Linux Audio

Check our new training course

Loading...
v5.9
  1/*
  2 * Copyright (C) 2003 Sistina Software
  3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
  4 *
  5 * This file is released under the LGPL.
  6 */
  7
  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>
 14
 15#include <linux/device-mapper.h>
 16
 17#define DM_MSG_PREFIX "dirty region log"
 18
 19static LIST_HEAD(_log_types);
 20static DEFINE_SPINLOCK(_lock);
 21
 22static struct dm_dirty_log_type *__find_dirty_log_type(const char *name)
 23{
 24	struct dm_dirty_log_type *log_type;
 25
 26	list_for_each_entry(log_type, &_log_types, list)
 27		if (!strcmp(name, log_type->name))
 28			return log_type;
 29
 30	return NULL;
 31}
 32
 33static struct dm_dirty_log_type *_get_dirty_log_type(const char *name)
 34{
 35	struct dm_dirty_log_type *log_type;
 36
 37	spin_lock(&_lock);
 38
 39	log_type = __find_dirty_log_type(name);
 40	if (log_type && !try_module_get(log_type->module))
 41		log_type = NULL;
 42
 43	spin_unlock(&_lock);
 44
 45	return log_type;
 46}
 47
 48/*
 49 * get_type
 50 * @type_name
 51 *
 52 * Attempt to retrieve the dm_dirty_log_type by name.  If not already
 53 * available, attempt to load the appropriate module.
 54 *
 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.
 59 *
 60 * For example, if type_name was "clustered-disk", it would search
 61 * 'dm-log-clustered-disk' then 'dm-log-clustered'.
 62 *
 63 * Returns: dirty_log_type* on success, NULL on failure
 64 */
 65static struct dm_dirty_log_type *get_type(const char *type_name)
 66{
 67	char *p, *type_name_dup;
 68	struct dm_dirty_log_type *log_type;
 69
 70	if (!type_name)
 71		return NULL;
 72
 73	log_type = _get_dirty_log_type(type_name);
 74	if (log_type)
 75		return log_type;
 76
 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;
 82	}
 83
 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';
 90	}
 91
 92	if (!log_type)
 93		DMWARN("Module for logging type \"%s\" not found.", type_name);
 94
 95	kfree(type_name_dup);
 96
 97	return log_type;
 98}
 99
100static void put_type(struct dm_dirty_log_type *type)
101{
102	if (!type)
103		return;
104
105	spin_lock(&_lock);
106	if (!__find_dirty_log_type(type->name))
107		goto out;
108
109	module_put(type->module);
110
111out:
112	spin_unlock(&_lock);
113}
114
115int dm_dirty_log_type_register(struct dm_dirty_log_type *type)
116{
117	int r = 0;
118
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);
125
126	return r;
127}
128EXPORT_SYMBOL(dm_dirty_log_type_register);
129
130int dm_dirty_log_type_unregister(struct dm_dirty_log_type *type)
131{
132	spin_lock(&_lock);
133
134	if (!__find_dirty_log_type(type->name)) {
135		spin_unlock(&_lock);
136		return -EINVAL;
137	}
138
139	list_del(&type->list);
140
141	spin_unlock(&_lock);
142
143	return 0;
144}
145EXPORT_SYMBOL(dm_dirty_log_type_unregister);
146
147struct 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)
151{
152	struct dm_dirty_log_type *type;
153	struct dm_dirty_log *log;
154
155	log = kmalloc(sizeof(*log), GFP_KERNEL);
156	if (!log)
157		return NULL;
158
159	type = get_type(type_name);
160	if (!type) {
161		kfree(log);
162		return NULL;
163	}
164
165	log->flush_callback_fn = flush_callback_fn;
166	log->type = type;
167	if (type->ctr(log, ti, argc, argv)) {
168		kfree(log);
169		put_type(type);
170		return NULL;
171	}
172
173	return log;
174}
175EXPORT_SYMBOL(dm_dirty_log_create);
176
177void dm_dirty_log_destroy(struct dm_dirty_log *log)
178{
179	log->type->dtr(log);
180	put_type(log->type);
181	kfree(log);
182}
183EXPORT_SYMBOL(dm_dirty_log_destroy);
184
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 *---------------------------------------------------------------*/
189/*
190 * Magic for persistent mirrors: "MiRr"
191 */
192#define MIRROR_MAGIC 0x4D695272
193
194/*
195 * The on-disk version of the metadata.
196 */
197#define MIRROR_DISK_VERSION 2
198#define LOG_OFFSET 2
199
200struct log_header_disk {
201	__le32 magic;
202
203	/*
204	 * Simple, incrementing version. no backward
205	 * compatibility.
206	 */
207	__le32 version;
208	__le64 nr_regions;
209} __packed;
210
211struct log_header_core {
212	uint32_t magic;
213	uint32_t version;
214	uint64_t nr_regions;
215};
216
217struct log_c {
218	struct dm_target *ti;
219	int touched_dirtied;
220	int touched_cleaned;
221	int flush_failed;
222	uint32_t region_size;
223	unsigned int region_count;
224	region_t sync_count;
225
226	unsigned bitset_uint32_count;
227	uint32_t *clean_bits;
228	uint32_t *sync_bits;
229	uint32_t *recovering_bits;	/* FIXME: this seems excessive */
230
231	int sync_search;
232
233	/* Resync flag */
234	enum sync {
235		DEFAULTSYNC,	/* Synchronize if necessary */
236		NOSYNC,		/* Devices known to be already in sync */
237		FORCESYNC,	/* Force a sync to happen */
238	} sync;
239
240	struct dm_io_request io_req;
241
242	/*
243	 * Disk log fields
244	 */
245	int log_dev_failed;
246	int log_dev_flush_failed;
247	struct dm_dev *log_dev;
248	struct log_header_core header;
249
250	struct dm_io_region header_location;
251	struct log_header_disk *disk_header;
252};
253
254/*
255 * The touched member needs to be updated every time we access
256 * one of the bitsets.
