blob: 6fa8ccf91c70e70174d578e10b8793b884a86406 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright (C) 2003 Sistina Software
Heinz Mauelshagenb7fd54a2008-04-24 21:43:06 +01003 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
5 * This file is released under the LGPL.
6 */
7
8#include <linux/init.h>
9#include <linux/slab.h>
10#include <linux/module.h>
11#include <linux/vmalloc.h>
Alasdair G Kergona765e202008-04-24 22:02:01 +010012#include <linux/dm-io.h>
13#include <linux/dm-dirty-log.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014
Mikulas Patocka586e80e2008-10-21 17:44:59 +010015#include <linux/device-mapper.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016
Heinz Mauelshagenb7fd54a2008-04-24 21:43:06 +010017#define DM_MSG_PREFIX "dirty region log"
Alasdair G Kergon72d94862006-06-26 00:27:35 -070018
Linus Torvalds1da177e2005-04-16 15:20:36 -070019static LIST_HEAD(_log_types);
20static DEFINE_SPINLOCK(_lock);
21
Mike Snitzerec44ab92009-04-02 19:55:30 +010022static struct dm_dirty_log_type *__find_dirty_log_type(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -070023{
Mike Snitzerec44ab92009-04-02 19:55:30 +010024 struct dm_dirty_log_type *log_type;
Jonathan Brassow2a23aa12008-04-24 21:43:41 +010025
26 list_for_each_entry(log_type, &_log_types, list)
Mike Snitzerec44ab92009-04-02 19:55:30 +010027 if (!strcmp(name, log_type->name))
Jonathan Brassow2a23aa12008-04-24 21:43:41 +010028 return log_type;
29
30 return NULL;
31}
32
Mike Snitzerec44ab92009-04-02 19:55:30 +010033static struct dm_dirty_log_type *_get_dirty_log_type(const char *name)
Jonathan Brassow2a23aa12008-04-24 21:43:41 +010034{
Mike Snitzerec44ab92009-04-02 19:55:30 +010035 struct dm_dirty_log_type *log_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
37 spin_lock(&_lock);
Jonathan Brassow2a23aa12008-04-24 21:43:41 +010038
39 log_type = __find_dirty_log_type(name);
Mike Snitzerec44ab92009-04-02 19:55:30 +010040 if (log_type && !try_module_get(log_type->module))
Mike Snitzer84e67c92009-04-02 19:55:29 +010041 log_type = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
43 spin_unlock(&_lock);
Jonathan Brassow2a23aa12008-04-24 21:43:41 +010044
45 return log_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -070046}
47
Jonathan Brassowfb8b2842008-02-08 02:11:19 +000048/*
49 * get_type
50 * @type_name
51 *
Jonathan Brassow2a23aa12008-04-24 21:43:41 +010052 * Attempt to retrieve the dm_dirty_log_type by name. If not already
Jonathan Brassowfb8b2842008-02-08 02:11:19 +000053 * available, attempt to load the appropriate module.
54 *
55 * Log modules are named "dm-log-" followed by the 'type_name'.
56 * Modules may contain multiple types.
57 * This function will first try the module "dm-log-<type_name>",
58 * then truncate 'type_name' on the last '-' and try again.
59 *
60 * For example, if type_name was "clustered-disk", it would search
61 * 'dm-log-clustered-disk' then 'dm-log-clustered'.
