2 * Copyright (C) 2003 Sistina Software
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the LGPL.
8 #include <linux/init.h>
9 #include <linux/slab.h>
10 #include <linux/module.h>
11 #include <linux/vmalloc.h>
12 #include <linux/dm-io.h>
13 #include <linux/dm-dirty-log.h>
15 #include <linux/device-mapper.h>
17 #define DM_MSG_PREFIX "dirty region log"
19 static LIST_HEAD(_log_types);
20 static DEFINE_SPINLOCK(_lock);
22 static struct dm_dirty_log_type *__find_dirty_log_type(const char *name)
24 struct dm_dirty_log_type *log_type;
26 list_for_each_entry(log_type, &_log_types, list)
27 if (!strcmp(name, log_type->name))
33 static struct dm_dirty_log_type *_get_dirty_log_type(const char *name)
35 struct dm_dirty_log_type *log_type;
39 log_type = __find_dirty_log_type(name);
40 if (log_type && !try_module_get(log_type->module))
52 * Attempt to retrieve the dm_dirty_log_type by name. If not already
53 * available, attempt to load the appropriate module.
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.
60 * For example, if type_name was "clustered-disk", it would search
61 * 'dm-log-clustered-disk' then 'dm-log-clustered'.
63 * Returns: dirty_log_type* on success, NULL on failure
65 static struct dm_dirty_log_type *get_type(const char *type_name)
67 char *p, *type_name_dup;
68 struct dm_dirty_log_type *log_type;
73 log_type = _get_dirty_log_type(type_name);
77 type_name_dup = kstrdup(type_name, GFP_KERNEL);
79 DMWARN("No memory left to attempt log module load for \"%s\"",
84 while (request_module("dm-log-%s", type_name_dup) ||
85 !(log_type = _get_dirty_log_type(type_name))) {
86 p = strrchr(type_name_dup, '-');
93 DMWARN("Module for logging type \"%s\" not found.", type_name);
100 static void put_type(struct dm_dirty_log_type *type)
106 if (!__find_dirty_log_type(type->name))
109 module_put(type->module);
115 int dm_dirty_log_type_register(struct dm_dirty_log_type *type)
120 if (!__find_dirty_log_type(type->name))
121 list_add(&type->list, &_log_types);
128 EXPORT_SYMBOL(dm_dirty_log_type_register);
130 int dm_dirty_log_type_unregister(struct dm_dirty_log_type *type)
134 if (!__find_dirty_log_type(type->name)) {
139 list_del(&type->list);
145 EXPORT_SYMBOL(dm_dirty_log_type_unregister);
147 struct dm_dirty_log *dm_dirty_log_create(const char *type_name,
148 struct dm_target *ti,
149 unsigned int argc, char **argv)
151 struct dm_dirty_log_type *type;
152 struct dm_dirty_log *log;
154 log = kmalloc(sizeof(*log), GFP_KERNEL);
158 type = get_type(type_name);
165 if (type->ctr(log, ti, argc, argv)) {
173 EXPORT_SYMBOL(dm_dirty_log_create);
175 void dm_dirty_log_destroy(struct dm_dirty_log *log)
181 EXPORT_SYMBOL(dm_dirty_log_destroy);
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 *---------------------------------------------------------------*/
188 * Magic for persistent mirrors: "MiRr"
190 #define MIRROR_MAGIC 0x4D695272
193 * The on-disk version of the metadata.
195 #define MIRROR_DISK_VERSION 2
202 * Simple, incrementing version. no backward
210 struct dm_target *ti;
212 uint32_t region_size;
213 unsigned int region_count;
216 unsigned bitset_uint32_count;
217 uint32_t *clean_bits;
219 uint32_t *recovering_bits; /* FIXME: this seems excessive */
225 DEFAULTSYNC, /* Synchronize if necessary */
226 NOSYNC, /* Devices known to be already in sync */
227 FORCESYNC, /* Force a sync to happen */
230 struct dm_io_request io_req;
236 struct dm_dev *log_dev;
237 struct log_header header;
239 struct dm_io_region header_location;
240 struct log_header *disk_header;
244 * The touched member needs to be updated every time we access
245 * one of the bitsets.