257 */
258static inline int log_test_bit(uint32_t *bs, unsigned bit)
259{
260	return test_bit_le(bit, bs) ? 1 : 0;
261}
262
263static inline void log_set_bit(struct log_c *l,
264			       uint32_t *bs, unsigned bit)
265{
266	__set_bit_le(bit, bs);
267	l->touched_cleaned = 1;
268}
269
270static inline void log_clear_bit(struct log_c *l,
271				 uint32_t *bs, unsigned bit)
272{
273	__clear_bit_le(bit, bs);
274	l->touched_dirtied = 1;
275}
276
277/*----------------------------------------------------------------
278 * Header IO
279 *--------------------------------------------------------------*/
280static void header_to_disk(struct log_header_core *core, struct log_header_disk *disk)
281{
282	disk->magic = cpu_to_le32(core->magic);
283	disk->version = cpu_to_le32(core->version);
284	disk->nr_regions = cpu_to_le64(core->nr_regions);
285}
286
287static void header_from_disk(struct log_header_core *core, struct log_header_disk *disk)
288{
289	core->magic = le32_to_cpu(disk->magic);
290	core->version = le32_to_cpu(disk->version);
291	core->nr_regions = le64_to_cpu(disk->nr_regions);
292}
293
294static int rw_header(struct log_c *lc, int op)
295{
296	lc->io_req.bi_op = op;
297	lc->io_req.bi_op_flags = 0;
298
299	return dm_io(&lc->io_req, 1, &lc->header_location, NULL);
300}
301
302static int flush_header(struct log_c *lc)
303{
304	struct dm_io_region null_location = {
305		.bdev = lc->header_location.bdev,
306		.sector = 0,
307		.count = 0,
308	};
309
310	lc->io_req.bi_op = REQ_OP_WRITE;
311	lc->io_req.bi_op_flags = REQ_PREFLUSH;
312
313	return dm_io(&lc->io_req, 1, &null_location, NULL);
314}
315
316static int read_header(struct log_c *log)
317{
318	int r;
319
320	r = rw_header(log, REQ_OP_READ);
321	if (r)
322		return r;
323
324	header_from_disk(&log->header, log->disk_header);
325
326	/* New log required? */
327	if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) {
328		log->header.magic = MIRROR_MAGIC;
329		log->header.version = MIRROR_DISK_VERSION;
330		log->header.nr_regions = 0;
331	}
332
333#ifdef __LITTLE_ENDIAN
334	if (log->header.version == 1)
335		log->header.version = 2;
336#endif
337
338	if (log->header.version != MIRROR_DISK_VERSION) {
339		DMWARN("incompatible disk log version");
340		return -EINVAL;
341	}
342
343	return 0;
344}
345
346static int _check_region_size(struct dm_target *ti, uint32_t region_size)
347{
348	if (region_size < 2 || region_size > ti->len)
349		return 0;
350
351	if (!is_power_of_2(region_size))
352		return 0;
353
354	return 1;
355}
356
357/*----------------------------------------------------------------
358 * core log constructor/destructor
359 *
360 * argv contains region_size followed optionally by [no]sync
361 *--------------------------------------------------------------*/
362#define BYTE_SHIFT 3
363static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti,
364			      unsigned int argc, char **argv,
365			      struct dm_dev *dev)
366{
367	enum sync sync = DEFAULTSYNC;
368
369	struct log_c *lc;
370	uint32_t region_size;
371	unsigned int region_count;
372	size_t bitset_size, buf_size;
373	int r;
374	char dummy;
375
376	if (argc < 1 || argc > 2) {
377		DMWARN("wrong number of arguments to dirty region log");
378		return -EINVAL;
379	}
380
381	if (argc > 1) {
382		if (!strcmp(argv[1], "sync"))
383			sync = FORCESYNC;
384		else if (!strcmp(argv[1], "nosync"))
385			sync = NOSYNC;
386		else {
387			DMWARN("unrecognised sync argument to "
388			       "dirty region log: %s", argv[1]);
389			return -EINVAL;
390		}
391	}
392
393	if (sscanf(argv[0], "%u%c", &region_size, &dummy) != 1 ||
394	    !_check_region_size(ti, region_size)) {
395		DMWARN("invalid region size %s", argv[0]);
396		return -EINVAL;
397	}
398
399	region_count = dm_sector_div_up(ti->len, region_size);
400
401	lc = kmalloc(sizeof(*lc), GFP_KERNEL);
402	if (!lc) {
403		DMWARN("couldn't allocate core log");
404		return -ENOMEM;
405	}
406
407	lc->ti = ti;
408	lc->touched_dirtied = 0;
409	lc->touched_cleaned = 0;
410	lc->flush_failed = 0;
411	lc->region_size = region_size;
412	lc->region_count = region_count;
413	lc->sync = sync;
414
415	/*
416	 * Work out how many "unsigned long"s we need to hold the bitset.
417	 */
418	bitset_size = dm_round_up(region_count,
419				  sizeof(*lc->clean_bits) << BYTE_SHIFT);
420	bitset_size >>= BYTE_SHIFT;
421
422	lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits);
423
424	/*
425	 * Disk log?
426	 */
427	if (!dev) {
428		lc->clean_bits = vmalloc(bitset_size);
429		if (!lc->clean_bits) {
430			DMWARN("couldn't allocate clean bitset");
431			kfree(lc);
432			return -ENOMEM;
433		}
434		lc->disk_header = NULL;
435	} else {
436		lc->log_dev = dev;
437		lc->log_dev_failed = 0;
438		lc->log_dev_flush_failed = 0;
439		lc->header_location.bdev = lc->log_dev->bdev;
440		lc->header_location.sector = 0;
441
442		/*
443		 * Buffer holds both header and bitset.
444		 */
445		buf_size =
446		    dm_round_up((LOG_OFFSET << SECTOR_SHIFT) + bitset_size,
447				bdev_logical_block_size(lc->header_location.
