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 struct dm_dirty_log_internal {
20 struct dm_dirty_log_type *type;
22 struct list_head list;
25 static LIST_HEAD(_log_types);
26 static DEFINE_SPINLOCK(_lock);
28 static struct dm_dirty_log_internal *__find_dirty_log_type(const char *name)
30 struct dm_dirty_log_internal *log_type;
32 list_for_each_entry(log_type, &_log_types, list)
33 if (!strcmp(name, log_type->type->name))
39 static struct dm_dirty_log_internal *_get_dirty_log_type(const char *name)
41 struct dm_dirty_log_internal *log_type;
45 log_type = __find_dirty_log_type(name);
46 if (log_type && !try_module_get(log_type->type->module))
58 * Attempt to retrieve the dm_dirty_log_type by name. If not already
59 * available, attempt to load the appropriate module.
61 * Log modules are named "dm-log-" followed by the 'type_name'.
62 * Modules may contain multiple types.
63 * This function will first try the module "dm-log-<type_name>",
64 * then truncate 'type_name' on the last '-' and try again.
66 * For example, if type_name was "clustered-disk", it would search
67 * 'dm-log-clustered-disk' then 'dm-log-clustered'.
69 * Returns: dirty_log_type* on success, NULL on failure
71 static struct dm_dirty_log_type *get_type(const char *type_name)
73 char *p, *type_name_dup;
74 struct dm_dirty_log_internal *log_type;
79 log_type = _get_dirty_log_type(type_name);
81 return log_type->type;
83 type_name_dup = kstrdup(type_name, GFP_KERNEL);
85 DMWARN("No memory left to attempt log module load for \"%s\"",
90 while (request_module("dm-log-%s", type_name_dup) ||
91 !(log_type = _get_dirty_log_type(type_name))) {
92 p = strrchr(type_name_dup, '-');
99 DMWARN("Module for logging type \"%s\" not found.", type_name);
101 kfree(type_name_dup);
103 return log_type ? log_type->type : NULL;
106 static void put_type(struct dm_dirty_log_type *type)
108 struct dm_dirty_log_internal *log_type;
114 log_type = __find_dirty_log_type(type->name);
118 module_put(type->module);
124 static struct dm_dirty_log_internal *_alloc_dirty_log_type(struct dm_dirty_log_type *type)
126 struct dm_dirty_log_internal *log_type = kzalloc(sizeof(*log_type),
130 log_type->type = type;
135 int dm_dirty_log_type_register(struct dm_dirty_log_type *type)
137 struct dm_dirty_log_internal *log_type = _alloc_dirty_log_type(type);
144 if (!__find_dirty_log_type(type->name))
145 list_add(&log_type->list, &_log_types);
154 EXPORT_SYMBOL(dm_dirty_log_type_register);
156 int dm_dirty_log_type_unregister(struct dm_dirty_log_type *type)
158 struct dm_dirty_log_internal *log_type;
162 log_type = __find_dirty_log_type(type->name);
168 list_del(&log_type->list);
175 EXPORT_SYMBOL(dm_dirty_log_type_unregister);
177 struct dm_dirty_log *dm_dirty_log_create(const char *type_name,
178 struct dm_target *ti,
179 unsigned int argc, char **argv)
181 struct dm_dirty_log_type *type;
182 struct dm_dirty_log *log;
184 log = kmalloc(sizeof(*log), GFP_KERNEL);
188 type = get_type(type_name);
195 if (type->ctr(log, ti, argc, argv)) {
203 EXPORT_SYMBOL(dm_dirty_log_create);
205 void dm_dirty_log_destroy(struct dm_dirty_log *log)
211 EXPORT_SYMBOL(dm_dirty_log_destroy);
213 /*-----------------------------------------------------------------
214 * Persistent and core logs share a lot of their implementation.