62 *
63 * Returns: dirty_log_type* on success, NULL on failure
64 */
Heinz Mauelshagen416cd172008-04-24 21:43:35 +010065static struct dm_dirty_log_type *get_type(const char *type_name)
Jonathan Brassowfb8b2842008-02-08 02:11:19 +000066{
67 char *p, *type_name_dup;
Mike Snitzerec44ab92009-04-02 19:55:30 +010068 struct dm_dirty_log_type *log_type;
Jonathan Brassowfb8b2842008-02-08 02:11:19 +000069
Jonathan Brassow2a23aa12008-04-24 21:43:41 +010070 if (!type_name)
71 return NULL;
72
73 log_type = _get_dirty_log_type(type_name);
74 if (log_type)
Mike Snitzerec44ab92009-04-02 19:55:30 +010075 return log_type;
Jonathan Brassowfb8b2842008-02-08 02:11:19 +000076
77 type_name_dup = kstrdup(type_name, GFP_KERNEL);
78 if (!type_name_dup) {
79 DMWARN("No memory left to attempt log module load for \"%s\"",
80 type_name);
81 return NULL;
82 }
83
84 while (request_module("dm-log-%s", type_name_dup) ||
Jonathan Brassow2a23aa12008-04-24 21:43:41 +010085 !(log_type = _get_dirty_log_type(type_name))) {
Jonathan Brassowfb8b2842008-02-08 02:11:19 +000086 p = strrchr(type_name_dup, '-');
87 if (!p)
88 break;
89 p[0] = '\0';
90 }
91
Jonathan Brassow2a23aa12008-04-24 21:43:41 +010092 if (!log_type)
Jonathan Brassowfb8b2842008-02-08 02:11:19 +000093 DMWARN("Module for logging type \"%s\" not found.", type_name);
94
95 kfree(type_name_dup);
96
Mike Snitzerec44ab92009-04-02 19:55:30 +010097 return log_type;
Jonathan Brassowfb8b2842008-02-08 02:11:19 +000098}
99
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100100static void put_type(struct dm_dirty_log_type *type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101{
Jonathan Brassow2a23aa12008-04-24 21:43:41 +0100102 if (!type)
103 return;
104
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 spin_lock(&_lock);
Mike Snitzerec44ab92009-04-02 19:55:30 +0100106 if (!__find_dirty_log_type(type->name))
Jonathan Brassow2a23aa12008-04-24 21:43:41 +0100107 goto out;
108
Mike Snitzer84e67c92009-04-02 19:55:29 +0100109 module_put(type->module);
Jonathan Brassow2a23aa12008-04-24 21:43:41 +0100110
111out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 spin_unlock(&_lock);
113}
114
Alasdair G Kergonb8206bc2008-04-24 21:43:38 +0100115int dm_dirty_log_type_register(struct dm_dirty_log_type *type)
116{
Jonathan Brassow2a23aa12008-04-24 21:43:41 +0100117 int r = 0;
118
Alasdair G Kergonb8206bc2008-04-24 21:43:38 +0100119 spin_lock(&_lock);
Jonathan Brassow2a23aa12008-04-24 21:43:41 +0100120 if (!__find_dirty_log_type(type->name))
Mike Snitzerec44ab92009-04-02 19:55:30 +0100121 list_add(&type->list, &_log_types);
122 else
Jonathan Brassow2a23aa12008-04-24 21:43:41 +0100123 r = -EEXIST;
Alasdair G Kergonb8206bc2008-04-24 21:43:38 +0100124 spin_unlock(&_lock);
125
Jonathan Brassow2a23aa12008-04-24 21:43:41 +0100126 return r;
Alasdair G Kergonb8206bc2008-04-24 21:43:38 +0100127}
128EXPORT_SYMBOL(dm_dirty_log_type_register);
129
130int dm_dirty_log_type_unregister(struct dm_dirty_log_type *type)
131{
132 spin_lock(&_lock);
133
Mike Snitzerec44ab92009-04-02 19:55:30 +0100134 if (!__find_dirty_log_type(type->name)) {
Jonathan Brassow2a23aa12008-04-24 21:43:41 +0100135 spin_unlock(&_lock);
136 return -EINVAL;
137 }
138
Mike Snitzerec44ab92009-04-02 19:55:30 +0100139 list_del(&type->list);
Alasdair G Kergonb8206bc2008-04-24 21:43:38 +0100140
141 spin_unlock(&_lock);
142
143 return 0;
144}
145EXPORT_SYMBOL(dm_dirty_log_type_unregister);
146
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100147struct dm_dirty_log *dm_dirty_log_create(const char *type_name,
148 struct dm_target *ti,
149 unsigned int argc, char **argv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150{
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100151 struct dm_dirty_log_type *type;
152 struct dm_dirty_log *log;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
154 log = kmalloc(sizeof(*log), GFP_KERNEL);
155 if (!log)
156 return NULL;
157
158 type = get_type(type_name);
159 if (!type) {
160 kfree(log);
161 return NULL;
162 }
163
164 log->type = type;
165 if (type->ctr(log, ti, argc, argv)) {
166 kfree(log);
167 put_type(type);
168 return NULL;
169 }
170
171 return log;
172}
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100173EXPORT_SYMBOL(dm_dirty_log_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100175void dm_dirty_log_destroy(struct dm_dirty_log *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176{
177 log->type->dtr(log);
178 put_type(log->type);
179 kfree(log);
180}
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100181EXPORT_SYMBOL(dm_dirty_log_destroy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
183/*-----------------------------------------------------------------
184 * Persistent and core logs share a lot of their implementation.