448							    bdev));
449
450		if (buf_size > i_size_read(dev->bdev->bd_inode)) {
451			DMWARN("log device %s too small: need %llu bytes",
452				dev->name, (unsigned long long)buf_size);
453			kfree(lc);
454			return -EINVAL;
455		}
456
457		lc->header_location.count = buf_size >> SECTOR_SHIFT;
458
459		lc->io_req.mem.type = DM_IO_VMA;
460		lc->io_req.notify.fn = NULL;
461		lc->io_req.client = dm_io_client_create();
462		if (IS_ERR(lc->io_req.client)) {
463			r = PTR_ERR(lc->io_req.client);
464			DMWARN("couldn't allocate disk io client");
465			kfree(lc);
466			return r;
467		}
468
469		lc->disk_header = vmalloc(buf_size);
470		if (!lc->disk_header) {
471			DMWARN("couldn't allocate disk log buffer");
472			dm_io_client_destroy(lc->io_req.client);
473			kfree(lc);
474			return -ENOMEM;
475		}
476
477		lc->io_req.mem.ptr.vma = lc->disk_header;
478		lc->clean_bits = (void *)lc->disk_header +
479				 (LOG_OFFSET << SECTOR_SHIFT);
480	}
481
482	memset(lc->clean_bits, -1, bitset_size);
483
484	lc->sync_bits = vmalloc(bitset_size);
485	if (!lc->sync_bits) {
486		DMWARN("couldn't allocate sync bitset");
487		if (!dev)
488			vfree(lc->clean_bits);
489		else
490			dm_io_client_destroy(lc->io_req.client);
491		vfree(lc->disk_header);
492		kfree(lc);
493		return -ENOMEM;
494	}
495	memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
496	lc->sync_count = (sync == NOSYNC) ? region_count : 0;
497
498	lc->recovering_bits = vzalloc(bitset_size);
499	if (!lc->recovering_bits) {
500		DMWARN("couldn't allocate sync bitset");
501		vfree(lc->sync_bits);
502		if (!dev)
503			vfree(lc->clean_bits);
504		else
505			dm_io_client_destroy(lc->io_req.client);
506		vfree(lc->disk_header);
507		kfree(lc);
508		return -ENOMEM;
509	}
510	lc->sync_search = 0;
511	log->context = lc;
512
513	return 0;
514}
515
516static int core_ctr(struct dm_dirty_log *log, struct dm_target *ti,
517		    unsigned int argc, char **argv)
518{
519	return create_log_context(log, ti, argc, argv, NULL);
520}
521
522static void destroy_log_context(struct log_c *lc)
523{
524	vfree(lc->sync_bits);
525	vfree(lc->recovering_bits);
526	kfree(lc);
527}
528
529static void core_dtr(struct dm_dirty_log *log)
530{
531	struct log_c *lc = (struct log_c *) log->context;
532
533	vfree(lc->clean_bits);
534	destroy_log_context(lc);
535}
536
537/*----------------------------------------------------------------
538 * disk log constructor/destructor
539 *
540 * argv contains log_device region_size followed optionally by [no]sync
541 *--------------------------------------------------------------*/
542static int disk_ctr(struct dm_dirty_log *log, struct dm_target *ti,
543		    unsigned int argc, char **argv)
544{
545	int r;
546	struct dm_dev *dev;
547
548	if (argc < 2 || argc > 3) {
549		DMWARN("wrong number of arguments to disk dirty region log");
550		return -EINVAL;
551	}
552
553	r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev);
554	if (r)
555		return r;
556
557	r = create_log_context(log, ti, argc - 1, argv + 1, dev);
558	if (r) {
559		dm_put_device(ti, dev);
560		return r;
561	}
562
563	return 0;
564}
565
566static void disk_dtr(struct dm_dirty_log *log)
567{
568	struct log_c *lc = (struct log_c *) log->context;
569
570	dm_put_device(lc->ti, lc->log_dev);
571	vfree(lc->disk_header);
572	dm_io_client_destroy(lc->io_req.client);
573	destroy_log_context(lc);
574}
575
576static void fail_log_device(struct log_c *lc)
577{
578	if (lc->log_dev_failed)
579		return;
580
581	lc->log_dev_failed = 1;
582	dm_table_event(lc->ti->table);
583}
584
585static int disk_resume(struct dm_dirty_log *log)
586{
587	int r;
588	unsigned i;
589	struct log_c *lc = (struct log_c *) log->context;
590	size_t size = lc->bitset_uint32_count * sizeof(uint32_t);
591
592	/* read the disk header */
593	r = read_header(lc);
594	if (r) {
595		DMWARN("%s: Failed to read header on dirty region log device",
596		       lc->log_dev->name);
597		fail_log_device(lc);
598		/*
599		 * If the log device cannot be read, we must assume
600		 * all regions are out-of-sync.  If we simply return
601		 * here, the state will be uninitialized and could
602		 * lead us to return 'in-sync' status for regions
603		 * that are actually 'out-of-sync'.
604		 */
605		lc->header.nr_regions = 0;
606	}
607
608	/* set or clear any new bits -- device has grown */
609	if (lc->sync == NOSYNC)
610		for (i = lc->header.nr_regions; i < lc->region_count; i++)
611			/* FIXME: amazingly inefficient */
612			log_set_bit(lc, lc->clean_bits, i);
613	else
614		for (i = lc->header.nr_regions; i < lc->region_count; i++)
615			/* FIXME: amazingly inefficient */
616			log_clear_bit(lc, lc->clean_bits, i);
617
618	/* clear any old bits -- device has shrunk */
619	for (i = lc->region_count; i % (sizeof(*lc->clean_bits) << BYTE_SHIFT); i++)
620		log_clear_bit(lc, lc->clean_bits, i);
621
622	/* copy clean across to sync */
623	memcpy(lc->sync_bits, lc->clean_bits, size);
624	lc->sync_count = memweight(lc->clean_bits,
625				lc->bitset_uint32_count * sizeof(uint32_t));
626	lc->sync_search = 0;
627
628	/* set the correct number of regions in the header */
629	lc->header.nr_regions = lc->region_count;
630
631	header_to_disk(&lc->header, lc->disk_header);
632
633	/* write the new header */
634	r = rw_header(lc, REQ_OP_WRITE);
635	if (!r) {
636		r = flush_header(lc);
637		if (r)
638			lc->log_dev_flush_failed = 1;
639	}
640	if (r) {
641		DMWARN("%s: Failed to write header on dirty region log device",
642		       lc->log_dev->name);
643		fail_log_device(lc);
644	}
645
646	return r;
647}
648
649static uint32_t core_get_region_size(struct dm_dirty_log *log)
650{
651	struct log_c *lc = (struct log_c *) log->context;
652	return lc->region_size;
653}
654
655static int core_resume(struct dm_dirty_log *log)
656{
657	struct log_c *lc = (struct log_c *) log->context;
658	lc->sync_search = 0;
659	return 0;
660}
661
662static int core_is_clean(struct dm_dirty_log *log, region_t region)
663{
664	struct log_c *lc = (struct log_c *) log->context;
665	return log_test_bit(lc->clean_bits, region);
666}
667
668static int core_in_sync(struct dm_dirty_log *log, region_t region, int block)
669{
670	struct log_c *lc = (struct log_c *) log->context;
671	return log_test_bit(lc->sync_bits, region);
672}
673
674static int core_flush(struct dm_dirty_log *log)
675{
676	/* no op */
677	return 0;
678}
679
680static int disk_flush(struct dm_dirty_log *log)
681{
682	int r, i;
683	struct log_c *lc = log->context;
684
685	/* only write if the log has changed */
686	if (!lc->touched_cleaned && !lc->touched_dirtied)
687		return 0;
688
689	if (lc->touched_cleaned && log->flush_callback_fn &&
690	    log->flush_callback_fn(lc->ti)) {
691		/*
692		 * At this point it is impossible to determine which
693		 * regions are clean and which are dirty (without
694		 * re-reading the log off disk). So mark all of them
695		 * dirty.