215 * FIXME: need a reload method to be called from a resume
216 *---------------------------------------------------------------*/
218 * Magic for persistent mirrors: "MiRr"
220 #define MIRROR_MAGIC 0x4D695272
223 * The on-disk version of the metadata.
225 #define MIRROR_DISK_VERSION 2
232 * Simple, incrementing version. no backward
240 struct dm_target *ti;
242 uint32_t region_size;
243 unsigned int region_count;
246 unsigned bitset_uint32_count;
247 uint32_t *clean_bits;
249 uint32_t *recovering_bits; /* FIXME: this seems excessive */
255 DEFAULTSYNC, /* Synchronize if necessary */
256 NOSYNC, /* Devices known to be already in sync */
257 FORCESYNC, /* Force a sync to happen */
260 struct dm_io_request io_req;
266 struct dm_dev *log_dev;
267 struct log_header header;
269 struct dm_io_region header_location;
270 struct log_header *disk_header;
274 * The touched member needs to be updated every time we access
275 * one of the bitsets.
277 static inline int log_test_bit(uint32_t *bs, unsigned bit)
279 return ext2_test_bit(bit, (unsigned long *) bs) ? 1 : 0;
282 static inline void log_set_bit(struct log_c *l,
283 uint32_t *bs, unsigned bit)
285 ext2_set_bit(bit, (unsigned long *) bs);
289 static inline void log_clear_bit(struct log_c *l,
290 uint32_t *bs, unsigned bit)
292 ext2_clear_bit(bit, (unsigned long *) bs);
296 /*----------------------------------------------------------------
298 *--------------------------------------------------------------*/
299 static void header_to_disk(struct log_header *core, struct log_header *disk)
301 disk->magic = cpu_to_le32(core->magic);
302 disk->version = cpu_to_le32(core->version);
303 disk->nr_regions = cpu_to_le64(core->nr_regions);
306 static void header_from_disk(struct log_header *core, struct log_header *disk)
308 core->magic = le32_to_cpu(disk->magic);
309 core->version = le32_to_cpu(disk->version);
310 core->nr_regions = le64_to_cpu(disk->nr_regions);
313 static int rw_header(struct log_c *lc, int rw)
315 lc->io_req.bi_rw = rw;
317 return dm_io(&lc->io_req, 1, &lc->header_location, NULL);
320 static int read_header(struct log_c *log)
324 r = rw_header(log, READ);
328 header_from_disk(&log->header, log->disk_header);
330 /* New log required? */
331 if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) {
332 log->header.magic = MIRROR_MAGIC;
333 log->header.version = MIRROR_DISK_VERSION;
334 log->header.nr_regions = 0;
337 #ifdef __LITTLE_ENDIAN
338 if (log->header.version == 1)
339 log->header.version = 2;
342 if (log->header.version != MIRROR_DISK_VERSION) {
343 DMWARN("incompatible disk log version");
350 static int _check_region_size(struct dm_target *ti, uint32_t region_size)
352 if (region_size < 2 || region_size > ti->len)
355 if (!is_power_of_2(region_size))
361 /*----------------------------------------------------------------
362 * core log constructor/destructor
364 * argv contains region_size followed optionally by [no]sync
365 *--------------------------------------------------------------*/
367 static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti,
368 unsigned int argc, char **argv,
371 enum sync sync = DEFAULTSYNC;
374 uint32_t region_size;
375 unsigned int region_count;
376 size_t bitset_size, buf_size;
379 if (argc < 1 || argc > 2) {
380 DMWARN("wrong number of arguments to dirty region log");
385 if (!strcmp(argv[1], "sync"))
387 else if (!strcmp(argv[1], "nosync"))
390 DMWARN("unrecognised sync argument to "
391 "dirty region log: %s", argv[1]);
396 if (sscanf(argv[0], "%u", ®ion_size) != 1 ||
397 !_check_region_size(ti, region_size)) {
398 DMWARN("invalid region size %s", argv[0]);
402 region_count = dm_sector_div_up(ti->len, region_size);
404 lc = kmalloc(sizeof(*lc), GFP_KERNEL);
406 DMWARN("couldn't allocate core log");
412 lc->region_size = region_size;
413 lc->region_count = region_count;
417 * Work out how many "unsigned long"s we need to hold the bitset.
