1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2003 Sistina Software
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
6 * This file is released under the LGPL.
9 #include <linux/init.h>
10 #include <linux/slab.h>
11 #include <linux/module.h>
12 #include <linux/vmalloc.h>
13 #include <linux/dm-io.h>
14 #include <linux/dm-dirty-log.h>
16 #include <linux/device-mapper.h>
18 #define DM_MSG_PREFIX "dirty region log"
20 static LIST_HEAD(_log_types
);
21 static DEFINE_SPINLOCK(_lock
);
23 static struct dm_dirty_log_type
*__find_dirty_log_type(const char *name
)
25 struct dm_dirty_log_type
*log_type
;
27 list_for_each_entry(log_type
, &_log_types
, list
)
28 if (!strcmp(name
, log_type
->name
))
34 static struct dm_dirty_log_type
*_get_dirty_log_type(const char *name
)
36 struct dm_dirty_log_type
*log_type
;
40 log_type
= __find_dirty_log_type(name
);
41 if (log_type
&& !try_module_get(log_type
->module
))
53 * Attempt to retrieve the dm_dirty_log_type by name. If not already
54 * available, attempt to load the appropriate module.
56 * Log modules are named "dm-log-" followed by the 'type_name'.
57 * Modules may contain multiple types.
58 * This function will first try the module "dm-log-<type_name>",
59 * then truncate 'type_name' on the last '-' and try again.
61 * For example, if type_name was "clustered-disk", it would search
62 * 'dm-log-clustered-disk' then 'dm-log-clustered'.
64 * Returns: dirty_log_type* on success, NULL on failure
66 static struct dm_dirty_log_type
*get_type(const char *type_name
)
68 char *p
, *type_name_dup
;
69 struct dm_dirty_log_type
*log_type
;
74 log_type
= _get_dirty_log_type(type_name
);
78 type_name_dup
= kstrdup(type_name
, GFP_KERNEL
);
80 DMWARN("No memory left to attempt log module load for \"%s\"",
85 while (request_module("dm-log-%s", type_name_dup
) ||
86 !(log_type
= _get_dirty_log_type(type_name
))) {
87 p
= strrchr(type_name_dup
, '-');
94 DMWARN("Module for logging type \"%s\" not found.", type_name
);
101 static void put_type(struct dm_dirty_log_type
*type
)
107 if (!__find_dirty_log_type(type
->name
))
110 module_put(type
->module
);
116 int dm_dirty_log_type_register(struct dm_dirty_log_type
*type
)
121 if (!__find_dirty_log_type(type
->name
))
122 list_add(&type
->list
, &_log_types
);
129 EXPORT_SYMBOL(dm_dirty_log_type_register
);
131 int dm_dirty_log_type_unregister(struct dm_dirty_log_type
*type
)
135 if (!__find_dirty_log_type(type
->name
)) {
140 list_del(&type
->list
);
146 EXPORT_SYMBOL(dm_dirty_log_type_unregister
);
148 struct dm_dirty_log
*dm_dirty_log_create(const char *type_name
,
149 struct dm_target
*ti
,
150 int (*flush_callback_fn
)(struct dm_target
*ti
),
151 unsigned int argc
, char **argv
)
153 struct dm_dirty_log_type
*type
;
154 struct dm_dirty_log
*log
;
156 log
= kmalloc(sizeof(*log
), GFP_KERNEL
);
160 type
= get_type(type_name
);
166 log
->flush_callback_fn
= flush_callback_fn
;
168 if (type
->ctr(log
, ti
, argc
, argv
)) {
176 EXPORT_SYMBOL(dm_dirty_log_create
);
178 void dm_dirty_log_destroy(struct dm_dirty_log
*log
)
184 EXPORT_SYMBOL(dm_dirty_log_destroy
);
187 *---------------------------------------------------------------
188 * Persistent and core logs share a lot of their implementation.
189 * FIXME: need a reload method to be called from a resume
190 *---------------------------------------------------------------
193 * Magic for persistent mirrors: "MiRr"
195 #define MIRROR_MAGIC 0x4D695272
198 * The on-disk version of the metadata.
