First Support on Ginger and OMAP TI
[linux-ginger.git] / drivers / md / dm-log.c
blob9443896ede070152020046d15979bb9f1dc55b29
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;
212 uint32_t region_size;
213 unsigned int region_count;
214 region_t sync_count;
216 unsigned bitset_uint32_count;
217 uint32_t *clean_bits;
218 uint32_t *sync_bits;
219 uint32_t *recovering_bits; /* FIXME: this seems excessive */
221 int sync_search;
223 /* Resync flag */
224 enum sync {
225 DEFAULTSYNC, /* Synchronize if necessary */
226 NOSYNC, /* Devices known to be already in sync */
227 FORCESYNC, /* Force a sync to happen */
228 } sync;
230 struct dm_io_request io_req;
233 * Disk log fields
235 int log_dev_failed;
236 struct dm_dev *log_dev;
237 struct log_header header;
239 struct dm_io_region header_location;
240 struct log_header *disk_header;
244 * The touched member needs to be updated every time we access
245 * one of the bitsets.
247 static inline int log_test_bit(uint32_t *bs, unsigned bit)
249 return ext2_test_bit(bit, (unsigned long *) bs) ? 1 : 0;
252 static inline void log_set_bit(struct log_c *l,
253 uint32_t *bs, unsigned bit)
255 ext2_set_bit(bit, (unsigned long *) bs);
256 l->touched = 1;
259 static inline void log_clear_bit(struct log_c *l,
260 uint32_t *bs, unsigned bit)
262 ext2_clear_bit(bit, (unsigned long *) bs);
263 l->touched = 1;
266 /*----------------------------------------------------------------
267 * Header IO
268 *--------------------------------------------------------------*/
269 static void header_to_disk(struct log_header *core, struct log_header *disk)
271 disk->magic = cpu_to_le32(core->magic);
272 disk->version = cpu_to_le32(core->version);
273 disk->nr_regions = cpu_to_le64(core->nr_regions);
276 static void header_from_disk(struct log_header *core, struct log_header *disk)
278 core->magic = le32_to_cpu(disk->magic);
279 core->version = le32_to_cpu(disk->version);
280 core->nr_regions = le64_to_cpu(disk->nr_regions);
283 static int rw_header(struct log_c *lc, int rw)
285 lc->io_req.bi_rw = rw;
287 return dm_io(&lc->io_req, 1, &lc->header_location, NULL);
290 static int read_header(struct log_c *log)
292 int r;
294 r = rw_header(log, READ);
295 if (r)
296 return r;
298 header_from_disk(&log->header, log->disk_header);
300 /* New log required? */
301 if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) {
302 log->header.magic = MIRROR_MAGIC;
303 log->header.version = MIRROR_DISK_VERSION;
304 log->header.nr_regions = 0;
307 #ifdef __LITTLE_ENDIAN
308 if (log->header.version == 1)
309 log->header.version = 2;
310 #endif
312 if (log->header.version != MIRROR_DISK_VERSION) {
313 DMWARN("incompatible disk log version");
314 return -EINVAL;
317 return 0;
320 static int _check_region_size(struct dm_target *ti, uint32_t region_size)
322 if (region_size < 2 || region_size > ti->len)
323 return 0;
325 if (!is_power_of_2(region_size))
326 return 0;
328 return 1;
331 /*----------------------------------------------------------------
332 * core log constructor/destructor
334 * argv contains region_size followed optionally by [no]sync
335 *--------------------------------------------------------------*/
336 #define BYTE_SHIFT 3
337 static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti,
338 unsigned int argc, char **argv,
339 struct dm_dev *dev)
341 enum sync sync = DEFAULTSYNC;
343 struct log_c *lc;
344 uint32_t region_size;
345 unsigned int region_count;
346 size_t bitset_size, buf_size;
347 int r;
349 if (argc < 1 || argc > 2) {
350 DMWARN("wrong number of arguments to dirty region log");
351 return -EINVAL;
354 if (argc > 1) {
355 if (!strcmp(argv[1], "sync"))
356 sync = FORCESYNC;
357 else if (!strcmp(argv[1], "nosync"))
358 sync = NOSYNC;
359 else {
360 DMWARN("unrecognised sync argument to "
361 "dirty region log: %s", argv[1]);
362 return -EINVAL;
366 if (sscanf(argv[0], "%u", &region_size) != 1 ||
367 !_check_region_size(ti, region_size)) {
368 DMWARN("invalid region size %s", argv[0]);
369 return -EINVAL;
372 region_count = dm_sector_div_up(ti->len, region_size);
374 lc = kmalloc(sizeof(*lc), GFP_KERNEL);
375 if (!lc) {
376 DMWARN("couldn't allocate core log");
377 return -ENOMEM;
380 lc->ti = ti;
381 lc->touched = 0;
382 lc->region_size = region_size;
383 lc->region_count = region_count;
384 lc->sync = sync;
387 * Work out how many "unsigned long"s we need to hold the bitset.