696		 */
697		lc->flush_failed = 1;
698		for (i = 0; i < lc->region_count; i++)
699			log_clear_bit(lc, lc->clean_bits, i);
700	}
701
702	r = rw_header(lc, REQ_OP_WRITE);
703	if (r)
704		fail_log_device(lc);
705	else {
706		if (lc->touched_dirtied) {
707			r = flush_header(lc);
708			if (r) {
709				lc->log_dev_flush_failed = 1;
710				fail_log_device(lc);
711			} else
712				lc->touched_dirtied = 0;
713		}
714		lc->touched_cleaned = 0;
715	}
716
717	return r;
718}
719
720static void core_mark_region(struct dm_dirty_log *log, region_t region)
721{
722	struct log_c *lc = (struct log_c *) log->context;
723	log_clear_bit(lc, lc->clean_bits, region);
724}
725
726static void core_clear_region(struct dm_dirty_log *log, region_t region)
727{
728	struct log_c *lc = (struct log_c *) log->context;
729	if (likely(!lc->flush_failed))
730		log_set_bit(lc, lc->clean_bits, region);
731}
732
733static int core_get_resync_work(struct dm_dirty_log *log, region_t *region)
734{
735	struct log_c *lc = (struct log_c *) log->context;
736
737	if (lc->sync_search >= lc->region_count)
738		return 0;
739
740	do {
741		*region = find_next_zero_bit_le(lc->sync_bits,
742					     lc->region_count,
743					     lc->sync_search);
744		lc->sync_search = *region + 1;
745
746		if (*region >= lc->region_count)
747			return 0;
748
749	} while (log_test_bit(lc->recovering_bits, *region));
750
751	log_set_bit(lc, lc->recovering_bits, *region);
752	return 1;
753}
754
755static void core_set_region_sync(struct dm_dirty_log *log, region_t region,
756				 int in_sync)
757{
758	struct log_c *lc = (struct log_c *) log->context;
759
760	log_clear_bit(lc, lc->recovering_bits, region);
761	if (in_sync) {
762		log_set_bit(lc, lc->sync_bits, region);
763                lc->sync_count++;
764        } else if (log_test_bit(lc->sync_bits, region)) {
765		lc->sync_count--;
766		log_clear_bit(lc, lc->sync_bits, region);
767	}
768}
769
770static region_t core_get_sync_count(struct dm_dirty_log *log)
771{
772        struct log_c *lc = (struct log_c *) log->context;
773
774        return lc->sync_count;
775}
776
777#define	DMEMIT_SYNC \
778	if (lc->sync != DEFAULTSYNC) \
779		DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
780
781static int core_status(struct dm_dirty_log *log, status_type_t status,
782		       char *result, unsigned int maxlen)
783{
784	int sz = 0;
785	struct log_c *lc = log->context;
786
787	switch(status) {
788	case STATUSTYPE_INFO:
789		DMEMIT("1 %s", log->type->name);
790		break;
791
792	case STATUSTYPE_TABLE:
793		DMEMIT("%s %u %u ", log->type->name,
794		       lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size);
795		DMEMIT_SYNC;
 
 
 
 
 
796	}
797
798	return sz;
799}
800
801static int disk_status(struct dm_dirty_log *log, status_type_t status,
802		       char *result, unsigned int maxlen)
803{
804	int sz = 0;
805	struct log_c *lc = log->context;
806
807	switch(status) {
808	case STATUSTYPE_INFO:
809		DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name,
810		       lc->log_dev_flush_failed ? 'F' :
811		       lc->log_dev_failed ? 'D' :
812		       'A');
813		break;
814
815	case STATUSTYPE_TABLE:
816		DMEMIT("%s %u %s %u ", log->type->name,
817		       lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name,
818		       lc->region_size);
819		DMEMIT_SYNC;
 
 
 
 
 
820	}
821
822	return sz;
823}
824
825static struct dm_dirty_log_type _core_type = {
826	.name = "core",
827	.module = THIS_MODULE,
828	.ctr = core_ctr,
829	.dtr = core_dtr,
830	.resume = core_resume,
831	.get_region_size = core_get_region_size,
832	.is_clean = core_is_clean,
833	.in_sync = core_in_sync,
834	.flush = core_flush,
835	.mark_region = core_mark_region,
836	.clear_region = core_clear_region,
837	.get_resync_work = core_get_resync_work,
838	.set_region_sync = core_set_region_sync,
839	.get_sync_count = core_get_sync_count,
840	.status = core_status,
841};
842
843static struct dm_dirty_log_type _disk_type = {
844	.name = "disk",
845	.module = THIS_MODULE,
846	.ctr = disk_ctr,
847	.dtr = disk_dtr,
848	.postsuspend = disk_flush,
849	.resume = disk_resume,
850	.get_region_size = core_get_region_size,
851	.is_clean = core_is_clean,
852	.in_sync = core_in_sync,
853	.flush = disk_flush,
854	.mark_region = core_mark_region,
855	.clear_region = core_clear_region,
856	.get_resync_work = core_get_resync_work,
857	.set_region_sync = core_set_region_sync,
858	.get_sync_count = core_get_sync_count,
859	.status = disk_status,
860};
861
862static int __init dm_dirty_log_init(void)
863{
864	int r;
865
866	r = dm_dirty_log_type_register(&_core_type);
867	if (r)
868		DMWARN("couldn't register core log");
869
870	r = dm_dirty_log_type_register(&_disk_type);
871	if (r) {
872		DMWARN("couldn't register disk type");
873		dm_dirty_log_type_unregister(&_core_type);
874	}
875
876	return r;
877}
878
879static void __exit dm_dirty_log_exit(void)
880{
881	dm_dirty_log_type_unregister(&_disk_type);
882	dm_dirty_log_type_unregister(&_core_type);
883}
884
885module_init(dm_dirty_log_init);
886module_exit(dm_dirty_log_exit);
887
888MODULE_DESCRIPTION(DM_NAME " dirty region log");
889MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@redhat.com>");
890MODULE_LICENSE("GPL");
v6.2
  1/*
  2 * Copyright (C) 2003 Sistina Software
  3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
  4 *
  5 * This file is released under the LGPL.