200 #define MIRROR_DISK_VERSION 2
203 struct log_header_disk
{
207 * Simple, incrementing version. no backward
214 struct log_header_core
{
221 struct dm_target
*ti
;
225 uint32_t region_size
;
226 unsigned int region_count
;
229 unsigned int bitset_uint32_count
;
230 uint32_t *clean_bits
;
232 uint32_t *recovering_bits
; /* FIXME: this seems excessive */
238 DEFAULTSYNC
, /* Synchronize if necessary */
239 NOSYNC
, /* Devices known to be already in sync */
240 FORCESYNC
, /* Force a sync to happen */
243 struct dm_io_request io_req
;
249 int log_dev_flush_failed
;
250 struct dm_dev
*log_dev
;
251 struct log_header_core header
;
253 struct dm_io_region header_location
;
254 struct log_header_disk
*disk_header
;
258 * The touched member needs to be updated every time we access
259 * one of the bitsets.
261 static inline int log_test_bit(uint32_t *bs
, unsigned int bit
)
263 return test_bit_le(bit
, bs
) ? 1 : 0;
266 static inline void log_set_bit(struct log_c
*l
,
267 uint32_t *bs
, unsigned int bit
)
269 __set_bit_le(bit
, bs
);
270 l
->touched_cleaned
= 1;
273 static inline void log_clear_bit(struct log_c
*l
,
274 uint32_t *bs
, unsigned int bit
)
276 __clear_bit_le(bit
, bs
);
277 l
->touched_dirtied
= 1;
281 *---------------------------------------------------------------
283 *--------------------------------------------------------------
285 static void header_to_disk(struct log_header_core
*core
, struct log_header_disk
*disk
)
287 disk
->magic
= cpu_to_le32(core
->magic
);
288 disk
->version
= cpu_to_le32(core
->version
);
289 disk
->nr_regions
= cpu_to_le64(core
->nr_regions
);
292 static void header_from_disk(struct log_header_core
*core
, struct log_header_disk
*disk
)
294 core
->magic
= le32_to_cpu(disk
->magic
);
295 core
->version
= le32_to_cpu(disk
->version
);
296 core
->nr_regions
= le64_to_cpu(disk
->nr_regions
);
299 static int rw_header(struct log_c
*lc
, enum req_op op
)
301 lc
->io_req
.bi_opf
= op
;
303 return dm_io(&lc
->io_req
, 1, &lc
->header_location
, NULL
, IOPRIO_DEFAULT
);
306 static int flush_header(struct log_c
*lc
)
308 struct dm_io_region null_location
= {
309 .bdev
= lc
->header_location
.bdev
,
314 lc
->io_req
.bi_opf
= REQ_OP_WRITE
| REQ_PREFLUSH
;
316 return dm_io(&lc
->io_req
, 1, &null_location
, NULL
, IOPRIO_DEFAULT
);
319 static int read_header(struct log_c
*log
)
323 r
= rw_header(log
, REQ_OP_READ
);
327 header_from_disk(&log
->header
, log
->disk_header
);
329 /* New log required? */
330 if (log
->sync
!= DEFAULTSYNC
|| log
->header
.magic
!= MIRROR_MAGIC
) {
331 log
->header
.magic
= MIRROR_MAGIC
;
332 log
->header
.version
= MIRROR_DISK_VERSION
;
333 log
->header
.nr_regions
= 0;
336 #ifdef __LITTLE_ENDIAN
337 if (log
->header
.version
== 1)
338 log
->header
.version
= 2;
341 if (log
->header
.version
!= MIRROR_DISK_VERSION
) {
342 DMWARN("incompatible disk log version");
349 static int _check_region_size(struct dm_target
*ti
, uint32_t region_size
)
351 if (region_size
< 2 || region_size
> ti
->len
)
354 if (!is_power_of_2(region_size
))
361 *--------------------------------------------------------------
362 * core log constructor/destructor
364 * argv contains region_size followed optionally by [no]sync
365 *--------------------------------------------------------------
368 static int create_log_context(struct dm_dirty_log
*log
, struct dm_target
*ti
,
369 unsigned int argc
, char **argv
,
372 enum sync sync
= DEFAULTSYNC
;
375 uint32_t region_size
;
376 unsigned int region_count
;
377 size_t bitset_size
, buf_size
;
381 if (argc
< 1 || argc
> 2) {
382 DMWARN("wrong number of arguments to dirty region log");
387 if (!strcmp(argv
[1], "sync"))
389 else if (!strcmp(argv
[1], "nosync"))
392 DMWARN("unrecognised sync argument to dirty region log: %s", argv
[1]);
397 if (sscanf(argv
[0], "%u%c", ®ion_size
, &dummy
) != 1 ||
398 !_check_region_size(ti
, region_size
)) {
399 DMWARN("invalid region size %s", argv
[0]);
403 region_count
= dm_sector_div_up(ti
->len
, region_size
);
405 lc
= kmalloc(sizeof(*lc
), GFP_KERNEL
);
407 DMWARN("couldn't allocate core log");
412 lc
->touched_dirtied
= 0;
413 lc
->touched_cleaned
= 0;
414 lc
->flush_failed
= 0;
415 lc
->region_size
= region_size
;
416 lc
->region_count
= region_count
;
420 * Work out how many "unsigned long"s we need to hold the bitset.
