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 int (*flush_callback_fn)(struct dm_target *ti),
150 unsigned int argc, char **argv)
152 struct dm_dirty_log_type *type;
153 struct dm_dirty_log *log;
155 log = kmalloc(sizeof(*log), GFP_KERNEL);
159 type = get_type(type_name);
165 log->flush_callback_fn = flush_callback_fn;
167 if (type->ctr(log, ti, argc, argv)) {
175 EXPORT_SYMBOL(dm_dirty_log_create);
177 void dm_dirty_log_destroy(struct dm_dirty_log *log)
183 EXPORT_SYMBOL(dm_dirty_log_destroy);
185 /*-----------------------------------------------------------------
186 * Persistent and core logs share a lot of their implementation.
187 * FIXME: need a reload method to be called from a resume
188 *---------------------------------------------------------------*/
190 * Magic for persistent mirrors: "MiRr"
192 #define MIRROR_MAGIC 0x4D695272
195 * The on-disk version of the metadata.
197 #define MIRROR_DISK_VERSION 2
200 struct log_header_disk {
204 * Simple, incrementing version. no backward
211 struct log_header_core {
218 struct dm_target *ti;
222 uint32_t region_size;
223 unsigned int region_count;
226 unsigned bitset_uint32_count;
227 uint32_t *clean_bits;
229 uint32_t *recovering_bits; /* FIXME: this seems excessive */
235 DEFAULTSYNC, /* Synchronize if necessary */
236 NOSYNC, /* Devices known to be already in sync */
237 FORCESYNC, /* Force a sync to happen */
240 struct dm_io_request io_req;
246 int log_dev_flush_failed;
247 struct dm_dev *log_dev;
248 struct log_header_core header;
250 struct dm_io_region header_location;
251 struct log_header_disk *disk_header;
255 * The touched member needs to be updated every time we access
256 * one of the bitsets.
258 static inline int log_test_bit(uint32_t *bs, unsigned bit)
260 return test_bit_le(bit, bs) ? 1 : 0;
263 static inline void log_set_bit(struct log_c *l,
264 uint32_t *bs, unsigned bit)
266 __set_bit_le(bit, bs);
267 l->touched_cleaned = 1;
270 static inline void log_clear_bit(struct log_c *l,
271 uint32_t *bs, unsigned bit)
273 __clear_bit_le(bit, bs);
274 l->touched_dirtied = 1;
277 /*----------------------------------------------------------------
279 *--------------------------------------------------------------*/
280 static void header_to_disk(struct log_header_core *core, struct log_header_disk *disk)
282 disk->magic = cpu_to_le32(core->magic);
283 disk->version = cpu_to_le32(core->version);
284 disk->nr_regions = cpu_to_le64(core->nr_regions);
287 static void header_from_disk(struct log_header_core *core, struct log_header_disk *disk)
289 core->magic = le32_to_cpu(disk->magic);
290 core->version = le32_to_cpu(disk->version);
291 core->nr_regions = le64_to_cpu(disk->nr_regions);
294 static int rw_header(struct log_c *lc, enum req_op op)
296 lc->io_req.bi_opf = op;
298 return dm_io(&lc->io_req, 1, &lc->header_location, NULL);
301 static int flush_header(struct log_c *lc)
303 struct dm_io_region null_location = {
304 .bdev = lc->header_location.bdev,
309 lc->io_req.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
311 return dm_io(&lc->io_req, 1, &null_location, NULL);
314 static int read_header(struct log_c *log)
318 r = rw_header(log, REQ_OP_READ);
322 header_from_disk(&log->header, log->disk_header);
324 /* New log required? */
325 if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) {
326 log->header.magic = MIRROR_MAGIC;
327 log->header.version = MIRROR_DISK_VERSION;
328 log->header.nr_regions = 0;
331 #ifdef __LITTLE_ENDIAN
332 if (log->header.version == 1)
333 log->header.version = 2;
336 if (log->header.version != MIRROR_DISK_VERSION) {
337 DMWARN("incompatible disk log version");
344 static int _check_region_size(struct dm_target *ti, uint32_t region_size)
346 if (region_size < 2 || region_size > ti->len)
349 if (!is_power_of_2(region_size))
355 /*----------------------------------------------------------------
356 * core log constructor/destructor
358 * argv contains region_size followed optionally by [no]sync
359 *--------------------------------------------------------------*/
361 static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti,
362 unsigned int argc, char **argv,
365 enum sync sync = DEFAULTSYNC;
368 uint32_t region_size;
369 unsigned int region_count;
370 size_t bitset_size, buf_size;
374 if (argc < 1 || argc > 2) {
375 DMWARN("wrong number of arguments to dirty region log");
380 if (!strcmp(argv[1], "sync"))
382 else if (!strcmp(argv[1], "nosync"))
385 DMWARN("unrecognised sync argument to "
386 "dirty region log: %s", argv[1]);
391 if (sscanf(argv[0], "%u%c", ®ion_size, &dummy) != 1 ||
392 !_check_region_size(ti, region_size)) {
393 DMWARN("invalid region size %s", argv[0]);
397 region_count = dm_sector_div_up(ti->len, region_size);
399 lc = kmalloc(sizeof(*lc), GFP_KERNEL);
401 DMWARN("couldn't allocate core log");
406 lc->touched_dirtied = 0;
407 lc->touched_cleaned = 0;
408 lc->flush_failed = 0;
409 lc->region_size = region_size;
410 lc->region_count = region_count;
414 * Work out how many "unsigned long"s we need to hold the bitset.
