dm raid1: split touched state into two
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / md / dm-log.c
blob31dc33df95c79ed914d0bc0b9c3f1efa62489a13
1 /*
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.
6 */
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))
28 return log_type;
30 return NULL;
33 static struct dm_dirty_log_type *_get_dirty_log_type(const char *name)
35 struct dm_dirty_log_type *log_type;
37 spin_lock(&_lock);
39 log_type = __find_dirty_log_type(name);
40 if (log_type && !try_module_get(log_type->module))
41 log_type = NULL;
43 spin_unlock(&_lock);
45 return log_type;
49 * get_type
50 * @type_name
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;
70 if (!type_name)
71 return NULL;
73 log_type = _get_dirty_log_type(type_name);
74 if (log_type)
75 return log_type;
77 type_name_dup = kstrdup(type_name, GFP_KERNEL);
78 if (!type_name_dup) {
79 DMWARN("No memory left to attempt log module load for \"%s\"",
80 type_name);
81 return NULL;
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, '-');
87 if (!p)
88 break;
89 p[0] = '\0';
92 if (!log_type)
93 DMWARN("Module for logging type \"%s\" not found.", type_name);
95 kfree(type_name_dup);
97 return log_type;
100 static void put_type(struct dm_dirty_log_type *type)
102 if (!type)
103 return;
105 spin_lock(&_lock);
106 if (!__find_dirty_log_type(type->name))
107 goto out;
109 module_put(type->module);
111 out:
112 spin_unlock(&_lock);
115 int dm_dirty_log_type_register(struct dm_dirty_log_type *type)
117 int r = 0;
119 spin_lock(&_lock);
120 if (!__find_dirty_log_type(type->name))
121 list_add(&type->list, &_log_types);
122 else
123 r = -EEXIST;
124 spin_unlock(&_lock);
126 return r;
128 EXPORT_SYMBOL(dm_dirty_log_type_register);
130 int dm_dirty_log_type_unregister(struct dm_dirty_log_type *type)
132 spin_lock(&_lock);
134 if (!__find_dirty_log_type(type->name)) {
135 spin_unlock(&_lock);
136 return -EINVAL;
139 list_del(&type->list);
141 spin_unlock(&_lock);
143 return 0;
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);
155 if (!log)
156 return NULL;
158 type = get_type(type_name);
159 if (!type) {
160 kfree(log);
161 return NULL;
164 log->type = type;
165 if (type->ctr(log, ti, argc, argv)) {
166 kfree(log);
167 put_type(type);
168 return NULL;
171 return log;
173 EXPORT_SYMBOL(dm_dirty_log_create);
175 void dm_dirty_log_destroy(struct dm_dirty_log *log)
177 log->type->dtr(log);
178 put_type(log->type);
179 kfree(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
196 #define LOG_OFFSET 2
198 struct log_header {
199 uint32_t magic;
202 * Simple, incrementing version. no backward
203 * compatibility.
205 uint32_t version;
206 sector_t nr_regions;
209 struct log_c {
210 struct dm_target *ti;
211 int touched_dirtied;
212 int touched_cleaned;
213 uint32_t region_size;
214 unsigned int region_count;
215 region_t sync_count;
217 unsigned bitset_uint32_count;
218 uint32_t *clean_bits;
219 uint32_t *sync_bits;
220 uint32_t *recovering_bits; /* FIXME: this seems excessive */
222 int sync_search;
224 /* Resync flag */
225 enum sync {
226 DEFAULTSYNC, /* Synchronize if necessary */
227 NOSYNC, /* Devices known to be already in sync */
228 FORCESYNC, /* Force a sync to happen */
229 } sync;
231 struct dm_io_request io_req;
234 * Disk log fields
236 int log_dev_failed;
237 struct dm_dev *log_dev;
238 struct log_header header;
240 struct dm_io_region header_location;
241 struct log_header *disk_header;
245 * The touched member needs to be updated every time we access
246 * one of the bitsets.
