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 ext2_test_bit(bit
, (unsigned long *) bs
) ? 1 : 0;
257 static inline void log_set_bit(struct log_c
*l
,
258 uint32_t *bs
, unsigned bit
)
260 ext2_set_bit(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 ext2_clear_bit(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_BARRIER
;
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(dm_div_up(buf_size
,
454 if (IS_ERR(lc
->io_req
.client
)) {
455 r
= PTR_ERR(lc
->io_req
.client
);
456 DMWARN("couldn't allocate disk io client");
461 lc
->disk_header
= vmalloc(buf_size
);
462 if (!lc
->disk_header
) {
463 DMWARN("couldn't allocate disk log buffer");
464 dm_io_client_destroy(lc
->io_req
.client
);
469 lc
->io_req
.mem
.ptr
.vma
= lc
->disk_header
;
470 lc
->clean_bits
= (void *)lc
->disk_header
+
471 (LOG_OFFSET
<< SECTOR_SHIFT
);
474 memset(lc
->clean_bits
, -1, bitset_size
);
476 lc
->sync_bits
= vmalloc(bitset_size
);
477 if (!lc
->sync_bits
) {
478 DMWARN("couldn't allocate sync bitset");
480 vfree(lc
->clean_bits
);
482 dm_io_client_destroy(lc
->io_req
.client
);
483 vfree(lc
->disk_header
);
487 memset(lc
->sync_bits
, (sync
== NOSYNC
) ? -1 : 0, bitset_size
);
488 lc
->sync_count
= (sync
== NOSYNC
) ? region_count
: 0;
490 lc
->recovering_bits
= vmalloc(bitset_size
);
491 if (!lc
->recovering_bits
) {
492 DMWARN("couldn't allocate sync bitset");
493 vfree(lc
->sync_bits
);
495 vfree(lc
->clean_bits
);
497 dm_io_client_destroy(lc
->io_req
.client
);
498 vfree(lc
->disk_header
);
502 memset(lc
->recovering_bits
, 0, bitset_size
);
509 static int core_ctr(struct dm_dirty_log
*log
, struct dm_target
*ti
,
510 unsigned int argc
, char **argv
)
512 return create_log_context(log
, ti
, argc
, argv
, NULL
);
515 static void destroy_log_context(struct log_c
*lc
)
517 vfree(lc
->sync_bits
);
518 vfree(lc
->recovering_bits
);
522 static void core_dtr(struct dm_dirty_log
*log
)
524 struct log_c
*lc
= (struct log_c
*) log
->context
;
526 vfree(lc
->clean_bits
);
527 destroy_log_context(lc
);
530 /*----------------------------------------------------------------
531 * disk log constructor/destructor
533 * argv contains log_device region_size followed optionally by [no]sync
534 *--------------------------------------------------------------*/
535 static int disk_ctr(struct dm_dirty_log
*log
, struct dm_target
*ti
,
536 unsigned int argc
, char **argv
)
541 if (argc
< 2 || argc
> 3) {
542 DMWARN("wrong number of arguments to disk dirty region log");
546 r
= dm_get_device(ti
, argv
[0], FMODE_READ
| FMODE_WRITE
, &dev
);
550 r
= create_log_context(log
, ti
, argc
- 1, argv
+ 1, dev
);
552 dm_put_device(ti
, dev
);
559 static void disk_dtr(struct dm_dirty_log
*log
)
561 struct log_c
*lc
= (struct log_c
*) log
->context
;
563 dm_put_device(lc
->ti
, lc
->log_dev
);
564 vfree(lc
->disk_header
);
565 dm_io_client_destroy(lc
->io_req
.client
);
566 destroy_log_context(lc
);
569 static int count_bits32(uint32_t *addr
, unsigned size
)
573 for (i
= 0; i
< size
; i
++) {
574 count
+= hweight32(*(addr
+i
));
579 static void fail_log_device(struct log_c
*lc
)
581 if (lc
->log_dev_failed
)
584 lc
->log_dev_failed
= 1;
585 dm_table_event(lc
->ti
->table
);
588 static int disk_resume(struct dm_dirty_log
*log
)
592 struct log_c
*lc
= (struct log_c
*) log
->context
;
593 size_t size
= lc
->bitset_uint32_count
* sizeof(uint32_t);
595 /* read the disk header */
598 DMWARN("%s: Failed to read header on dirty region log device",
602 * If the log device cannot be read, we must assume
603 * all regions are out-of-sync. If we simply return
604 * here, the state will be uninitialized and could
605 * lead us to return 'in-sync' status for regions
606 * that are actually 'out-of-sync'.
