mm owner: fix race between swapoff and exit
[linux-2.6/next.git] / drivers / md / dm-log.c
blob5b48478c79f53037474579edf1930b4d103b160c
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 "dm.h"
17 #define DM_MSG_PREFIX "dirty region log"
19 struct dm_dirty_log_internal {
20 struct dm_dirty_log_type *type;
22 struct list_head list;
23 long use;
26 static LIST_HEAD(_log_types);
27 static DEFINE_SPINLOCK(_lock);
29 static struct dm_dirty_log_internal *__find_dirty_log_type(const char *name)
31 struct dm_dirty_log_internal *log_type;
33 list_for_each_entry(log_type, &_log_types, list)
34 if (!strcmp(name, log_type->type->name))
35 return log_type;
37 return NULL;
40 static struct dm_dirty_log_internal *_get_dirty_log_type(const char *name)
42 struct dm_dirty_log_internal *log_type;
44 spin_lock(&_lock);
46 log_type = __find_dirty_log_type(name);
47 if (log_type) {
48 if (!log_type->use && !try_module_get(log_type->type->module))
49 log_type = NULL;
50 else
51 log_type->use++;
54 spin_unlock(&_lock);
56 return log_type;
60 * get_type
61 * @type_name
63 * Attempt to retrieve the dm_dirty_log_type by name. If not already
64 * available, attempt to load the appropriate module.
66 * Log modules are named "dm-log-" followed by the 'type_name'.
67 * Modules may contain multiple types.
68 * This function will first try the module "dm-log-<type_name>",
69 * then truncate 'type_name' on the last '-' and try again.
71 * For example, if type_name was "clustered-disk", it would search
72 * 'dm-log-clustered-disk' then 'dm-log-clustered'.
74 * Returns: dirty_log_type* on success, NULL on failure
76 static struct dm_dirty_log_type *get_type(const char *type_name)
78 char *p, *type_name_dup;
79 struct dm_dirty_log_internal *log_type;
81 if (!type_name)
82 return NULL;
84 log_type = _get_dirty_log_type(type_name);
85 if (log_type)
86 return log_type->type;
88 type_name_dup = kstrdup(type_name, GFP_KERNEL);
89 if (!type_name_dup) {
90 DMWARN("No memory left to attempt log module load for \"%s\"",
91 type_name);
92 return NULL;
95 while (request_module("dm-log-%s", type_name_dup) ||
96 !(log_type = _get_dirty_log_type(type_name))) {
97 p = strrchr(type_name_dup, '-');
98 if (!p)
99 break;
100 p[0] = '\0';
103 if (!log_type)
104 DMWARN("Module for logging type \"%s\" not found.", type_name);
106 kfree(type_name_dup);
108 return log_type ? log_type->type : NULL;
111 static void put_type(struct dm_dirty_log_type *type)
113 struct dm_dirty_log_internal *log_type;
115 if (!type)
116 return;
118 spin_lock(&_lock);
119 log_type = __find_dirty_log_type(type->name);
120 if (!log_type)
121 goto out;
123 if (!--log_type->use)
124 module_put(type->module);
126 BUG_ON(log_type->use < 0);
128 out:
129 spin_unlock(&_lock);
132 static struct dm_dirty_log_internal *_alloc_dirty_log_type(struct dm_dirty_log_type *type)
134 struct dm_dirty_log_internal *log_type = kzalloc(sizeof(*log_type),
135 GFP_KERNEL);
137 if (log_type)
138 log_type->type = type;
140 return log_type;
143 int dm_dirty_log_type_register(struct dm_dirty_log_type *type)
145 struct dm_dirty_log_internal *log_type = _alloc_dirty_log_type(type);
146 int r = 0;
148 if (!log_type)
149 return -ENOMEM;
151 spin_lock(&_lock);
152 if (!__find_dirty_log_type(type->name))
153 list_add(&log_type->list, &_log_types);
154 else {
155 kfree(log_type);
156 r = -EEXIST;