419 bitset_size = dm_round_up(region_count,
420 sizeof(*lc->clean_bits) << BYTE_SHIFT);
421 bitset_size >>= BYTE_SHIFT;
423 lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits);
429 lc->clean_bits = vmalloc(bitset_size);
430 if (!lc->clean_bits) {
431 DMWARN("couldn't allocate clean bitset");
435 lc->disk_header = NULL;
438 lc->log_dev_failed = 0;
439 lc->header_location.bdev = lc->log_dev->bdev;
440 lc->header_location.sector = 0;
443 * Buffer holds both header and bitset.
445 buf_size = dm_round_up((LOG_OFFSET << SECTOR_SHIFT) +
446 bitset_size, ti->limits.hardsect_size);
448 if (buf_size > dev->bdev->bd_inode->i_size) {
449 DMWARN("log device %s too small: need %llu bytes",
450 dev->name, (unsigned long long)buf_size);
455 lc->header_location.count = buf_size >> SECTOR_SHIFT;
457 lc->io_req.mem.type = DM_IO_VMA;
458 lc->io_req.notify.fn = NULL;
459 lc->io_req.client = dm_io_client_create(dm_div_up(buf_size,
461 if (IS_ERR(lc->io_req.client)) {
462 r = PTR_ERR(lc->io_req.client);
463 DMWARN("couldn't allocate disk io client");
468 lc->disk_header = vmalloc(buf_size);
469 if (!lc->disk_header) {
470 DMWARN("couldn't allocate disk log buffer");
471 dm_io_client_destroy(lc->io_req.client);
476 lc->io_req.mem.ptr.vma = lc->disk_header;
477 lc->clean_bits = (void *)lc->disk_header +
478 (LOG_OFFSET << SECTOR_SHIFT);
481 memset(lc->clean_bits, -1, bitset_size);
483 lc->sync_bits = vmalloc(bitset_size);
484 if (!lc->sync_bits) {
485 DMWARN("couldn't allocate sync bitset");
487 vfree(lc->clean_bits);
489 dm_io_client_destroy(lc->io_req.client);
490 vfree(lc->disk_header);
494 memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
495 lc->sync_count = (sync == NOSYNC) ? region_count : 0;
497 lc->recovering_bits = vmalloc(bitset_size);
498 if (!lc->recovering_bits) {
499 DMWARN("couldn't allocate sync bitset");
500 vfree(lc->sync_bits);
502 vfree(lc->clean_bits);
504 dm_io_client_destroy(lc->io_req.client);
505 vfree(lc->disk_header);
509 memset(lc->recovering_bits, 0, bitset_size);
516 static int core_ctr(struct dm_dirty_log *log, struct dm_target *ti,
517 unsigned int argc, char **argv)
519 return create_log_context(log, ti, argc, argv, NULL);
522 static void destroy_log_context(struct log_c *lc)
524 vfree(lc->sync_bits);
525 vfree(lc->recovering_bits);
529 static void core_dtr(struct dm_dirty_log *log)
531 struct log_c *lc = (struct log_c *) log->context;
533 vfree(lc->clean_bits);
534 destroy_log_context(lc);
537 /*----------------------------------------------------------------
538 * disk log constructor/destructor
540 * argv contains log_device region_size followed optionally by [no]sync
541 *--------------------------------------------------------------*/
542 static int disk_ctr(struct dm_dirty_log *log, struct dm_target *ti,
543 unsigned int argc, char **argv)
548 if (argc < 2 || argc > 3) {
549 DMWARN("wrong number of arguments to disk dirty region log");
553 r = dm_get_device(ti, argv[0], 0, 0 /* FIXME */,
554 FMODE_READ | FMODE_WRITE, &dev);
558 r = create_log_context(log, ti, argc - 1, argv + 1, dev);
560 dm_put_device(ti, dev);
567 static void disk_dtr(struct dm_dirty_log *log)
569 struct log_c *lc = (struct log_c *) log->context;
571 dm_put_device(lc->ti, lc->log_dev);