608 lc
->header
.nr_regions
= 0;
611 /* set or clear any new bits -- device has grown */
612 if (lc
->sync
== NOSYNC
)
613 for (i
= lc
->header
.nr_regions
; i
< lc
->region_count
; i
++)
614 /* FIXME: amazingly inefficient */
615 log_set_bit(lc
, lc
->clean_bits
, i
);
617 for (i
= lc
->header
.nr_regions
; i
< lc
->region_count
; i
++)
618 /* FIXME: amazingly inefficient */
619 log_clear_bit(lc
, lc
->clean_bits
, i
);
621 /* clear any old bits -- device has shrunk */
622 for (i
= lc
->region_count
; i
% (sizeof(*lc
->clean_bits
) << BYTE_SHIFT
); i
++)
623 log_clear_bit(lc
, lc
->clean_bits
, i
);
625 /* copy clean across to sync */
626 memcpy(lc
->sync_bits
, lc
->clean_bits
, size
);
627 lc
->sync_count
= count_bits32(lc
->clean_bits
, lc
->bitset_uint32_count
);
630 /* set the correct number of regions in the header */
631 lc
->header
.nr_regions
= lc
->region_count
;
633 header_to_disk(&lc
->header
, lc
->disk_header
);
635 /* write the new header */
636 r
= rw_header(lc
, WRITE
);
638 r
= flush_header(lc
);
640 lc
->log_dev_flush_failed
= 1;
643 DMWARN("%s: Failed to write header on dirty region log device",
651 static uint32_t core_get_region_size(struct dm_dirty_log
*log
)
653 struct log_c
*lc
= (struct log_c
*) log
->context
;
654 return lc
->region_size
;
657 static int core_resume(struct dm_dirty_log
*log
)
659 struct log_c
*lc
= (struct log_c
*) log
->context
;
664 static int core_is_clean(struct dm_dirty_log
*log
, region_t region
)
666 struct log_c
*lc
= (struct log_c
*) log
->context
;
667 return log_test_bit(lc
->clean_bits
, region
);
670 static int core_in_sync(struct dm_dirty_log
*log
, region_t region
, int block
)
672 struct log_c
*lc
= (struct log_c
*) log
->context
;
673 return log_test_bit(lc
->sync_bits
, region
);
676 static int core_flush(struct dm_dirty_log
*log
)
682 static int disk_flush(struct dm_dirty_log
*log
)
685 struct log_c
*lc
= log
->context
;
687 /* only write if the log has changed */
688 if (!lc
->touched_cleaned
&& !lc
->touched_dirtied
)
691 if (lc
->touched_cleaned
&& log
->flush_callback_fn
&&
692 log
->flush_callback_fn(lc
->ti
)) {
694 * At this point it is impossible to determine which
695 * regions are clean and which are dirty (without
696 * re-reading the log off disk). So mark all of them
699 lc
->flush_failed
= 1;
700 for (i
= 0; i
< lc
->region_count
; i
++)
701 log_clear_bit(lc
, lc
->clean_bits
, i
);
704 r
= rw_header(lc
, WRITE
);
708 if (lc
->touched_dirtied
) {
709 r
= flush_header(lc
);
711 lc
->log_dev_flush_failed
= 1;
714 lc
->touched_dirtied
= 0;
716 lc
->touched_cleaned
= 0;
722 static void core_mark_region(struct dm_dirty_log
*log
, region_t region
)
724 struct log_c
*lc
= (struct log_c
*) log
->context
;
725 log_clear_bit(lc
, lc
->clean_bits
, region
);
728 static void core_clear_region(struct dm_dirty_log
*log
, region_t region
)
730 struct log_c
*lc
= (struct log_c
*) log
->context
;
731 if (likely(!lc
->flush_failed
))
732 log_set_bit(lc
, lc
->clean_bits
, region
);
735 static int core_get_resync_work(struct dm_dirty_log
*log
, region_t
*region
)
737 struct log_c
*lc
= (struct log_c
*) log
->context
;
739 if (lc
->sync_search
>= lc
->region_count
)
743 *region
= ext2_find_next_zero_bit(
744 (unsigned long *) lc
