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
204 * Simple, incrementing version. no backward
212 struct dm_target *ti;
216 uint32_t region_size;
217 unsigned int region_count;
220 unsigned bitset_uint32_count;
221 uint32_t *clean_bits;
223 uint32_t *recovering_bits; /* FIXME: this seems excessive */
229 DEFAULTSYNC, /* Synchronize if necessary */
230 NOSYNC, /* Devices known to be already in sync */
231 FORCESYNC, /* Force a sync to happen */
234 struct dm_io_request io_req;
240 struct dm_dev *log_dev;
241 struct log_header header;
243 struct dm_io_region header_location;
244 struct log_header *disk_header;
248 * The touched member needs to be updated every time we access
249 * one of the bitsets.
251 static inline int log_test_bit(uint32_t *bs, unsigned bit)
253 return ext2_test_bit(bit, (unsigned long *) bs) ? 1 : 0;
256 static inline void log_set_bit(struct log_c *l,
257 uint32_t *bs, unsigned bit)
259 ext2_set_bit(bit, (unsigned long *) bs);
260 l->touched_cleaned = 1;
263 static inline void log_clear_bit(struct log_c *l,
264 uint32_t *bs, unsigned bit)
266 ext2_clear_bit(bit, (unsigned long *) bs);
267 l->touched_dirtied = 1;
270 /*----------------------------------------------------------------
272 *--------------------------------------------------------------*/
273 static void header_to_disk(struct log_header *core, struct log_header *disk)
275 disk->magic = cpu_to_le32(core->magic);
276 disk->version = cpu_to_le32(core->version);
277 disk->nr_regions = cpu_to_le64(core->nr_regions);
280 static void header_from_disk(struct log_header *core, struct log_header *disk)
282 core->magic = le32_to_cpu(disk->magic);
283 core->version = le32_to_cpu(disk->version);
284 core->nr_regions = le64_to_cpu(disk->nr_regions);
287 static int rw_header(struct log_c *lc, int rw)
289 lc->io_req.bi_rw = rw;
291 return dm_io(&lc->io_req, 1, &lc->header_location, NULL);
294 static int flush_header(struct log_c *lc)
296 struct dm_io_region null_location = {
297 .bdev = lc->header_location.bdev,
302 lc->io_req.bi_rw = WRITE_BARRIER;
304 return dm_io(&lc->io_req, 1, &null_location, NULL);
307 static int read_header(struct log_c *log)
311 r = rw_header(log, READ);
315 header_from_disk(&log->header, log->disk_header);
317 /* New log required? */
318 if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) {
319 log->header.magic = MIRROR_MAGIC;
320 log->header.version = MIRROR_DISK_VERSION;
321 log->header.nr_regions = 0;
324 #ifdef __LITTLE_ENDIAN
325 if (log->header.version == 1)
326 log->header.version = 2;
329 if (log->header.version != MIRROR_DISK_VERSION) {
330 DMWARN("incompatible disk log version");
337 static int _check_region_size(struct dm_target *ti, uint32_t region_size)
339 if (region_size < 2 || region_size > ti->len)
342 if (!is_power_of_2(region_size))
348 /*----------------------------------------------------------------
349 * core log constructor/destructor
351 * argv contains region_size followed optionally by [no]sync
352 *--------------------------------------------------------------*/
354 static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti,
355 unsigned int argc, char **argv,
358 enum sync sync = DEFAULTSYNC;
361 uint32_t region_size;
362 unsigned int region_count;
363 size_t bitset_size, buf_size;
366 if (argc < 1 || argc > 2) {
367 DMWARN("wrong number of arguments to dirty region log");
372 if (!strcmp(argv[1], "sync"))
374 else if (!strcmp(argv[1], "nosync"))
377 DMWARN("unrecognised sync argument to "
378 "dirty region log: %s", argv[1]);
383 if (sscanf(argv[0], "%u", ®ion_size) != 1 ||
384 !_check_region_size(ti, region_size)) {
385 DMWARN("invalid region size %s", argv[0]);
389 region_count = dm_sector_div_up(ti->len, region_size);
391 lc = kmalloc(sizeof(*lc), GFP_KERNEL);
393 DMWARN("couldn't allocate core log");
398 lc->touched_dirtied = 0;
399 lc->touched_cleaned = 0;
400 lc->flush_failed = 0;
401 lc->region_size = region_size;
402 lc->region_count = region_count;
406 * Work out how many "unsigned long"s we need to hold the bitset.
