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, int op)
296 lc->io_req.bi_op = op;
297 lc->io_req.bi_op_flags = 0;
299 return dm_io(&lc->io_req, 1, &lc->header_location, NULL);
302 static int flush_header(struct log_c *lc)
304 struct dm_io_region null_location = {
305 .bdev = lc->header_location.bdev,
310 lc->io_req.bi_op = REQ_OP_WRITE;
311 lc->io_req.bi_op_flags = REQ_PREFLUSH;
313 return dm_io(&lc->io_req, 1, &null_location, NULL);
316 static int read_header(struct log_c *log)
320 r = rw_header(log, REQ_OP_READ);
324 header_from_disk(&log->header, log->disk_header);
326 /* New log required? */
327 if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) {
328 log->header.magic = MIRROR_MAGIC;
329 log->header.version = MIRROR_DISK_VERSION;
330 log->header.nr_regions = 0;
333 #ifdef __LITTLE_ENDIAN
334 if (log->header.version == 1)
335 log->header.version = 2;
338 if (log->header.version != MIRROR_DISK_VERSION) {
339 DMWARN("incompatible disk log version");
346 static int _check_region_size(struct dm_target *ti, uint32_t region_size)
348 if (region_size < 2 || region_size > ti->len)
351 if (!is_power_of_2(region_size))
357 /*----------------------------------------------------------------
358 * core log constructor/destructor
360 * argv contains region_size followed optionally by [no]sync
361 *--------------------------------------------------------------*/
363 static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti,
364 unsigned int argc, char **argv,
367 enum sync sync = DEFAULTSYNC;
370 uint32_t region_size;
371 unsigned int region_count;
372 size_t bitset_size, buf_size;
376 if (argc < 1 || argc > 2) {
377 DMWARN("wrong number of arguments to dirty region log");
382 if (!strcmp(argv[1], "sync"))
384 else if (!strcmp(argv[1], "nosync"))
387 DMWARN("unrecognised sync argument to "
388 "dirty region log: %s", argv[1]);
393 if (sscanf(argv[0], "%u%c", ®ion_size, &dummy) != 1 ||
394 !_check_region_size(ti, region_size)) {
395 DMWARN("invalid region size %s", argv[0]);
399 region_count = dm_sector_div_up(ti->len, region_size);
401 lc = kmalloc(sizeof(*lc), GFP_KERNEL);
403 DMWARN("couldn't allocate core log");
408 lc->touched_dirtied = 0;
409 lc->touched_cleaned = 0;
410 lc->flush_failed = 0;
411 lc->region_size = region_size;
412 lc->region_count = region_count;
416 * Work out how many "unsigned long"s we need to hold the bitset.
418 bitset_size = dm_round_up(region_count,
419 sizeof(*lc->clean_bits) << BYTE_SHIFT);
420 bitset_size >>= BYTE_SHIFT;
422 lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits);
428 lc->clean_bits = vmalloc(bitset_size);
429 if (!lc->clean_bits) {
430 DMWARN("couldn't allocate clean bitset");
434 lc->disk_header = NULL;
437 lc->log_dev_failed = 0;
438 lc->log_dev_flush_failed = 0;
439 lc->header_location.bdev = lc->log_dev->bdev;
440 lc->header_location.sector = 0;
443 * Buffer holds both header and bitset.
446 dm_round_up((LOG_OFFSET << SECTOR_SHIFT) + bitset_size,
447 bdev_logical_block_size(lc->header_location.
