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;
214 uint32_t region_size;
215 unsigned int region_count;
218 unsigned bitset_uint32_count;
219 uint32_t *clean_bits;
221 uint32_t *recovering_bits; /* FIXME: this seems excessive */
227 DEFAULTSYNC, /* Synchronize if necessary */
228 NOSYNC, /* Devices known to be already in sync */
229 FORCESYNC, /* Force a sync to happen */
232 struct dm_io_request io_req;
238 struct dm_dev *log_dev;
239 struct log_header header;
241 struct dm_io_region header_location;
242 struct log_header *disk_header;
246 * The touched member needs to be updated every time we access
247 * one of the bitsets.
249 static inline int log_test_bit(uint32_t *bs, unsigned bit)
251 return ext2_test_bit(bit, (unsigned long *) bs) ? 1 : 0;
254 static inline void log_set_bit(struct log_c *l,
255 uint32_t *bs, unsigned bit)
257 ext2_set_bit(bit, (unsigned long *) bs);
258 l->touched_cleaned = 1;
261 static inline void log_clear_bit(struct log_c *l,
262 uint32_t *bs, unsigned bit)
264 ext2_clear_bit(bit, (unsigned long *) bs);
265 l->touched_dirtied = 1;
268 /*----------------------------------------------------------------
270 *--------------------------------------------------------------*/
271 static void header_to_disk(struct log_header *core, struct log_header *disk)
273 disk->magic = cpu_to_le32(core->magic);
274 disk->version = cpu_to_le32(core->version);
275 disk->nr_regions = cpu_to_le64(core->nr_regions);
278 static void header_from_disk(struct log_header *core, struct log_header *disk)
280 core->magic = le32_to_cpu(disk->magic);
281 core->version = le32_to_cpu(disk->version);
282 core->nr_regions = le64_to_cpu(disk->nr_regions);
285 static int rw_header(struct log_c *lc, int rw)
287 lc->io_req.bi_rw = rw;
289 return dm_io(&lc->io_req, 1, &lc->header_location, NULL);
292 static int flush_header(struct log_c *lc)
294 struct dm_io_region null_location = {
295 .bdev = lc->header_location.bdev,
300 lc->io_req.bi_rw = WRITE_BARRIER;
302 return dm_io(&lc->io_req, 1, &null_location, NULL);
305 static int read_header(struct log_c *log)
309 r = rw_header(log, READ);
313 header_from_disk(&log->header, log->disk_header);
315 /* New log required? */
316 if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) {
317 log->header.magic = MIRROR_MAGIC;
318 log->header.version = MIRROR_DISK_VERSION;
319 log->header.nr_regions = 0;
322 #ifdef __LITTLE_ENDIAN
323 if (log->header.version == 1)
324 log->header.version = 2;
327 if (log->header.version != MIRROR_DISK_VERSION) {
328 DMWARN("incompatible disk log version");
335 static int _check_region_size(struct dm_target *ti, uint32_t region_size)
337 if (region_size < 2 || region_size > ti->len)
340 if (!is_power_of_2(region_size))
346 /*----------------------------------------------------------------
347 * core log constructor/destructor
349 * argv contains region_size followed optionally by [no]sync
350 *--------------------------------------------------------------*/
352 static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti,
353 unsigned int argc, char **argv,
356 enum sync sync = DEFAULTSYNC;
359 uint32_t region_size;
360 unsigned int region_count;
361 size_t bitset_size, buf_size;
364 if (argc < 1 || argc > 2) {
365 DMWARN("wrong number of arguments to dirty region log");
370 if (!strcmp(argv[1], "sync"))
372 else if (!strcmp(argv[1], "nosync"))
375 DMWARN("unrecognised sync argument to "
376 "dirty region log: %s", argv[1]);
381 if (sscanf(argv[0], "%u", ®ion_size) != 1 ||
382 !_check_region_size(ti, region_size)) {
383 DMWARN("invalid region size %s", argv[0]);
387 region_count = dm_sector_div_up(ti->len, region_size);
389 lc = kmalloc(sizeof(*lc), GFP_KERNEL);
391 DMWARN("couldn't allocate core log");
396 lc->touched_dirtied = 0;
397 lc->touched_cleaned = 0;
398 lc->flush_failed = 0;
399 lc->region_size = region_size;
400 lc->region_count = region_count;
404 * Work out how many "unsigned long"s we need to hold the bitset.