248 static inline int log_test_bit(uint32_t *bs, unsigned bit)
250 return ext2_test_bit(bit, (unsigned long *) bs) ? 1 : 0;
253 static inline void log_set_bit(struct log_c *l,
254 uint32_t *bs, unsigned bit)
256 ext2_set_bit(bit, (unsigned long *) bs);
257 l->touched_cleaned = 1;
260 static inline void log_clear_bit(struct log_c *l,
261 uint32_t *bs, unsigned bit)
263 ext2_clear_bit(bit, (unsigned long *) bs);
264 l->touched_dirtied = 1;
267 /*----------------------------------------------------------------
268 * Header IO
269 *--------------------------------------------------------------*/
270 static void header_to_disk(struct log_header *core, struct log_header *disk)
272 disk->magic = cpu_to_le32(core->magic);
273 disk->version = cpu_to_le32(core->version);
274 disk->nr_regions = cpu_to_le64(core->nr_regions);
277 static void header_from_disk(struct log_header *core, struct log_header *disk)
279 core->magic = le32_to_cpu(disk->magic);
280 core->version = le32_to_cpu(disk->version);
281 core->nr_regions = le64_to_cpu(disk->nr_regions);
284 static int rw_header(struct log_c *lc, int rw)
286 lc->io_req.bi_rw = rw;
288 return dm_io(&lc->io_req, 1, &lc->header_location, NULL);
291 static int read_header(struct log_c *log)
293 int r;
295 r = rw_header(log, READ);
296 if (r)
297 return r;
299 header_from_disk(&log->header, log->disk_header);
301 /* New log required? */
302 if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) {
303 log->header.magic = MIRROR_MAGIC;
304 log->header.version = MIRROR_DISK_VERSION;
305 log->header.nr_regions = 0;
308 #ifdef __LITTLE_ENDIAN
309 if (log->header.version == 1)
310 log->header.version = 2;
311 #endif
313 if (log->header.version != MIRROR_DISK_VERSION) {
314 DMWARN("incompatible disk log version");
315 return -EINVAL;
318 return 0;
321 static int _check_region_size(struct dm_target *ti, uint32_t region_size)
323 if (region_size < 2 || region_size > ti->len)
324 return 0;
326 if (!is_power_of_2(region_size))
327 return 0;
329 return 1;
332 /*----------------------------------------------------------------
333 * core log constructor/destructor
335 * argv contains region_size followed optionally by [no]sync
336 *--------------------------------------------------------------*/
337 #define BYTE_SHIFT 3
338 static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti,
339 unsigned int argc, char **argv,
340 struct dm_dev *dev)
342 enum sync sync = DEFAULTSYNC;
344 struct log_c *lc;
345 uint32_t region_size;
346 unsigned int region_count;
347 size_t bitset_size, buf_size;
348 int r;
350 if (argc < 1 || argc > 2) {
351 DMWARN("wrong number of arguments to dirty region log");
352 return -EINVAL;
355 if (argc > 1) {
356 if (!strcmp(argv[1], "sync"))
357 sync = FORCESYNC;
358 else if (!strcmp(argv[1], "nosync"))
359 sync = NOSYNC;
360 else {
361 DMWARN("unrecognised sync argument to "
362 "dirty region log: %s", argv[1]);
363 return -EINVAL;
367 if (sscanf(argv[0], "%u", &region_size) != 1 ||
368 !_check_region_size(ti, region_size)) {
369 DMWARN("invalid region size %s", argv[0]);
370 return -EINVAL;
373 region_count = dm_sector_div_up(ti->len, region_size);
375 lc = kmalloc(sizeof(*lc), GFP_KERNEL);
376 if (!lc) {
377 DMWARN("couldn't allocate core log");
378 return -ENOMEM;
381 lc->ti = ti;
382 lc->touched_dirtied = 0;
383 lc->touched_cleaned = 0;
384 lc->region_size = region_size;
385 lc->region_count = region_count;
386 lc->sync = sync;
389 * Work out how many "unsigned long"s we need to hold the bitset.
391 bitset_size = dm_round_up(region_count,
392 sizeof(*lc->clean_bits) << BYTE_SHIFT);
393 bitset_size >>= BYTE_SHIFT;
395 lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits);
398 * Disk log?
400 if (!dev) {
401 lc->clean_bits = vmalloc(bitset_size);
402 if (!lc->clean_bits) {
403 DMWARN("couldn't allocate clean bitset");
404 kfree(lc);
405 return -ENOMEM;
407 lc->disk_header = NULL;
408 } else {
409 lc->log_dev = dev;
410 lc->log_dev_failed = 0;
411 lc->header_location.bdev = lc->log_dev->bdev;
412 lc->header_location.sector = 0;
415 * Buffer holds both header and bitset.