450 if (buf_size > i_size_read(dev->bdev->bd_inode)) {
451 DMWARN("log device %s too small: need %llu bytes",
452 dev->name, (unsigned long long)buf_size);
457 lc->header_location.count = buf_size >> SECTOR_SHIFT;
459 lc->io_req.mem.type = DM_IO_VMA;
460 lc->io_req.notify.fn = NULL;
461 lc->io_req.client = dm_io_client_create();
462 if (IS_ERR(lc->io_req.client)) {
463 r = PTR_ERR(lc->io_req.client);
464 DMWARN("couldn't allocate disk io client");
469 lc->disk_header = vmalloc(buf_size);
470 if (!lc->disk_header) {
471 DMWARN("couldn't allocate disk log buffer");
472 dm_io_client_destroy(lc->io_req.client);
477 lc->io_req.mem.ptr.vma = lc->disk_header;
478 lc->clean_bits = (void *)lc->disk_header +
479 (LOG_OFFSET << SECTOR_SHIFT);
482 memset(lc->clean_bits, -1, bitset_size);
484 lc->sync_bits = vmalloc(bitset_size);
485 if (!lc->sync_bits) {
486 DMWARN("couldn't allocate sync bitset");
488 vfree(lc->clean_bits);
490 dm_io_client_destroy(lc->io_req.client);
491 vfree(lc->disk_header);
495 memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
496 lc->sync_count = (sync == NOSYNC) ? region_count : 0;
498 lc->recovering_bits = vzalloc(bitset_size);
499 if (!lc->recovering_bits) {
500 DMWARN("couldn't allocate sync bitset");
501 vfree(lc->sync_bits);
503 vfree(lc->clean_bits);
505 dm_io_client_destroy(lc->io_req.client);
506 vfree(lc->disk_header);
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], dm_table_get_mode(ti->table), &dev);
557 r = create_log_context(log, ti, argc - 1, argv + 1, dev);
559 dm_put_device(ti, dev);
566 static void disk_dtr(struct dm_dirty_log *log)
568 struct log_c *lc = (struct log_c *) log->context;
570 dm_put_device(lc->ti, lc->log_dev);
571 vfree(lc->disk_header);
572 dm_io_client_destroy(lc->io_req.client);
573 destroy_log_context(lc);
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 = memweight(lc->clean_bits,
625 lc->bitset_uint32_count * sizeof(uint32_t));
628 /* set the correct number of regions in the header */
629 lc->header.nr_regions = lc->region_count;
631 header_to_disk(&lc->header, lc->disk_header);
633 /* write the new header */
634 r = rw_header(lc, REQ_OP_WRITE);
636 r = flush_header(lc);
638 lc->log_dev_flush_failed = 1;
641 DMWARN("%s: Failed to write header on dirty region log device",
649 static uint32_t core_get_region_size(struct dm_dirty_log *log)
651 struct log_c *lc = (struct log_c *) log->context;
652 return lc->region_size;
655 static int core_resume(struct dm_dirty_log *log)
657 struct log_c *lc = (struct log_c *) log->context;
662 static int core_is_clean(struct dm_dirty_log *log, region_t region)
664 struct log_c *lc = (struct log_c *) log->context;
665 return log_test_bit(lc->clean_bits, region);
668 static int core_in_sync(struct dm_dirty_log *log, region_t region, int block)
670 struct log_c *lc = (struct log_c *) log->context;
671 return log_test_bit(lc->sync_bits, region);
674 static int core_flush(struct dm_dirty_log *log)
680 static int disk_flush(struct dm_dirty_log *log)
683 struct log_c *lc = log->context;
685 /* only write if the log has changed */
686 if (!lc->touched_cleaned && !lc->touched_dirtied)
689 if (lc->touched_cleaned && log->flush_callback_fn &&
690 log->flush_callback_fn(lc->ti)) {
692 * At this point it is impossible to determine which
693 * regions are clean and which are dirty (without
694 * re-reading the log off disk). So mark all of them
697 lc->flush_failed = 1;
698 for (i = 0; i < lc->region_count; i++)
699 log_clear_bit(lc, lc->clean_bits, i);
702 r = rw_header(lc, REQ_OP_WRITE);
706 if (lc->touched_dirtied) {
707 r = flush_header(lc);
709 lc->log_dev_flush_failed = 1;
712 lc->touched_dirtied = 0;
714 lc->touched_cleaned = 0;
