Merge git://git.infradead.org/iommu-2.6
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / md / dm-log.c
blob6fa8ccf91c70e70174d578e10b8793b884a86406
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 = dm_round_up((LOG_OFFSET << SECTOR_SHIFT) +
416 bitset_size,
417 ti->limits.logical_block_size);
419 if (buf_size > dev->bdev->bd_inode->i_size) {
420 DMWARN("log device %s too small: need %llu bytes",
421 dev->name, (unsigned long long)buf_size);
422 kfree(lc);
423 return -EINVAL;
426 lc->header_location.count = buf_size >> SECTOR_SHIFT;
428 lc->io_req.mem.type = DM_IO_VMA;
429 lc->io_req.notify.fn = NULL;
430 lc->io_req.client = dm_io_client_create(dm_div_up(buf_size,
431 PAGE_SIZE));
432 if (IS_ERR(lc->io_req.client)) {
433 r = PTR_ERR(lc->io_req.client);
434 DMWARN("couldn't allocate disk io client");
435 kfree(lc);
436 return -ENOMEM;
439 lc->disk_header = vmalloc(buf_size);
440 if (!lc->disk_header) {
441 DMWARN("couldn't allocate disk log buffer");
442 dm_io_client_destroy(lc->io_req.client);
443 kfree(lc);
444 return -ENOMEM;
447 lc->io_req.mem.ptr.vma = lc->disk_header;
448 lc->clean_bits = (void *)lc->disk_header +
449 (LOG_OFFSET << SECTOR_SHIFT);
452 memset(lc->clean_bits, -1, bitset_size);
454 lc->sync_bits = vmalloc(bitset_size);
455 if (!lc->sync_bits) {
456 DMWARN("couldn't allocate sync bitset");
457 if (!dev)
458 vfree(lc->clean_bits);
459 else
460 dm_io_client_destroy(lc->io_req.client);
461 vfree(lc->disk_header);
462 kfree(lc);
463 return -ENOMEM;
465 memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
466 lc->sync_count = (sync == NOSYNC) ? region_count : 0;
468 lc->recovering_bits = vmalloc(bitset_size);
469 if (!lc->recovering_bits) {
470 DMWARN("couldn't allocate sync bitset");
471 vfree(lc->sync_bits);
472 if (!dev)
473 vfree(lc->clean_bits);
474 else
475 dm_io_client_destroy(lc->io_req.client);
476 vfree(lc->disk_header);
477 kfree(lc);
478 return -ENOMEM;
480 memset(lc->recovering_bits, 0, bitset_size);
481 lc->sync_search = 0;
482 log->context = lc;
484 return 0;
487 static int core_ctr(struct dm_dirty_log *log, struct dm_target *ti,
488 unsigned int argc, char **argv)
490 return create_log_context(log, ti, argc, argv, NULL);
493 static void destroy_log_context(struct log_c *lc)
495 vfree(lc->sync_bits);
496 vfree(lc->recovering_bits);
497 kfree(lc);
500 static void core_dtr(struct dm_dirty_log *log)
502 struct log_c *lc = (struct log_c *) log->context;
504 vfree(lc->clean_bits);
505 destroy_log_context(lc);
508 /*----------------------------------------------------------------
509 * disk log constructor/destructor
511 * argv contains log_device region_size followed optionally by [no]sync
512 *--------------------------------------------------------------*/
513 static int disk_ctr(struct dm_dirty_log *log, struct dm_target *ti,
514 unsigned int argc, char **argv)
516 int r;
517 struct dm_dev *dev;
519 if (argc < 2 || argc > 3) {
520 DMWARN("wrong number of arguments to disk dirty region log");
521 return -EINVAL;
524 r = dm_get_device(ti, argv[0], 0, 0 /* FIXME */,
525 FMODE_READ | FMODE_WRITE, &dev);
526 if (r)
527 return r;
529 r = create_log_context(log, ti, argc - 1, argv + 1, dev);
530 if (r) {
531 dm_put_device(ti, dev);
532 return r;
535 return 0;
538 static void disk_dtr(struct dm_dirty_log *log)
540 struct log_c *lc = (struct log_c *) log->context;
542 dm_put_device(lc->ti, lc->log_dev);
543 vfree(lc->disk_header);
544 dm_io_client_destroy(lc->io_req.client);
545 destroy_log_context(lc);
548 static int count_bits32(uint32_t *addr, unsigned size)
550 int count = 0, i;
552 for (i = 0; i < size; i++) {
553 count += hweight32(*(addr+i));
555 return count;
558 static void fail_log_device(struct log_c *lc)
560 if (lc->log_dev_failed)
561 return;
563 lc->log_dev_failed = 1;
564 dm_table_event(lc->ti->table);
567 static int disk_resume(struct dm_dirty_log *log)
569 int r;
570 unsigned i;
571 struct log_c *lc = (struct log_c *) log->context;
572 size_t size = lc->bitset_uint32_count * sizeof(uint32_t);
574 /* read the disk header */
575 r = read_header(lc);
576 if (r) {
577 DMWARN("%s: Failed to read header on dirty region log device",
578 lc->log_dev->name);
579 fail_log_device(lc);
581 * If the log device cannot be read, we must assume
582 * all regions are out-of-sync. If we simply return
583 * here, the state will be uninitialized and could
584 * lead us to return 'in-sync' status for regions
585 * that are actually 'out-of-sync'.