185 * FIXME: need a reload method to be called from a resume
186 *---------------------------------------------------------------*/
187/*
188 * Magic for persistent mirrors: "MiRr"
189 */
190#define MIRROR_MAGIC 0x4D695272
191
192/*
193 * The on-disk version of the metadata.
194 */
Patrick Caulfielda4fc4712006-02-01 03:04:51 -0800195#define MIRROR_DISK_VERSION 2
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196#define LOG_OFFSET 2
197
198struct log_header {
199 uint32_t magic;
200
201 /*
202 * Simple, incrementing version. no backward
203 * compatibility.
204 */
205 uint32_t version;
206 sector_t nr_regions;
207};
208
209struct log_c {
210 struct dm_target *ti;
211 int touched;
212 uint32_t region_size;
213 unsigned int region_count;
214 region_t sync_count;
215
216 unsigned bitset_uint32_count;
217 uint32_t *clean_bits;
218 uint32_t *sync_bits;
219 uint32_t *recovering_bits; /* FIXME: this seems excessive */
220
221 int sync_search;
222
223 /* Resync flag */
224 enum sync {
225 DEFAULTSYNC, /* Synchronize if necessary */
226 NOSYNC, /* Devices known to be already in sync */
227 FORCESYNC, /* Force a sync to happen */
228 } sync;
229
Heinz Mauelshagen5d234d12007-05-09 02:33:03 -0700230 struct dm_io_request io_req;
231
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 /*
233 * Disk log fields
234 */
Jonathan E Brassow01d03a62007-05-09 02:32:57 -0700235 int log_dev_failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 struct dm_dev *log_dev;
237 struct log_header header;
238
Heinz Mauelshagen22a1ceb2008-04-24 21:43:17 +0100239 struct dm_io_region header_location;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 struct log_header *disk_header;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241};
242
243/*
244 * The touched member needs to be updated every time we access
245 * one of the bitsets.
246 */
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100247static inline int log_test_bit(uint32_t *bs, unsigned bit)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248{
Patrick Caulfielda4fc4712006-02-01 03:04:51 -0800249 return ext2_test_bit(bit, (unsigned long *) bs) ? 1 : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250}
251
252static inline void log_set_bit(struct log_c *l,
253 uint32_t *bs, unsigned bit)
254{
Patrick Caulfielda4fc4712006-02-01 03:04:51 -0800255 ext2_set_bit(bit, (unsigned long *) bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 l->touched = 1;
257}
258
259static inline void log_clear_bit(struct log_c *l,
260 uint32_t *bs, unsigned bit)
261{
Patrick Caulfielda4fc4712006-02-01 03:04:51 -0800262 ext2_clear_bit(bit, (unsigned long *) bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 l->touched = 1;
264}
265
266/*----------------------------------------------------------------
267 * Header IO
268 *--------------------------------------------------------------*/
269static void header_to_disk(struct log_header *core, struct log_header *disk)
270{
271 disk->magic = cpu_to_le32(core->magic);
272 disk->version = cpu_to_le32(core->version);
273 disk->nr_regions = cpu_to_le64(core->nr_regions);
274}
275
276static void header_from_disk(struct log_header *core, struct log_header *disk)
277{
278 core->magic = le32_to_cpu(disk->magic);
279 core->version = le32_to_cpu(disk->version);
280 core->nr_regions = le64_to_cpu(disk->nr_regions);
281}
282
Heinz Mauelshagen5d234d12007-05-09 02:33:03 -0700283static int rw_header(struct log_c *lc, int rw)
284{
285 lc->io_req.bi_rw = rw;
Heinz Mauelshagen5d234d12007-05-09 02:33:03 -0700286
287 return dm_io(&lc->io_req, 1, &lc->header_location, NULL);
288}
289
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290static int read_header(struct log_c *log)
291{
292 int r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293
Heinz Mauelshagen5d234d12007-05-09 02:33:03 -0700294 r = rw_header(log, READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 if (r)
296 return r;
297
298 header_from_disk(&log->header, log->disk_header);
299
300 /* New log required? */