  6 */
  7
  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>
 14
 15#include <linux/device-mapper.h>
 16
 17#define DM_MSG_PREFIX "dirty region log"
 18
 19static LIST_HEAD(_log_types);
 20static DEFINE_SPINLOCK(_lock);
 21
 22static struct dm_dirty_log_type *__find_dirty_log_type(const char *name)
 23{
 24	struct dm_dirty_log_type *log_type;
 25
 26	list_for_each_entry(log_type, &_log_types, list)
 27		if (!strcmp(name, log_type->name))
 28			return log_type;
 29
 30	return NULL;
 31}
 32
 33static struct dm_dirty_log_type *_get_dirty_log_type(const char *name)
 34{
 35	struct dm_dirty_log_type *log_type;
 36
 37	spin_lock(&_lock);
 38
 39	log_type = __find_dirty_log_type(name);
 40	if (log_type && !try_module_get(log_type->module))
 41		log_type = NULL;
 42
 43	spin_unlock(&_lock);
 44
 45	return log_type;
 46}
 47
 48/*
 49 * get_type
 50 * @type_name
 51 *
 52 * Attempt to retrieve the dm_dirty_log_type by name.  If not already
 53 * available, attempt to load the appropriate module.
 54 *
 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.
 59 *
 60 * For example, if type_name was "clustered-disk", it would search
 61 * 'dm-log-clustered-disk' then 'dm-log-clustered'.
 62 *
 63 * Returns: dirty_log_type* on success, NULL on failure
 64 */
 65static struct dm_dirty_log_type *get_type(const char *type_name)
 66{
 67	char *p, *type_name_dup;
 68	struct dm_dirty_log_type *log_type;
 69
 70	if (!type_name)
 71		return NULL;
 72
 73	log_type = _get_dirty_log_type(type_name);
 74	if (log_type)
 75		return log_type;
 76
 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;
 82	}
 83
 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';
 90	}
 91
 92	if (!log_type)
 93		DMWARN("Module for logging type \"%s\" not found.", type_name);
 94
 95	kfree(type_name_dup);
 96
 97	return log_type;
 98}
 99
100static void put_type(struct dm_dirty_log_type *type)
101{
102	if (!type)
103		return;
104
105	spin_lock(&_lock);
106	if (!__find_dirty_log_type(type->name))
107		goto out;
108
109	module_put(type->module);
110
111out:
112	spin_unlock(&_lock);
113}
114
115int dm_dirty_log_type_register(struct dm_dirty_log_type *type)
116{
117	int r = 0;
118
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);
125
126	return r;
127}
128EXPORT_SYMBOL(dm_dirty_log_type_register);
129
130int dm_dirty_log_type_unregister(struct dm_dirty_log_type *type)
131{
132	spin_lock(&_lock);
133
134	if (!__find_dirty_log_type(type->name)) {
135		spin_unlock(&_lock);
136		return -EINVAL;
137	}
138
139	list_del(&type->list);
140
141	spin_unlock(&_lock);
142
143	return 0;
144}
145EXPORT_SYMBOL(dm_dirty_log_type_unregister);
146
147struct 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)
151{
152	struct dm_dirty_log_type *type;
153	struct dm_dirty_log *log;
154
155	log = kmalloc(sizeof(*log), GFP_KERNEL);
156	if (!log)
157		return NULL;
158
159	type = get_type(type_name);
160	if (!type) {
161		kfree(log);
162		return NULL;
163	}
164
165	log->flush_callback_fn = flush_callback_fn;
166	log->type = type;
167	if (type->ctr(log, ti, argc, argv)) {
168		kfree(log);
169		put_type(type);
170		return NULL;
171	}
172
173	return log;
174}
175EXPORT_SYMBOL(dm_dirty_log_create);
176
177void dm_dirty_log_destroy(struct dm_dirty_log *log)
178{
179	log->type->dtr(log);
180	put_type(log->type);
181	kfree(log);
182}
183EXPORT_SYMBOL(dm_dirty_log_destroy);
184
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 *---------------------------------------------------------------*/
189/*
190 * Magic for persistent mirrors: "MiRr"
191 */
192#define MIRROR_MAGIC 0x4D695272
193
194/*
195 * The on-disk version of the metadata.
196 */
197#define MIRROR_DISK_VERSION 2
198#define LOG_OFFSET 2
199
200struct log_header_disk {
201	__le32 magic;
202
203	/*
204	 * Simple, incrementing version. no backward
205	 * compatibility.
206	 */
207	__le32 version;
208	__le64 nr_regions;
209} __packed;
210
211struct log_header_core {
212	uint32_t magic;
213	uint32_t version;
214	uint64_t nr_regions;
215};
216
217struct log_c {
218	struct dm_target *ti;
219	int touched_dirtied;
220	int touched_cleaned;
221	int flush_failed;
222	uint32_t region_size;
223	unsigned int region_count;
224	region_t sync_count;
225
226	unsigned bitset_uint32_count;
227	uint32_t *clean_bits;
228	uint32_t *sync_bits;
229	uint32_t *recovering_bits;	/* FIXME: this seems excessive */
230
231	int sync_search;
232
233	/* Resync flag */
234	enum sync {
235		DEFAULTSYNC,	/* Synchronize if necessary */
236		NOSYNC,		/* Devices known to be already in sync */
237		FORCESYNC,	/* Force a sync to happen */
238	} sync;
239
240	struct dm_io_request io_req;
241
242	/*
243	 * Disk log fields
244	 */
245	int log_dev_failed;
246	int log_dev_flush_failed;
247	struct dm_dev *log_dev;
248	struct log_header_core header;
249
250	struct dm_io_region header_location;
251	struct log_header_disk *disk_header;
252};
253
254/*
255 * The touched member needs to be updated every time we access
256 * one of the bitsets.
