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 int log_dev_flush_failed;
241 struct dm_dev *log_dev;
242 struct log_header header;
244 struct dm_io_region header_location;
245 struct log_header *disk_header;
249 * The touched member needs to be updated every time we access
250 * one of the bitsets.
252 static inline int log_test_bit(uint32_t *bs, unsigned bit)
254 return test_bit_le(bit, (unsigned long *) bs) ? 1 : 0;
257 static inline void log_set_bit(struct log_c *l,
258 uint32_t *bs, unsigned bit)
260 __test_and_set_bit_le(bit, (unsigned long *) bs);
261 l->touched_cleaned = 1;
264 static inline void log_clear_bit(struct log_c *l,
265 uint32_t *bs, unsigned bit)
267 __test_and_clear_bit_le(bit, (unsigned long *) bs);
268 l->touched_dirtied = 1;
271 /*----------------------------------------------------------------
273 *--------------------------------------------------------------*/
274 static void header_to_disk(struct log_header *core, struct log_header *disk)
276 disk->magic = cpu_to_le32(core->magic);
277 disk->version = cpu_to_le32(core->version);
278 disk->nr_regions = cpu_to_le64(core->nr_regions);
281 static void header_from_disk(struct log_header *core, struct log_header *disk)
283 core->magic = le32_to_cpu(disk->magic);
284 core->version = le32_to_cpu(disk->version);
285 core->nr_regions = le64_to_cpu(disk->nr_regions);
288 static int rw_header(struct log_c *lc, int rw)
290 lc->io_req.bi_rw = rw;
292 return dm_io(&lc->io_req, 1, &lc->header_location, NULL);
295 static int flush_header(struct log_c *lc)
297 struct dm_io_region null_location = {
298 .bdev = lc->header_location.bdev,
303 lc->io_req.bi_rw = WRITE_FLUSH;
305 return dm_io(&lc->io_req, 1, &null_location, NULL);
308 static int read_header(struct log_c *log)
312 r = rw_header(log, READ);
316 header_from_disk(&log->header, log->disk_header);
318 /* New log required? */
319 if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) {
320 log->header.magic = MIRROR_MAGIC;
321 log->header.version = MIRROR_DISK_VERSION;
322 log->header.nr_regions = 0;
325 #ifdef __LITTLE_ENDIAN
326 if (log->header.version == 1)
327 log->header.version = 2;
330 if (log->header.version != MIRROR_DISK_VERSION) {
331 DMWARN("incompatible disk log version");
338 static int _check_region_size(struct dm_target *ti, uint32_t region_size)
340 if (region_size < 2 || region_size > ti->len)
343 if (!is_power_of_2(region_size))
349 /*----------------------------------------------------------------
350 * core log constructor/destructor
352 * argv contains region_size followed optionally by [no]sync
353 *--------------------------------------------------------------*/
355 static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti,
356 unsigned int argc, char **argv,
359 enum sync sync = DEFAULTSYNC;
362 uint32_t region_size;
363 unsigned int region_count;
364 size_t bitset_size, buf_size;
367 if (argc < 1 || argc > 2) {
368 DMWARN("wrong number of arguments to dirty region log");
373 if (!strcmp(argv[1], "sync"))
375 else if (!strcmp(argv[1], "nosync"))
378 DMWARN("unrecognised sync argument to "
379 "dirty region log: %s", argv[1]);
384 if (sscanf(argv[0], "%u", ®ion_size) != 1 ||
385 !_check_region_size(ti, region_size)) {
386 DMWARN("invalid region size %s", argv[0]);
390 region_count = dm_sector_div_up(ti->len, region_size);
392 lc = kmalloc(sizeof(*lc), GFP_KERNEL);
394 DMWARN("couldn't allocate core log");
399 lc->touched_dirtied = 0;
400 lc->touched_cleaned = 0;
401 lc->flush_failed = 0;
402 lc->region_size = region_size;
403 lc->region_count = region_count;
407 * Work out how many "unsigned long"s we need to hold the bitset.