301 if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) {
302 log->header.magic = MIRROR_MAGIC;
303 log->header.version = MIRROR_DISK_VERSION;
304 log->header.nr_regions = 0;
305 }
306
Patrick Caulfielda4fc4712006-02-01 03:04:51 -0800307#ifdef __LITTLE_ENDIAN
308 if (log->header.version == 1)
309 log->header.version = 2;
310#endif
311
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 if (log->header.version != MIRROR_DISK_VERSION) {
313 DMWARN("incompatible disk log version");
314 return -EINVAL;
315 }
316
317 return 0;
318}
319
Milan Broz2045e882009-01-06 03:05:01 +0000320static int _check_region_size(struct dm_target *ti, uint32_t region_size)
321{
322 if (region_size < 2 || region_size > ti->len)
323 return 0;
324
325 if (!is_power_of_2(region_size))
326 return 0;
327
328 return 1;
329}
330
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331/*----------------------------------------------------------------
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 * core log constructor/destructor
333 *
334 * argv contains region_size followed optionally by [no]sync
335 *--------------------------------------------------------------*/
336#define BYTE_SHIFT 3
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100337static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti,
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700338 unsigned int argc, char **argv,
339 struct dm_dev *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340{
341 enum sync sync = DEFAULTSYNC;
342
343 struct log_c *lc;
344 uint32_t region_size;
345 unsigned int region_count;
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700346 size_t bitset_size, buf_size;
Heinz Mauelshagen5d234d12007-05-09 02:33:03 -0700347 int r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
349 if (argc < 1 || argc > 2) {
Heinz Mauelshagenb7fd54a2008-04-24 21:43:06 +0100350 DMWARN("wrong number of arguments to dirty region log");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 return -EINVAL;
352 }
353
354 if (argc > 1) {
355 if (!strcmp(argv[1], "sync"))
356 sync = FORCESYNC;
357 else if (!strcmp(argv[1], "nosync"))
358 sync = NOSYNC;
359 else {
Heinz Mauelshagenb7fd54a2008-04-24 21:43:06 +0100360 DMWARN("unrecognised sync argument to "
361 "dirty region log: %s", argv[1]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 return -EINVAL;
363 }
364 }
365
Milan Broz2045e882009-01-06 03:05:01 +0000366 if (sscanf(argv[0], "%u", &region_size) != 1 ||
367 !_check_region_size(ti, region_size)) {
368 DMWARN("invalid region size %s", argv[0]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 return -EINVAL;
370 }
371
372 region_count = dm_sector_div_up(ti->len, region_size);
373
374 lc = kmalloc(sizeof(*lc), GFP_KERNEL);
375 if (!lc) {
376 DMWARN("couldn't allocate core log");
377 return -ENOMEM;
378 }
379
380 lc->ti = ti;
381 lc->touched = 0;
382 lc->region_size = region_size;
383 lc->region_count = region_count;
384 lc->sync = sync;
385
386 /*
Alasdair G Kergon0e568222005-11-21 21:32:34 -0800387 * Work out how many "unsigned long"s we need to hold the bitset.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 */
389 bitset_size = dm_round_up(region_count,
Alasdair G Kergon29121bd2006-06-26 00:27:29 -0700390 sizeof(*lc->clean_bits) << BYTE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 bitset_size >>= BYTE_SHIFT;
392
Alasdair G Kergon29121bd2006-06-26 00:27:29 -0700393 lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits);
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700394
395 /*
396 * Disk log?
397 */
398 if (!dev) {
399 lc->clean_bits = vmalloc(bitset_size);
400 if (!lc->clean_bits) {
401 DMWARN("couldn't allocate clean bitset");
402 kfree(lc);
403 return -ENOMEM;
404 }
405 lc->disk_header = NULL;
406 } else {
407 lc->log_dev = dev;
Jonathan E Brassow01d03a62007-05-09 02:32:57 -0700408 lc->log_dev_failed = 0;
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700409 lc->header_location.bdev = lc->log_dev->bdev;
410 lc->header_location.sector = 0;
411
412 /*
413 * Buffer holds both header and bitset.