257 */
258static inline int log_test_bit(uint32_t *bs, unsigned bit)
259{
260	return test_bit_le(bit, bs) ? 1 : 0;
261}
262
263static inline void log_set_bit(struct log_c *l,
264			       uint32_t *bs, unsigned bit)
265{
266	__set_bit_le(bit, bs);
267	l->touched_cleaned = 1;
268}
269
270static inline void log_clear_bit(struct log_c *l,
271				 uint32_t *bs, unsigned bit)
272{
273	__clear_bit_le(bit, bs);
274	l->touched_dirtied = 1;
275}
276
277/*----------------------------------------------------------------
278 * Header IO
279 *--------------------------------------------------------------*/
280static void header_to_disk(struct log_header_core *core, struct log_header_disk *disk)
281{
282	disk->magic = cpu_to_le32(core->magic);
283	disk->version = cpu_to_le32(core->version);
284	disk->nr_regions = cpu_to_le64(core->nr_regions);
285}
286
287static void header_from_disk(struct log_header_core *core, struct log_header_disk *disk)
288{
289	core->magic = le32_to_cpu(disk->magic);
290	core->version = le32_to_cpu(disk->version);
291	core->nr_regions = le64_to_cpu(disk->nr_regions);
292}
293
294static int rw_header(struct log_c *lc, enum req_op op)
295{
296	lc->io_req.bi_opf = op;
 
297
298	return dm_io(&lc->io_req, 1, &lc->header_location, NULL);
299}
300
301static int flush_header(struct log_c *lc)
302{
303	struct dm_io_region null_location = {
304		.bdev = lc->header_location.bdev,
305		.sector = 0,
306		.count = 0,
307	};
308
309	lc->io_req.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
 
310
311	return dm_io(&lc->io_req, 1, &null_location, NULL);
312}
313
314static int read_header(struct log_c *log)
315{
316	int r;
317
318	r = rw_header(log, REQ_OP_READ);
319	if (r)
320		return r;
321
322	header_from_disk(&log->header, log->disk_header);
323
324	/* New log required? */
325	if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) {
326		log->header.magic = MIRROR_MAGIC;
327		log->header.version = MIRROR_DISK_VERSION;
328		log->header.nr_regions = 0;
329	}
330
331#ifdef __LITTLE_ENDIAN
332	if (log->header.version == 1)
333		log->header.version = 2;
334#endif
335
336	if (log->header.version != MIRROR_DISK_VERSION) {
337		DMWARN("incompatible disk log version");
338		return -EINVAL;
339	}
340
341	return 0;
342}
343
344static int _check_region_size(struct dm_target *ti, uint32_t region_size)
345{
346	if (region_size < 2 || region_size > ti->len)
347		return 0;
348
349	if (!is_power_of_2(region_size))
350		return 0;
351
352	return 1;
353}
354
355/*----------------------------------------------------------------
356 * core log constructor/destructor
357 *
358 * argv contains region_size followed optionally by [no]sync
359 *--------------------------------------------------------------*/
360#define BYTE_SHIFT 3
361static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti,
362			      unsigned int argc, char **argv,
363			      struct dm_dev *dev)
364{
365	enum sync sync = DEFAULTSYNC;
366
367	struct log_c *lc;
368	uint32_t region_size;
369	unsigned int region_count;
370	size_t bitset_size, buf_size;
371	int r;
372	char dummy;
373
374	if (argc < 1 || argc > 2) {
375		DMWARN("wrong number of arguments to dirty region log");
376		return -EINVAL;
377	}
378
379	if (argc > 1) {
380		if (!strcmp(argv[1], "sync"))
381			sync = FORCESYNC;
382		else if (!strcmp(argv[1], "nosync"))
383			sync = NOSYNC;
384		else {
385			DMWARN("unrecognised sync argument to "
386			       "dirty region log: %s", argv[1]);
387			return -EINVAL;
388		}
389	}
390
391	if (sscanf(argv[0], "%u%c", &region_size, &dummy) != 1 ||
392	    !_check_region_size(ti, region_size)) {
393		DMWARN("invalid region size %s", argv[0]);
394		return -EINVAL;
395	}
396
397	region_count = dm_sector_div_up(ti->len, region_size);
398
399	lc = kmalloc(sizeof(*lc), GFP_KERNEL);
400	if (!lc) {
401		DMWARN("couldn't allocate core log");
402		return -ENOMEM;
403	}
404
405	lc->ti = ti;
406	lc->touched_dirtied = 0;
407	lc->touched_cleaned = 0;
408	lc->flush_failed = 0;
409	lc->region_size = region_size;
410	lc->region_count = region_count;
411	lc->sync = sync;
412
413	/*
414	 * Work out how many "unsigned long"s we need to hold the bitset.
415	 */
416	bitset_size = dm_round_up(region_count, BITS_PER_LONG);
 
417	bitset_size >>= BYTE_SHIFT;
418
419	lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits);
420
421	/*
422	 * Disk log?
423	 */
424	if (!dev) {
425		lc->clean_bits = vmalloc(bitset_size);
426		if (!lc->clean_bits) {
427			DMWARN("couldn't allocate clean bitset");
428			kfree(lc);
429			return -ENOMEM;
430		}
431		lc->disk_header = NULL;
432	} else {
433		lc->log_dev = dev;
434		lc->log_dev_failed = 0;
435		lc->log_dev_flush_failed = 0;
436		lc->header_location.bdev = lc->log_dev->bdev;
437		lc->header_location.sector = 0;
438
439		/*
440		 * Buffer holds both header and bitset.
441		 */
442		buf_size =
443		    dm_round_up((LOG_OFFSET << SECTOR_SHIFT) + bitset_size,
444				bdev_logical_block_size(lc->header_location.