417 buf_size =
418 dm_round_up((LOG_OFFSET << SECTOR_SHIFT) + bitset_size,
419 bdev_logical_block_size(lc->header_location.
420 bdev));
422 if (buf_size > i_size_read(dev->bdev->bd_inode)) {
423 DMWARN("log device %s too small: need %llu bytes",
424 dev->name, (unsigned long long)buf_size);
425 kfree(lc);
426 return -EINVAL;
429 lc->header_location.count = buf_size >> SECTOR_SHIFT;
431 lc->io_req.mem.type = DM_IO_VMA;
432 lc->io_req.notify.fn = NULL;
433 lc->io_req.client = dm_io_client_create(dm_div_up(buf_size,
434 PAGE_SIZE));
435 if (IS_ERR(lc->io_req.client)) {
436 r = PTR_ERR(lc->io_req.client);
437 DMWARN("couldn't allocate disk io client");
438 kfree(lc);
439 return -ENOMEM;
442 lc->disk_header = vmalloc(buf_size);
443 if (!lc->disk_header) {
444 DMWARN("couldn't allocate disk log buffer");
445 dm_io_client_destroy(lc->io_req.client);
446 kfree(lc);
447 return -ENOMEM;
450 lc->io_req.mem.ptr.vma = lc->disk_header;
451 lc->clean_bits = (void *)lc->disk_header +
452 (LOG_OFFSET << SECTOR_SHIFT);
455 memset(lc->clean_bits, -1, bitset_size);
457 lc->sync_bits = vmalloc(bitset_size);
458 if (!lc->sync_bits) {
459 DMWARN("couldn't allocate sync bitset");
460 if (!dev)
461 vfree(lc->clean_bits);
462 else
463 dm_io_client_destroy(lc->io_req.client);
464 vfree(lc->disk_header);
465 kfree(lc);
466 return -ENOMEM;
468 memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
469 lc->sync_count = (sync == NOSYNC) ? region_count : 0;
471 lc->recovering_bits = vmalloc(bitset_size);
472 if (!lc->recovering_bits) {
473 DMWARN("couldn't allocate sync bitset");
474 vfree(lc->sync_bits);
475 if (!dev)
476 vfree(lc->clean_bits);
477 else
478 dm_io_client_destroy(lc->io_req.client);
479 vfree(lc->disk_header);
480 kfree(lc);
481 return -ENOMEM;
483 memset(lc->recovering_bits, 0, bitset_size);
484 lc->sync_search = 0;
485 log->context = lc;
487 return 0;
490 static int core_ctr(struct dm_dirty_log *log, struct dm_target *ti,
491 unsigned int argc, char **argv)
493 return create_log_context(log, ti, argc, argv, NULL);
496 static void destroy_log_context(struct log_c *lc)
498 vfree(lc->sync_bits);
499 vfree(lc->recovering_bits);
500 kfree(lc);
503 static void core_dtr(struct dm_dirty_log *log)
505 struct log_c *lc = (struct log_c *) log->context;
507 vfree(lc->clean_bits);
508 destroy_log_context(lc);
511 /*----------------------------------------------------------------
512 * disk log constructor/destructor
514 * argv contains log_device region_size followed optionally by [no]sync
515 *--------------------------------------------------------------*/
516 static int disk_ctr(struct dm_dirty_log *log, struct dm_target *ti,
517 unsigned int argc, char **argv)
519 int r;
520 struct dm_dev *dev;
522 if (argc < 2 || argc > 3) {
523 DMWARN("wrong number of arguments to disk dirty region log");
524 return -EINVAL;
527 r = dm_get_device(ti, argv[0], 0, 0 /* FIXME */,
528 FMODE_READ | FMODE_WRITE, &dev);
529 if (r)
530 return r;
532 r = create_log_context(log, ti, argc - 1, argv + 1, dev);
533 if (r) {
534 dm_put_device(ti, dev);
535 return r;
538 return 0;
541 static void disk_dtr(struct dm_dirty_log *log)
543 struct log_c *lc = (struct log_c *) log->context;
545 dm_put_device(lc->ti, lc->log_dev);
546 vfree(lc->disk_header);
547 dm_io_client_destroy(lc->io_req.client);
548 destroy_log_context(lc);
551 static int count_bits32(uint32_t *addr, unsigned size)
553 int count = 0, i;
555 for (i = 0; i < size; i++) {
556 count += hweight32(*(addr+i));
558 return count;
561 static void fail_log_device(struct log_c *lc)
563 if (lc->log_dev_failed)
564 return;
566 lc->log_dev_failed = 1;
567 dm_table_event(lc->ti->table);
570 static int disk_resume(struct dm_dirty_log *log)
572 int r;
573 unsigned i;
574 struct log_c *lc = (struct log_c *) log->context;
575 size_t size = lc->bitset_uint32_count * sizeof(uint32_t);
577 /* read the disk header */
578 r = read_header(lc);
579 if (r) {
580 DMWARN("%s: Failed to read header on dirty region log device",
581 lc->log_dev->name);
582 fail_log_device(lc);
584 * If the log device cannot be read, we must assume
585 * all regions are out-of-sync. If we simply return
586 * here, the state will be uninitialized and could
587 * lead us to return 'in-sync' status for regions
588 * that are actually 'out-of-sync'.