416 bitset_size = dm_round_up(region_count, BITS_PER_LONG);
417 bitset_size >>= BYTE_SHIFT;
419 lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits);
425 lc->clean_bits = vmalloc(bitset_size);
426 if (!lc->clean_bits) {
427 DMWARN("couldn't allocate clean bitset");
431 lc->disk_header = NULL;
434 lc->log_dev_failed = 0;
435 lc->log_dev_flush_failed = 0;
436 lc->header_location.bdev = lc->log_dev->bdev;
437 lc->header_location.sector = 0;
440 * Buffer holds both header and bitset.
443 dm_round_up((LOG_OFFSET << SECTOR_SHIFT) + bitset_size,
444 bdev_logical_block_size(lc->header_location.
447 if (buf_size > bdev_nr_bytes(dev->bdev)) {
448 DMWARN("log device %s too small: need %llu bytes",
449 dev->name, (unsigned long long)buf_size);
454 lc->header_location.count = buf_size >> SECTOR_SHIFT;
456 lc->io_req.mem.type = DM_IO_VMA;
457 lc->io_req.notify.fn = NULL;
458 lc->io_req.client = dm_io_client_create();
459 if (IS_ERR(lc->io_req.client)) {
460 r = PTR_ERR(lc->io_req.client);
461 DMWARN("couldn't allocate disk io client");
466 lc->disk_header = vmalloc(buf_size);
467 if (!lc->disk_header) {
468 DMWARN("couldn't allocate disk log buffer");
469 dm_io_client_destroy(lc->io_req.client);
474 lc->io_req.mem.ptr.vma = lc->disk_header;
475 lc->clean_bits = (void *)lc->disk_header +
476 (LOG_OFFSET << SECTOR_SHIFT);
479 memset(lc->clean_bits, -1, bitset_size);
481 lc->sync_bits = vmalloc(bitset_size);
482 if (!lc->sync_bits) {
483 DMWARN("couldn't allocate sync bitset");
485 vfree(lc->clean_bits);
487 dm_io_client_destroy(lc->io_req.client);
488 vfree(lc->disk_header);
492 memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
493 lc->sync_count = (sync == NOSYNC) ? region_count : 0;
495 lc->recovering_bits = vzalloc(bitset_size);
496 if (!lc->recovering_bits) {
497 DMWARN("couldn't allocate sync bitset");
498 vfree(lc->sync_bits);
500 vfree(lc->clean_bits);
502 dm_io_client_destroy(lc->io_req.client);
503 vfree(lc->disk_header);
513 static int core_ctr(struct dm_dirty_log *log, struct dm_target *ti,
514 unsigned int argc, char **argv)
516 return create_log_context(log, ti, argc, argv, NULL);
519 static void destroy_log_context(struct log_c *lc)
521 vfree(lc->sync_bits);
522 vfree(lc->recovering_bits);
526 static void core_dtr(struct dm_dirty_log *log)
528 struct log_c *lc = (struct log_c *) log->context;
530 vfree(lc->clean_bits);
531 destroy_log_context(lc);
534 /*----------------------------------------------------------------
535 * disk log constructor/destructor
537 * argv contains log_device region_size followed optionally by [no]sync
538 *--------------------------------------------------------------*/
539 static int disk_ctr(struct dm_dirty_log *log, struct dm_target *ti,
540 unsigned int argc, char **argv)
545 if (argc < 2 || argc > 3) {
546 DMWARN("wrong number of arguments to disk dirty region log");
550 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev);
554 r = create_log_context(log, ti, argc - 1, argv + 1, dev);
556 dm_put_device(ti, dev);
563 static void disk_dtr(struct dm_dirty_log *log)
565 struct log_c *lc = (struct log_c *) log->context;
567 dm_put_device(lc->ti, lc->log_dev);
568 vfree(lc->disk_header);
569 dm_io_client_destroy(lc->io_req.client);
570 destroy_log_context(lc);
573 static void fail_log_device(struct log_c *lc)
575 if (lc->log_dev_failed)
578 lc->log_dev_failed = 1;
579 dm_table_event(lc->ti->table);