720 static void core_mark_region(struct dm_dirty_log *log, region_t region)
722 struct log_c *lc = (struct log_c *) log->context;
723 log_clear_bit(lc, lc->clean_bits, region);
726 static void core_clear_region(struct dm_dirty_log *log, region_t region)
728 struct log_c *lc = (struct log_c *) log->context;
729 if (likely(!lc->flush_failed))
730 log_set_bit(lc, lc->clean_bits, region);
733 static int core_get_resync_work(struct dm_dirty_log *log, region_t *region)
735 struct log_c *lc = (struct log_c *) log->context;
737 if (lc->sync_search >= lc->region_count)
741 *region = find_next_zero_bit_le(lc->sync_bits,
744 lc->sync_search = *region + 1;
746 if (*region >= lc->region_count)
749 } while (log_test_bit(lc->recovering_bits, *region));
751 log_set_bit(lc, lc->recovering_bits, *region);
755 static void core_set_region_sync(struct dm_dirty_log *log, region_t region,
758 struct log_c *lc = (struct log_c *) log->context;
760 log_clear_bit(lc, lc->recovering_bits, region);
762 log_set_bit(lc, lc->sync_bits, region);
764 } else if (log_test_bit(lc->sync_bits, region)) {
766 log_clear_bit(lc, lc->sync_bits, region);
770 static region_t core_get_sync_count(struct dm_dirty_log *log)
772 struct log_c *lc = (struct log_c *) log->context;
774 return lc->sync_count;
777 #define DMEMIT_SYNC \
778 if (lc->sync != DEFAULTSYNC) \
779 DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
781 static int core_status(struct dm_dirty_log *log, status_type_t status,
782 char *result, unsigned int maxlen)
785 struct log_c *lc = log->context;
788 case STATUSTYPE_INFO:
789 DMEMIT("1 %s", log->type->name);
792 case STATUSTYPE_TABLE:
793 DMEMIT("%s %u %u ", log->type->name,
794 lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size);
806 static int disk_status(struct dm_dirty_log *log, status_type_t status,
807 char *result, unsigned int maxlen)
810 struct log_c *lc = log->context;
813 case STATUSTYPE_INFO:
814 DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name,
815 lc->log_dev_flush_failed ? 'F' :
816 lc->log_dev_failed ? 'D' :
820 case STATUSTYPE_TABLE:
821 DMEMIT("%s %u %s %u ", log->type->name,
822 lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name,
835 static struct dm_dirty_log_type _core_type = {
837 .module = THIS_MODULE,
840 .resume = core_resume,
841 .get_region_size = core_get_region_size,
842 .is_clean = core_is_clean,
843 .in_sync = core_in_sync,
845 .mark_region = core_mark_region,
846 .clear_region = core_clear_region,
847 .get_resync_work = core_get_resync_work,
848 .set_region_sync = core_set_region_sync,
849 .get_sync_count = core_get_sync_count,
850 .status = core_status,
853 static struct dm_dirty_log_type _disk_type = {
855 .module = THIS_MODULE,
858 .postsuspend = disk_flush,
859 .resume = disk_resume,
860 .get_region_size = core_get_region_size,
861 .is_clean = core_is_clean,
862 .in_sync = core_in_sync,
864 .mark_region = core_mark_region,
865 .clear_region = core_clear_region,
866 .get_resync_work = core_get_resync_work,
867 .set_region_sync = core_set_region_sync,
868 .get_sync_count = core_get_sync_count,
869 .status = disk_status,
872 static int __init dm_dirty_log_init(void)
876 r = dm_dirty_log_type_register(&_core_type);
878 DMWARN("couldn't register core log");
880 r = dm_dirty_log_type_register(&_disk_type);
882 DMWARN("couldn't register disk type");
883 dm_dirty_log_type_unregister(&_core_type);
889 static void __exit dm_dirty_log_exit(void)
891 dm_dirty_log_type_unregister(&_disk_type);
892 dm_dirty_log_type_unregister(&_core_type);
895 module_init(dm_dirty_log_init);
896 module_exit(dm_dirty_log_exit);
898 MODULE_DESCRIPTION(DM_NAME " dirty region log");
899 MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@redhat.com>");
900 MODULE_LICENSE("GPL");