408 bitset_size = dm_round_up(region_count,
409 sizeof(*lc->clean_bits) << BYTE_SHIFT);
410 bitset_size >>= BYTE_SHIFT;
412 lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits);
418 lc->clean_bits = vmalloc(bitset_size);
419 if (!lc->clean_bits) {
420 DMWARN("couldn't allocate clean bitset");
424 lc->disk_header = NULL;
427 lc->log_dev_failed = 0;
428 lc->header_location.bdev = lc->log_dev->bdev;
429 lc->header_location.sector = 0;
432 * Buffer holds both header and bitset.
435 dm_round_up((LOG_OFFSET << SECTOR_SHIFT) + bitset_size,
436 bdev_logical_block_size(lc->header_location.
439 if (buf_size > i_size_read(dev->bdev->bd_inode)) {
440 DMWARN("log device %s too small: need %llu bytes",
441 dev->name, (unsigned long long)buf_size);
446 lc->header_location.count = buf_size >> SECTOR_SHIFT;
448 lc->io_req.mem.type = DM_IO_VMA;
449 lc->io_req.notify.fn = NULL;
450 lc->io_req.client = dm_io_client_create(dm_div_up(buf_size,
452 if (IS_ERR(lc->io_req.client)) {
453 r = PTR_ERR(lc->io_req.client);
454 DMWARN("couldn't allocate disk io client");
459 lc->disk_header = vmalloc(buf_size);
460 if (!lc->disk_header) {
461 DMWARN("couldn't allocate disk log buffer");
462 dm_io_client_destroy(lc->io_req.client);
467 lc->io_req.mem.ptr.vma = lc->disk_header;
468 lc->clean_bits = (void *)lc->disk_header +
469 (LOG_OFFSET << SECTOR_SHIFT);
472 memset(lc->clean_bits, -1, bitset_size);
474 lc->sync_bits = vmalloc(bitset_size);
475 if (!lc->sync_bits) {
476 DMWARN("couldn't allocate sync bitset");
478 vfree(lc->clean_bits);
480 dm_io_client_destroy(lc->io_req.client);
481 vfree(lc->disk_header);
485 memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
486 lc->sync_count = (sync == NOSYNC) ? region_count : 0;
488 lc->recovering_bits = vmalloc(bitset_size);
489 if (!lc->recovering_bits) {
490 DMWARN("couldn't allocate sync bitset");
491 vfree(lc->sync_bits);
493 vfree(lc->clean_bits);
495 dm_io_client_destroy(lc->io_req.client);
496 vfree(lc->disk_header);
500 memset(lc->recovering_bits, 0, bitset_size);
507 static int core_ctr(struct dm_dirty_log *log, struct dm_target *ti,
508 unsigned int argc, char **argv)
510 return create_log_context(log, ti, argc, argv, NULL);
513 static void destroy_log_context(struct log_c *lc)
515 vfree(lc->sync_bits);
516 vfree(lc->recovering_bits);
520 static void core_dtr(struct dm_dirty_log *log)
522 struct log_c *lc = (struct log_c *) log->context;
524 vfree(lc->clean_bits);
525 destroy_log_context(lc);
528 /*----------------------------------------------------------------
529 * disk log constructor/destructor
531 * argv contains log_device region_size followed optionally by [no]sync
532 *--------------------------------------------------------------*/
533 static int disk_ctr(struct dm_dirty_log *log, struct dm_target *ti,
534 unsigned int argc, char **argv)
539 if (argc < 2 || argc > 3) {
540 DMWARN("wrong number of arguments to disk dirty region log");
544 r = dm_get_device(ti, argv[0], 0, 0 /* FIXME */,
545 FMODE_READ | FMODE_WRITE, &dev);
549 r = create_log_context(log, ti, argc - 1, argv + 1, dev);
551 dm_put_device(ti, dev);
558 static void disk_dtr(struct dm_dirty_log *log)
560 struct log_c *lc = (struct log_c *) log->context;
562 dm_put_device(lc->ti, lc->log_dev);
563 vfree(lc->disk_header);
564 dm_io_client_destroy(lc->io_req.client);