->sync_bits
,
747 lc
->sync_search
= *region
+ 1;
749 if (*region
>= lc
->region_count
)
752 } while (log_test_bit(lc
->recovering_bits
, *region
));
754 log_set_bit(lc
, lc
->recovering_bits
, *region
);
758 static void core_set_region_sync(struct dm_dirty_log
*log
, region_t region
,
761 struct log_c
*lc
= (struct log_c
*) log
->context
;
763 log_clear_bit(lc
, lc
->recovering_bits
, region
);
765 log_set_bit(lc
, lc
->sync_bits
, region
);
767 } else if (log_test_bit(lc
->sync_bits
, region
)) {
769 log_clear_bit(lc
, lc
->sync_bits
, region
);
773 static region_t
core_get_sync_count(struct dm_dirty_log
*log
)
775 struct log_c
*lc
= (struct log_c
*) log
->context
;
777 return lc
->sync_count
;
780 #define DMEMIT_SYNC \
781 if (lc->sync != DEFAULTSYNC) \
782 DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
784 static int core_status(struct dm_dirty_log
*log
, status_type_t status
,
785 char *result
, unsigned int maxlen
)
788 struct log_c
*lc
= log
->context
;
791 case STATUSTYPE_INFO
:
792 DMEMIT("1 %s", log
->type
->name
);
795 case STATUSTYPE_TABLE
:
796 DMEMIT("%s %u %u ", log
->type
->name
,
797 lc
->sync
== DEFAULTSYNC
? 1 : 2, lc
->region_size
);
804 static int disk_status(struct dm_dirty_log
*log
, status_type_t status
,
805 char *result
, unsigned int maxlen
)
808 struct log_c
*lc
= log
->context
;
811 case STATUSTYPE_INFO
:
812 DMEMIT("3 %s %s %c", log
->type
->name
, lc
->log_dev
->name
,
813 lc
->log_dev_flush_failed
? 'F' :
814 lc
->log_dev_failed
? 'D' :
818 case STATUSTYPE_TABLE
:
819 DMEMIT("%s %u %s %u ", log
->type
->name
,
820 lc
->sync
== DEFAULTSYNC
? 2 : 3, lc
->log_dev
->name
,
828 static struct dm_dirty_log_type _core_type
= {
830 .module
= THIS_MODULE
,
833 .resume
= core_resume
,
834 .get_region_size
= core_get_region_size
,
835 .is_clean
= core_is_clean
,
836 .in_sync
= core_in_sync
,
838 .mark_region
= core_mark_region
,
839 .clear_region
= core_clear_region
,
840 .get_resync_work
= core_get_resync_work
,
841 .set_region_sync
= core_set_region_sync
,
842 .get_sync_count
= core_get_sync_count
,
843 .status
= core_status
,
846 static struct dm_dirty_log_type _disk_type
= {
848 .module
= THIS_MODULE
,
851 .postsuspend
= disk_flush
,
852 .resume
= disk_resume
,
853 .get_region_size
= core_get_region_size
,
854 .is_clean
= core_is_clean
,
855 .in_sync
= core_in_sync
,
857 .mark_region
= core_mark_region
,
858 .clear_region
= core_clear_region
,
859 .get_resync_work
= core_get_resync_work
,
860 .set_region_sync
= core_set_region_sync
,
861 .get_sync_count
= core_get_sync_count
,
862 .status
= disk_status
,
865 static int __init
dm_dirty_log_init(void)
869 r
= dm_dirty_log_type_register(&_core_type
);
871 DMWARN("couldn't register core log");
873 r
= dm_dirty_log_type_register(&_disk_type
);
875 DMWARN("couldn't register disk type");
876 dm_dirty_log_type_unregister(&_core_type
);
882 static void __exit
dm_dirty_log_exit(void)
884 dm_dirty_log_type_unregister(&_disk_type
);
885 dm_dirty_log_type_unregister(&_core_type
);
888 module_init(dm_dirty_log_init
);
889 module_exit(dm_dirty_log_exit
);
891 MODULE_DESCRIPTION(DM_NAME
" dirty region log");
892 MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@redhat.com>");
893 MODULE_LICENSE("GPL");