414 */
415 buf_size = dm_round_up((LOG_OFFSET << SECTOR_SHIFT) +
Martin K. Petersene1defc42009-05-22 17:17:49 -0400416 bitset_size,
417 ti->limits.logical_block_size);
Milan Brozac1f0ac2009-01-06 03:05:02 +0000418
419 if (buf_size > dev->bdev->bd_inode->i_size) {
420 DMWARN("log device %s too small: need %llu bytes",
421 dev->name, (unsigned long long)buf_size);
422 kfree(lc);
423 return -EINVAL;
424 }
425
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700426 lc->header_location.count = buf_size >> SECTOR_SHIFT;
Takahiro Yasui6f3af012009-01-06 03:04:59 +0000427
Heinz Mauelshagen5d234d12007-05-09 02:33:03 -0700428 lc->io_req.mem.type = DM_IO_VMA;
Takahiro Yasui6f3af012009-01-06 03:04:59 +0000429 lc->io_req.notify.fn = NULL;
Heinz Mauelshagen5d234d12007-05-09 02:33:03 -0700430 lc->io_req.client = dm_io_client_create(dm_div_up(buf_size,
431 PAGE_SIZE));
432 if (IS_ERR(lc->io_req.client)) {
433 r = PTR_ERR(lc->io_req.client);
434 DMWARN("couldn't allocate disk io client");
435 kfree(lc);
436 return -ENOMEM;
437 }
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700438
439 lc->disk_header = vmalloc(buf_size);
440 if (!lc->disk_header) {
441 DMWARN("couldn't allocate disk log buffer");
Takahiro Yasuic7a2bd12009-01-06 03:04:56 +0000442 dm_io_client_destroy(lc->io_req.client);
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700443 kfree(lc);
444 return -ENOMEM;
445 }
446
Takahiro Yasui6f3af012009-01-06 03:04:59 +0000447 lc->io_req.mem.ptr.vma = lc->disk_header;
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700448 lc->clean_bits = (void *)lc->disk_header +
449 (LOG_OFFSET << SECTOR_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 }
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700451
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 memset(lc->clean_bits, -1, bitset_size);
453
454 lc->sync_bits = vmalloc(bitset_size);
455 if (!lc->sync_bits) {
456 DMWARN("couldn't allocate sync bitset");
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700457 if (!dev)
458 vfree(lc->clean_bits);
Takahiro Yasuic7a2bd12009-01-06 03:04:56 +0000459 else
460 dm_io_client_destroy(lc->io_req.client);
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700461 vfree(lc->disk_header);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 kfree(lc);
463 return -ENOMEM;
464 }
465 memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
466 lc->sync_count = (sync == NOSYNC) ? region_count : 0;
467
468 lc->recovering_bits = vmalloc(bitset_size);
469 if (!lc->recovering_bits) {
470 DMWARN("couldn't allocate sync bitset");
471 vfree(lc->sync_bits);
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700472 if (!dev)
473 vfree(lc->clean_bits);
Takahiro Yasuic7a2bd12009-01-06 03:04:56 +0000474 else
475 dm_io_client_destroy(lc->io_req.client);
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700476 vfree(lc->disk_header);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 kfree(lc);
478 return -ENOMEM;
479 }
480 memset(lc->recovering_bits, 0, bitset_size);
481 lc->sync_search = 0;
482 log->context = lc;
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700483
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 return 0;
485}
486
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100487static int core_ctr(struct dm_dirty_log *log, struct dm_target *ti,
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700488 unsigned int argc, char **argv)
489{
490 return create_log_context(log, ti, argc, argv, NULL);
491}
492
493static void destroy_log_context(struct log_c *lc)
494{
495 vfree(lc->sync_bits);
496 vfree(lc->recovering_bits);
497 kfree(lc);
498}
499
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100500static void core_dtr(struct dm_dirty_log *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501{
502 struct log_c *lc = (struct log_c *) log->context;
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700503
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 vfree(lc->clean_bits);
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700505 destroy_log_context(lc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506}
507
508/*----------------------------------------------------------------
509 * disk log constructor/destructor
510 *
511 * argv contains log_device region_size followed optionally by [no]sync
512 *--------------------------------------------------------------*/
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100513static int disk_ctr(struct dm_dirty_log *log, struct dm_target *ti,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 unsigned int argc, char **argv)