422 bitset_size
= dm_round_up(region_count
, BITS_PER_LONG
);
423 bitset_size
>>= BYTE_SHIFT
;
425 lc
->bitset_uint32_count
= bitset_size
/ sizeof(*lc
->clean_bits
);
431 lc
->clean_bits
= vmalloc(bitset_size
);
432 if (!lc
->clean_bits
) {
433 DMWARN("couldn't allocate clean bitset");
437 lc
->disk_header
= NULL
;
440 lc
->log_dev_failed
= 0;
441 lc
->log_dev_flush_failed
= 0;
442 lc
->header_location
.bdev
= lc
->log_dev
->bdev
;
443 lc
->header_location
.sector
= 0;
446 * Buffer holds both header and bitset.
449 dm_round_up((LOG_OFFSET
<< SECTOR_SHIFT
) + bitset_size
,
450 bdev_logical_block_size(lc
->header_location
.bdev
));
452 if (buf_size
> bdev_nr_bytes(dev
->bdev
)) {
453 DMWARN("log device %s too small: need %llu bytes",
454 dev
->name
, (unsigned long long)buf_size
);
459 lc
->header_location
.count
= buf_size
>> SECTOR_SHIFT
;
461 lc
->io_req
.mem
.type
= DM_IO_VMA
;
462 lc
->io_req
.notify
.fn
= NULL
;
463 lc
->io_req
.client
= dm_io_client_create();
464 if (IS_ERR(lc
->io_req
.client
)) {
465 r
= PTR_ERR(lc
->io_req
.client
);
466 DMWARN("couldn't allocate disk io client");
471 lc
->disk_header
= vmalloc(buf_size
);
472 if (!lc
->disk_header
) {
473 DMWARN("couldn't allocate disk log buffer");
474 dm_io_client_destroy(lc
->io_req
.client
);
479 lc
->io_req
.mem
.ptr
.vma
= lc
->disk_header
;
480 lc
->clean_bits
= (void *)lc
->disk_header
+
481 (LOG_OFFSET
<< SECTOR_SHIFT
);
484 memset(lc
->clean_bits
, -1, bitset_size
);
486 lc
->sync_bits
= vmalloc(bitset_size
);
487 if (!lc
->sync_bits
) {
488 DMWARN("couldn't allocate sync bitset");
490 vfree(lc
->clean_bits
);
492 dm_io_client_destroy(lc
->io_req
.client
);
493 vfree(lc
->disk_header
);
497 memset(lc
->sync_bits
, (sync
== NOSYNC
) ? -1 : 0, bitset_size
);
498 lc
->sync_count
= (sync
== NOSYNC
) ? region_count
: 0;
500 lc
->recovering_bits
= vzalloc(bitset_size
);
501 if (!lc
->recovering_bits
) {
502 DMWARN("couldn't allocate sync bitset");
503 vfree(lc
->sync_bits
);
505 vfree(lc
->clean_bits
);
507 dm_io_client_destroy(lc
->io_req
.client
);
508 vfree(lc
->disk_header
);
518 static int core_ctr(struct dm_dirty_log
*log
, struct dm_target
*ti
,
519 unsigned int argc
, char **argv
)
521 return create_log_context(log
, ti
, argc
, argv
, NULL
);
524 static void destroy_log_context(struct log_c
*lc
)
526 vfree(lc
->sync_bits
);
527 vfree(lc
->recovering_bits
);
531 static void core_dtr(struct dm_dirty_log
*log
)
533 struct log_c
*lc
= log
->context
;
535 vfree(lc
->clean_bits
);
536 destroy_log_context(lc
);
540 *---------------------------------------------------------------------
541 * disk log constructor/destructor
543 * argv contains log_device region_size followed optionally by [no]sync
544 *---------------------------------------------------------------------
546 static int disk_ctr(struct dm_dirty_log
*log
, struct dm_target
*ti
,
547 unsigned int argc
, char **argv
)
552 if (argc
< 2 || argc
> 3) {
553 DMWARN("wrong number of arguments to disk dirty region log");
557 r
= dm_get_device(ti
, argv
[0], dm_table_get_mode(ti
->table
), &dev
);
561 r
= create_log_context(log
, ti
, argc
- 1, argv
+ 1, dev
);
563 dm_put_device(ti
, dev
);
570 static void disk_dtr(struct dm_dirty_log
*log
)
572 struct log_c
*lc
= log
->context
;
574 dm_put_device(lc
->ti
, lc
->log_dev
);
575 vfree(lc
->disk_header
);
576 dm_io_client_destroy(lc
->io_req
.client
);
577 destroy_log_context(lc
);
580 static void fail_log_device(struct log_c
*lc
)
582 if (lc
->log_dev_failed
)
585 lc
->log_dev_failed
= 1;
586 dm_table_event(lc
->ti
->table
);
589 static int disk_resume(struct dm_dirty_log
*log
)
593 struct log_c
*lc
= log
->context
;
594 size_t size
= lc
->bitset_uint32_count
* sizeof(uint32_t);
596 /* read the disk header */
599 DMWARN("%s: Failed to read header on dirty region log device",
603 * If the log device cannot be read, we must assume
604 * all regions are out-of-sync. If we simply return
605 * here, the state will be uninitialized and could
606 * lead us to return 'in-sync' status for regions
607 * that are actually 'out-of-sync'.