582 static int disk_resume(struct dm_dirty_log *log)
586 struct log_c *lc = (struct log_c *) log->context;
587 size_t size = lc->bitset_uint32_count * sizeof(uint32_t);
589 /* read the disk header */
592 DMWARN("%s: Failed to read header on dirty region log device",
596 * If the log device cannot be read, we must assume
597 * all regions are out-of-sync. If we simply return
598 * here, the state will be uninitialized and could
599 * lead us to return 'in-sync' status for regions
600 * that are actually 'out-of-sync'.
602 lc->header.nr_regions = 0;
605 /* set or clear any new bits -- device has grown */
606 if (lc->sync == NOSYNC)
607 for (i = lc->header.nr_regions; i < lc->region_count; i++)
608 /* FIXME: amazingly inefficient */
609 log_set_bit(lc, lc->clean_bits, i);
611 for (i = lc->header.nr_regions; i < lc->region_count; i++)
612 /* FIXME: amazingly inefficient */
613 log_clear_bit(lc, lc->clean_bits, i);
615 /* clear any old bits -- device has shrunk */
616 for (i = lc->region_count; i % BITS_PER_LONG; i++)
617 log_clear_bit(lc, lc->clean_bits, i);
619 /* copy clean across to sync */
620 memcpy(lc->sync_bits, lc->clean_bits, size);
621 lc->sync_count = memweight(lc->clean_bits,
622 lc->bitset_uint32_count * sizeof(uint32_t));
625 /* set the correct number of regions in the header */
626 lc->header.nr_regions = lc->region_count;
628 header_to_disk(&lc->header, lc->disk_header);
630 /* write the new header */
631 r = rw_header(lc, REQ_OP_WRITE);
633 r = flush_header(lc);
635 lc->log_dev_flush_failed = 1;
638 DMWARN("%s: Failed to write header on dirty region log device",
646 static uint32_t core_get_region_size(struct dm_dirty_log *log)
648 struct log_c *lc = (struct log_c *) log->context;
649 return lc->region_size;
652 static int core_resume(struct dm_dirty_log *log)
654 struct log_c *lc = (struct log_c *) log->context;
659 static int core_is_clean(struct dm_dirty_log *log, region_t region)
661 struct log_c *lc = (struct log_c *) log->context;
662 return log_test_bit(lc->clean_bits, region);
665 static int core_in_sync(struct dm_dirty_log *log, region_t region, int block)
667 struct log_c *lc = (struct log_c *) log->context;
668 return log_test_bit(lc->sync_bits, region);
671 static int core_flush(struct dm_dirty_log *log)
677 static int disk_flush(struct dm_dirty_log *log)
680 struct log_c *lc = log->context;
682 /* only write if the log has changed */
683 if (!lc->touched_cleaned && !lc->touched_dirtied)
686 if (lc->touched_cleaned && log->flush_callback_fn &&
687 log->flush_callback_fn(lc->ti)) {
689 * At this point it is impossible to determine which
690 * regions are clean and which are dirty (without
691 * re-reading the log off disk). So mark all of them
694 lc->flush_failed = 1;
695 for (i = 0; i < lc->region_count; i++)
696 log_clear_bit(lc, lc->clean_bits, i);
699 r = rw_header(lc, REQ_OP_WRITE);
703 if (lc->touched_dirtied) {
704 r = flush_header(lc);
706 lc->log_dev_flush_failed = 1;
709 lc->touched_dirtied = 0;
711 lc->touched_cleaned = 0;
717 static void core_mark_region(struct dm_dirty_log *log, region_t region)
719 struct log_c *lc = (struct log_c *) log->context;
720 log_clear_bit(lc, lc->clean_bits, region);
723 static void core_clear_region(struct dm_dirty_log *log, region_t region)
725 struct log_c *lc = (struct log_c *) log->context;
726 if (likely(!lc->flush_failed))