158 spin_unlock(&_lock);
160 return r;
162 EXPORT_SYMBOL(dm_dirty_log_type_register);
164 int dm_dirty_log_type_unregister(struct dm_dirty_log_type *type)
166 struct dm_dirty_log_internal *log_type;
168 spin_lock(&_lock);
170 log_type = __find_dirty_log_type(type->name);
171 if (!log_type) {
172 spin_unlock(&_lock);
173 return -EINVAL;
176 if (log_type->use) {
177 spin_unlock(&_lock);
178 return -ETXTBSY;
181 list_del(&log_type->list);
183 spin_unlock(&_lock);
184 kfree(log_type);
186 return 0;
188 EXPORT_SYMBOL(dm_dirty_log_type_unregister);
190 struct dm_dirty_log *dm_dirty_log_create(const char *type_name,
191 struct dm_target *ti,
192 unsigned int argc, char **argv)
194 struct dm_dirty_log_type *type;
195 struct dm_dirty_log *log;
197 log = kmalloc(sizeof(*log), GFP_KERNEL);
198 if (!log)
199 return NULL;
201 type = get_type(type_name);
202 if (!type) {
203 kfree(log);
204 return NULL;
207 log->type = type;
208 if (type->ctr(log, ti, argc, argv)) {
209 kfree(log);
210 put_type(type);
211 return NULL;
214 return log;
216 EXPORT_SYMBOL(dm_dirty_log_create);
218 void dm_dirty_log_destroy(struct dm_dirty_log *log)
220 log->type->dtr(log);
221 put_type(log->type);
222 kfree(log);
224 EXPORT_SYMBOL(dm_dirty_log_destroy);
226 /*-----------------------------------------------------------------
227 * Persistent and core logs share a lot of their implementation.
228 * FIXME: need a reload method to be called from a resume
229 *---------------------------------------------------------------*/
231 * Magic for persistent mirrors: "MiRr"
233 #define MIRROR_MAGIC 0x4D695272
236 * The on-disk version of the metadata.
238 #define MIRROR_DISK_VERSION 2
239 #define LOG_OFFSET 2
241 struct log_header {
242 uint32_t magic;
245 * Simple, incrementing version. no backward
246 * compatibility.
248 uint32_t version;
249 sector_t nr_regions;
252 struct log_c {
253 struct dm_target *ti;
254 int touched;
255 uint32_t region_size;
256 unsigned int region_count;
257 region_t sync_count;
259 unsigned bitset_uint32_count;
260 uint32_t *clean_bits;
261 uint32_t *sync_bits;
262 uint32_t *recovering_bits; /* FIXME: this seems excessive */
264 int sync_search;
266 /* Resync flag */
267 enum sync {
268 DEFAULTSYNC, /* Synchronize if necessary */
269 NOSYNC, /* Devices known to be already in sync */
270 FORCESYNC, /* Force a sync to happen */
271 } sync;
273 struct dm_io_request io_req;
276 * Disk log fields
278 int log_dev_failed;
279 struct dm_dev *log_dev;
280 struct log_header header;
282 struct dm_io_region header_location;
283 struct log_header *disk_header;
287 * The touched member needs to be updated every time we access
288 * one of the bitsets.
290 static inline int log_test_bit(uint32_t *bs, unsigned bit)
292 return ext2_test_bit(bit, (unsigned long *) bs) ? 1 : 0;
295 static inline void log_set_bit(struct log_c *l,
296 uint32_t *bs, unsigned bit)
298 ext2_set_bit(bit, (unsigned long *) bs);
299 l->touched = 1;
302 static inline void log_clear_bit(struct log_c *l,
303 uint32_t *bs, unsigned bit)
305 ext2_clear_bit(bit, (unsigned long *) bs);
306 l->touched = 1;
309 /*----------------------------------------------------------------
310 * Header IO
311 *--------------------------------------------------------------*/
312 static void header_to_disk(struct log_header *core, struct log_header *disk)