389 bitset_size = dm_round_up(region_count,
390 sizeof(*lc->clean_bits) << BYTE_SHIFT);
391 bitset_size >>= BYTE_SHIFT;
393 lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits);
396 * Disk log?
398 if (!dev) {
399 lc->clean_bits = vmalloc(bitset_size);
400 if (!lc->clean_bits) {
401 DMWARN("couldn't allocate clean bitset");
402 kfree(lc);
403 return -ENOMEM;
405 lc->disk_header = NULL;
406 } else {
407 lc->log_dev = dev;
408 lc->log_dev_failed = 0;
409 lc->header_location.bdev = lc->log_dev->bdev;
410 lc->header_location.sector = 0;
413 * Buffer holds both header and bitset.
415 buf_size =
416 dm_round_up((LOG_OFFSET << SECTOR_SHIFT) + bitset_size,
417 bdev_logical_block_size(lc->header_location.
418 bdev));
420 if (buf_size > i_size_read(dev->bdev->bd_inode)) {
421 DMWARN("log device %s too small: need %llu bytes",
422 dev->name, (unsigned long long)buf_size);
423 kfree(lc);
424 return -EINVAL;
427 lc->header_location.count = buf_size >> SECTOR_SHIFT;
429 lc->io_req.mem.type = DM_IO_VMA;
430 lc->io_req.notify.fn = NULL;
431 lc->io_req.client = dm_io_client_create(dm_div_up(buf_size,
432 PAGE_SIZE));
433 if (IS_ERR(lc->io_req.client)) {
434 r = PTR_ERR(lc->io_req.client);
435 DMWARN("couldn't allocate disk io client");
436 kfree(lc);
437 return -ENOMEM;
440 lc->disk_header = vmalloc(buf_size);
441 if (!lc->disk_header) {
442 DMWARN("couldn't allocate disk log buffer");
443 dm_io_client_destroy(lc->io_req.client);
444 kfree(lc);
445 return -ENOMEM;
448 lc->io_req.mem.ptr.vma = lc->disk_header;
449 lc->clean_bits = (void *)lc->disk_header +
450 (LOG_OFFSET << SECTOR_SHIFT);
453 memset(lc->clean_bits, -1, bitset_size);
455 lc->sync_bits = vmalloc(bitset_size);
456 if (!lc->sync_bits) {
457 DMWARN("couldn't allocate sync bitset");
458 if (!dev)
459 vfree(lc->clean_bits);
460 else
461 dm_io_client_destroy(lc->io_req.client);
462 vfree(lc->disk_header);
463 kfree(lc);
464 return -ENOMEM;
466 memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
467 lc->sync_count = (sync == NOSYNC) ? region_count : 0;
469 lc->recovering_bits = vmalloc(bitset_size);
470 if (!lc->recovering_bits) {
471 DMWARN("couldn't allocate sync bitset");
472 vfree(lc->sync_bits);
473 if (!dev)
474 vfree(lc->clean_bits);
475 else
476 dm_io_client_destroy(lc->io_req.client);
477 vfree(lc->disk_header);
478 kfree(lc);
479 return -ENOMEM;
481 memset(lc->recovering_bits, 0, bitset_size);
482 lc->sync_search = 0;
483 log->context = lc;
485 return 0;
488 static int core_ctr(struct dm_dirty_log *log, struct dm_target *ti,
489 unsigned int argc, char **argv)
491 return create_log_context(log, ti, argc, argv, NULL);
494 static void destroy_log_context(struct log_c *lc)
496 vfree(lc->sync_bits);
497 vfree(lc->recovering_bits);
498 kfree(lc);
501 static void core_dtr(struct dm_dirty_log *log)
503 struct log_c *lc = (struct log_c *) log->context;
505 vfree(lc->clean_bits);
506 destroy_log_context(lc);
509 /*----------------------------------------------------------------
510 * disk log constructor/destructor
512 * argv contains log_device region_size followed optionally by [no]sync
513 *--------------------------------------------------------------*/
514 static int disk_ctr(struct dm_dirty_log *log, struct dm_target *ti,
515 unsigned int argc, char **argv)
517 int r;
518 struct dm_dev *dev;
520 if (argc < 2 || argc > 3) {
521 DMWARN("wrong number of arguments to disk dirty region log");
522 return -EINVAL;
525 r = dm_get_device(ti, argv[0], 0, 0 /* FIXME */,
526 FMODE_READ | FMODE_WRITE, &dev);
527 if (r)
528 return r;
530 r = create_log_context(log, ti, argc - 1, argv + 1, dev);
531 if (r) {
532 dm_put_device(ti, dev);
533 return r;
536 return 0;
539 static void disk_dtr(struct dm_dirty_log *log)
541 struct log_c *lc = (struct log_c *) log->context;