247 static inline int log_test_bit(uint32_t *bs, unsigned bit)
249 return ext2_test_bit(bit, (unsigned long *) bs) ? 1 : 0;
252 static inline void log_set_bit(struct log_c *l,
253 uint32_t *bs, unsigned bit)
255 ext2_set_bit(bit, (unsigned long *) bs);
259 static inline void log_clear_bit(struct log_c *l,
260 uint32_t *bs, unsigned bit)
262 ext2_clear_bit(bit, (unsigned long *) bs);
266 /*----------------------------------------------------------------
268 *--------------------------------------------------------------*/
269 static void header_to_disk(struct log_header *core, struct log_header *disk)
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);
276 static void header_from_disk(struct log_header *core, struct log_header *disk)
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);
283 static int rw_header(struct log_c *lc, int rw)
285 lc->io_req.bi_rw = rw;
287 return dm_io(&lc->io_req, 1, &lc->header_location, NULL);
290 static int read_header(struct log_c *log)
294 r = rw_header(log, READ);
298 header_from_disk(&log->header, log->disk_header);
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;
307 #ifdef __LITTLE_ENDIAN
308 if (log->header.version == 1)
309 log->header.version = 2;
312 if (log->header.version != MIRROR_DISK_VERSION) {
313 DMWARN("incompatible disk log version");
320 static int _check_region_size(struct dm_target *ti, uint32_t region_size)
322 if (region_size < 2 || region_size > ti->len)
325 if (!is_power_of_2(region_size))
331 /*----------------------------------------------------------------
332 * core log constructor/destructor
334 * argv contains region_size followed optionally by [no]sync
335 *--------------------------------------------------------------*/
337 static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti,
338 unsigned int argc, char **argv,
341 enum sync sync = DEFAULTSYNC;
344 uint32_t region_size;
345 unsigned int region_count;
346 size_t bitset_size, buf_size;
349 if (argc < 1 || argc > 2) {
350 DMWARN("wrong number of arguments to dirty region log");
355 if (!strcmp(argv[1], "sync"))
357 else if (!strcmp(argv[1], "nosync"))
360 DMWARN("unrecognised sync argument to "
361 "dirty region log: %s", argv[1]);
366 if (sscanf(argv[0], "%u", ®ion_size) != 1 ||
367 !_check_region_size(ti, region_size)) {
368 DMWARN("invalid region size %s", argv[0]);
372 region_count = dm_sector_div_up(ti->len, region_size);
374 lc = kmalloc(sizeof(*lc), GFP_KERNEL);
376 DMWARN("couldn't allocate core log");
382 lc->region_size = region_size;
383 lc->region_count = region_count;
387 * Work out how many "unsigned long"s we need to hold the bitset.
389 bitset_size = dm_round_up(region_count,
390 sizeof(*lc->clean_bits) << BYTE_SHIFT);
391 bitset_size >>= BYTE_SHIFT;
393 lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits);
399 lc->clean_bits = vmalloc(bitset_size);
400 if (!lc->clean_bits) {
401 DMWARN("couldn't allocate clean bitset");
405 lc->disk_header = NULL;
408 lc->log_dev_failed = 0;
409 lc->header_location.bdev = lc->log_dev->bdev;
410 lc->header_location.sector = 0;
413 * Buffer holds both header and bitset.
415 buf_size = dm_round_up((LOG_OFFSET << SECTOR_SHIFT) +
416 bitset_size, ti->limits.hardsect_size);
418 if (buf_size > dev->bdev->bd_inode->i_size) {
419 DMWARN("log device %s too small: need %llu bytes",
420 dev->name, (unsigned long long)buf_size);
425 lc->header_location.count = buf_size >> SECTOR_SHIFT;
427 lc->io_req.mem.type = DM_IO_VMA;
428 lc->io_req.notify.fn = NULL;
429 lc->io_req.client = dm_io_client_create(dm_div_up(buf_size,
431 if (IS_ERR(lc->io_req.client)) {
432 r = PTR_ERR(lc->io_req.client);
433 DMWARN("couldn't allocate disk io client");
438 lc->disk_header = vmalloc(buf_size);
439 if (!lc->disk_header) {
440 DMWARN("couldn't allocate disk log buffer");
441 dm_io_client_destroy(lc->io_req.client);
446 lc->io_req.mem.ptr.vma = lc->disk_header;
447 lc->clean_bits = (void *)lc->disk_header +
448 (LOG_OFFSET << SECTOR_SHIFT);
451 memset(lc->clean_bits, -1, bitset_size);
453 lc->sync_bits = vmalloc(bitset_size);
454 if (!lc->sync_bits) {
455 DMWARN("couldn't allocate sync bitset");