515{
516 int r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 struct dm_dev *dev;
518
519 if (argc < 2 || argc > 3) {
Heinz Mauelshagenb7fd54a2008-04-24 21:43:06 +0100520 DMWARN("wrong number of arguments to disk dirty region log");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 return -EINVAL;
522 }
523
524 r = dm_get_device(ti, argv[0], 0, 0 /* FIXME */,
525 FMODE_READ | FMODE_WRITE, &dev);
526 if (r)
527 return r;
528
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700529 r = create_log_context(log, ti, argc - 1, argv + 1, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 if (r) {
531 dm_put_device(ti, dev);
532 return r;
533 }
534
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536}
537
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100538static void disk_dtr(struct dm_dirty_log *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539{
540 struct log_c *lc = (struct log_c *) log->context;
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700541
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 dm_put_device(lc->ti, lc->log_dev);
543 vfree(lc->disk_header);
Heinz Mauelshagen5d234d12007-05-09 02:33:03 -0700544 dm_io_client_destroy(lc->io_req.client);
Alasdair G Kergonb7cca192006-06-26 00:27:29 -0700545 destroy_log_context(lc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546}
547
548static int count_bits32(uint32_t *addr, unsigned size)
549{
550 int count = 0, i;
551
552 for (i = 0; i < size; i++) {
553 count += hweight32(*(addr+i));
554 }
555 return count;
556}
557
Jonathan E Brassow01d03a62007-05-09 02:32:57 -0700558static void fail_log_device(struct log_c *lc)
559{
560 if (lc->log_dev_failed)
561 return;
562
563 lc->log_dev_failed = 1;
564 dm_table_event(lc->ti->table);
565}
566
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100567static int disk_resume(struct dm_dirty_log *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568{
569 int r;
570 unsigned i;
571 struct log_c *lc = (struct log_c *) log->context;
572 size_t size = lc->bitset_uint32_count * sizeof(uint32_t);
573
574 /* read the disk header */
575 r = read_header(lc);
Jonathan E Brassow01d03a62007-05-09 02:32:57 -0700576 if (r) {
Heinz Mauelshagenb7fd54a2008-04-24 21:43:06 +0100577 DMWARN("%s: Failed to read header on dirty region log device",
Jonathan E Brassow01d03a62007-05-09 02:32:57 -0700578 lc->log_dev->name);
579 fail_log_device(lc);
Jonathan Brassowba8b45c2007-05-09 02:33:08 -0700580 /*
581 * If the log device cannot be read, we must assume
582 * all regions are out-of-sync. If we simply return
583 * here, the state will be uninitialized and could
584 * lead us to return 'in-sync' status for regions
585 * that are actually 'out-of-sync'.
586 */
587 lc->header.nr_regions = 0;
Jonathan E Brassow01d03a62007-05-09 02:32:57 -0700588 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
Alasdair G Kergon8a835f12006-06-26 00:27:30 -0700590 /* set or clear any new bits -- device has grown */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 if (lc->sync == NOSYNC)
592 for (i = lc->header.nr_regions; i < lc->region_count; i++)
593 /* FIXME: amazingly inefficient */
594 log_set_bit(lc, lc->clean_bits, i);
595 else
596 for (i = lc->header.nr_regions; i < lc->region_count; i++)
597 /* FIXME: amazingly inefficient */
598 log_clear_bit(lc, lc->clean_bits, i);
599
Alasdair G Kergon8a835f12006-06-26 00:27:30 -0700600 /* clear any old bits -- device has shrunk */
601 for (i = lc->region_count; i % (sizeof(*lc->clean_bits) << BYTE_SHIFT); i++)
602 log_clear_bit(lc, lc->clean_bits, i);
603
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 /* copy clean across to sync */
605 memcpy(lc->sync_bits, lc->clean_bits, size);
606 lc->sync_count = count_bits32(lc->clean_bits, lc->bitset_uint32_count);
Jonathan E Brassow88b20a12006-12-08 02:41:12 -0800607 lc->sync_search = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 /* set the correct number of regions in the header */
610 lc->header.nr_regions = lc->region_count;
611
Takahiro Yasui6f3af012009-01-06 03:04:59 +0000612 header_to_disk(&lc->header, lc->disk_header);
613
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 /* write the new header */
Takahiro Yasui6f3af012009-01-06 03:04:59 +0000615 r = rw_header(lc, WRITE);
Jonathan E Brassow01d03a62007-05-09 02:32:57 -0700616 if (r) {
Heinz Mauelshagenb7fd54a2008-04-24 21:43:06 +0100617 DMWARN("%s: Failed to write header on dirty region log device",
Jonathan E Brassow01d03a62007-05-09 02:32:57 -0700618 lc->log_dev->name);
619 fail_log_device(lc);
620 }
621