445							    bdev));
446
447		if (buf_size > bdev_nr_bytes(dev->bdev)) {
448			DMWARN("log device %s too small: need %llu bytes",
449				dev->name, (unsigned long long)buf_size);
450			kfree(lc);
451			return -EINVAL;
452		}
453
454		lc->header_location.count = buf_size >> SECTOR_SHIFT;
455
456		lc->io_req.mem.type = DM_IO_VMA;
457		lc->io_req.notify.fn = NULL;
458		lc->io_req.client = dm_io_client_create();
459		if (IS_ERR(lc->io_req.client)) {
460			r = PTR_ERR(lc->io_req.client);
461			DMWARN("couldn't allocate disk io client");
462			kfree(lc);
463			return r;
464		}
465
466		lc->disk_header = vmalloc(buf_size);
467		if (!lc->disk_header) {
468			DMWARN("couldn't allocate disk log buffer");
469			dm_io_client_destroy(lc->io_req.client);
470			kfree(lc);
471			return -ENOMEM;
472		}
473
474		lc->io_req.mem.ptr.vma = lc->disk_header;
475		lc->clean_bits = (void *)lc->disk_header +
476				 (LOG_OFFSET << SECTOR_SHIFT);
477	}
478
479	memset(lc->clean_bits, -1, bitset_size);
480
481	lc->sync_bits = vmalloc(bitset_size);
482	if (!lc->sync_bits) {
483		DMWARN("couldn't allocate sync bitset");
484		if (!dev)
485			vfree(lc->clean_bits);
486		else
487			dm_io_client_destroy(lc->io_req.client);
488		vfree(lc->disk_header);
489		kfree(lc);
490		return -ENOMEM;
491	}
492	memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
493	lc->sync_count = (sync == NOSYNC) ? region_count : 0;
494
495	lc->recovering_bits = vzalloc(bitset_size);
496	if (!lc->recovering_bits) {
497		DMWARN("couldn't allocate sync bitset");
498		vfree(lc->sync_bits);
499		if (!dev)
500			vfree(lc->clean_bits);
501		else
502			dm_io_client_destroy(lc->io_req.client);
503		vfree(lc->disk_header);
504		kfree(lc);
505		return -ENOMEM;
506	}
507	lc->sync_search = 0;
508	log->context = lc;
509
510	return 0;
511}
512
513static int core_ctr(struct dm_dirty_log *log, struct dm_target *ti,
514		    unsigned int argc, char **argv)
515{
516	return create_log_context(log, ti, argc, argv, NULL);
517}
518
519static void destroy_log_context(struct log_c *lc)
520{
521	vfree(lc->sync_bits);
522	vfree(lc->recovering_bits);
523	kfree(lc);
524}
525
526static void core_dtr(struct dm_dirty_log *log)
527{
528	struct log_c *lc = (struct log_c *) log->context;
529
530	vfree(lc->clean_bits);
531	destroy_log_context(lc);
532}
533
534/*----------------------------------------------------------------
535 * disk log constructor/destructor
536 *
537 * argv contains log_device region_size followed optionally by [no]sync
538 *--------------------------------------------------------------*/
539static int disk_ctr(struct dm_dirty_log *log, struct dm_target *ti,
540		    unsigned int argc, char **argv)
541{
542	int r;
543	struct dm_dev *dev;
544
545	if (argc < 2 || argc > 3) {
546		DMWARN("wrong number of arguments to disk dirty region log");
547		return -EINVAL;
548	}
549
550	r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev);
551	if (r)
552		return r;
553
554	r = create_log_context(log, ti, argc - 1, argv + 1, dev);
555	if (r) {
556		dm_put_device(ti, dev);
557		return r;
558	}
559
560	return 0;
561}
562
563static void disk_dtr(struct dm_dirty_log *log)
564{
565	struct log_c *lc = (struct log_c *) log->context;
566
567	dm_put_device(lc->ti, lc->log_dev);
568	vfree(lc->disk_header);
569	dm_io_client_destroy(lc->io_req.client);
570	destroy_log_context(lc);
571}
572
573static void fail_log_device(struct log_c *lc)
574{
575	if (lc->log_dev_failed)
576		return;
577
578	lc->log_dev_failed = 1;
579	dm_table_event(lc->ti->table);
580}
581
582static int disk_resume(struct dm_dirty_log *log)
583{
584	int r;
585	unsigned i;
586	struct log_c *lc = (struct log_c *) log->context;
587	size_t size = lc->bitset_uint32_count * sizeof(uint32_t);
588
589	/* read the disk header */
590	r = read_header(lc);
591	if (r) {
592		DMWARN("%s: Failed to read header on dirty region log device",
593		       lc->log_dev->name);
594		fail_log_device(lc);
595		/*
596		 * If the log device cannot be read, we must assume
597		 * all regions are out-of-sync.  If we simply return
598		 * here, the state will be uninitialized and could
599		 * lead us to return 'in-sync' status for regions
600		 * that are actually 'out-of-sync'.
601		 */
602		lc->header.nr_regions = 0;
603	}
604
605	/* set or clear any new bits -- device has grown */
606	if (lc->sync == NOSYNC)
607		for (i = lc->header.nr_regions; i < lc->region_count; i++)
608			/* FIXME: amazingly inefficient */
609			log_set_bit(lc, lc->clean_bits, i);
610	else
611		for (i = lc->header.nr_regions; i < lc->region_count; i++)
612			/* FIXME: amazingly inefficient */
613			log_clear_bit(lc, lc->clean_bits, i);
614
615	/* clear any old bits -- device has shrunk */
616	for (i = lc->region_count; i % BITS_PER_LONG; i++)
617		log_clear_bit(lc, lc->clean_bits, i);
618
619	/* copy clean across to sync */
620	memcpy(lc->sync_bits, lc->clean_bits, size);
621	lc->sync_count = memweight(lc->clean_bits,
622				lc->bitset_uint32_count * sizeof(uint32_t));
623	lc->sync_search = 0;
624
625	/* set the correct number of regions in the header */
626	lc->header.nr_regions = lc->region_count;
627
628	header_to_disk(&lc->header, lc->disk_header);
629
630	/* write the new header */
631	r = rw_header(lc, REQ_OP_WRITE);
632	if (!r) {
633		r = flush_header(lc);
634		if (r)
635			lc->log_dev_flush_failed = 1;
636	}
637	if (r) {
638		DMWARN("%s: Failed to write header on dirty region log device",
639		       lc->log_dev->name);
640		fail_log_device(lc);
641	}
642
643	return r;
644}
645
646static uint32_t core_get_region_size(struct dm_dirty_log *log)
647{
648	struct log_c *lc = (struct log_c *) log->context;
649	return lc->region_size;
650}
651
652static int core_resume(struct dm_dirty_log *log)
653{
654	struct log_c *lc = (struct log_c *) log->context;
655	lc->sync_search = 0;
656	return 0;
657}
658
659static int core_is_clean(struct dm_dirty_log *log, region_t region)
660{
661	struct log_c *lc = (struct log_c *) log->context;
662	return log_test_bit(lc->clean_bits, region);
663}
664
665static int core_in_sync(struct dm_dirty_log *log, region_t region, int block)
666{
667	struct log_c *lc = (struct log_c *) log->context;
668	return log_test_bit(lc->sync_bits, region);
669}
670
671static int core_flush(struct dm_dirty_log *log)
672{
673	/* no op */
674	return 0;
675}
676
677static int disk_flush(struct dm_dirty_log *log)
678{
679	int r, i;
680	struct log_c *lc = log->context;
681
682	/* only write if the log has changed */
683	if (!lc->touched_cleaned && !lc->touched_dirtied)
684		return 0;
685
686	if (lc->touched_cleaned && log->flush_callback_fn &&
687	    log->flush_callback_fn(lc->ti)) {
688		/*
689		 * At this point it is impossible to determine which
690		 * regions are clean and which are dirty (without
691		 * re-reading the log off disk). So mark all of them
692		 * dirty.