314 disk->magic = cpu_to_le32(core->magic);
315 disk->version = cpu_to_le32(core->version);
316 disk->nr_regions = cpu_to_le64(core->nr_regions);
319 static void header_from_disk(struct log_header *core, struct log_header *disk)
321 core->magic = le32_to_cpu(disk->magic);
322 core->version = le32_to_cpu(disk->version);
323 core->nr_regions = le64_to_cpu(disk->nr_regions);
326 static int rw_header(struct log_c *lc, int rw)
328 lc->io_req.bi_rw = rw;
329 lc->io_req.mem.ptr.vma = lc->disk_header;
330 lc->io_req.notify.fn = NULL;
332 return dm_io(&lc->io_req, 1, &lc->header_location, NULL);
335 static int read_header(struct log_c *log)
337 int r;
339 r = rw_header(log, READ);
340 if (r)
341 return r;
343 header_from_disk(&log->header, log->disk_header);
345 /* New log required? */
346 if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) {
347 log->header.magic = MIRROR_MAGIC;
348 log->header.version = MIRROR_DISK_VERSION;
349 log->header.nr_regions = 0;
352 #ifdef __LITTLE_ENDIAN
353 if (log->header.version == 1)
354 log->header.version = 2;
355 #endif
357 if (log->header.version != MIRROR_DISK_VERSION) {
358 DMWARN("incompatible disk log version");
359 return -EINVAL;
362 return 0;
365 static inline int write_header(struct log_c *log)
367 header_to_disk(&log->header, log->disk_header);
368 return rw_header(log, WRITE);
371 /*----------------------------------------------------------------
372 * core log constructor/destructor
374 * argv contains region_size followed optionally by [no]sync
375 *--------------------------------------------------------------*/
376 #define BYTE_SHIFT 3
377 static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti,
378 unsigned int argc, char **argv,
379 struct dm_dev *dev)
381 enum sync sync = DEFAULTSYNC;
383 struct log_c *lc;
384 uint32_t region_size;
385 unsigned int region_count;
386 size_t bitset_size, buf_size;
387 int r;
389 if (argc < 1 || argc > 2) {
390 DMWARN("wrong number of arguments to dirty region log");
391 return -EINVAL;
394 if (argc > 1) {
395 if (!strcmp(argv[1], "sync"))
396 sync = FORCESYNC;
397 else if (!strcmp(argv[1], "nosync"))
398 sync = NOSYNC;
399 else {
400 DMWARN("unrecognised sync argument to "
401 "dirty region log: %s", argv[1]);
402 return -EINVAL;
406 if (sscanf(argv[0], "%u", &region_size) != 1) {
407 DMWARN("invalid region size string");
408 return -EINVAL;
411 region_count = dm_sector_div_up(ti->len, region_size);
413 lc = kmalloc(sizeof(*lc), GFP_KERNEL);
414 if (!lc) {
415 DMWARN("couldn't allocate core log");
416 return -ENOMEM;
419 lc->ti = ti;
420 lc->touched = 0;
421 lc->region_size = region_size;
422 lc->region_count = region_count;
423 lc->sync = sync;
426 * Work out how many "unsigned long"s we need to hold the bitset.
428 bitset_size = dm_round_up(region_count,
429 sizeof(*lc->clean_bits) << BYTE_SHIFT);
430 bitset_size >>= BYTE_SHIFT;
432 lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits);
435 * Disk log?
437 if (!dev) {
438 lc->clean_bits = vmalloc(bitset_size);
439 if (!lc->clean_bits) {
440 DMWARN("couldn't allocate clean bitset");
441 kfree(lc);
442 return -ENOMEM;
444 lc->disk_header = NULL;
445 } else {
446 lc->log_dev = dev;
447 lc->log_dev_failed = 0;
448 lc->header_location.bdev = lc->log_dev->bdev;
449 lc->header_location.sector = 0;
452 * Buffer holds both header and bitset.