609 lc
->header
.nr_regions
= 0;
612 /* set or clear any new bits -- device has grown */
613 if (lc
->sync
== NOSYNC
)
614 for (i
= lc
->header
.nr_regions
; i
< lc
->region_count
; i
++)
615 /* FIXME: amazingly inefficient */
616 log_set_bit(lc
, lc
->clean_bits
, i
);
618 for (i
= lc
->header
.nr_regions
; i
< lc
->region_count
; i
++)
619 /* FIXME: amazingly inefficient */
620 log_clear_bit(lc
, lc
->clean_bits
, i
);
622 /* clear any old bits -- device has shrunk */
623 for (i
= lc
->region_count
; i
% BITS_PER_LONG
; i
++)
624 log_clear_bit(lc
, lc
->clean_bits
, i
);
626 /* copy clean across to sync */
627 memcpy(lc
->sync_bits
, lc
->clean_bits
, size
);
628 lc
->sync_count
= memweight(lc
->clean_bits
,
629 lc
->bitset_uint32_count
* sizeof(uint32_t));
632 /* set the correct number of regions in the header */
633 lc
->header
.nr_regions
= lc
->region_count
;
635 header_to_disk(&lc
->header
, lc
->disk_header
);
637 /* write the new header */
638 r
= rw_header(lc
, REQ_OP_WRITE
);
640 r
= flush_header(lc
);
642 lc
->log_dev_flush_failed
= 1;
645 DMWARN("%s: Failed to write header on dirty region log device",
653 static uint32_t core_get_region_size(struct dm_dirty_log
*log
)
655 struct log_c
*lc
= log
->context
;
657 return lc
->region_size
;
660 static int core_resume(struct dm_dirty_log
*log
)
662 struct log_c
*lc
= log
->context
;
668 static int core_is_clean(struct dm_dirty_log
*log
, region_t region
)
670 struct log_c
*lc
= log
->context
;
672 return log_test_bit(lc
->clean_bits
, region
);
675 static int core_in_sync(struct dm_dirty_log
*log
, region_t region
, int block
)
677 struct log_c
*lc
= log
->context
;
679 return log_test_bit(lc
->sync_bits
, region
);
682 static int core_flush(struct dm_dirty_log
*log
)
688 static int disk_flush(struct dm_dirty_log
*log
)
691 struct log_c
*lc
= log
->context
;
693 /* only write if the log has changed */
694 if (!lc
->touched_cleaned
&& !lc
->touched_dirtied
)
697 if (lc
->touched_cleaned
&& log
->flush_callback_fn
&&
698 log
->flush_callback_fn(lc
->ti
)) {
700 * At this point it is impossible to determine which
701 * regions are clean and which are dirty (without
702 * re-reading the log off disk). So mark all of them
705 lc
->flush_failed
= 1;
706 for (i
= 0; i
< lc
->region_count
; i
++)
707 log_clear_bit(lc
, lc
->clean_bits
, i
);
710 r
= rw_header(lc
, REQ_OP_WRITE
);
714 if (lc
->touched_dirtied
) {
715 r
= flush_header(lc
);
717 lc
->log_dev_flush_failed
= 1;
720 lc
->touched_dirtied
= 0;
722 lc
->touched_cleaned
= 0;
728 static void core_mark_region(struct dm_dirty_log
*log
, region_t region
)
730 struct log_c
*lc
= log
->context
;
732 log_clear_bit(lc
, lc
->clean_bits
, region
);
735 static void core_clear_region(struct dm_dirty_log
*log
, region_t region
)
737 struct log_c
*lc
= log
->context
;
739 if (likely(!lc
->flush_failed
))
740 log_set_bit(lc
, lc
->clean_bits
, region
);
743 static int core_get_resync_work(struct dm_dirty_log
*log
, region_t
*region
)
745 struct log_c
*lc
= log
->context