409 bitset_size = dm_round_up(region_count,
410 sizeof(*lc->clean_bits) << BYTE_SHIFT);
411 bitset_size >>= BYTE_SHIFT;
413 lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits);
419 lc->clean_bits = vmalloc(bitset_size);
420 if (!lc->clean_bits) {
421 DMWARN("couldn't allocate clean bitset");
425 lc->disk_header = NULL;
428 lc->log_dev_failed = 0;
429 lc->log_dev_flush_failed = 0;
430 lc->header_location.bdev = lc->log_dev->bdev;
431 lc->header_location.sector = 0;
434 * Buffer holds both header and bitset.
437 dm_round_up((LOG_OFFSET << SECTOR_SHIFT) + bitset_size,
438 bdev_logical_block_size(lc->header_location.
441 if (buf_size > i_size_read(dev->bdev->bd_inode)) {
442 DMWARN("log device %s too small: need %llu bytes",
443 dev->name, (unsigned long long)buf_size);
448 lc->header_location.count = buf_size >> SECTOR_SHIFT;
450 lc->io_req.mem.type = DM_IO_VMA;
451 lc->io_req.notify.fn = NULL;
452 lc->io_req.client = dm_io_client_create();
453 if (IS_ERR(lc->io_req.client)) {
454 r = PTR_ERR(lc->io_req.client);
455 DMWARN("couldn't allocate disk io client");
460 lc->disk_header = vmalloc(buf_size);
461 if (!lc->disk_header) {
462 DMWARN("couldn't allocate disk log buffer");
463 dm_io_client_destroy(lc->io_req.client);
468 lc->io_req.mem.ptr.vma = lc->disk_header;
469 lc->clean_bits = (void *)lc->disk_header +
470 (LOG_OFFSET << SECTOR_SHIFT);
473 memset(lc->clean_bits, -1, bitset_size);
475 lc->sync_bits = vmalloc(bitset_size);
476 if (!lc->sync_bits) {
477 DMWARN("couldn't allocate sync bitset");
479 vfree(lc->clean_bits);
481 dm_io_client_destroy(lc->io_req.client);
482 vfree(lc->disk_header);
486 memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
487 lc->sync_count = (sync == NOSYNC) ? region_count : 0;
489 lc->recovering_bits = vmalloc(bitset_size);
490 if (!lc->recovering_bits) {
491 DMWARN("couldn't allocate sync bitset");
492 vfree(lc->sync_bits);
494 vfree(lc->clean_bits);
496 dm_io_client_destroy(lc->io_req.client);
497 vfree(lc->disk_header);
501 memset(lc->recovering_bits, 0, bitset_size);
508 static int core_ctr(struct dm_dirty_log *log, struct dm_target *ti,
509 unsigned int argc, char **argv)
511 return create_log_context(log, ti, argc, argv, NULL);
514 static void destroy_log_context(struct log_c *lc)
516 vfree(lc->sync_bits);
517 vfree(lc->recovering_bits);
521 static void core_dtr(struct dm_dirty_log *log)
523 struct log_c *lc = (struct log_c *) log->context;
525 vfree(lc->clean_bits);
526 destroy_log_context(lc);
529 /*----------------------------------------------------------------
530 * disk log constructor/destructor
532 * argv contains log_device region_size followed optionally by [no]sync
533 *--------------------------------------------------------------*/
534 static int disk_ctr(struct dm_dirty_log *log, struct dm_target *ti,
535 unsigned int argc, char **argv)
540 if (argc < 2 || argc > 3) {
541 DMWARN("wrong number of arguments to disk dirty region log");
545 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &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 lc->log_dev_flush_failed = 1;
642 DMWARN("%s: Failed to write header on dirty region log device",
650 static uint32_t core_get_region_size(struct dm_dirty_log *log)
652 struct log_c *lc = (struct log_c *) log->context;
653 return lc->region_size;
656 static int core_resume(struct dm_dirty_log *log)
658 struct log_c *lc = (struct log_c *) log->context;
663 static int core_is_clean(struct dm_dirty_log *log, region_t region)
665 struct log_c *lc = (struct log_c *) log->context;
666 return log_test_bit(lc->clean_bits, region);
669 static int core_in_sync(struct dm_dirty_log *log, region_t region, int block)
671 struct log_c *lc = (struct log_c *) log->context;
672 return log_test_bit(lc->sync_bits, region);
675 static int core_flush(struct dm_dirty_log *log)
