consecutive chunks within the current area,
794 * working backwards.
795 */
796 for (nr_consecutive = 1; nr_consecutive < ps->current_committed;
797 nr_consecutive++) {
Mikulas Patocka2cadabd2014-01-13 19:14:04 -0500798 read_exception(ps, ps->area,
799 ps->current_committed - 1 - nr_consecutive, &ce);
Alasdair G Kergon283a8322011-08-02 12:32:01 +0100800 if (ce.old_chunk != *last_old_chunk - nr_consecutive ||
801 ce.new_chunk != *last_new_chunk - nr_consecutive)
Mikulas Patocka4454a622009-12-10 23:52:29 +0000802 break;
803 }
804
805 return nr_consecutive;
806}
807
808static int persistent_commit_merge(struct dm_exception_store *store,
809 int nr_merged)
810{
811 int r, i;
812 struct pstore *ps = get_info(store);
813
814 BUG_ON(nr_merged > ps->current_committed);
815
816 for (i = 0; i < nr_merged; i++)
817 clear_exception(ps, ps->current_committed - 1 - i);
818
Mikulas Patocka762a80d2011-08-02 12:32:00 +0100819 r = area_io(ps, WRITE_FLUSH_FUA);
Mikulas Patocka4454a622009-12-10 23:52:29 +0000820 if (r < 0)
821 return r;
822
823 ps->current_committed -= nr_merged;
824
825 /*
826 * At this stage, only persistent_usage() uses ps->next_free, so
827 * we make no attempt to keep ps->next_free strictly accurate
828 * as exceptions may have been committed out-of-order originally.
829 * Once a snapshot has become merging, we set it to the value it
830 * would have held had all the exceptions been committed in order.
831 *
832 * ps->current_area does not get reduced by prepare_merge() until
833 * after commit_merge() has removed the nr_merged previous exceptions.
834 */
Tomohiro Kusumi87c961c2010-08-12 04:13:59 +0100835 ps->next_free = area_location(ps, ps->current_area) +
836 ps->current_committed + 1;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000837
838 return 0;
839}
840
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000841static void persistent_drop_snapshot(struct dm_exception_store *store)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000842{
843 struct pstore *ps = get_info(store);
844
845 ps->valid = 0;
846 if (write_header(ps))
847 DMWARN("write header failed");
848}
849
Mike Snitzerb0d3cc02015-10-08 18:05:41 -0400850static int persistent_ctr(struct dm_exception_store *store, char *options)
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000851{
852 struct pstore *ps;
Sudip Mukherjeea2a678e2015-10-13 13:33:08 +0530853 int r;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000854
855 /* allocate the pstore */
Jonathan Brassowa32079c2009-04-02 19:55:35 +0100856 ps = kzalloc(sizeof(*ps), GFP_KERNEL);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000857 if (!ps)
858 return -ENOMEM;
859
Jonathan Brassow71fab002009-04-02 19:55:33 +0100860 ps->store = store;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000861 ps->valid = 1;
862 ps->version = SNAPSHOT_DISK_VERSION;
863 ps->area = NULL;
Mikulas Patocka61578dc2009-09-04 20:40:39 +0100864 ps->zero_area = NULL;
865 ps->header_area = NULL;
Mikulas Patocka4454a622009-12-10 23:52:29 +0000866 ps->next_free = NUM_SNAPSHOT_HDR_CHUNKS + 1; /* header and 1st area */
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000867 ps->current_committed = 0;
868
869 ps->callback_count = 0;
870 atomic_set(&ps->pending_count, 0);
871 ps->callbacks = NULL;
872
Tejun Heo239c8dd2011-01-13 19:59:59 +0000873 ps->metadata_wq = alloc_workqueue("ksnaphd", WQ_MEM_RECLAIM, 0);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000874 if (!ps->metadata_wq) {
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000875 DMERR("couldn't start header metadata update thread");
Sudip Mukherjeea2a678e2015-10-13 13:33:08 +0530876 r = -ENOMEM;
877 goto err_workqueue;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000878 }
879