590 lc->header.nr_regions = 0;
593 /* set or clear any new bits -- device has grown */
594 if (lc->sync == NOSYNC)
595 for (i = lc->header.nr_regions; i < lc->region_count; i++)
596 /* FIXME: amazingly inefficient */
597 log_set_bit(lc, lc->clean_bits, i);
598 else
599 for (i = lc->header.nr_regions; i < lc->region_count; i++)
600 /* FIXME: amazingly inefficient */
601 log_clear_bit(lc, lc->clean_bits, i);
603 /* clear any old bits -- device has shrunk */
604 for (i = lc->region_count; i % (sizeof(*lc->clean_bits) << BYTE_SHIFT); i++)
605 log_clear_bit(lc, lc->clean_bits, i);
607 /* copy clean across to sync */
608 memcpy(lc->sync_bits, lc->clean_bits, size);
609 lc->sync_count = count_bits32(lc->clean_bits, lc->bitset_uint32_count);
610 lc->sync_search = 0;
612 /* set the correct number of regions in the header */
613 lc->header.nr_regions = lc->region_count;
615 header_to_disk(&lc->header, lc->disk_header);
617 /* write the new header */
618 r = rw_header(lc, WRITE);
619 if (r) {
620 DMWARN("%s: Failed to write header on dirty region log device",
621 lc->log_dev->name);
622 fail_log_device(lc);
625 return r;
628 static uint32_t core_get_region_size(struct dm_dirty_log *log)
630 struct log_c *lc = (struct log_c *) log->context;
631 return lc->region_size;
634 static int core_resume(struct dm_dirty_log *log)
636 struct log_c *lc = (struct log_c *) log->context;
637 lc->sync_search = 0;
638 return 0;
641 static int core_is_clean(struct dm_dirty_log *log, region_t region)
643 struct log_c *lc = (struct log_c *) log->context;
644 return log_test_bit(lc->clean_bits, region);
647 static int core_in_sync(struct dm_dirty_log *log, region_t region, int block)
649 struct log_c *lc = (struct log_c *) log->context;
650 return log_test_bit(lc->sync_bits, region);
653 static int core_flush(struct dm_dirty_log *log)
655 /* no op */
656 return 0;
659 static int disk_flush(struct dm_dirty_log *log)
661 int r;
662 struct log_c *lc = (struct log_c *) log->context;
664 /* only write if the log has changed */
665 if (!lc->touched_cleaned && !lc->touched_dirtied)
666 return 0;
668 r = rw_header(lc, WRITE);
669 if (r)
670 fail_log_device(lc);
671 else {
672 lc->touched_dirtied = 0;
673 lc->touched_cleaned = 0;
676 return r;
679 static void core_mark_region(struct dm_dirty_log *log, region_t region)
681 struct log_c *lc = (struct log_c *) log->context;
682 log_clear_bit(lc, lc->clean_bits, region);
685 static void core_clear_region(struct dm_dirty_log *log, region_t region)
687 struct log_c *lc = (struct log_c *) log->context;
688 log_set_bit(lc, lc->clean_bits, region);
691 static int core_get_resync_work(struct dm_dirty_log *log, region_t *region)
693 struct log_c *lc = (struct log_c *) log->context;
695 if (lc->sync_search >= lc->region_count)
696 return 0;
698 do {
699 *region = ext2_find_next_zero_bit(
700 (unsigned long *) lc->sync_bits,
701 lc->region_count,
702 lc->sync_search);
703 lc->sync_search = *region + 1;
705 if (*region >= lc->region_count)
706 return 0;
708 } while (log_test_bit(lc->recovering_bits, *region));
710 log_set_bit(lc, lc->recovering_bits, *region);
711 return 1;
714 static void core_set_region_sync(struct dm_dirty_log *log, region_t region,
715 int in_sync)
717 struct log_c *lc = (struct log_c *) log->context;