587 lc->header.nr_regions = 0;
590 /* set or clear any new bits -- device has grown */
591 if (lc->sync == NOSYNC)
592 for (i = lc->header.nr_regions; i < lc->region_count; i++)
593 /* FIXME: amazingly inefficient */
594 log_set_bit(lc, lc->clean_bits, i);
595 else
596 for (i = lc->header.nr_regions; i < lc->region_count; i++)
597 /* FIXME: amazingly inefficient */
598 log_clear_bit(lc, lc->clean_bits, i);
600 /* clear any old bits -- device has shrunk */
601 for (i = lc->region_count; i % (sizeof(*lc->clean_bits) << BYTE_SHIFT); i++)
602 log_clear_bit(lc, lc->clean_bits, i);
604 /* copy clean across to sync */
605 memcpy(lc->sync_bits, lc->clean_bits, size);
606 lc->sync_count = count_bits32(lc->clean_bits, lc->bitset_uint32_count);
607 lc->sync_search = 0;
609 /* set the correct number of regions in the header */
610 lc->header.nr_regions = lc->region_count;
612 header_to_disk(&lc->header, lc->disk_header);
614 /* write the new header */
615 r = rw_header(lc, WRITE);
616 if (r) {
617 DMWARN("%s: Failed to write header on dirty region log device",
618 lc->log_dev->name);
619 fail_log_device(lc);
622 return r;
625 static uint32_t core_get_region_size(struct dm_dirty_log *log)
627 struct log_c *lc = (struct log_c *) log->context;
628 return lc->region_size;
631 static int core_resume(struct dm_dirty_log *log)
633 struct log_c *lc = (struct log_c *) log->context;
634 lc->sync_search = 0;
635 return 0;
638 static int core_is_clean(struct dm_dirty_log *log, region_t region)
640 struct log_c *lc = (struct log_c *) log->context;
641 return log_test_bit(lc->clean_bits, region);
644 static int core_in_sync(struct dm_dirty_log *log, region_t region, int block)
646 struct log_c *lc = (struct log_c *) log->context;
647 return log_test_bit(lc->sync_bits, region);
650 static int core_flush(struct dm_dirty_log *log)
652 /* no op */
653 return 0;
656 static int disk_flush(struct dm_dirty_log *log)
658 int r;
659 struct log_c *lc = (struct log_c *) log->context;
661 /* only write if the log has changed */
662 if (!lc->touched)
663 return 0;
665 r = rw_header(lc, WRITE);
666 if (r)
667 fail_log_device(lc);
668 else
669 lc->touched = 0;
671 return r;
674 static void core_mark_region(struct dm_dirty_log *log, region_t region)
676 struct log_c *lc = (struct log_c *) log->context;
677 log_clear_bit(lc, lc->clean_bits, region);
680 static void core_clear_region(struct dm_dirty_log *log, region_t region)
682 struct log_c *lc = (struct log_c *) log->context;
683 log_set_bit(lc, lc->clean_bits, region);
686 static int core_get_resync_work(struct dm_dirty_log *log, region_t *region)
688 struct log_c *lc = (struct log_c *) log->context;
690 if (lc->sync_search >= lc->region_count)
691 return 0;
693 do {
694 *region = ext2_find_next_zero_bit(
695 (unsigned long *) lc->sync_bits,
696 lc->region_count,
697 lc->sync_search);
698 lc->sync_search = *region + 1;
700 if (*region >= lc->region_count)
701 return 0;
703 } while (log_test_bit(lc->recovering_bits, *region));
705 log_set_bit(lc, lc->recovering_bits, *region);
706 return 1;
709 static void core_set_region_sync(struct dm_dirty_log *log, region_t region,
710 int in_sync)
712 struct log_c *lc = (struct log_c *) log->context;