454 buf_size = dm_round_up((LOG_OFFSET << SECTOR_SHIFT) +
455 bitset_size, ti->limits.hardsect_size);
456 lc->header_location.count = buf_size >> SECTOR_SHIFT;
457 lc->io_req.mem.type = DM_IO_VMA;
458 lc->io_req.client = dm_io_client_create(dm_div_up(buf_size,
459 PAGE_SIZE));
460 if (IS_ERR(lc->io_req.client)) {
461 r = PTR_ERR(lc->io_req.client);
462 DMWARN("couldn't allocate disk io client");
463 kfree(lc);
464 return -ENOMEM;
467 lc->disk_header = vmalloc(buf_size);
468 if (!lc->disk_header) {
469 DMWARN("couldn't allocate disk log buffer");
470 kfree(lc);
471 return -ENOMEM;
474 lc->clean_bits = (void *)lc->disk_header +
475 (LOG_OFFSET << SECTOR_SHIFT);
478 memset(lc->clean_bits, -1, bitset_size);
480 lc->sync_bits = vmalloc(bitset_size);
481 if (!lc->sync_bits) {
482 DMWARN("couldn't allocate sync bitset");
483 if (!dev)
484 vfree(lc->clean_bits);
485 vfree(lc->disk_header);
486 kfree(lc);
487 return -ENOMEM;
489 memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
490 lc->sync_count = (sync == NOSYNC) ? region_count : 0;
492 lc->recovering_bits = vmalloc(bitset_size);
493 if (!lc->recovering_bits) {
494 DMWARN("couldn't allocate sync bitset");
495 vfree(lc->sync_bits);
496 if (!dev)
497 vfree(lc->clean_bits);
498 vfree(lc->disk_header);
499 kfree(lc);
500 return -ENOMEM;
502 memset(lc->recovering_bits, 0, bitset_size);
503 lc->sync_search = 0;
504 log->context = lc;
506 return 0;
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);
519 kfree(lc);
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)
538 int r;
539 struct dm_dev *dev;
541 if (argc < 2 || argc > 3) {
542 DMWARN("wrong number of arguments to disk dirty region log");
543 return -EINVAL;
546 r = dm_get_device(ti, argv[0], 0, 0 /* FIXME */,
547 FMODE_READ | FMODE_WRITE, &dev);
548 if (r)
549 return r;
551 r = create_log_context(log, ti, argc - 1, argv + 1, dev);
552 if (r) {
553 dm_put_device(ti, dev);
554 return r;
557 return 0;
560 static void disk_dtr(struct dm_dirty_log *log)
562 struct log_c *lc = (struct log_c *) log->context;
564 dm_put_device(lc->ti, lc->log_dev);
565 vfree(lc->disk_header);
566 dm_io_client_destroy(lc->io_req.client);
567 destroy_log_context(lc);
570 static int count_bits32(uint32_t *addr, unsigned size)
572 int count = 0, i;
574 for (i = 0; i < size; i++) {
575 count += hweight32(*(addr+i));
577 return count;
580 static void fail_log_device(struct log_c *lc)
582 if (lc->log_dev_failed)
583 return;
585 lc->log_dev_failed = 1;
586 dm_table_event(lc->ti->table);
589 static int disk_resume(struct dm_dirty_log *log)
591 int r;
592 unsigned i;
593 struct log_c *lc = (struct log_c *) log->context;
594 size_t size = lc->bitset_uint32_count * sizeof(uint32_t);
596 /* read the disk header */
597 r = read_header(lc);
598 if (r) {
599 DMWARN("%s: Failed to read header on dirty region log device",
600 lc->log_dev->name);
601 fail_log_device(lc);
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);
617 else
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 % (sizeof(*lc->clean_bits) << BYTE_SHIFT); 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 = count_bits32(lc->clean_bits, lc->bitset_uint32_count);
629 lc->sync_search = 0;
631 /* set the correct number of regions in the header */
632 lc->header.nr_regions = lc->region_count;
634 /* write the new header */
635 r = write_header(lc);
636 if (r) {
637 DMWARN("%s: Failed to write header on dirty region log device",
638 lc->log_dev->name);
639 fail_log_device(lc);
642 return r;
645 static uint32_t core_get_region_size(struct dm_dirty_log *log)
647 struct log_c *lc = (struct log_c *) log->context;
648 return lc->region_size;
651 static int core_resume(struct dm_dirty_log *log)
653 struct log_c *lc = (struct log_c *) log->context;
654 lc->sync_search = 0;
655 return 0;
658 static int core_is_clean(struct dm_dirty_log *log, region_t region)
660 struct log_c *lc = (struct log_c *) log->context;
661 return log_test_bit(lc->clean_bits, region);
664 static int core_in_sync(struct dm_dirty_log *log, region_t region, int block)
666 struct log_c *lc = (struct log_c *) log->context;
667 return log_test_bit(lc->sync_bits, region);
670 static int core_flush(struct dm_dirty_log *log)
672 /* no op */
673 return 0;