457 vfree(lc->clean_bits);
459 dm_io_client_destroy(lc->io_req.client);
460 vfree(lc->disk_header);
464 memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
465 lc->sync_count = (sync == NOSYNC) ? region_count : 0;
467 lc->recovering_bits = vmalloc(bitset_size);
468 if (!lc->recovering_bits) {
469 DMWARN("couldn't allocate sync bitset");
470 vfree(lc->sync_bits);
472 vfree(lc->clean_bits);
474 dm_io_client_destroy(lc->io_req.client);
475 vfree(lc->disk_header);
479 memset(lc->recovering_bits, 0, bitset_size);
486 static int core_ctr(struct dm_dirty_log *log, struct dm_target *ti,
487 unsigned int argc, char **argv)
489 return create_log_context(log, ti, argc, argv, NULL);
492 static void destroy_log_context(struct log_c *lc)
494 vfree(lc->sync_bits);
495 vfree(lc->recovering_bits);
499 static void core_dtr(struct dm_dirty_log *log)
501 struct log_c *lc = (struct log_c *) log->context;
503 vfree(lc->clean_bits);
504 destroy_log_context(lc);
507 /*----------------------------------------------------------------
508 * disk log constructor/destructor
510 * argv contains log_device region_size followed optionally by [no]sync
511 *--------------------------------------------------------------*/
512 static int disk_ctr(struct dm_dirty_log *log, struct dm_target *ti,
513 unsigned int argc, char **argv)
518 if (argc < 2 || argc > 3) {
519 DMWARN("wrong number of arguments to disk dirty region log");
523 r = dm_get_device(ti, argv[0], 0, 0 /* FIXME */,
524 FMODE_READ | FMODE_WRITE, &dev);
528 r = create_log_context(log, ti, argc - 1, argv + 1, dev);
530 dm_put_device(ti, dev);
537 static void disk_dtr(struct dm_dirty_log *log)
539 struct log_c *lc = (struct log_c *) log->context;
541 dm_put_device(lc->ti, lc->log_dev);
542 vfree(lc->disk_header);
543 dm_io_client_destroy(lc->io_req.client);
544 destroy_log_context(lc);
547 static int count_bits32(uint32_t *addr, unsigned size)
551 for (i = 0; i < size; i++) {
552 count += hweight32(*(addr+i));
557 static void fail_log_device(struct log_c *lc)
559 if (lc->log_dev_failed)
562 lc->log_dev_failed = 1;
563 dm_table_event(lc->ti->table);
566 static int disk_resume(struct dm_dirty_log *log)
570 struct log_c *lc = (struct log_c *) log->context;
571 size_t size = lc->bitset_uint32_count * sizeof(uint32_t);
573 /* read the disk header */
576 DMWARN("%s: Failed to read header on dirty region log device",
580 * If the log device cannot be read, we must assume
581 * all regions are out-of-sync. If we simply return
582 * here, the state will be uninitialized and could
583 * lead us to return 'in-sync' status for regions
584 * that are actually 'out-of-sync'.
586 lc->header.nr_regions = 0;
589 /* set or clear any new bits -- device has grown */
590 if (lc->sync == NOSYNC)
591 for (i = lc->header.nr_regions; i < lc->region_count; i++)
592 /* FIXME: amazingly inefficient */
593 log_set_bit(lc, lc->clean_bits, i);
595 for (i = lc->header.nr_regions; i < lc->region_count; i++)
596 /* FIXME: amazingly inefficient */
597 log_clear_bit(lc, lc->clean_bits, i);
599 /* clear any old bits -- device has shrunk */
600 for (i = lc->region_count; i % (sizeof(*lc->clean_bits) << BYTE_SHIFT); i++)
601 log_clear_bit(lc, lc->clean_bits, i);
603 /* copy clean across to sync */
604 memcpy(lc->sync_bits, lc->clean_bits, size);
605 lc->sync_count = count_bits32(lc->clean_bits, lc->bitset_uint32_count);
608 /* set the correct number of regions in the header */
609 lc->header.nr_regions = lc->region_count;
611 header_to_disk(&lc->header, lc->disk_header);
613 /* write the new header */
614 r = rw_header(lc, WRITE);
616 DMWARN("%s: Failed to write header on dirty region log device",
624 static uint32_t core_get_region_size(struct dm_dirty_log *log)
626 struct log_c *lc = (struct log_c *) log->context;
627 return lc->region_size;
630 static int core_resume(struct dm_dirty_log *log)
632 struct log_c *lc = (struct log_c *) log->context;
637 static int core_is_clean(struct dm_dirty_log *log, region_t region)
639 struct log_c *lc = (struct log_c *) log->context;
640 return log_test_bit(lc->clean_bits, region);