714 log_clear_bit(lc, lc->recovering_bits, region);
715 if (in_sync) {
716 log_set_bit(lc, lc->sync_bits, region);
717 lc->sync_count++;
718 } else if (log_test_bit(lc->sync_bits, region)) {
719 lc->sync_count--;
720 log_clear_bit(lc, lc->sync_bits, region);
724 static region_t core_get_sync_count(struct dm_dirty_log *log)
726 struct log_c *lc = (struct log_c *) log->context;
728 return lc->sync_count;
731 #define DMEMIT_SYNC \
732 if (lc->sync != DEFAULTSYNC) \
733 DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
735 static int core_status(struct dm_dirty_log *log, status_type_t status,
736 char *result, unsigned int maxlen)
738 int sz = 0;
739 struct log_c *lc = log->context;
741 switch(status) {
742 case STATUSTYPE_INFO:
743 DMEMIT("1 %s", log->type->name);
744 break;
746 case STATUSTYPE_TABLE:
747 DMEMIT("%s %u %u ", log->type->name,
748 lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size);
749 DMEMIT_SYNC;
752 return sz;
755 static int disk_status(struct dm_dirty_log *log, status_type_t status,
756 char *result, unsigned int maxlen)
758 int sz = 0;
759 struct log_c *lc = log->context;
761 switch(status) {
762 case STATUSTYPE_INFO:
763 DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name,
764 lc->log_dev_failed ? 'D' : 'A');
765 break;
767 case STATUSTYPE_TABLE:
768 DMEMIT("%s %u %s %u ", log->type->name,
769 lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name,
770 lc->region_size);
771 DMEMIT_SYNC;
774 return sz;
777 static struct dm_dirty_log_type _core_type = {
778 .name = "core",
779 .module = THIS_MODULE,
780 .ctr = core_ctr,
781 .dtr = core_dtr,
782 .resume = core_resume,
783 .get_region_size = core_get_region_size,
784 .is_clean = core_is_clean,
785 .in_sync = core_in_sync,
786 .flush = core_flush,
787 .mark_region = core_mark_region,
788 .clear_region = core_clear_region,
789 .get_resync_work = core_get_resync_work,
790 .set_region_sync = core_set_region_sync,
791 .get_sync_count = core_get_sync_count,
792 .status = core_status,
795 static struct dm_dirty_log_type _disk_type = {
796 .name = "disk",
797 .module = THIS_MODULE,
798 .ctr = disk_ctr,
799 .dtr = disk_dtr,
800 .postsuspend = disk_flush,
801 .resume = disk_resume,
802 .get_region_size = core_get_region_size,
803 .is_clean = core_is_clean,
804 .in_sync = core_in_sync,
805 .flush = disk_flush,
806 .mark_region = core_mark_region,
807 .clear_region = core_clear_region,
808 .get_resync_work = core_get_resync_work,
809 .set_region_sync = core_set_region_sync,
810 .get_sync_count = core_get_sync_count,
811 .status = disk_status,
814 static int __init dm_dirty_log_init(void)
816 int r;
818 r = dm_dirty_log_type_register(&_core_type);
819 if (r)
820 DMWARN("couldn't register core log");
822 r = dm_dirty_log_type_register(&_disk_type);
823 if (r) {
824 DMWARN("couldn't register disk type");
825 dm_dirty_log_type_unregister(&_core_type);
828 return r;
831 static void __exit dm_dirty_log_exit(void)
833 dm_dirty_log_type_unregister(&_disk_type);
834 dm_dirty_log_type_unregister(&_core_type);
837 module_init(dm_dirty_log_init);
838 module_exit(dm_dirty_log_exit);
840 MODULE_DESCRIPTION(DM_NAME " dirty region log");
841 MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@redhat.com>");
842 MODULE_LICENSE("GPL");