Mike Snitzerb0d3cc02015-10-08 18:05:41 -0400880 if (options) {
881 char overflow = toupper(options[0]);
882 if (overflow == 'O')
883 store->userspace_supports_overflow = true;
884 else {
885 DMERR("Unsupported persistent store option: %s", options);
Sudip Mukherjeea2a678e2015-10-13 13:33:08 +0530886 r = -EINVAL;
887 goto err_options;
Mike Snitzerb0d3cc02015-10-08 18:05:41 -0400888 }
889 }
890
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000891 store->context = ps;
892
893 return 0;
Sudip Mukherjeea2a678e2015-10-13 13:33:08 +0530894
895err_options:
896 destroy_workqueue(ps->metadata_wq);
897err_workqueue:
898 kfree(ps);
899
900 return r;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000901}
902
Jonathan Brassow1e302a92009-04-02 19:55:35 +0100903static unsigned persistent_status(struct dm_exception_store *store,
904 status_type_t status, char *result,
905 unsigned maxlen)
Jonathan Brassow493df712009-04-02 19:55:31 +0100906{
Jonathan Brassow1e302a92009-04-02 19:55:35 +0100907 unsigned sz = 0;
908
909 switch (status) {
910 case STATUSTYPE_INFO:
911 break;
912 case STATUSTYPE_TABLE:
Mike Snitzerb0d3cc02015-10-08 18:05:41 -0400913 DMEMIT(" %s %llu", store->userspace_supports_overflow ? "PO" : "P",
914 (unsigned long long)store->chunk_size);
Jonathan Brassow1e302a92009-04-02 19:55:35 +0100915 }
Jonathan Brassow493df712009-04-02 19:55:31 +0100916
917 return sz;
918}
919
920static struct dm_exception_store_type _persistent_type = {
921 .name = "persistent",
922 .module = THIS_MODULE,
923 .ctr = persistent_ctr,
924 .dtr = persistent_dtr,
925 .read_metadata = persistent_read_metadata,
926 .prepare_exception = persistent_prepare_exception,
927 .commit_exception = persistent_commit_exception,
Mikulas Patocka4454a622009-12-10 23:52:29 +0000928 .prepare_merge = persistent_prepare_merge,
929 .commit_merge = persistent_commit_merge,
Jonathan Brassow493df712009-04-02 19:55:31 +0100930 .drop_snapshot = persistent_drop_snapshot,
Mike Snitzer985903b2009-12-10 23:52:11 +0000931 .usage = persistent_usage,
Jonathan Brassow493df712009-04-02 19:55:31 +0100932 .status = persistent_status,
933};
934
935static struct dm_exception_store_type _persistent_compat_type = {
936 .name = "P",
937 .module = THIS_MODULE,
938 .ctr = persistent_ctr,
939 .dtr = persistent_dtr,
940 .read_metadata = persistent_read_metadata,
941 .prepare_exception = persistent_prepare_exception,
942 .commit_exception = persistent_commit_exception,
Mikulas Patocka4454a622009-12-10 23:52:29 +0000943 .prepare_merge = persistent_prepare_merge,
944 .commit_merge = persistent_commit_merge,
Jonathan Brassow493df712009-04-02 19:55:31 +0100945 .drop_snapshot = persistent_drop_snapshot,
Mike Snitzer985903b2009-12-10 23:52:11 +0000946 .usage = persistent_usage,
Jonathan Brassow493df712009-04-02 19:55:31 +0100947 .status = persistent_status,
948};
949
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000950int dm_persistent_snapshot_init(void)
951{
Jonathan Brassow493df712009-04-02 19:55:31 +0100952 int r;
953
954 r = dm_exception_store_type_register(&_persistent_type);
955 if (r) {
956 DMERR("Unable to register persistent exception store type");
957 return r;
958 }
959
960 r = dm_exception_store_type_register(&_persistent_compat_type);
961 if (r) {
962 DMERR("Unable to register old-style persistent exception "
963 "store type");
964 dm_exception_store_type_unregister(&_persistent_type);
965 return r;
966 }
967
968 return r;
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000969}
970
971void dm_persistent_snapshot_exit(void)
972{
Jonathan Brassow493df712009-04-02 19:55:31 +0100973 dm_exception_store_type_unregister(&_persistent_type);
974 dm_exception_store_type_unregister(&_persistent_compat_type);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +0000975}