;
747 if (lc
->sync_search
>= lc
->region_count
)
751 *region
= find_next_zero_bit_le(lc
->sync_bits
,
754 lc
->sync_search
= *region
+ 1;
756 if (*region
>= lc
->region_count
)
759 } while (log_test_bit(lc
->recovering_bits
, *region
));
761 log_set_bit(lc
, lc
->recovering_bits
, *region
);
765 static void core_set_region_sync(struct dm_dirty_log
*log
, region_t region
,
768 struct log_c
*lc
= log
->context
;
770 log_clear_bit(lc
, lc
->recovering_bits
, region
);
772 log_set_bit(lc
, lc
->sync_bits
, region
);
774 } else if (log_test_bit(lc
->sync_bits
, region
)) {
776 log_clear_bit(lc
, lc
->sync_bits
, region
);
780 static region_t
core_get_sync_count(struct dm_dirty_log
*log
)
782 struct log_c
*lc
= log
->context
;
784 return lc
->sync_count
;
787 #define DMEMIT_SYNC \
789 if (lc->sync != DEFAULTSYNC) \
790 DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : ""); \
793 static int core_status(struct dm_dirty_log
*log
, status_type_t status
,
794 char *result
, unsigned int maxlen
)
797 struct log_c
*lc
= log
->context
;
800 case STATUSTYPE_INFO
:
801 DMEMIT("1 %s", log
->type
->name
);
804 case STATUSTYPE_TABLE
:
805 DMEMIT("%s %u %u ", log
->type
->name
,
806 lc
->sync
== DEFAULTSYNC
? 1 : 2, lc
->region_size
);
818 static int disk_status(struct dm_dirty_log
*log
, status_type_t status
,
819 char *result
, unsigned int maxlen
)
822 struct log_c
*lc
= log
->context
;
825 case STATUSTYPE_INFO
:
826 DMEMIT("3 %s %s %c", log
->type
->name
, lc
->log_dev
->name
,
827 lc
->log_dev_flush_failed
? 'F' :
828 lc
->log_dev_failed
? 'D' :
832 case STATUSTYPE_TABLE
:
833 DMEMIT("%s %u %s %u ", log
->type
->name
,
834 lc
->sync
== DEFAULTSYNC
? 2 : 3, lc
->log_dev
->name
,
847 static struct dm_dirty_log_type _core_type
= {
849 .module
= THIS_MODULE
,
852 .resume
= core_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
= core_status
,
865 static struct dm_dirty_log_type _disk_type
= {
867 .module
= THIS_MODULE
,
870 .postsuspend
= disk_flush
,
871 .resume
= disk_resume
,
872 .get_region_size
= core_get_region_size
,
873 .is_clean
= core_is_clean
,
874 .in_sync
= core_in_sync
,
876 .mark_region
= core_mark_region
,
877 .clear_region
= core_clear_region
,
878 .get_resync_work
= core_get_resync_work
,
879 .set_region_sync
= core_set_region_sync
,
880 .get_sync_count
= core_get_sync_count
,
881 .status
= disk_status
,
884 static int __init
dm_dirty_log_init(void)
888 r
= dm_dirty_log_type_register(&_core_type
);
890 DMWARN("couldn't register core log");
892 r
= dm_dirty_log_type_register(&_disk_type
);
894 DMWARN("couldn't register disk type");
895 dm_dirty_log_type_unregister(&_core_type
);
901 static void __exit
dm_dirty_log_exit(void)
903 dm_dirty_log_type_unregister(&_disk_type
);
904 dm_dirty_log_type_unregister(&_core_type
);
907 module_init(dm_dirty_log_init
);
908 module_exit(dm_dirty_log_exit
);
910 MODULE_DESCRIPTION(DM_NAME
" dirty region log");
911 MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@lists.linux.dev>");
912 MODULE_LICENSE("GPL");