543 dm_put_device(lc->ti, lc->log_dev);
544 vfree(lc->disk_header);
545 dm_io_client_destroy(lc->io_req.client);
546 destroy_log_context(lc);
549 static int count_bits32(uint32_t *addr, unsigned size)
551 int count = 0, i;
553 for (i = 0; i < size; i++) {
554 count += hweight32(*(addr+i));
556 return count;
559 static void fail_log_device(struct log_c *lc)
561 if (lc->log_dev_failed)
562 return;
564 lc->log_dev_failed = 1;
565 dm_table_event(lc->ti->table);
568 static int disk_resume(struct dm_dirty_log *log)
570 int r;
571 unsigned i;
572 struct log_c *lc = (struct log_c *) log->context;
573 size_t size = lc->bitset_uint32_count * sizeof(uint32_t);
575 /* read the disk header */
576 r = read_header(lc);
577 if (r) {
578 DMWARN("%s: Failed to read header on dirty region log device",
579 lc->log_dev->name);
580 fail_log_device(lc);
582 * If the log device cannot be read, we must assume
583 * all regions are out-of-sync. If we simply return
584 * here, the state will be uninitialized and could
585 * lead us to return 'in-sync' status for regions
586 * that are actually 'out-of-sync'.
588 lc->header.nr_regions = 0;
591 /* set or clear any new bits -- device has grown */
592 if (lc->sync == NOSYNC)
593 for (i = lc->header.nr_regions; i < lc->region_count; i++)
594 /* FIXME: amazingly inefficient */
595 log_set_bit(lc, lc->clean_bits, i);
596 else
597 for (i = lc->header.nr_regions; i < lc->region_count; i++)
598 /* FIXME: amazingly inefficient */
599 log_clear_bit(lc, lc->clean_bits, i);
601 /* clear any old bits -- device has shrunk */
602 for (i = lc->region_count; i % (sizeof(*lc->clean_bits) << BYTE_SHIFT); i++)
603 log_clear_bit(lc, lc->clean_bits, i);
605 /* copy clean across to sync */
606 memcpy(lc->sync_bits, lc->clean_bits, size);
607 lc->sync_count = count_bits32(lc->clean_bits, lc->bitset_uint32_count);
608 lc->sync_search = 0;
610 /* set the correct number of regions in the header */
611 lc->header.nr_regions = lc->region_count;
613 header_to_disk(&lc->header, lc->disk_header);
615 /* write the new header */
616 r = rw_header(lc, WRITE);
617 if (r) {
618 DMWARN("%s: Failed to write header on dirty region log device",
619 lc->log_dev->name);
620 fail_log_device(lc);
623 return r;
626 static uint32_t core_get_region_size(struct dm_dirty_log *log)
628 struct log_c *lc = (struct log_c *) log->context;
629 return lc->region_size;
632 static int core_resume(struct dm_dirty_log *log)
634 struct log_c *lc = (struct log_c *) log->context;
635 lc->sync_search = 0;
636 return 0;
639 static int core_is_clean(struct dm_dirty_log *log, region_t region)
641 struct log_c *lc = (struct log_c *) log->context;
642 return log_test_bit(lc->clean_bits, region);
645 static int core_in_sync(struct dm_dirty_log *log, region_t region, int block)
647 struct log_c *lc = (struct log_c *) log->context;
648 return log_test_bit(lc->sync_bits, region);
651 static int core_flush(struct dm_dirty_log *log)
653 /* no op */
654 return 0;
657 static int disk_flush(struct dm_dirty_log *log)
659 int r;
660 struct log_c *lc = (struct log_c *) log->context;
662 /* only write if the log has changed */
663 if (!lc->touched)
664 return 0;
666 r = rw_header(lc, WRITE);
667 if (r)
668 fail_log_device(lc);
669 else
670 lc->touched = 0;
672 return r;
675 static void core_mark_region(struct dm_dirty_log *log, region_t region)
677 struct log_c *lc = (struct log_c *) log->context;
678 log_clear_bit(lc, lc->clean_bits, region);
681 static void core_clear_region(struct dm_dirty_log *log, region_t region)
683 struct log_c *lc = (struct log_c *) log->context;
684 log_set_bit(lc, lc->clean_bits, region);
687 static int core_get_resync_work(struct dm_dirty_log *log, region_t *region)
689 struct log_c *lc = (struct log_c *) log->context;
691 if (lc->sync_search >= lc->region_count)