406 bitset_size = dm_round_up(region_count,
407 sizeof(*lc->clean_bits) << BYTE_SHIFT);
408 bitset_size >>= BYTE_SHIFT;
410 lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits);
416 lc->clean_bits = vmalloc(bitset_size);
417 if (!lc->clean_bits) {
418 DMWARN("couldn't allocate clean bitset");
422 lc->disk_header = NULL;
425 lc->log_dev_failed = 0;
426 lc->header_location.bdev = lc->log_dev->bdev;
427 lc->header_location.sector = 0;
430 * Buffer holds both header and bitset.
433 dm_round_up((LOG_OFFSET << SECTOR_SHIFT) + bitset_size,
434 bdev_logical_block_size(lc->header_location.
437 if (buf_size > i_size_read(dev->bdev->bd_inode)) {
438 DMWARN("log device %s too small: need %llu bytes",
439 dev->name, (unsigned long long)buf_size);
444 lc->header_location.count = buf_size >> SECTOR_SHIFT;
446 lc->io_req.mem.type = DM_IO_VMA;
447 lc->io_req.notify.fn = NULL;
448 lc->io_req.client = dm_io_client_create(dm_div_up(buf_size,
450 if (IS_ERR(lc->io_req.client)) {
451 r = PTR_ERR(lc->io_req.client);
452 DMWARN("couldn't allocate disk io client");
457 lc->disk_header = vmalloc(buf_size);
458 if (!lc->disk_header) {
459 DMWARN("couldn't allocate disk log buffer");
460 dm_io_client_destroy(lc->io_req.client);
465 lc->io_req.mem.ptr.vma = lc->disk_header;
466 lc->clean_bits = (void *)lc->disk_header +
467 (LOG_OFFSET << SECTOR_SHIFT);
470 memset(lc->clean_bits, -1, bitset_size);
472 lc->sync_bits = vmalloc(bitset_size);
473 if (!lc->sync_bits) {
474 DMWARN("couldn't allocate sync bitset");
476 vfree(lc->clean_bits);
478 dm_io_client_destroy(lc->io_req.client);
479 vfree(lc->disk_header);
483 memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
484 lc->sync_count = (sync == NOSYNC) ? region_count : 0;
486 lc->recovering_bits = vmalloc(bitset_size);
487 if (!lc->recovering_bits) {
488 DMWARN("couldn't allocate sync bitset");
489 vfree(lc->sync_bits);
491 vfree(lc->clean_bits);
493 dm_io_client_destroy(lc->io_req.client);
494 vfree(lc->disk_header);
498 memset(lc->recovering_bits, 0, bitset_size);
505 static int core_ctr(struct dm_dirty_log *log, struct dm_target *ti,
506 unsigned int argc, char **argv)
508 return create_log_context(log, ti, argc, argv, NULL);
511 static void destroy_log_context(struct log_c *lc)
513 vfree(lc->sync_bits);
514 vfree(lc->recovering_bits);
518 static void core_dtr(struct dm_dirty_log *log)
520 struct log_c *lc = (struct log_c *) log->context;
522 vfree(lc->clean_bits);
523 destroy_log_context(lc);
526 /*----------------------------------------------------------------
527 * disk log constructor/destructor
529 * argv contains log_device region_size followed optionally by [no]sync
530 *--------------------------------------------------------------*/
531 static int disk_ctr(struct dm_dirty_log *log, struct dm_target *ti,
532 unsigned int argc, char **argv)
537 if (argc < 2 || argc > 3) {
538 DMWARN("wrong number of arguments to disk dirty region log");
542 r = dm_get_device(ti, argv[0], 0, 0 /* FIXME */,
543 FMODE_READ | FMODE_WRITE, &dev);
547 r = create_log_context(log, ti, argc - 1, argv + 1, dev);
549 dm_put_device(ti, dev);
556 static void disk_dtr(struct dm_dirty_log *log)
558 struct log_c *lc = (struct log_c *) log->context;
560 dm_put_device(lc->ti, lc->log_dev);
561 vfree(lc->disk_header);