681 static int disk_flush(struct dm_dirty_log *log)
684 struct log_c *lc = log->context;
686 /* only write if the log has changed */
687 if (!lc->touched_cleaned && !lc->touched_dirtied)
690 if (lc->touched_cleaned && log->flush_callback_fn &&
691 log->flush_callback_fn(lc->ti)) {
693 * At this point it is impossible to determine which
694 * regions are clean and which are dirty (without
695 * re-reading the log off disk). So mark all of them
698 lc->flush_failed = 1;
699 for (i = 0; i < lc->region_count; i++)
700 log_clear_bit(lc, lc->clean_bits, i);
703 r = rw_header(lc, WRITE);
707 if (lc->touched_dirtied) {
708 r = flush_header(lc);
710 lc->log_dev_flush_failed = 1;
713 lc->touched_dirtied = 0;
715 lc->touched_cleaned = 0;
721 static void core_mark_region(struct dm_dirty_log *log, region_t region)
723 struct log_c *lc = (struct log_c *) log->context;
724 log_clear_bit(lc, lc->clean_bits, region);
727 static void core_clear_region(struct dm_dirty_log *log, region_t region)
729 struct log_c *lc = (struct log_c *) log->context;
730 if (likely(!lc->flush_failed))
731 log_set_bit(lc, lc->clean_bits, region);
734 static int core_get_resync_work(struct dm_dirty_log *log, region_t *region)
736 struct log_c *lc = (struct log_c *) log->context;
738 if (lc->sync_search >= lc->region_count)
742 *region = find_next_zero_bit_le(
743 (unsigned long *) lc->sync_bits,
746 lc->sync_search = *region + 1;
748 if (*region >= lc->region_count)
751 } while (log_test_bit(lc->recovering_bits, *region));
753 log_set_bit(lc, lc->recovering_bits, *region);
757 static void core_set_region_sync(struct dm_dirty_log *log, region_t region,
760 struct log_c *lc = (struct log_c *) log->context;
762 log_clear_bit(lc, lc->recovering_bits, region);
764 log_set_bit(lc, lc->sync_bits, region);
766 } else if (log_test_bit(lc->sync_bits, region)) {
768 log_clear_bit(lc, lc->sync_bits, region);
772 static region_t core_get_sync_count(struct dm_dirty_log *log)
774 struct log_c *lc = (struct log_c *) log->context;
776 return lc->sync_count;
779 #define DMEMIT_SYNC \
780 if (lc->sync != DEFAULTSYNC) \
781 DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
783 static int core_status(struct dm_dirty_log *log, status_type_t status,
784 char *result, unsigned int maxlen)
787 struct log_c *lc = log->context;
790 case STATUSTYPE_INFO:
791 DMEMIT("1 %s", log->type->name);
794 case STATUSTYPE_TABLE:
795 DMEMIT("%s %u %u ", log->type->name,
796 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,
827 static struct dm_dirty_log_type _core_type = {
829 .module = THIS_MODULE,
832 .resume = core_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 = core_status,
845 static struct dm_dirty_log_type _disk_type = {
847 .module = THIS_MODULE,
850 .postsuspend = disk_flush,
851 .resume = disk_resume,
852 .get_region_size = core_get_region_size,
853 .is_clean = core_is_clean,
854 .in_sync = core_in_sync,
856 .mark_region = core_mark_region,
857 .clear_region = core_clear_region,
858 .get_resync_work = core_get_resync_work,
859 .set_region_sync = core_set_region_sync,
860 .get_sync_count = core_get_sync_count,
861 .status = disk_status,
864 static int __init dm_dirty_log_init(void)
868 r = dm_dirty_log_type_register(&_core_type);
870 DMWARN("couldn't register core log");
872 r = dm_dirty_log_type_register(&_disk_type);
874 DMWARN("couldn't register disk type");
875 dm_dirty_log_type_unregister(&_core_type);
881 static void __exit dm_dirty_log_exit(void)
883 dm_dirty_log_type_unregister(&_disk_type);
884 dm_dirty_log_type_unregister(&_core_type);
887 module_init(dm_dirty_log_init);
888 module_exit(dm_dirty_log_exit);
890 MODULE_DESCRIPTION(DM_NAME " dirty region log");
891 MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@redhat.com>");
892 MODULE_LICENSE("GPL");