622 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623}
624
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100625static uint32_t core_get_region_size(struct dm_dirty_log *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626{
627 struct log_c *lc = (struct log_c *) log->context;
628 return lc->region_size;
629}
630
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100631static int core_resume(struct dm_dirty_log *log)
Jonathan E Brassow88b20a12006-12-08 02:41:12 -0800632{
633 struct log_c *lc = (struct log_c *) log->context;
634 lc->sync_search = 0;
635 return 0;
636}
637
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100638static int core_is_clean(struct dm_dirty_log *log, region_t region)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639{
640 struct log_c *lc = (struct log_c *) log->context;
641 return log_test_bit(lc->clean_bits, region);
642}
643
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100644static int core_in_sync(struct dm_dirty_log *log, region_t region, int block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645{
646 struct log_c *lc = (struct log_c *) log->context;
647 return log_test_bit(lc->sync_bits, region);
648}
649
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100650static int core_flush(struct dm_dirty_log *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651{
652 /* no op */
653 return 0;
654}
655
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100656static int disk_flush(struct dm_dirty_log *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657{
658 int r;
659 struct log_c *lc = (struct log_c *) log->context;
660
661 /* only write if the log has changed */
662 if (!lc->touched)
663 return 0;
664
Takahiro Yasui6f3af012009-01-06 03:04:59 +0000665 r = rw_header(lc, WRITE);
Jonathan E Brassow01d03a62007-05-09 02:32:57 -0700666 if (r)
667 fail_log_device(lc);
668 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 lc->touched = 0;
670
671 return r;
672}
673
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100674static void core_mark_region(struct dm_dirty_log *log, region_t region)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675{
676 struct log_c *lc = (struct log_c *) log->context;
677 log_clear_bit(lc, lc->clean_bits, region);
678}
679
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100680static void core_clear_region(struct dm_dirty_log *log, region_t region)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
682 struct log_c *lc = (struct log_c *) log->context;
683 log_set_bit(lc, lc->clean_bits, region);
684}
685
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100686static int core_get_resync_work(struct dm_dirty_log *log, region_t *region)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687{
688 struct log_c *lc = (struct log_c *) log->context;
689
690 if (lc->sync_search >= lc->region_count)
691 return 0;
692
693 do {
Stefan Bader1113a7e2006-02-02 14:28:07 -0800694 *region = ext2_find_next_zero_bit(
695 (unsigned long *) lc->sync_bits,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 lc->region_count,
697 lc->sync_search);
698 lc->sync_search = *region + 1;
699
Darrick J. Wongac81b2e2006-01-06 00:20:11 -0800700 if (*region >= lc->region_count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 return 0;
702
703 } while (log_test_bit(lc->recovering_bits, *region));
704
705 log_set_bit(lc, lc->recovering_bits, *region);
706 return 1;
707}
708
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100709static void core_set_region_sync(struct dm_dirty_log *log, region_t region,
Jonathan E Brassowf3ee6b22006-12-08 02:41:11 -0800710 int in_sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711{
712 struct log_c *lc = (struct log_c *) log->context;
713
714 log_clear_bit(lc, lc->recovering_bits, region);
Jonathan E Brassowf3ee6b22006-12-08 02:41:11 -0800715 if (in_sync) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 log_set_bit(lc, lc->sync_bits, region);
717 lc->sync_count++;
Jonathan E Brassowf3ee6b22006-12-08 02:41:11 -0800718 } else if (log_test_bit(lc->sync_bits, region)) {
719 lc->sync_count--;
720 log_clear_bit(lc, lc->sync_bits, region);
721 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722}
723
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100724static region_t core_get_sync_count(struct dm_dirty_log *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725{
726 struct log_c *lc = (struct log_c *) log->context;
727
728 return lc->sync_count;
729}
730
731#define DMEMIT_SYNC \
732 if (lc->sync != DEFAULTSYNC) \
733 DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
734