676 static int disk_flush(struct dm_dirty_log *log)
678 int r;
679 struct log_c *lc = (struct log_c *) log->context;
681 /* only write if the log has changed */
682 if (!lc->touched)
683 return 0;
685 r = write_header(lc);
686 if (r)
687 fail_log_device(lc);
688 else
689 lc->touched = 0;
691 return r;
694 static void core_mark_region(struct dm_dirty_log *log, region_t region)
696 struct log_c *lc = (struct log_c *) log->context;
697 log_clear_bit(lc, lc->clean_bits, region);
700 static void core_clear_region(struct dm_dirty_log *log, region_t region)
702 struct log_c *lc = (struct log_c *) log->context;
703 log_set_bit(lc, lc->clean_bits, region);
706 static int core_get_resync_work(struct dm_dirty_log *log, region_t *region)
708 struct log_c *lc = (struct log_c *) log->context;
710 if (lc->sync_search >= lc->region_count)
711 return 0;
713 do {
714 *region = ext2_find_next_zero_bit(
715 (unsigned long *) lc->sync_bits,
716 lc->region_count,
717 lc->sync_search);
718 lc->sync_search = *region + 1;
720 if (*region >= lc->region_count)
721 return 0;
723 } while (log_test_bit(lc->recovering_bits, *region));
725 log_set_bit(lc, lc->recovering_bits, *region);
726 return 1;
729 static void core_set_region_sync(struct dm_dirty_log *log, region_t region,
730 int in_sync)
732 struct log_c *lc = (struct log_c *) log->context;
734 log_clear_bit(lc, lc->recovering_bits, region);
735 if (in_sync) {
736 log_set_bit(lc, lc->sync_bits, region);
737 lc->sync_count++;
738 } else if (log_test_bit(lc->sync_bits, region)) {
739 lc->sync_count--;
740 log_clear_bit(lc, lc->sync_bits, region);
744 static region_t core_get_sync_count(struct dm_dirty_log *log)
746 struct log_c *lc = (struct log_c *) log->context;
748 return lc->sync_count;
751 #define DMEMIT_SYNC \
752 if (lc->sync != DEFAULTSYNC) \
753 DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
755 static int core_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("1 %s", log->type->name);
764 break;
766 case STATUSTYPE_TABLE:
767 DMEMIT("%s %u %u ", log->type->name,
768 lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size);
769 DMEMIT_SYNC;
772 return sz;
775 static int disk_status(struct dm_dirty_log *log, status_type_t status,
776 char *result, unsigned int maxlen)
778 int sz = 0;
779 struct log_c *lc = log->context;
781 switch(status) {
782 case STATUSTYPE_INFO:
783 DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name,
784 lc->log_dev_failed ? 'D' : 'A');
785 break;
787 case STATUSTYPE_TABLE:
788 DMEMIT("%s %u %s %u ", log->type->name,
789 lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name,
790 lc->region_size);
791 DMEMIT_SYNC;
794 return sz;
797 static struct dm_dirty_log_type _core_type = {
798 .name = "core",
799 .module = THIS_MODULE,
800 .ctr = core_ctr,
801 .dtr = core_dtr,
802 .resume = core_resume,
803 .get_region_size = core_get_region_size,
804 .is_clean = core_is_clean,
805 .in_sync = core_in_sync,
806 .flush = core_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 = core_status,
815 static struct dm_dirty_log_type _disk_type = {
816 .name = "disk",
817 .module = THIS_MODULE,
818 .ctr = disk_ctr,
819 .dtr = disk_dtr,
820 .postsuspend = disk_flush,
821 .resume = disk_resume,
822 .get_region_size = core_get_region_size,
823 .is_clean = core_is_clean,
824 .in_sync = core_in_sync,
825 .flush = disk_flush,
826 .mark_region = core_mark_region,
827 .clear_region = core_clear_region,
828 .get_resync_work = core_get_resync_work,
829 .set_region_sync = core_set_region_sync,
830 .get_sync_count = core_get_sync_count,
831 .status = disk_status,
834 static int __init dm_dirty_log_init(void)
836 int r;
838 r = dm_dirty_log_type_register(&_core_type);
839 if (r)
840 DMWARN("couldn't register core log");
842 r = dm_dirty_log_type_register(&_disk_type);
843 if (r) {
844 DMWARN("couldn't register disk type");
845 dm_dirty_log_type_unregister(&_core_type);
848 return r;
851 static void __exit dm_dirty_log_exit(void)
853 dm_dirty_log_type_unregister(&_disk_type);
854 dm_dirty_log_type_unregister(&_core_type);
857 module_init(dm_dirty_log_init);
858 module_exit(dm_dirty_log_exit);
860 MODULE_DESCRIPTION(DM_NAME " dirty region log");
861 MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@redhat.com>");
862 MODULE_LICENSE("GPL");