565 destroy_log_context(lc);
568 static int count_bits32(uint32_t *addr, unsigned size)
572 for (i = 0; i < size; i++) {
573 count += hweight32(*(addr+i));
578 static void fail_log_device(struct log_c *lc)
580 if (lc->log_dev_failed)
583 lc->log_dev_failed = 1;
584 dm_table_event(lc->ti->table);
587 static int disk_resume(struct dm_dirty_log *log)
591 struct log_c *lc = (struct log_c *) log->context;
592 size_t size = lc->bitset_uint32_count * sizeof(uint32_t);
594 /* read the disk header */
597 DMWARN("%s: Failed to read header on dirty region log device",
601 * If the log device cannot be read, we must assume
602 * all regions are out-of-sync. If we simply return
603 * here, the state will be uninitialized and could
604 * lead us to return 'in-sync' status for regions
605 * that are actually 'out-of-sync'.
607 lc->header.nr_regions = 0;
610 /* set or clear any new bits -- device has grown */
611 if (lc->sync == NOSYNC)
612 for (i = lc->header.nr_regions; i < lc->region_count; i++)
613 /* FIXME: amazingly inefficient */
614 log_set_bit(lc, lc->clean_bits, i);
616 for (i = lc->header.nr_regions; i < lc->region_count; i++)
617 /* FIXME: amazingly inefficient */
618 log_clear_bit(lc, lc->clean_bits, i);
620 /* clear any old bits -- device has shrunk */
621 for (i = lc->region_count; i % (sizeof(*lc->clean_bits) << BYTE_SHIFT); i++)
622 log_clear_bit(lc, lc->clean_bits, i);
624 /* copy clean across to sync */
625 memcpy(lc->sync_bits, lc->clean_bits, size);
626 lc->sync_count = count_bits32(lc->clean_bits, lc->bitset_uint32_count);
629 /* set the correct number of regions in the header */
630 lc->header.nr_regions = lc->region_count;
632 header_to_disk(&lc->header, lc->disk_header);
634 /* write the new header */
635 r = rw_header(lc, WRITE);
637 r = flush_header(lc);
639 DMWARN("%s: Failed to write header on dirty region log device",
647 static uint32_t core_get_region_size(struct dm_dirty_log *log)
649 struct log_c *lc = (struct log_c *) log->context;
650 return lc->region_size;
653 static int core_resume(struct dm_dirty_log *log)
655 struct log_c *lc = (struct log_c *) log->context;
660 static int core_is_clean(struct dm_dirty_log *log, region_t region)
662 struct log_c *lc = (struct log_c *) log->context;
663 return log_test_bit(lc->clean_bits, region);
666 static int core_in_sync(struct dm_dirty_log *log, region_t region, int block)
668 struct log_c *lc = (struct log_c *) log->context;
669 return log_test_bit(lc->sync_bits, region);
672 static int core_flush(struct dm_dirty_log *log)
678 static int disk_flush(struct dm_dirty_log *log)
681 struct log_c *lc = (struct log_c *) log->context;
683 /* only write if the log has changed */
684 if (!lc->touched_cleaned && !lc->touched_dirtied)
687 r = rw_header(lc, WRITE);
691 if (lc->touched_dirtied) {
692 r = flush_header(lc);
696 lc->touched_dirtied = 0;
698 lc->touched_cleaned = 0;
704 static void core_mark_region(struct dm_dirty_log *log, region_t region)
706 struct log_c *lc = (struct log_c *) log->context;
707 log_clear_bit(lc, lc->clean_bits, region);
710 static void core_clear_region(struct dm_dirty_log *log, region_t region)
712 struct log_c *lc = (struct log_c *) log->context;
713 if (likely(!lc->flush_failed))
714 log_set_bit(lc, lc->clean_bits, region);
717 static int core_get_resync_work(struct dm_dirty_log *log, region_t *region)