719 log_clear_bit(lc, lc->recovering_bits, region);
720 if (in_sync) {
721 log_set_bit(lc, lc->sync_bits, region);
722 lc->sync_count++;
723 } else if (log_test_bit(lc->sync_bits, region)) {
724 lc->sync_count--;
725 log_clear_bit(lc, lc->sync_bits, region);
729 static region_t core_get_sync_count(struct dm_dirty_log *log)
731 struct log_c *lc = (struct log_c *) log->context;
733 return lc->sync_count;
736 #define DMEMIT_SYNC \
737 if (lc->sync != DEFAULTSYNC) \
738 DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
740 static int core_status(struct dm_dirty_log *log, status_type_t status,
741 char *result, unsigned int maxlen)
743 int sz = 0;
744 struct log_c *lc = log->context;
746 switch(status) {
747 case STATUSTYPE_INFO:
748 DMEMIT("1 %s", log->type->name);
749 break;
751 case STATUSTYPE_TABLE:
752 DMEMIT("%s %u %u ", log->type->name,
753 lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size);
754 DMEMIT_SYNC;
757 return sz;
760 static int disk_status(struct dm_dirty_log *log, status_type_t status,
761 char *result, unsigned int maxlen)
763 int sz = 0;
764 struct log_c *lc = log->context;
766 switch(status) {
767 case STATUSTYPE_INFO:
768 DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name,
769 lc->log_dev_failed ? 'D' : 'A');
770 break;
772 case STATUSTYPE_TABLE:
773 DMEMIT("%s %u %s %u ", log->type->name,
774 lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name,
775 lc->region_size);
776 DMEMIT_SYNC;
779 return sz;
782 static struct dm_dirty_log_type _core_type = {
783 .name = "core",
784 .module = THIS_MODULE,
785 .ctr = core_ctr,
786 .dtr = core_dtr,
787 .resume = core_resume,
788 .get_region_size = core_get_region_size,
789 .is_clean = core_is_clean,
790 .in_sync = core_in_sync,
791 .flush = core_flush,
792 .mark_region = core_mark_region,
793 .clear_region = core_clear_region,
794 .get_resync_work = core_get_resync_work,
795 .set_region_sync = core_set_region_sync,
796 .get_sync_count = core_get_sync_count,
797 .status = core_status,
800 static struct dm_dirty_log_type _disk_type = {
801 .name = "disk",
802 .module = THIS_MODULE,
803 .ctr = disk_ctr,
804 .dtr = disk_dtr,
805 .postsuspend = disk_flush,
806 .resume = disk_resume,
807 .get_region_size = core_get_region_size,
808 .is_clean = core_is_clean,
809 .in_sync = core_in_sync,
810 .flush = disk_flush,
811 .mark_region = core_mark_region,
812 .clear_region = core_clear_region,
813 .get_resync_work = core_get_resync_work,
814 .set_region_sync = core_set_region_sync,
815 .get_sync_count = core_get_sync_count,
816 .status = disk_status,
819 static int __init dm_dirty_log_init(void)
821 int r;
823 r = dm_dirty_log_type_register(&_core_type);
824 if (r)
825 DMWARN("couldn't register core log");
827 r = dm_dirty_log_type_register(&_disk_type);
828 if (r) {
829 DMWARN("couldn't register disk type");
830 dm_dirty_log_type_unregister(&_core_type);
833 return r;
836 static void __exit dm_dirty_log_exit(void)
838 dm_dirty_log_type_unregister(&_disk_type);
839 dm_dirty_log_type_unregister(&_core_type);
842 module_init(dm_dirty_log_init);
843 module_exit(dm_dirty_log_exit);
845 MODULE_DESCRIPTION(DM_NAME " dirty region log");
846 MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@redhat.com>");
847 MODULE_LICENSE("GPL");