692 return 0;
694 do {
695 *region = ext2_find_next_zero_bit(
696 (unsigned long *) lc->sync_bits,
697 lc->region_count,
698 lc->sync_search);
699 lc->sync_search = *region + 1;
701 if (*region >= lc->region_count)
702 return 0;
704 } while (log_test_bit(lc->recovering_bits, *region));
706 log_set_bit(lc, lc->recovering_bits, *region);
707 return 1;
710 static void core_set_region_sync(struct dm_dirty_log *log, region_t region,
711 int in_sync)
713 struct log_c *lc = (struct log_c *) log->context;
715 log_clear_bit(lc, lc->recovering_bits, region);
716 if (in_sync) {
717 log_set_bit(lc, lc->sync_bits, region);
718 lc->sync_count++;
719 } else if (log_test_bit(lc->sync_bits, region)) {
720 lc->sync_count--;
721 log_clear_bit(lc, lc->sync_bits, region);
725 static region_t core_get_sync_count(struct dm_dirty_log *log)
727 struct log_c *lc = (struct log_c *) log->context;
729 return lc->sync_count;
732 #define DMEMIT_SYNC \
733 if (lc->sync != DEFAULTSYNC) \
734 DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
736 static int core_status(struct dm_dirty_log *log, status_type_t status,
737 char *result, unsigned int maxlen)
739 int sz = 0;
740 struct log_c *lc = log->context;
742 switch(status) {
743 case STATUSTYPE_INFO:
744 DMEMIT("1 %s", log->type->name);
745 break;
747 case STATUSTYPE_TABLE:
748 DMEMIT("%s %u %u ", log->type->name,
749 lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size);
750 DMEMIT_SYNC;
753 return sz;
756 static int disk_status(struct dm_dirty_log *log, status_type_t status,
757 char *result, unsigned int maxlen)
759 int sz = 0;
760 struct log_c *lc = log->context;
762 switch(status) {
763 case STATUSTYPE_INFO:
764 DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name,
765 lc->log_dev_failed ? 'D' : 'A');
766 break;
768 case STATUSTYPE_TABLE:
769 DMEMIT("%s %u %s %u ", log->type->name,
770 lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name,
771 lc->region_size);
772 DMEMIT_SYNC;
775 return sz;
778 static struct dm_dirty_log_type _core_type = {
779 .name = "core",
780 .module = THIS_MODULE,
781 .ctr = core_ctr,
782 .dtr = core_dtr,
783 .resume = core_resume,
784 .get_region_size = core_get_region_size,
785 .is_clean = core_is_clean,
786 .in_sync = core_in_sync,
787 .flush = core_flush,
788 .mark_region = core_mark_region,
789 .clear_region = core_clear_region,
790 .get_resync_work = core_get_resync_work,
791 .set_region_sync = core_set_region_sync,
792 .get_sync_count = core_get_sync_count,
793 .status = core_status,
796 static struct dm_dirty_log_type _disk_type = {
797 .name = "disk",
798 .module = THIS_MODULE,
799 .ctr = disk_ctr,
800 .dtr = disk_dtr,
801 .postsuspend = disk_flush,
802 .resume = disk_resume,
803 .get_region_size = core_get_region_size,
804 .is_clean = core_is_clean,
805 .in_sync = core_in_sync,
806 .flush = disk_flush,
807 .mark_region = core_mark_region,
808 .clear_region = core_clear_region,
809 .get_resync_work = core_get_resync_work,
810 .set_region_sync = core_set_region_sync,
811 .get_sync_count = core_get_sync_count,
812 .status = disk_status,
815 static int __init dm_dirty_log_init(void)
817 int r;
819 r = dm_dirty_log_type_register(&_core_type);
820 if (r)
821 DMWARN("couldn't register core log");
823 r = dm_dirty_log_type_register(&_disk_type);
824 if (r) {
825 DMWARN("couldn't register disk type");
826 dm_dirty_log_type_unregister(&_core_type);
829 return r;
832 static void __exit dm_dirty_log_exit(void)
834 dm_dirty_log_type_unregister(&_disk_type);
835 dm_dirty_log_type_unregister(&_core_type);
838 module_init(dm_dirty_log_init);
839 module_exit(dm_dirty_log_exit);
841 MODULE_DESCRIPTION(DM_NAME " dirty region log");
842 MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@redhat.com>");
843 MODULE_LICENSE("GPL");