719 struct log_c *lc = (struct log_c *) log->context;
721 if (lc->sync_search >= lc->region_count)
725 *region = ext2_find_next_zero_bit(
726 (unsigned long *) lc->sync_bits,
729 lc->sync_search = *region + 1;
731 if (*region >= lc->region_count)
734 } while (log_test_bit(lc->recovering_bits, *region));
736 log_set_bit(lc, lc->recovering_bits, *region);
740 static void core_set_region_sync(struct dm_dirty_log *log, region_t region,
743 struct log_c *lc = (struct log_c *) log->context;
745 log_clear_bit(lc, lc->recovering_bits, region);
747 log_set_bit(lc, lc->sync_bits, region);
749 } else if (log_test_bit(lc->sync_bits, region)) {
751 log_clear_bit(lc, lc->sync_bits, region);
755 static region_t core_get_sync_count(struct dm_dirty_log *log)
757 struct log_c *lc = (struct log_c *) log->context;
759 return lc->sync_count;
762 #define DMEMIT_SYNC \
763 if (lc->sync != DEFAULTSYNC) \
764 DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
766 static int core_status(struct dm_dirty_log *log, status_type_t status,
767 char *result, unsigned int maxlen)
770 struct log_c *lc = log->context;
773 case STATUSTYPE_INFO:
774 DMEMIT("1 %s", log->type->name);
777 case STATUSTYPE_TABLE:
778 DMEMIT("%s %u %u ", log->type->name,
779 lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size);
786 static int disk_status(struct dm_dirty_log *log, status_type_t status,
787 char *result, unsigned int maxlen)
790 struct log_c *lc = log->context;
793 case STATUSTYPE_INFO:
794 DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name,
795 lc->log_dev_failed ? 'D' : 'A');
798 case STATUSTYPE_TABLE:
799 DMEMIT("%s %u %s %u ", log->type->name,
800 lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name,
808 static struct dm_dirty_log_type _core_type = {
810 .module = THIS_MODULE,
813 .resume = core_resume,
814 .get_region_size = core_get_region_size,
815 .is_clean = core_is_clean,
816 .in_sync = core_in_sync,
818 .mark_region = core_mark_region,
819 .clear_region = core_clear_region,
820 .get_resync_work = core_get_resync_work,
821 .set_region_sync = core_set_region_sync,
822 .get_sync_count = core_get_sync_count,
823 .status = core_status,
826 static struct dm_dirty_log_type _disk_type = {
828 .module = THIS_MODULE,
831 .postsuspend = disk_flush,
832 .resume = disk_resume,
833 .get_region_size = core_get_region_size,
834 .is_clean = core_is_clean,
835 .in_sync = core_in_sync,
837 .mark_region = core_mark_region,
838 .clear_region = core_clear_region,
839 .get_resync_work = core_get_resync_work,
840 .set_region_sync = core_set_region_sync,
841 .get_sync_count = core_get_sync_count,
842 .status = disk_status,
845 static int __init dm_dirty_log_init(void)
849 r = dm_dirty_log_type_register(&_core_type);
851 DMWARN("couldn't register core log");
853 r = dm_dirty_log_type_register(&_disk_type);
855 DMWARN("couldn't register disk type");
856 dm_dirty_log_type_unregister(&_core_type);
862 static void __exit dm_dirty_log_exit(void)
864 dm_dirty_log_type_unregister(&_disk_type);
865 dm_dirty_log_type_unregister(&_core_type);
868 module_init(dm_dirty_log_init);
869 module_exit(dm_dirty_log_exit);
871 MODULE_DESCRIPTION(DM_NAME " dirty region log");
872 MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@redhat.com>");
873 MODULE_LICENSE("GPL");