643 static int core_in_sync(struct dm_dirty_log *log, region_t region, int block)
645 struct log_c *lc = (struct log_c *) log->context;
646 return log_test_bit(lc->sync_bits, region);
649 static int core_flush(struct dm_dirty_log *log)
655 static int disk_flush(struct dm_dirty_log *log)
658 struct log_c *lc = (struct log_c *) log->context;
660 /* only write if the log has changed */
664 r = rw_header(lc, WRITE);
673 static void core_mark_region(struct dm_dirty_log *log, region_t region)
675 struct log_c *lc = (struct log_c *) log->context;
676 log_clear_bit(lc, lc->clean_bits, region);
679 static void core_clear_region(struct dm_dirty_log *log, region_t region)
681 struct log_c *lc = (struct log_c *) log->context;
682 log_set_bit(lc, lc->clean_bits, region);
685 static int core_get_resync_work(struct dm_dirty_log *log, region_t *region)
687 struct log_c *lc = (struct log_c *) log->context;
689 if (lc->sync_search >= lc->region_count)
693 *region = ext2_find_next_zero_bit(
694 (unsigned long *) lc->sync_bits,
697 lc->sync_search = *region + 1;
699 if (*region >= lc->region_count)
702 } while (log_test_bit(lc->recovering_bits, *region));
704 log_set_bit(lc, lc->recovering_bits, *region);
708 static void core_set_region_sync(struct dm_dirty_log *log, region_t region,
711 struct log_c *lc = (struct log_c *) log->context;
713 log_clear_bit(lc, lc->recovering_bits, region);
715 log_set_bit(lc, lc->sync_bits, region);
717 } else if (log_test_bit(lc->sync_bits, region)) {
719 log_clear_bit(lc, lc->sync_bits, region);
723 static region_t core_get_sync_count(struct dm_dirty_log *log)
725 struct log_c *lc = (struct log_c *) log->context;
727 return lc->sync_count;
730 #define DMEMIT_SYNC \
731 if (lc->sync != DEFAULTSYNC) \
732 DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
734 static int core_status(struct dm_dirty_log *log, status_type_t status,
735 char *result, unsigned int maxlen)
738 struct log_c *lc = log->context;
741 case STATUSTYPE_INFO:
742 DMEMIT("1 %s", log->type->name);
745 case STATUSTYPE_TABLE:
746 DMEMIT("%s %u %u ", log->type->name,
747 lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size);
754 static int disk_status(struct dm_dirty_log *log, status_type_t status,
755 char *result, unsigned int maxlen)
758 struct log_c *lc = log->context;
761 case STATUSTYPE_INFO:
762 DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name,
763 lc->log_dev_failed ? 'D' : 'A');
766 case STATUSTYPE_TABLE:
767 DMEMIT("%s %u %s %u ", log->type->name,
768 lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name,
776 static struct dm_dirty_log_type _core_type = {
778 .module = THIS_MODULE,
781 .resume = core_resume,
782 .get_region_size = core_get_region_size,
783 .is_clean = core_is_clean,
784 .in_sync = core_in_sync,
786 .mark_region = core_mark_region,
787 .clear_region = core_clear_region,
788 .get_resync_work = core_get_resync_work,
789 .set_region_sync = core_set_region_sync,
790 .get_sync_count = core_get_sync_count,
791 .status = core_status,
794 static struct dm_dirty_log_type _disk_type = {
796 .module = THIS_MODULE,
799 .postsuspend = disk_flush,
800 .resume = disk_resume,
801 .get_region_size = core_get_region_size,
802 .is_clean = core_is_clean,
803 .in_sync = core_in_sync,
805 .mark_region = core_mark_region,
806 .clear_region = core_clear_region,
807 .get_resync_work = core_get_resync_work,
808 .set_region_sync = core_set_region_sync,
809 .get_sync_count = core_get_sync_count,
810 .status = disk_status,
813 static int __init dm_dirty_log_init(void)
817 r = dm_dirty_log_type_register(&_core_type);
819 DMWARN("couldn't register core log");
821 r = dm_dirty_log_type_register(&_disk_type);
823 DMWARN("couldn't register disk type");
824 dm_dirty_log_type_unregister(&_core_type);
830 static void __exit dm_dirty_log_exit(void)
832 dm_dirty_log_type_unregister(&_disk_type);
833 dm_dirty_log_type_unregister(&_core_type);
836 module_init(dm_dirty_log_init);
837 module_exit(dm_dirty_log_exit);
839 MODULE_DESCRIPTION(DM_NAME " dirty region log");
840 MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@redhat.com>");
841 MODULE_LICENSE("GPL");