693		 */
694		lc->flush_failed = 1;
695		for (i = 0; i < lc->region_count; i++)
696			log_clear_bit(lc, lc->clean_bits, i);
697	}
698
699	r = rw_header(lc, REQ_OP_WRITE);
700	if (r)
701		fail_log_device(lc);
702	else {
703		if (lc->touched_dirtied) {
704			r = flush_header(lc);
705			if (r) {
706				lc->log_dev_flush_failed = 1;
707				fail_log_device(lc);
708			} else
709				lc->touched_dirtied = 0;
710		}
711		lc->touched_cleaned = 0;
712	}
713
714	return r;
715}
716
717static void core_mark_region(struct dm_dirty_log *log, region_t region)
718{
719	struct log_c *lc = (struct log_c *) log->context;
720	log_clear_bit(lc, lc->clean_bits, region);
721}
722
723static void core_clear_region(struct dm_dirty_log *log, region_t region)
724{
725	struct log_c *lc = (struct log_c *) log->context;
726	if (likely(!lc->flush_failed))
727		log_set_bit(lc, lc->clean_bits, region);
728}
729
730static int core_get_resync_work(struct dm_dirty_log *log, region_t *region)
731{
732	struct log_c *lc = (struct log_c *) log->context;
733
734	if (lc->sync_search >= lc->region_count)
735		return 0;
736
737	do {
738		*region = find_next_zero_bit_le(lc->sync_bits,
739					     lc->region_count,
740					     lc->sync_search);
741		lc->sync_search = *region + 1;
742
743		if (*region >= lc->region_count)
744			return 0;
745
746	} while (log_test_bit(lc->recovering_bits, *region));
747
748	log_set_bit(lc, lc->recovering_bits, *region);
749	return 1;
750}
751
752static void core_set_region_sync(struct dm_dirty_log *log, region_t region,
753				 int in_sync)
754{
755	struct log_c *lc = (struct log_c *) log->context;
756
757	log_clear_bit(lc, lc->recovering_bits, region);
758	if (in_sync) {
759		log_set_bit(lc, lc->sync_bits, region);
760                lc->sync_count++;
761        } else if (log_test_bit(lc->sync_bits, region)) {
762		lc->sync_count--;
763		log_clear_bit(lc, lc->sync_bits, region);
764	}
765}
766
767static region_t core_get_sync_count(struct dm_dirty_log *log)
768{
769        struct log_c *lc = (struct log_c *) log->context;
770
771        return lc->sync_count;
772}
773
774#define	DMEMIT_SYNC \
775	if (lc->sync != DEFAULTSYNC) \
776		DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
777
778static int core_status(struct dm_dirty_log *log, status_type_t status,
779		       char *result, unsigned int maxlen)
780{
781	int sz = 0;
782	struct log_c *lc = log->context;
783
784	switch(status) {
785	case STATUSTYPE_INFO:
786		DMEMIT("1 %s", log->type->name);
787		break;
788
789	case STATUSTYPE_TABLE:
790		DMEMIT("%s %u %u ", log->type->name,
791		       lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size);
792		DMEMIT_SYNC;
793		break;
794
795	case STATUSTYPE_IMA:
796		*result = '\0';
797		break;
798	}
799
800	return sz;
801}
802
803static int disk_status(struct dm_dirty_log *log, status_type_t status,
804		       char *result, unsigned int maxlen)
805{
806	int sz = 0;
807	struct log_c *lc = log->context;
808
809	switch(status) {
810	case STATUSTYPE_INFO:
811		DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name,
812		       lc->log_dev_flush_failed ? 'F' :
813		       lc->log_dev_failed ? 'D' :
814		       'A');
815		break;
816
817	case STATUSTYPE_TABLE:
818		DMEMIT("%s %u %s %u ", log->type->name,
819		       lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name,
820		       lc->region_size);
821		DMEMIT_SYNC;
822		break;
823
824	case STATUSTYPE_IMA:
825		*result = '\0';
826		break;
827	}
828
829	return sz;
830}
831
832static struct dm_dirty_log_type _core_type = {
833	.name = "core",
834	.module = THIS_MODULE,
835	.ctr = core_ctr,
836	.dtr = core_dtr,
837	.resume = core_resume,
838	.get_region_size = core_get_region_size,
839	.is_clean = core_is_clean,
840	.in_sync = core_in_sync,
841	.flush = core_flush,
842	.mark_region = core_mark_region,
843	.clear_region = core_clear_region,
844	.get_resync_work = core_get_resync_work,
845	.set_region_sync = core_set_region_sync,
846	.get_sync_count = core_get_sync_count,
847	.status = core_status,
848};
849
850static struct dm_dirty_log_type _disk_type = {
851	.name = "disk",
852	.module = THIS_MODULE,
853	.ctr = disk_ctr,
854	.dtr = disk_dtr,
855	.postsuspend = disk_flush,
856	.resume = disk_resume,
857	.get_region_size = core_get_region_size,
858	.is_clean = core_is_clean,
859	.in_sync = core_in_sync,
860	.flush = disk_flush,
861	.mark_region = core_mark_region,
862	.clear_region = core_clear_region,
863	.get_resync_work = core_get_resync_work,
864	.set_region_sync = core_set_region_sync,
865	.get_sync_count = core_get_sync_count,
866	.status = disk_status,
867};
868
869static int __init dm_dirty_log_init(void)
870{
871	int r;
872
873	r = dm_dirty_log_type_register(&_core_type);
874	if (r)
875		DMWARN("couldn't register core log");
876
877	r = dm_dirty_log_type_register(&_disk_type);
878	if (r) {
879		DMWARN("couldn't register disk type");
880		dm_dirty_log_type_unregister(&_core_type);
881	}
882
883	return r;
884}
885
886static void __exit dm_dirty_log_exit(void)
887{
888	dm_dirty_log_type_unregister(&_disk_type);
889	dm_dirty_log_type_unregister(&_core_type);
890}
891
892module_init(dm_dirty_log_init);
893module_exit(dm_dirty_log_exit);
894
895MODULE_DESCRIPTION(DM_NAME " dirty region log");
896MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@redhat.com>");
897MODULE_LICENSE("GPL");