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100735static int core_status(struct dm_dirty_log *log, status_type_t status,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 char *result, unsigned int maxlen)
737{
738 int sz = 0;
739 struct log_c *lc = log->context;
740
741 switch(status) {
742 case STATUSTYPE_INFO:
Jonathan E Brassow315dcc22007-05-09 02:32:58 -0700743 DMEMIT("1 %s", log->type->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 break;
745
746 case STATUSTYPE_TABLE:
747 DMEMIT("%s %u %u ", log->type->name,
748 lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size);
749 DMEMIT_SYNC;
750 }
751
752 return sz;
753}
754
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100755static int disk_status(struct dm_dirty_log *log, status_type_t status,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 char *result, unsigned int maxlen)
757{
758 int sz = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 struct log_c *lc = log->context;
760
761 switch(status) {
762 case STATUSTYPE_INFO:
Jonathan E Brassow315dcc22007-05-09 02:32:58 -0700763 DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name,
764 lc->log_dev_failed ? 'D' : 'A');
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 break;
766
767 case STATUSTYPE_TABLE:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 DMEMIT("%s %u %s %u ", log->type->name,
Jonathan E Brassow315dcc22007-05-09 02:32:58 -0700769 lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 lc->region_size);
771 DMEMIT_SYNC;
772 }
773
774 return sz;
775}
776
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100777static struct dm_dirty_log_type _core_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 .name = "core",
779 .module = THIS_MODULE,
780 .ctr = core_ctr,
781 .dtr = core_dtr,
Jonathan E Brassow88b20a12006-12-08 02:41:12 -0800782 .resume = core_resume,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 .get_region_size = core_get_region_size,
784 .is_clean = core_is_clean,
785 .in_sync = core_in_sync,
786 .flush = core_flush,
787 .mark_region = core_mark_region,
788 .clear_region = core_clear_region,
789 .get_resync_work = core_get_resync_work,
Jonathan E Brassowf3ee6b22006-12-08 02:41:11 -0800790 .set_region_sync = core_set_region_sync,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 .get_sync_count = core_get_sync_count,
792 .status = core_status,
793};
794
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100795static struct dm_dirty_log_type _disk_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 .name = "disk",
797 .module = THIS_MODULE,
798 .ctr = disk_ctr,
799 .dtr = disk_dtr,
Jonathan Brassow6b3df0d2007-10-19 22:47:57 +0100800 .postsuspend = disk_flush,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 .resume = disk_resume,
802 .get_region_size = core_get_region_size,
803 .is_clean = core_is_clean,
804 .in_sync = core_in_sync,
805 .flush = disk_flush,
806 .mark_region = core_mark_region,
807 .clear_region = core_clear_region,
808 .get_resync_work = core_get_resync_work,
Jonathan E Brassowf3ee6b22006-12-08 02:41:11 -0800809 .set_region_sync = core_set_region_sync,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 .get_sync_count = core_get_sync_count,
811 .status = disk_status,
812};
813
Adrian Bunkc8da2f82008-07-21 12:00:27 +0100814static int __init dm_dirty_log_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815{
816 int r;
817
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100818 r = dm_dirty_log_type_register(&_core_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 if (r)
820 DMWARN("couldn't register core log");
821
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100822 r = dm_dirty_log_type_register(&_disk_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 if (r) {
824 DMWARN("couldn't register disk type");
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100825 dm_dirty_log_type_unregister(&_core_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 }
827
828 return r;
829}
830
Adrian Bunkc8da2f82008-07-21 12:00:27 +0100831static void __exit dm_dirty_log_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832{
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100833 dm_dirty_log_type_unregister(&_disk_type);
834 dm_dirty_log_type_unregister(&_core_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835}
836
Heinz Mauelshagen769aef32008-04-24 21:43:09 +0100837module_init(dm_dirty_log_init);
838module_exit(dm_dirty_log_exit);
839
840MODULE_DESCRIPTION(DM_NAME " dirty region log");
841MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@redhat.com>");
842MODULE_LICENSE("GPL");