562 dm_io_client_destroy(lc->io_req.client);
563 destroy_log_context(lc);
566 static int count_bits32(uint32_t *addr, unsigned size)
570 for (i = 0; i < size; i++) {
571 count += hweight32(*(addr+i));
576 static void fail_log_device(struct log_c *lc)
578 if (lc->log_dev_failed)
581 lc->log_dev_failed = 1;
582 dm_table_event(lc->ti->table);
585 static int disk_resume(struct dm_dirty_log *log)
589 struct log_c *lc = (struct log_c *) log->context;
590 size_t size = lc->bitset_uint32_count * sizeof(uint32_t);
592 /* read the disk header */
595 DMWARN("%s: Failed to read header on dirty region log device",
599 * If the log device cannot be read, we must assume
600 * all regions are out-of-sync. If we simply return
601 * here, the state will be uninitialized and could
602 * lead us to return 'in-sync' status for regions
603 * that are actually 'out-of-sync'.
605 lc->header.nr_regions = 0;
608 /* set or clear any new bits -- device has grown */
609 if (lc->sync == NOSYNC)
610 for (i = lc->header.nr_regions; i < lc->region_count; i++)
611 /* FIXME: amazingly inefficient */
612 log_set_bit(lc, lc->clean_bits, i);
614 for (i = lc->header.nr_regions; i < lc->region_count; i++)
615 /* FIXME: amazingly inefficient */
616 log_clear_bit(lc, lc->clean_bits, i);
618 /* clear any old bits -- device has shrunk */
619 for (i = lc->region_count; i % (sizeof(*lc->clean_bits) << BYTE_SHIFT); i++)
620 log_clear_bit(lc, lc->clean_bits, i);
622 /* copy clean across to sync */
623 memcpy(lc->sync_bits, lc->clean_bits, size);
624 lc->sync_count = count_bits32(lc->clean_bits, lc->bitset_uint32_count);
627 /* set the correct number of regions in the header */
628 lc->header.nr_regions = lc->region_count;
630 header_to_disk(&lc->header, lc->disk_header);
632 /* write the new header */
633 r = rw_header(lc, WRITE);
635 r = flush_header(lc);
637 DMWARN("%s: Failed to write header on dirty region log device",
645 static uint32_t core_get_region_size(struct dm_dirty_log *log)
647 struct log_c *lc = (struct log_c *) log->context;
648 return lc->region_size;
651 static int core_resume(struct dm_dirty_log *log)
653 struct log_c *lc = (struct log_c *) log->context;
658 static int core_is_clean(struct dm_dirty_log *log, region_t region)
660 struct log_c *lc = (struct log_c *) log->context;
661 return log_test_bit(lc->clean_bits, region);
664 static int core_in_sync(struct dm_dirty_log *log, region_t region, int block)
666 struct log_c *lc = (struct log_c *) log->context;
667 return log_test_bit(lc->sync_bits, region);
670 static int core_flush(struct dm_dirty_log *log)
676 static int disk_flush(struct dm_dirty_log *log)
679 struct log_c *lc = (struct log_c *) log->context;
681 /* only write if the log has changed */
682 if (!lc->touched_cleaned && !lc->touched_dirtied)
685 r = rw_header(lc, WRITE);
689 if (lc->touched_dirtied) {
690 r = flush_header(lc);
694 lc->touched_dirtied = 0;
696 lc->touched_cleaned = 0;
702 static void core_mark_region(struct dm_dirty_log *log, region_t region)
704 struct log_c *lc = (struct log_c *) log->context;
705 log_clear_bit(lc, lc->clean_bits, region);
708 static void core_clear_region(struct dm_dirty_log *log, region_t region)
710 struct log_c *lc = (struct log_c *) log->context;
711 if (likely(!lc->flush_failed))
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");