572 vfree(lc->disk_header);
573 dm_io_client_destroy(lc->io_req.client);
574 destroy_log_context(lc);
577 static int count_bits32(uint32_t *addr, unsigned size)
581 for (i = 0; i < size; i++) {
582 count += hweight32(*(addr+i));
587 static void fail_log_device(struct log_c *lc)
589 if (lc->log_dev_failed)
592 lc->log_dev_failed = 1;
593 dm_table_event(lc->ti->table);
596 static int disk_resume(struct dm_dirty_log *log)
600 struct log_c *lc = (struct log_c *) log->context;
601 size_t size = lc->bitset_uint32_count * sizeof(uint32_t);
603 /* read the disk header */
606 DMWARN("%s: Failed to read header on dirty region log device",
610 * If the log device cannot be read, we must assume
611 * all regions are out-of-sync. If we simply return
612 * here, the state will be uninitialized and could
613 * lead us to return 'in-sync' status for regions
614 * that are actually 'out-of-sync'.
616 lc->header.nr_regions = 0;
619 /* set or clear any new bits -- device has grown */
620 if (lc->sync == NOSYNC)
621 for (i = lc->header.nr_regions; i < lc->region_count; i++)
622 /* FIXME: amazingly inefficient */
623 log_set_bit(lc, lc->clean_bits, i);
625 for (i = lc->header.nr_regions; i < lc->region_count; i++)
626 /* FIXME: amazingly inefficient */
627 log_clear_bit(lc, lc->clean_bits, i);
629 /* clear any old bits -- device has shrunk */
630 for (i = lc->region_count; i % (sizeof(*lc->clean_bits) << BYTE_SHIFT); i++)
631 log_clear_bit(lc, lc->clean_bits, i);
633 /* copy clean across to sync */
634 memcpy(lc->sync_bits, lc->clean_bits, size);
635 lc->sync_count = count_bits32(lc->clean_bits, lc->bitset_uint32_count);
638 /* set the correct number of regions in the header */
639 lc->header.nr_regions = lc->region_count;
641 header_to_disk(&lc->header, lc->disk_header);
643 /* write the new header */
644 r = rw_header(lc, WRITE);
646 DMWARN("%s: Failed to write header on dirty region log device",
654 static uint32_t core_get_region_size(struct dm_dirty_log *log)
656 struct log_c *lc = (struct log_c *) log->context;
657 return lc->region_size;
660 static int core_resume(struct dm_dirty_log *log)
662 struct log_c *lc = (struct log_c *) log->context;
667 static int core_is_clean(struct dm_dirty_log *log, region_t region)
669 struct log_c *lc = (struct log_c *) log->context;
670 return log_test_bit(lc->clean_bits, region);
673 static int core_in_sync(struct dm_dirty_log *log, region_t region, int block)
675 struct log_c *lc = (struct log_c *) log->context;
676 return log_test_bit(lc->sync_bits, region);
679 static int core_flush(struct dm_dirty_log *log)
685 static int disk_flush(struct dm_dirty_log *log)
688 struct log_c *lc = (struct log_c *) log->context;
690 /* only write if the log has changed */
694 r = rw_header(lc, WRITE);
703 static void core_mark_region(struct dm_dirty_log *log, region_t region)
705 struct log_c *lc = (struct log_c *) log->context;
706 log_clear_bit(lc, lc->clean_bits, region);
709 static void core_clear_region(struct dm_dirty_log *log, region_t region)
711 struct log_c *lc = (struct log_c *) log->context;
712 log_set_bit(lc, lc->clean_bits, region);
715 static int core_get_resync_work(struct dm_dirty_log *log, region_t *region)
717 struct log_c *lc = (struct log_c *) log->context;