727 log_set_bit(lc, lc->clean_bits, region);
730 static int core_get_resync_work(struct dm_dirty_log *log, region_t *region)
732 struct log_c *lc = (struct log_c *) log->context;
734 if (lc->sync_search >= lc->region_count)
738 *region = find_next_zero_bit_le(lc->sync_bits,
741 lc->sync_search = *region + 1;
743 if (*region >= lc->region_count)
746 } while (log_test_bit(lc->recovering_bits, *region));
748 log_set_bit(lc, lc->recovering_bits, *region);
752 static void core_set_region_sync(struct dm_dirty_log *log, region_t region,
755 struct log_c *lc = (struct log_c *) log->context;
757 log_clear_bit(lc, lc->recovering_bits, region);
759 log_set_bit(lc, lc->sync_bits, region);
761 } else if (log_test_bit(lc->sync_bits, region)) {
763 log_clear_bit(lc, lc->sync_bits, region);
767 static region_t core_get_sync_count(struct dm_dirty_log *log)
769 struct log_c *lc = (struct log_c *) log->context;
771 return lc->sync_count;
774 #define DMEMIT_SYNC \
775 if (lc->sync != DEFAULTSYNC) \
776 DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
778 static int core_status(struct dm_dirty_log *log, status_type_t status,
779 char *result, unsigned int maxlen)
782 struct log_c *lc = log->context;
785 case STATUSTYPE_INFO:
786 DMEMIT("1 %s", log->type->name);
789 case STATUSTYPE_TABLE:
790 DMEMIT("%s %u %u ", log->type->name,
791 lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size);
803 static int disk_status(struct dm_dirty_log *log, status_type_t status,
804 char *result, unsigned int maxlen)
807 struct log_c *lc = log->context;
810 case STATUSTYPE_INFO:
811 DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name,
812 lc->log_dev_flush_failed ? 'F' :
813 lc->log_dev_failed ? 'D' :
817 case STATUSTYPE_TABLE:
818 DMEMIT("%s %u %s %u ", log->type->name,
819 lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name,
832 static struct dm_dirty_log_type _core_type = {
834 .module = THIS_MODULE,
837 .resume = core_resume,
838 .get_region_size = core_get_region_size,
839 .is_clean = core_is_clean,
840 .in_sync = core_in_sync,
842 .mark_region = core_mark_region,
843 .clear_region = core_clear_region,
844 .get_resync_work = core_get_resync_work,
845 .set_region_sync = core_set_region_sync,
846 .get_sync_count = core_get_sync_count,
847 .status = core_status,
850 static struct dm_dirty_log_type _disk_type = {
852 .module = THIS_MODULE,
855 .postsuspend = disk_flush,
856 .resume = disk_resume,
857 .get_region_size = core_get_region_size,
858 .is_clean = core_is_clean,
859 .in_sync = core_in_sync,
861 .mark_region = core_mark_region,
862 .clear_region = core_clear_region,
863 .get_resync_work = core_get_resync_work,
864 .set_region_sync = core_set_region_sync,
865 .get_sync_count = core_get_sync_count,
866 .status = disk_status,
869 static int __init dm_dirty_log_init(void)
873 r = dm_dirty_log_type_register(&_core_type);
875 DMWARN("couldn't register core log");
877 r = dm_dirty_log_type_register(&_disk_type);
879 DMWARN("couldn't register disk type");
880 dm_dirty_log_type_unregister(&_core_type);
886 static void __exit dm_dirty_log_exit(void)
888 dm_dirty_log_type_unregister(&_disk_type);
889 dm_dirty_log_type_unregister(&_core_type);
892 module_init(dm_dirty_log_init);
893 module_exit(dm_dirty_log_exit);
895 MODULE_DESCRIPTION(DM_NAME " dirty region log");
896 MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@redhat.com>");
897 MODULE_LICENSE("GPL");