719 if (lc->sync_search >= lc->region_count)
723 *region = ext2_find_next_zero_bit(
724 (unsigned long *) lc->sync_bits,
727 lc->sync_search = *region + 1;
729 if (*region >= lc->region_count)
732 } while (log_test_bit(lc->recovering_bits, *region));
734 log_set_bit(lc, lc->recovering_bits, *region);
738 static void core_set_region_sync(struct dm_dirty_log *log, region_t region,
741 struct log_c *lc = (struct log_c *) log->context;
743 log_clear_bit(lc, lc->recovering_bits, region);
745 log_set_bit(lc, lc->sync_bits, region);
747 } else if (log_test_bit(lc->sync_bits, region)) {
749 log_clear_bit(lc, lc->sync_bits, region);
753 static region_t core_get_sync_count(struct dm_dirty_log *log)
755 struct log_c *lc = (struct log_c *) log->context;
757 return lc->sync_count;
760 #define DMEMIT_SYNC \
761 if (lc->sync != DEFAULTSYNC) \
762 DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
764 static int core_status(struct dm_dirty_log *log, status_type_t status,
765 char *result, unsigned int maxlen)
768 struct log_c *lc = log->context;
771 case STATUSTYPE_INFO:
772 DMEMIT("1 %s", log->type->name);
775 case STATUSTYPE_TABLE:
776 DMEMIT("%s %u %u ", log->type->name,
777 lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size);
784 static int disk_status(struct dm_dirty_log *log, status_type_t status,
785 char *result, unsigned int maxlen)
788 struct log_c *lc = log->context;
791 case STATUSTYPE_INFO:
792 DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name,
793 lc->log_dev_failed ? 'D' : 'A');
796 case STATUSTYPE_TABLE:
797 DMEMIT("%s %u %s %u ", log->type->name,
798 lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name,
806 static struct dm_dirty_log_type _core_type = {
808 .module = THIS_MODULE,
811 .resume = core_resume,
812 .get_region_size = core_get_region_size,
813 .is_clean = core_is_clean,
814 .in_sync = core_in_sync,
816 .mark_region = core_mark_region,
817 .clear_region = core_clear_region,
818 .get_resync_work = core_get_resync_work,
819 .set_region_sync = core_set_region_sync,
820 .get_sync_count = core_get_sync_count,
821 .status = core_status,
824 static struct dm_dirty_log_type _disk_type = {
826 .module = THIS_MODULE,
829 .postsuspend = disk_flush,
830 .resume = disk_resume,
831 .get_region_size = core_get_region_size,
832 .is_clean = core_is_clean,
833 .in_sync = core_in_sync,
835 .mark_region = core_mark_region,
836 .clear_region = core_clear_region,
837 .get_resync_work = core_get_resync_work,
838 .set_region_sync = core_set_region_sync,
839 .get_sync_count = core_get_sync_count,
840 .status = disk_status,
843 static int __init dm_dirty_log_init(void)
847 r = dm_dirty_log_type_register(&_core_type);
849 DMWARN("couldn't register core log");
851 r = dm_dirty_log_type_register(&_disk_type);
853 DMWARN("couldn't register disk type");
854 dm_dirty_log_type_unregister(&_core_type);
860 static void __exit dm_dirty_log_exit(void)
862 dm_dirty_log_type_unregister(&_disk_type);
863 dm_dirty_log_type_unregister(&_core_type);
866 module_init(dm_dirty_log_init);
867 module_exit(dm_dirty_log_exit);
869 MODULE_DESCRIPTION(DM_NAME " dirty region log");
870 MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@redhat.com>");
871 MODULE_LICENSE("GPL");